Multi-variate model for predicting cytokine release syndrome

Abstract

To provide techniques for predicting the risk of a subject experiencing a cytokine release syndrome of at least a threshold grade subsequent to receiving a treatment.SOLUTION: A risk may be predicted based on, for example, a set of baseline characteristics, a risk-score generation model, an on-treatment cytokine level, and / or a treatment dosage. The risk may be used to generate an output corresponding to a recommendation as to whether to monitor the subject via in-patient monitoring.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of, and priority to, U.S. Provisional Patent Application No. 63 / 221,323, filed July 13, 2021, U.S. Provisional Patent Application No. 63 / 263,787, filed November 9, 2021, and U.S. Provisional Patent Application No. 63 / 341,208, filed May 12, 2022. Each of these applications is incorporated herein by reference in its entirety for all purposes. [Background technology]

[0002] Cytokine release syndrome (and cytokine release storm) is a potentially fatal condition that can be caused by viral infections, autoimmune diseases, and immunotherapy. Cytokine release syndrome is characterized by a dramatic increase in cytokine levels and immune system dysfunction. Under normal circumstances, a balance is maintained between anti-inflammatory and pro-inflammatory cytokines. However, an overactivated immune response can lead to significantly increased secretion of pro-inflammatory cytokines from lymphocytes (T cells, B cells, and natural killer cells) and myeloid cells (monocytes, macrophages, and dendritic cells).

[0003] The incidence of cytokine release syndrome in subjects receiving cancer immunotherapy varies widely depending on the type of immunotherapy agent. Cytokine release syndrome can occur within hours, or in the case of CAR-T cell therapy, up to several weeks after drug infusion. While the incidence of cytokine release syndrome is relatively low with most conventional monoclonal antibodies, T cell-engaged cancer immunotherapies pose a particularly high risk of triggering cytokine release syndrome. Therefore, the standard of care is to monitor subjects receiving immunotherapy for symptoms of cytokine release syndrome immediately after and after treatment.

[0004] The risk of cytokine release syndrome is influenced by factors related to the type of therapy and underlying disease. Many drugs that can induce cytokine release syndrome exhibit a first-dose effect, i.e., the most severe symptoms occur only after the first dose and do not recur after subsequent doses (Klinger et al., Blood 119:6226-33 (2012)).

[0005] Despite efforts to that effect, it remains impossible to predict which subjects will experience cytokine release syndrome, let alone predict any grade-specific susceptibility to such occurrence. Rather, a wide variety of clinical manifestations and severe incidences of cytokine release syndrome continue to be observed, and the default is to institute consistent inpatient monitoring after administration of select therapy to promptly detect and treat any cytokine release syndrome.

[0006] Cytokine release syndrome can cause fever, chills, fatigue, nausea, headache, myalgia, dyspnea, tachycardia, hypotension, liver dysfunction, respiratory distress syndrome, acute vascular leak syndrome, disseminated intravascular coagulopathy, neurotoxicity, cardiac dysfunction, renal failure, and / or multiple organ failure. Mild symptoms, such as fever, nausea, fatigue, headache, and malaise, can be treated with fluid therapy and analgesics while the patient is monitored. More severe symptoms, resulting from excessive pro-inflammatory cytokine production (i.e., cytokine release syndrome), require prompt intervention with corticosteroids and / or anti-cytokine therapy to prevent organ damage and death. Therefore, improving the identification of risk factors for cytokine release syndrome is crucial. Summary of the Invention

[0007] In some embodiments, a method is provided that includes identifying a set of baseline characteristics of a subject diagnosed with cancer, the set of baseline characteristics relating to one or more baseline time points before the start of treatment, each of the set of baseline characteristics characterizing the stage of the cancer, demographic attributes, one or more tumor sizes, white blood cell count, and / or lactate dehydrogenase level. A numerical cytokine release syndrome risk score is generated by processing the set of baseline characteristics using a risk score generation model. The risk of the subject experiencing cytokine release syndrome of at least a threshold grade after receiving treatment is predicted based on the numerical cytokine release syndrome risk score. A result is determined based on the predicted risk, which corresponds to a recommendation whether to monitor the subject with inpatient monitoring after completion of treatment. The result is output.

[0008] The method includes determining an outcome based on the predicted risk, the outcome corresponding to a recommendation whether to monitor the subject with inpatient monitoring after completion of treatment. The outcome may correspond to a recommendation to monitor the subject with inpatient monitoring after completion of treatment, and the method further includes monitoring the subject with inpatient monitoring at a medical facility for at least 24 hours after completion of treatment if the result indicates that the subject is at high risk of experiencing cytokine release syndrome.

[0009] The method can include identifying an on-treatment level of a cytokine, where the on-treatment level of the cytokine indicates the level of the cytokine in an on-treatment sample collected from the subject while the treatment was being administered or within one hour of completion of the treatment, and determining an on-treatment cytokine fold change of the cytokine based on the on-treatment level of the cytokine and a baseline level of the cytokine indicating the level of the cytokine in a baseline sample collected from the subject prior to initiation of the treatment, wherein the predicted risk is further based on the on-treatment cytokine fold change.

[0010] The method may include specifying a dosage for at least a portion of the treatment, wherein the predicted risk is further based on the dosage.

[0011] The risk score generation can include a regression model.

[0012] The treatment can include administering a T cell immunotherapy.

[0013] Treatment can include administering glofitamab or mosunetuzumab.

[0014] In some embodiments, a method is provided that includes identifying an on-treatment level of a cytokine, the on-treatment level of the cytokine representing the level of the cytokine in an on-treatment sample collected from the subject while the treatment was being administered or within one hour of completion of the treatment. An on-treatment cytokine fold change of the cytokine is determined based on the on-treatment level of the cytokine and a baseline level of the cytokine representing the level of the cytokine in a baseline sample collected from the subject before the start of the treatment. At least a partial dosage of the treatment is identified. Based on the on-treatment cytokine fold change and the dosage, a risk of the subject experiencing cytokine release syndrome of at least a threshold grade after receiving at least a partial dosage of the treatment is predicted. A result is determined based on the predicted risk, which corresponds to a recommendation of whether to monitor the subject with inpatient monitoring after completion of the treatment. The result is output.

[0015] The method can include identifying a set of baseline characteristics of the subject, the set of baseline characteristics relating to one or more baseline time points before treatment initiation, each of the set of baseline characteristics characterizing tumor burden, cancer stage, tumor spread, one or more tumor sizes, demographic attributes, white blood cell count and / or lactate dehydrogenase level, and the predicted risk further depends on the set of baseline characteristics.

[0016] The method can include generating a cytokine release syndrome risk score by processing the set of baseline characteristics with a risk score generation model, and the predicted risk is based on the cytokine release syndrome risk score.

[0017] The risk score generation can include a regression model.

[0018] The one or more parameters may include a set of weights.

[0019] Risk can be determined based on a linear combination of cytokine release syndrome risk score and dosage.

[0020] Predicting the risk that a subject will experience cytokine release syndrome can include performing one or more threshold comparisons.

[0021] The results may correspond to a recommendation to monitor the subject with inpatient monitoring after completion of treatment, and the method may include monitoring the subject with inpatient monitoring at a medical facility for at least 24 hours after completion of treatment if the results indicate that the subject is at high risk of experiencing cytokine release syndrome.

[0022] The results may correspond to a recommendation to monitor the subject with outpatient monitoring after completion of treatment, and the method may include monitoring the subject with outpatient monitoring if the results indicate that the subject is at low risk of experiencing cytokine release syndrome.

[0023] The subject may have been diagnosed with cancer and the treatment may include administering T cell immunotherapy.

[0024] The subject may have been diagnosed with cancer and the treatment may include administering glofitamab or mosunetuzumab.

[0025] Determining an on-treatment cytokine fold change of a cytokine based on a baseline level of the cytokine may include calculating the logarithm of the baseline level of the cytokine or a treated version thereof to generate a baseline logarithm, calculating the logarithm of the on-treatment level of the cytokine or a treated version thereof to generate an on-treatment logarithm, and subtracting the baseline logarithm from the on-treatment logarithm.

[0026] Determining the on-treatment cytokine fold change of the cytokine based on the baseline level of the cytokine may include calculating the logarithm of the difference between the baseline level of the cytokine and a constant to generate a baseline logarithm; calculating the logarithm of the difference between the on-treatment level of the cytokine and the constant to generate an on-treatment logarithm; and subtracting the baseline logarithm from the on-treatment logarithm.

[0027] Identifying the on-treatment level of the cytokine may include identifying a plurality of preliminary on-treatment levels of the cytokine indicative of the levels of the cytokine in a plurality of on-treatment samples collected from the subject while the treatment was being administered or within one day of completion of the treatment, wherein each of the plurality of on-treatment samples is collected at a different time point, and defining the on-treatment level of the cytokine as the maximum value of the plurality of preliminary on-treatment levels of the cytokine.

[0028] The treatment may involve administering an active ingredient, or the treatment may be preceded by the administration of a pre-treatment with another agent.

[0029] On-treatment levels may be determined using samples collected after administration of the active ingredient.

[0030] The cytokines can include tumor necrosis factor alpha, interleukin 6, interleukin 8, interleukin 10, or macrophage inflammatory protein 1 beta.

[0031] On-treatment levels of cytokines can be determined by collecting blood samples from the subject while the treatment is being administered and processing the blood samples with capture and detection antibodies for the cytokine.

[0032] In some embodiments, a method is provided that includes determining a baseline level of a cytokine, the level of the cytokine being in a baseline sample collected from the subject before the start of treatment; determining an on-treatment level of the cytokine, the on-treatment level of the cytokine being indicative of the level of the cytokine in an on-treatment sample collected from the subject while the treatment was being administered or within one hour of completion of the treatment; and identifying at least a portion of the dosage of the treatment. Further, the baseline level of the cytokine and the on-treatment level of the cytokine are input into a calculation system. A result corresponding to a recommendation to monitor the subject with in-patient monitoring after completion of the treatment is received, and the subject is monitored with in-patient monitoring after completion of the treatment.

[0033] Subjects may be monitored by self-monitoring for at least four hours after completion of treatment.

[0034] Results can be generated by the computational system by determining an on-treatment cytokine fold change for the cytokine based on the baseline level of the cytokine and the on-treatment level of the cytokine, and predicting the risk of the subject experiencing cytokine release syndrome of at least a threshold grade after receiving at least some doses of the treatment based on the on-treatment cytokine fold change and the dose.

[0035] In some embodiments, a method is provided that includes determining a baseline level of a cytokine indicative of the level of the cytokine in a baseline sample collected from the subject prior to initiation of treatment, determining an on-treatment level of the cytokine, the on-treatment level of the cytokine indicative of the level of the cytokine in an on-treatment sample collected from the subject while the treatment was being administered or within one hour of completion of the treatment, and identifying a dosage of at least a portion of the treatment. Further, the baseline level of the cytokine and the on-treatment level of the cytokine are input into a computing system, and a result corresponding to a recommendation to monitor the subject with outpatient monitoring after completion of the treatment is received, and the subject is monitored with outpatient monitoring after completion of the treatment.

[0036] Subjects may be monitored by self-monitoring for at least four hours after completion of treatment.

[0037] Results can be generated by the computational system by determining an on-treatment cytokine fold change of the cytokine based on the baseline level of the cytokine and the on-treatment level of the cytokine, and predicting the risk of the subject to experience cytokine release syndrome of at least a threshold grade after receiving at least some doses of the treatment based on the on-treatment cytokine fold change and the dose.

[0038] In some embodiments, there is provided a use of a computational prediction to determine whether to monitor a subject with inpatient monitoring for cytokine release syndrome after administration of a treatment, the computational prediction being provided by a computing device implementing a risk score generation model that determines an on-treatment cytokine fold-change for a cytokine based on a baseline level of the cytokine indicative of the level of the cytokine in a baseline sample collected from the subject prior to the start of the treatment, and an on-treatment level of the cytokine indicative of the level of the cytokine in an on-treatment sample collected from the subject while the treatment was being administered or within one hour of completion of the treatment, and predicts the risk of the subject experiencing cytokine release syndrome of at least a threshold grade after administration of the treatment based on the on-treatment cytokine fold-change.

[0039] In some embodiments, a system is provided that includes one or more data processors and a non-transitory computer-readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of one or more of the methods disclosed herein.

[0040] In some embodiments, a computer program product is provided that is tangibly embodied in a non-transitory machine-readable storage medium and includes instructions configured to cause one or more data processors to perform some or all of one or more of the methods disclosed herein.

[0041] Some embodiments of the present disclosure include a system including one or more data processors. In some embodiments, the system includes a non-transitory computer-readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein. Some embodiments of the present disclosure include a computer program product tangibly embodied in a non-transitory machine-readable storage medium containing instructions configured to cause one or more data processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein.

[0042] For any method, use, system, or computer program product disclosed herein, treating can include administering a therapy comprising an antibody or a small molecule.

[0043] The administered therapy may include an antibody.

[0044] The antibody can specifically bind to CD20, CD52, CD30, CD40, or PD-1.

[0045] The antibody can be rituximab, obinutuzumab, alemtuzumab, brentuximab, dacetuzumab, or nivolumab.

[0046] The antibody may be a multispecific antibody that engages T cells when bound to at least one of the antigens.

[0047] The multispecific antibody is capable of specifically binding to at least CD3.

[0048] The multispecific antibody is capable of binding at least more specifically to CD20.

[0049] The multispecific antibody may be a bispecific antibody.

[0050] The bispecific antibody can specifically bind to CD3 and / or CD20.

[0051] The bispecific antibody can be mosunetuzumab or glofitamab.

[0052] The therapy can include a small molecule, for example, oxaliplatin or lenalidomide.

[0053] The terms and expressions which have been employed are used as terms of description rather than of limitation, and in the use of such terms and expressions there is no intention to exclude any equivalents of the features shown and described or portions thereof, recognizing that various modifications are possible within the scope of the claimed invention. Thus, while the claimed invention has been specifically disclosed by embodiments and optional features, it is to be understood that modifications and variations of the concepts disclosed herein may be adopted by those skilled in the art, and that such modifications and variations are deemed to be within the scope of the invention as defined by the appended claims.

[0054] The present disclosure is described in conjunction with the accompanying drawings, in which: [Brief explanation of the drawings]

[0055] [Figure 1] 1 shows an exemplary network for stratifying subjects for differential monitoring or treatment that predicts the risk of one or more individual subjects experiencing a cytokine risk syndrome event according to some embodiments. [Figure 2A] 1 illustrates a flow chart of a process for predicting a subject's risk of experiencing cytokine release syndrome. [Figure 2B] 1 illustrates a process for using predicted risk to determine whether to recommend a subject for hospitalization or outpatient monitoring for cytokine release syndrome. [Figure 3]1 depicts the timing of administration in various cohorts receiving treatments including obinutuzumab and glofitamab. [Figure 4] Depictions of exemplary data used to train and validate a feature selection model (which identifies a reduced feature set and threshold to convert any non-binary baseline characteristics into binary variables), a risk score generation model (which converts the binary value of the reduced feature set into a risk score), and a decision tree model (which converts the risk score and cytokine fold change into a prediction of whether grade 2+ cytokine release syndrome will occur). [Figure 5] The timing of cytokine release syndrome in each exemplary analysis cohort is shown. [Figure 6] The percentage of subjects from exemplary training and validation datasets within each cohort who experienced a cytokine release syndrome event during week 1 of cycle 1 is shown. [Figure 7] An exemplary workflow is shown that is used to identify the extent to which various baseline characteristics (or "risk factors") contribute to predicting the occurrence of cytokine release syndrome and how the parameters of the model are learned. [Figure 8] 1 is a forest plot showing the extent to which each of several baseline characteristics was predictive of the occurrence of cytokine release syndrome (grade 2+ after the first glofitamab dose) in an exemplary dataset. [Figure 9A] Illustrates how a multivariate logistic regression model can be used to predict risk of cytokine release. [Figure 9B] Illustrates how cytokine release risk can be calculated and used together with dosage in predictive models. [Figure 10] 1 shows exemplary negative predictive values ​​(NPVs) for predicted negative cases corresponding to risk scores from two versions of the risk score generation model. [Figure 11] 1 shows exemplary negative predictive values ​​for predicted negative cases in the validation dataset for the 2.5 / 10 / 30 mg step-up dose cohort. [Figure 12A] 1 shows an exemplary probability of cytokine release syndrome (grade 2 or greater after the first glofitamab dose) occurring according to the cytokine release syndrome risk score (CRSRS) at each of three exemplary thresholds that identify whether the event is predicted to occur or not. [Figure 12B] 10 shows statistics about predictions made using a trained decision tree model processing a validation data set. [Figure 13] 1 shows an exemplary baseline cytokine release syndrome risk score (CRSRS) distribution corresponding to clinical study NP30179. [Figure 14] A and B show exemplary fold changes in IL-6 and TNF-α (respectively) during the first glofitamab treatment cycle. [Figure 15] Cytokine fold change in IL-6 contrast between an exemplary subject who did not experience cytokine release syndrome (left plot) and an exemplary subject who did experience cytokine release syndrome (right plot). [Figure 16] A and B show boxplots illustrating the dependence of exemplary on-treatment cytokine fold changes on the presence or severity of initial cytokine release syndrome. [Figure 17A] 1 shows exemplary on-treatment levels of the cytokine IL-6 during the first cycle of glofitamab treatment. [Figure 17B] 1 shows exemplary on-treatment levels of the cytokine TNF-α during the first cycle of glofitamab treatment. [Figure 18] A and B show the maximum log2 fold change of an exemplary subject in IL-6 and TNF-α, respectively. [Figure 19A] FIG. 1 shows an exemplary time course of cytokine fold change over various treatment-related times, and was generated to simultaneously stratify subjects according to time of onset of cytokine release syndrome relative to treatment initiation. [Figure 19B]FIG. 1 shows an exemplary time course of cytokine fold change over various treatment-related times, and was generated to simultaneously stratify subjects according to time of onset of cytokine release syndrome relative to treatment initiation. [Figure 20A] 1 shows an exemplary time course of cytokine fold changes and box plots comparing cytokine changes in cases identified based on whether any type of cytokine release syndrome occurs or whether at least grade 2 cytokine release syndrome occurs. [Figure 20B] 1 shows an exemplary time course of cytokine fold changes and box plots comparing cytokine changes in cases identified based on whether any type of cytokine release syndrome occurs or whether at least grade 2 cytokine release syndrome occurs. [Figure 21A] 1 is a comparison of exemplary fold changes in cytokines and cytokine release syndrome risk scores in different dose groups. [Figure 21B] 1 is a comparison of exemplary fold changes in cytokines and cytokine release syndrome risk scores in different dose groups. [Figure 22] 1 shows the results of a landmark analysis of how the probability of developing grade 2 or higher cytokine release syndrome (adjusted for the first glofitamab dose) relates to a normalized version of the cytokine release syndrome risk score. [Figure 23] 1 illustrates how the aggressiveness of any observed cytokine release syndrome relates to both the cytokine release syndrome risk score and cytokine fold change for TNF-α. [Figure 24] The negative predictive value and low-risk detection rate for the step-up (validation model) dose cohort cutoff value set by CRSRS.5p for the full 8-parameter score and the reduced 5-parameter score are shown.

[0056] In the accompanying drawings, similar components and / or features may have the same reference label. Furthermore, various components of the same type may be distinguished by following the reference label with a dash and a second label that distinguishes between the similar components. When only a first reference label is used in this specification, the description is applicable to any one of the similar components having the same first reference label, regardless of the second reference label. DETAILED DESCRIPTION OF THE INVENTION

[0057] I. Overview The technology disclosed herein relates to using multivariate analysis to predict whether a subject will experience cytokine release syndrome (e.g., of at least a predetermined severity) based on baseline or on-treatment data points. Prediction can include predicting whether the subject will be determined to be at low risk of experiencing cytokine release syndrome (e.g., of at least a threshold severity) and / or a candidate for outpatient monitoring for cytokine release syndrome. This prediction can be made after optimizing the negative predictive value of a model used to generate an output predicting the occurrence of cytokine release syndrome (e.g., of at least a threshold severity). Alternatively or additionally, prediction can include predicting whether the subject will be determined to be at risk of experiencing cytokine release syndrome (e.g., of at least a threshold severity) and / or a candidate for inpatient monitoring for cytokine release syndrome. This prediction can be made after optimizing the positive predictive value of a model used to generate an output predicting the occurrence of cytokine release syndrome (e.g., of at least a threshold severity).

[0058] A baseline data point may relate to one or more time points before treatment is initiated and / or between pre-treatment and the first non-pre-treatment dose of another active ingredient. For example, a baseline data point may be generated by processing a sample collected before the administration of the first dose of an active ingredient, and / or the baseline data point may be retrieved from medical records associated with an assessment performed before the first non-pre-treatment dose of an active ingredient. An on-treatment data point may relate to one or more time points between the start of treatment and the end of treatment (potentially with a buffer). For example, an on-treatment data point may be defined as equal to the maximum concentration level of a particular cytokine in all measurements taken after the first (e.g., non-priming) dose of an active ingredient and at the end of treatment (potentially with a buffer).

[0059] Multivariate analysis and risk predictors can include the change in the level of a particular cytokine at on-treatment (or a treated version thereof) from the level of a particular cytokine at baseline (or a treated version thereof). For example, the treated version of the cytokine level can be defined as including the logarithm (e.g., log base 2) of the sum of the cytokine levels and a non-zero positive constant (e.g., 1).

[0060] The multivariate analysis can include generating a cytokine release syndrome risk score based on one or more baseline characteristics of the subject (associated with one or more baseline time points). The baseline characteristics can include one or more measurements characterizing tumor burden, one or more measurements characterizing tumor spread, one or more measurements characterizing the presence or extent of malignant cells in a defined body component (e.g., bone marrow or peripheral blood), one or more demographic attributes (e.g., age), and / or one or more measurements characterizing the incidence or severity of comorbidities. The cytokine release syndrome risk score can also or alternatively be generated based on the dosage of the active ingredient in the treatment.

[0061] Generating a cytokine release syndrome risk score can include using a multivariate regression model (e.g., a linear regression model or a logistic regression model) to convert one or more baseline characteristics (or multiple baseline characteristics) into a model output. The model output can include a scaled or unscaled representation of the risk of experiencing cytokine release syndrome. The model output can be normalized. For example, the model output can be a number between 0 and 1, where a value of 1 represents the highest predicted risk of developing cytokine release syndrome and a value of 0 represents the lowest predicted risk of developing cytokine release syndrome.

[0062] The multivariate model can incorporate outputs from other machine learning modules (e.g., random forest models). The multivariate model can include a set of parameters, the values ​​of each parameter being learned by training the multivariate model and the multivariate machine learning model using a training dataset. The set of parameters (e.g., a set of model weights) can include one or more associated parameters for each of one or more baseline characteristics, and the one or more parameters can identify the degree to which the model output depends on the baseline characteristics and / or can identify a significance value representing the degree to which the baseline characteristics are predictive of the model output.

[0063] The final cytokine release syndrome risk predictor can be generated based on terms identifying or derived from baseline characteristics (e.g., parameters of a cytokine release syndrome risk score) and terms identifying or derived from dosage (e.g., dosage of a treatment or active ingredient) or drug exposure. For example, a combined cytokine release syndrome risk score can be defined as a linear combination, sum, or weighted sum of the cytokine release syndrome risk score and dosage / exposure.

[0064] With the help of a predictive model that combines risk factors with dose / exposure information, for all subjects with a predetermined (e.g., obtained at baseline) value of the cytokine release syndrome risk score, the dose or exposure can be adjusted to limit the expected risk of cytokine release syndrome.

[0065] The cytokine release syndrome prediction model can be expanded using one or more cytokine fold changes. For example, cytokine release syndrome risk can be predicted based on cytokine release syndrome risk score and cytokine fold change. As another example, cytokine release syndrome risk can be predicted based on cytokine release syndrome risk score, dosage / exposure, and cytokine fold change. In another example, cytokine release syndrome risk can be predicted based on dosage and cytokine fold change. As yet another example, for a particular subject, cytokine release syndrome risk can be optimized (limited) by selecting a maximum dose / exposure at which the cytokine release syndrome risk predicted based on the score, dosage, and cytokine fold change does not exceed a certain predefined value. Risk can be a numerical risk (e.g., representing probability), a categorical risk (e.g., high, moderate, or low), or a binary risk (e.g., at risk or not). Categorical or binary risks can be generated based on one or more threshold comparisons. For example, it can be determined that a subject has a high risk of experiencing cytokine release syndrome if the risk score exceeds a threshold, and / or if the cytokine fold change exceeds a cytokine threshold, the subject has a low risk of experiencing cytokine release syndrome otherwise.

[0066] Results corresponding to the prediction can be output (e.g., displayed or transmitted). The results may include a predicted cytokine release syndrome risk. The results may include a recommended action, a default action, or an action to be implemented.

[0067] The cytokine release syndrome risk can be used to identify recommended monitoring actions for a subject at the end of treatment, or to identify recommendations for such procedures. For example, a given subject may be recommended to have inpatient monitoring (e.g., admitting the subject to a medical facility) if an inpatient monitoring condition is met. The inpatient monitoring condition can be configured to be met (for example) if the subject is defined as being at high risk, if the risk is defined as being in a category other than low, or if the risk exceeds a predefined (e.g., numerical or categorical) threshold. In some cases, inpatient monitoring of a subject is provided if the subject is at high risk, if the risk is in a category other than low, or if the risk exceeds a predefined threshold.

[0068] A given subject may alternatively or additionally be recommended to be discharged and / or have outpatient monitoring (e.g., admitting the subject to a medical facility) if inpatient monitoring conditions are not met. In some examples, a subject's discharge and / or outpatient monitoring is provided if they are at low risk, in a category other than high risk, or if their risk does not exceed a predefined threshold.

[0069] Determining cytokine release syndrome risk (e.g., with few false negatives) has the advantage of promoting efficient resource allocation for inpatient monitoring. Carefully monitoring all patients with inpatient monitoring is expensive, consumes considerable physical resources, and is time-intensive. Meanwhile, over-inclusion of subjects who should have outpatient monitoring may result in the selection of patients who cannot receive prompt treatment for cytokine release syndrome. Thus, the technology disclosed herein can promote efficient resource use by prioritizing providing resource-intensive monitoring to subjects at a relatively high risk of experiencing cytokine release syndrome, while reserving less resource-intensive monitoring for subjects who are unlikely to require prompt intervention (e.g., cytokine release syndrome response).

[0070] II. Definition The term "cytokine," as used herein, refers to a signaling molecule that is transiently produced after cell activation and helps mediate and regulate immunity, inflammation, and hematopoiesis. Cytokines can be any of a large group of proteins, peptides, and glycoproteins secreted by specific cells of the immune system. These molecules act as regulators that modulate the function of individual cells. Cytokines can act locally as autocrine, paracrine, or endocrine response modifiers, exerting their effects through specific cell surface receptors on their target cells. As used herein, "autocrine" or "autocrine action" means that a cytokine exerts its effect by binding to a receptor on the membrane of the same cell from which it is secreted. "Paracrine" or "paracrine action" means that a cytokine binds to a receptor on a target cell in close proximity to the cell that produced the cytokine. "Endocrine" or "endocrine action" means that a cytokine travels through the circulation and acts on target cells in parts of the body. Elevated levels of cytokines, such as one or more cytokines selected from the group consisting of IL-1β, IL-2, IL-6, IL-8, MIP1b, MCP1, IL-10, IFN-γ, TGF-β, and TNF-α, are often associated with cytokine release syndrome.

[0071] The term "cytokine release syndrome" or "CRS," as used herein, refers to an acute systemic inflammatory syndrome characterized by fever and multiple organ failure associated with immunotherapy, such as T cell immunotherapy, therapeutic antibodies, chimeric antigen receptor (CAR) T cell therapy, and stem cell transplantation. CRS is a potentially fatal cytokine-related toxicity that can occur as a result of cancer immunotherapy. CRS is an acute systemic inflammatory syndrome resulting from high levels of immune activation and characterized by elevated circulating cytokine levels when large numbers of lymphocytes and myeloid cells release inflammatory cytokines upon activation. The severity and timing of symptom onset of CRS vary depending on the degree of immune cell activation, the type of therapy administered, and the tumor burden. Symptoms of CRS can include neurotoxicity, cardiac dysfunction, disseminated intravascular coagulation, adult respiratory distress syndrome, renal failure, and liver failure. Symptoms include fever (with or without chills ("shivering chills" - elevated body temperature accompanied by shivering and chills), fatigue, malaise, myalgia (muscle pain), vomiting, headache, nausea, loss of appetite, arthalgia (joint pain), diarrhea, rash, hypoxemia (low blood oxygen), tachypnea (rapid breathing), hypotension, widened pulse pressure (the difference between systolic and diastolic blood pressure), potentially decreased cardiac output (late stage), increased cardiac output (early stage), and azotemia (high levels of nitrogen in the blood). ), hypofibrinogenemia (blood clotting disorder, with or without bleeding), elevated D-dimer (correlated with blood clotting), hyperbilirubinemia (excess bilirubin in the blood due to red blood cell breakdown), hypertransaminasemia (elevated blood transaminases correlated with liver disease and hepatitis), confusion, delirium, altered mental status, hallucinations, tremors, seizures, abnormal gait, word finding difficulty, frank aphasia (language impairment affecting speaking and / or comprehension and writing), or It can include dymetria (the inability to accurately adapt movements without visual aids). CRS is characterized by inflammation beyond that which can be attributed to a normal response to a pathogen (if a pathogen is present) or any cytokine-induced organ failure (if a pathogen is absent).

[0072] The term "inpatient monitoring," as used herein, refers to monitoring performed by one or more healthcare providers (e.g., one or more physicians and / or one or more nurses) concurrently at a healthcare facility and provided to a subject at the healthcare facility. Thus, the subject and at least one healthcare provider can be physically present at the same healthcare facility at the same time. The healthcare facility can include (for example) a hospital, medical clinic, doctor's office, or drug infusion center. The subject can be admitted to the healthcare facility during inpatient monitoring. The duration of inpatient monitoring can be at least (for example) 2 hours, 4 hours, 8 hours, 12 hours, 24 hours, 36 hours, 48 ​​hours, 72 hours, or 96 hours. The duration of inpatient monitoring can be (for example) 2 weeks, 1 week, 5 days, 4 days, 3 days, or less than 2 days. For example, a subject may undergo inpatient monitoring for 2 to 4 days after treatment administration has ended.

[0073] The term "antibody" as used herein is used in the broadest sense and encompasses a variety of antibody structures, including, but not limited to, monoclonal antibodies, polyclonal antibodies, multispecific antibodies (e.g., bispecific antibodies), and antibody fragments, so long as they exhibit the desired antigen-binding activity.

[0074] "Antibody fragment" refers to a molecule other than an intact antibody that contains a portion of an intact antibody that binds to the antigen to which the intact antibody binds. Examples of antibody fragments include, but are not limited to, Fv, Fab, Fab', Fab'-SH, F(ab')2; diabodies; linear antibodies; single-chain antibody molecules (e.g., scFv); and multispecific antibodies formed from antibody fragments.

[0075] The terms "full-length antibody," "intact antibody," and "whole antibody" are used interchangeably herein to refer to an antibody having a heavy chain that has a structure substantially similar to a native antibody structure or that contains an Fc region as defined herein.

[0076] "Binding domain" refers to a portion of a compound or molecule that specifically binds to a target epitope, antigen, ligand, or receptor. Binding domains include, but are not limited to, antibodies (e.g., monoclonal, polyclonal, recombinant, humanized, and chimeric antibodies), antibody fragments or portions thereof (e.g., Fab fragments, Fab'2, scFv antibodies, SMIPs, domain antibodies, diabodies, minibodies, scFv-Fc, affibodies, nanobodies, and antibody VH and / or VL domains), receptors, ligands, aptamers, and other molecules with identified binding partners.

[0077] The term "Fc region" as used herein defines the C-terminal region of an immunoglobulin heavy chain containing at least a portion of the constant region. This term includes native-sequence Fc regions and variant Fc regions. In one embodiment, a human IgG heavy chain Fc region extends from Cys226 or Pro230 to the carboxyl terminus of the heavy chain. However, the C-terminal lysine (Lys447) of the Fc region may or may not be present. Unless otherwise specified herein, the numbering of amino acid residues within the Fc region or constant region follows EU numbering, also referred to as the EU index, as described in Kabat et al., Sequences of Proteins of Immunological Interest, 5th Ed. Public Health Service, National Institutes of Health, Bethesda, MD, 1991.

[0078] The "class" of an antibody refers to the type of constant domain or region carried by its heavy chain. There are five major classes of antibodies: IgA, IgD, IgE, IgG, and IgM, several of which can be further divided into subclasses (isotypes), e.g., IgG1, IgG2, IgG3, IgG4, IgA1, and IgA2. The heavy-chain constant domains that correspond to the different classes of immunoglobulins are called α, δ, ε, γ, and μ, respectively.

[0079] The term "variable region" or "variable domain" refers to the domain of an antibody heavy or light chain that is involved in binding the antibody to an antigen. The variable domains of the heavy and light chains (VH and VL, respectively) of natural antibodies generally have similar structures, with each domain containing four conserved framework regions (FR) and three hypervariable regions (HVR). (See, for example, Kindt et al., Kuby Immunology, 6th ed., W.H. Freeman and Co., page 91 (2007)). A single VH or VL domain may be sufficient to confer antigen-binding specificity. Furthermore, antibodies that bind to a specific antigen may be isolated using the VH or VL domain from an antibody that binds the antigen, followed by screening a library of complementary VL or VH domains, respectively. See, for example, Portolano et al., J. Immunol. 150:880-887 (1993); Clarkson et al., Nature 352:624-628 (1991).

[0080] The term "hypervariable region" or "HVR," as used herein, refers to each of the regions of an antibody variable domain whose sequences are hypervariable ("complementarity determining regions" or "CDRs") and / or which form structurally defined loops ("hypervariable loops") and / or contain residues that contact the antigen ("antigen contacts"). Generally, antibodies contain six HVRs, three in the VH (H1, H2, H3) and three in the VL (L1, L2, L3). Exemplary HVRs herein include the following: (a) Hypervariable loops present at amino acid residues 26–32 (L1), 50–52 (L2), 91–96 (L3), 26–32 (H1), 53–55 (H2), and 96–101 (H3) (Chothia and Lesk, J. Mol. Biol. 196:901–917 (1987)); (b) CDRs located at amino acid residues 24–34 (L1), 50–56 (L2), 89–97 (L3), 31–35b (H1), 50–65 (H2), and 95–102 (H3) (Kabat et al., Sequences of Proteins of Immunological Interest, 5th Ed. Public Health Service, National Institutes of Health, Bethesda, MD (1991)); (c) antigenic contacts present at amino acid residues 27c-36 (L1), 46-55 (L2), 89-96 (L3), 30-35b (H1), 47-58 (H2), and 93-101 (H3) (MacCallum et al., J. Mol. Biol. 262:732-745 (1996)), and (d) A combination of (a), (b) and / or (c) comprising HVR amino acid residues 46-56 (L2), 47-56 (L2), 48-56 (L2), 49-56 (L2), 26-35 (H1), 26-35b (H1), 49-65 (H2), 93-102 (H3) and 94-102 (H3).

[0081] Unless otherwise indicated, HVR residues and other residues in the variable domain (e.g., FR residues) are numbered herein according to Kabat et al., supra.

[0082] The term "monoclonal antibody," as used herein, refers to an antibody obtained from a population of substantially homogeneous antibodies, i.e., the individual antibodies comprising the population are identical and / or bind to the same epitope, except for possible variant antibodies (e.g., containing naturally occurring mutations or arising during production of the monoclonal antibody preparation, such variants generally being present in small amounts). In contrast to polyclonal antibody preparations, which typically include different antibodies directed against different determinants (epitopes), each monoclonal antibody of a monoclonal antibody preparation is directed against a single determinant on an antigen. Thus, the modifier "monoclonal" indicates the character of the antibody as being obtained from a substantially homogeneous population of antibodies and should not be construed as requiring production of the antibody by any particular method. For example, monoclonal antibodies used in accordance with the present invention may be produced by a variety of techniques, including, but not limited to, hybridoma methods, recombinant DNA methods, phage display methods, and methods utilizing transgenic animals containing all or part of the human immunoglobulin loci; such methods and other exemplary methods for producing monoclonal antibodies are described herein.

[0083] "Affinity" refers to the strength of the sum total of non-covalent interactions between a single binding site of a molecule (e.g., an antibody) and its binding partner (e.g., an antigen). Unless otherwise indicated, as used herein, "binding affinity" refers to the intrinsic binding affinity that reflects a 1:1 interaction between members of a binding pair (e.g., an antibody and an antigen). The affinity of a molecule X for its partner Y can generally be represented by the dissociation constant (Kd). Affinity can be measured by common methods known in the art, including those described herein. Specific illustrative and exemplary embodiments for measuring binding affinity are described below.

[0084] In certain embodiments, the antibody is a multispecific antibody, e.g., a bispecific antibody. A "multispecific antibody" is a monoclonal antibody that has binding specificities for at least two different sites, i.e., different epitopes on different antigens or different epitopes on the same antigen. A multispecific antibody can also have three or more binding specificities. Multispecific antibodies can be prepared as full-length antibodies or antibody fragments.

[0085] Techniques for producing multispecific antibodies include, but are not limited to, recombinant coexpression of two immunoglobulin heavy chain-light chain pairs with different specificities (see Milstein and Cuello, Nature 305:537 (1983)) and "knob-in-hole" engineering (see, e.g., U.S. Pat. No. 5,731,168 and Atwell et al., J. Mol. Biol. 270:26 (1997)). Multispecific antibodies can also be produced by modifying the electrostatic steering effect to create antibody Fc heterodimeric molecules (see, e.g., WO 2009 / 089004), cross-linking two or more antibodies or fragments (see, e.g., U.S. Pat. No. 4,676,9880 and Brennan et al., Science, 229:81 (1985)), using leucine zippers to produce bispecific antibodies (see, e.g., Kostelny et al., J. Immunol., 148(5):1547-1553 (1992) and WO 2011 / 034605), and circumventing the problem of light chain mispairing. Bispecific antibody fragments can also be made by using conventional light chain technology to produce bispecific antibody fragments (see, e.g., WO 98 / 504311), by using "diabody" technology to make bispecific antibody fragments (see, e.g., Hollinger et al., Proc. Natl. Acad. Sci. USA, 90:6444-6448 (1993)), by using single-chain Fv (sFv) dimers (see, e.g., Gruber et al., J. Immunol., 152:5368 (1994)), and by preparing trispecific antibodies as described, for example, in Tutt et al., J. Immunol. 147:60 (1991).

[0086] For example, engineered antibodies with three or more antigen-binding sites, including "octopus antibodies," or DVD-Igs, can also be used in the disclosed methods (see, e.g., WO 2001 / 77342 and WO 2008 / 024715). Other examples of multispecific antibodies with three or more antigen-binding sites can be found in WO 2010 / 115589, WO 2010 / 112193, WO 2010 / 136172, WO 2010 / 145792, and WO 2013 / 026831. Bispecific antibodies or antigen-binding fragments thereof also include "dual-acting FAbs" or "DAFs" (see, e.g., U.S. Patent Application Publication Nos. 2008 / 0069820 and WO 2015 / 095539).

[0087] Multispecific antibodies can also be provided by asymmetric configurations with domain crossovers in one or more binding arms of the same antigen specificity, i.e., by exchanging VH / VL domains (see, e.g., WO 2009 / 080252 and WO 2015 / 150447), CH1 / CL domains (see, e.g., WO 2009 / 080253), or complete Fab arms (see, WO 2009 / 080251, WO 2016 / 016299; see also Schaefer et al., USA, 108 (2011) 118-1191, and Klein et al., MAbs 8 (2016) 1010-20). In one embodiment, the multispecific antibody comprises a cross-Fab fragment. The term "cross-Fab fragment" or "xFab fragment" or "crossover Fab fragment" refers to a Fab fragment in which either the variable or constant regions of the heavy and light chains have been exchanged. A cross-Fab fragment contains a polypeptide chain composed of a light chain variable region (VL) and a heavy chain constant region 1 (CH1), and a polypeptide chain composed of a heavy chain variable region (VH) and a light chain constant region (CL). Asymmetric Fab arms can also be engineered by introducing charged or uncharged amino acid mutations at the domain interface to direct correct Fab pairing. See, for example, WO 2016 / 172485.

[0088] A variety of additional molecular formats for multispecific antibodies are known in the art (see, for example, Spiess et al., Mol Immunol 67 (2015) 95-106).

[0089] Certain types of multispecific antibodies can recruit T cells, T cell-engaging antibodies. "T cell bispecific antibodies" are one type of multispecific, bispecific antibodies engineered to bind to two different antigens, one targeting tumor cells and the other targeting effector cells, usually T lymphocytes. When a T cell bispecific antibody binds to a T cell and a tumor cell, the tumor cell and T cell are brought into close proximity, activating the T cell and mediating tumor cell destruction.

[0090] Examples of bispecific antibody formats are so-called "BiTE" (bispecific T cell engager) molecules in which two scFv molecules are fused by a flexible linker (see, e.g., WO 2004 / 106381, WO 2005 / 061547, WO 2007 / 042261, and WO 2008 / 119567; Nagorsen and Baeuerle, Exp Cell Res 317, 1255-1260 (2011)); diabodies (Holliger et al., Prot Eng 9, 299-305 (1996)) and derivatives thereof, such as tandem diabodies ("TandAb"; Kipriyanov et al., J Mol Biol 293, 41-56 (1999); "DART" (dual affinity retargeting) molecules (Johnson et al., J Mol Biol 399, 436-449 (2010)), which are based on the diabody format but feature a C-terminal disulfide bridge for further stabilization, as well as the so-called triomabs (reviewed in Seimetz et al., Cancer Treat Rev 36, 458-467 (2010)), which are whole mouse / rat IgG hybrid molecules. Specific T cell bispecific antibody formats are described in WO 2013 / 026833, WO 2013 / 026839, WO 2016 / 020309; Bacac et al., Oncoimmunology 5(8) (2016) e1203498.

[0091] The terms "anti-CD3 antibody" and "antibody that binds to CD3" refer to an antibody that can bind to CD3 with sufficient affinity such that the antibody is useful as a diagnostic and / or therapeutic agent in targeting CD3. In one embodiment, the extent of binding of an anti-CD3 antibody to an unrelated, non-CD3 protein is less than about 10% of the binding of the antibody to CD3, as measured, for example, by radioimmunoassay (RIA). In certain embodiments, an antibody that binds to CD3 has an affinity of ≦1 μM, ≦100 nM, ≦10 nM, ≦1 nM, ≦0.1 nM, ≦0.01 nM, or ≦0.001 nM (e.g., 10-8 M or less, e.g. 10 -8 ~10 -13 M, e.g. 10 -9 M~10 -13 In certain embodiments, the anti-CD3 antibody binds to an epitope of CD3 that is conserved among CD3 of different species.

[0092] The term "cluster of differentiation 3" or "CD3," as used herein, unless otherwise indicated, refers to any native CD3 of any vertebrate source, including mammals such as primates (e.g., humans) and rodents (e.g., mice and rats), including, for example, the CD3ε, CD3γ, CD3α, and CD3β chains. The term encompasses "full-length" native CD3 (e.g., native or unmodified CD3ε or CD3γ) as well as any form of CD3 resulting from intracellular processing. The term also encompasses naturally occurring variants of CD3, including, for example, splice variants or allelic variants. CD3 includes, for example, the human CD3ε protein (NCBI Reference SEQ ID NO: NP_000724), which is 207 amino acids long, and the human CD3γ protein (NCBI Reference SEQ ID NO: NP_000064), which is 182 amino acids long.

[0093] The terms "anti-CD20 antibody" and "antibody that binds to CD20," as used herein, refer to an antibody that is capable of binding to CD20 with sufficient affinity such that the antibody is useful as a therapeutic agent in targeting CD20. In one embodiment, the extent of binding of an anti-CD20 antibody to an unrelated, non-CD20 protein is less than about 10% of the binding of the antibody to CD20, as measured, for example, by radioimmunoassay (RIA). In certain embodiments, an antibody that binds to CD20 has an affinity of ≦1 μM, ≦100 nM, ≦10 nM, ≦1 nM, ≦0.1 nM, ≦0.01 nM, or ≦0.001 nM (e.g., 10 -8 M or less, e.g. 10 -8 ~10 -13 M, e.g. 10 -9 M~10 -13In certain embodiments, the anti-CD20 antibody binds to an epitope of CD20 that is conserved among CD20 of different species.

[0094] The term "cluster of differentiation 20" or "CD20," as used herein, unless otherwise indicated, refers to any native CD20 of any vertebrate source, including mammals such as primates (e.g., humans), and rodents (e.g., mice and rats). The term encompasses "full-length," unprocessed CD20, as well as any form of CD20 resulting from processing within a cell. The term also encompasses naturally occurring variants of CD20, including, for example, splice variants or allelic variants. CD20 includes, for example, the human CD20 protein (see, e.g., NCBI Reference SEQ ID NO: NP_068769.2 (SEQ ID NO: 47) and NP_690605.1 (SEQ ID NO: 48)), which may be generated from a variant mRNA transcript that is, for example, 297 amino acids in length and that lacks, for example, a portion of the 5'UTR (see, e.g., NCBI Reference SEQ ID NO: NM_021950.3 (SEQ ID NO: 49)), or a longer variant mRNA transcript (see, e.g., NCBI Reference SEQ ID NO: NM_152866.2 (SEQ ID NO: 50)).

[0095] The terms "anti-CD20 / anti-CD3 bispecific antibody," "bispecific anti-CD20 / anti-CD3 antibody," and "antibody that binds to CD20 and CD3" or variants thereof refer to a multispecific antibody (e.g., a bispecific antibody) that is capable of binding to CD20 and / or CD3 with sufficient affinity such that the antibody is useful as a diagnostic and / or therapeutic agent targeting CD20 and / or CD3. In one embodiment, the extent of binding of a bispecific antibody that binds CD20 and CD3 to unrelated non-CD3 and / or non-CD20 proteins is less than about 10% of the binding of the antibody to CD3 and / or CD20, as measured, for example, by radioimmunoassay (RIA). In certain embodiments, a bispecific antibody that binds CD20 and CD3 has an affinity of ≦1 μM, ≦100 nM, ≦10 nM, ≦1 nM, ≦0.1 nM, ≦0.01 nM, or ≦0.001 nM (e.g., 10 -8 M or less, e.g. 10 -8 ~10 -13 M, e.g. 10 -9 M~10 -13 In certain embodiments, a bispecific antibody that binds to CD20 and CD3 binds to an epitope of CD3 that is conserved among CD3 of different species and / or an epitope of CD20 that is conserved among CD20 of different species.

[0096] As used herein, the terms "bind," "specifically bind to," or "specific for" refer to a measurable and reproducible interaction, such as binding, between a target and an antibody, which is determinative of the presence of the target in the presence of a heterogeneous population of molecules, including biomolecules. For example, an antibody that specifically binds to a target (which may be an epitope) is one that binds to that target with higher affinity, avidity, more readily, and / or for a longer period of time than it binds to other targets. In one embodiment, the extent to which an antibody binds to an unrelated target is less than about 10% of the binding of the antibody to the target, as measured, for example, by radioimmunoassay (RIA). In certain embodiments, an antibody that specifically binds to a target has a dissociation constant (K) of ≦1 μM, ≦100 nM, ≦10 nM, ≦1 nM, or ≦0.1 nM. D In certain embodiments, the antibody specifically binds to an epitope on a protein that is conserved among proteins of different species. In another embodiment, specific binding can include, but is not required to include, exclusive binding. As used herein, the term "specific binding" refers to, for example, binding to an epitope on a protein that is conserved among proteins of different species. -4 M or less, or 10 -5 M or less, or 10 -6 M or less, or 10 -7 M or less, or 10 -8 M or less, or 10 -9 M or less, or 10 -10 M or less, or 10 -11 M or less, or 10 -12 KD for targets below M, or 10 -4 M~10 -6 M or 10 -6 M~10 -10 M or 10 -7 M~10 -9 K in the M range D As will be appreciated by those skilled in the art, affinity and K D The values ​​are inversely correlated: a high affinity for the antigen corresponds to a low K DIn one embodiment, the term "specific binding" refers to binding of a molecule to a particular polypeptide or an epitope on a particular polypeptide without substantially binding to any other polypeptides or polypeptide epitopes.

[0097] The disclosed methods may be used when a therapeutic bispecific antibody that binds CD20 and CD3 (i.e., an anti-CD20 / anti-CD3 antibody) is used to treat a CD20-positive cell proliferative disorder, such as a B-cell proliferative disorder (e.g., non-Hodgkin's lymphoma (NHL) (e.g., diffuse large B-cell lymphoma (DLBCL) (e.g., relapsed and / or refractory DLBCL, or Richter's transformation), follicular lymphoma (FL) (e.g., relapsed and / or refractory FL, or transformed FL), mantle cell lymphoma (MCL), high-grade B-cell lymphoma, or primary mediastinal (thymic) large B-cell lymphoma (PMLBCL)) or chronic lymphocytic leukemia (CLL).

[0098] In some examples, the anti-CD20 / anti-CD3 bispecific antibody is mosunetuzumab, which has International Nonproprietary Name (INN) List 117 (WHO Drug Information, Vol. 31, No. 2, 2017, p. 303) or CAS Registry Number 1905409-39-3, and has (1) an anti-CD20 arm comprising the heavy chain and light chain sequences of SEQ ID NOs: 17 and 18, respectively, and (2) an anti-CD3 arm comprising the heavy chain and light chain sequences of SEQ ID NOs: 19 and 20, respectively. In some examples, the anti-CD20 / anti-CD3 bispecific antibody comprises (1) an anti-CD20 arm comprising a first binding domain comprising a heavy chain comprising the amino acid sequence of SEQ ID NO: 17 and a light chain comprising the amino acid sequence of SEQ ID NO: 18, and (2) an anti-CD3 arm comprising a second binding domain comprising a heavy chain comprising the amino acid sequence of SEQ ID NO: 19 and a light chain comprising the amino acid sequence of SEQ ID NO: 20. The various elements of mosunetuzumab (HVR, VH, VL, HC and LC) are shown in Table 1.

[0099] Anti-CD20 / anti-CD3 bispecific antibodies can be produced using recombinant methods and compositions, for example, as described in US Pat. No. 4,816,567. TIFF2025169248000002.tif255164TIFF2025169248000003.tif221170

[0100] In some embodiments, the anti-CD20 / anti-CD3 bispecific antibody useful in the methods provided herein is glofitamab. Glofitamab (Proposed INN: List 121 WHO Drug Information, Vol. 33, No. 2, 2019, page 276; also known as CD20-TCB, RO7082859, or RG6026) is a novel T cell-engaging bispecific full-length antibody with a 2:1 molecular configuration for bivalent binding to CD20 on B cells and monovalent binding to CD3, particularly the CD3 epsilon chain (CD3e), on T cells. Its CD3-binding region is fused head-to-tail to one of the CD20-binding regions via a flexible linker. This structure allows glofitamab to have superior in vitro potency compared to other CD20-CD3 bispecific antibodies in a 1:1 configuration and to produce significant antitumor effects in preclinical DLBCL models. The bivalency of CD20 preserves its efficacy in the presence of competing anti-CD20 antibodies, providing the opportunity for pretreatment or cotreatment with these agents. Glofitamab contains an engineered heterodimerized Fc region that completely abolishes binding to FcgR and Clq. By simultaneously binding to CD3e in the T cell receptor (TCR) complex on human CD20-expressing tumor cells and T cells, glofitamab induces tumor cell lysis in addition to T cell activation, proliferation, and cytokine release. B cell-mediated lysis by glofitamab is CD20-specific and does not occur in the absence of CD20 expression or simultaneous binding (crosslinking) of T cells to CD20-expressing cells. In addition to killing, CD3 crosslinking activates T cells, which is detected by increased T cell activation markers (CD25 and CD69), cytokine release (IFNγ, TNFα, IL-2, IL-6, IL-10), cytotoxic granule release (granzyme B), and T cell proliferation. The amino acid sequences are shown in Tables 2 and 3. TIFF2025169248000004.tif255165TIFF2025169248000005.tif255165TIFF2025169248000006.tif91170TIFF2025169248000007.tif145170

[0101] The term "bispecific antibody therapy," as used herein, refers to therapy using bispecific antibodies.

[0102] The term "on-treatment" period, as used herein, refers to the period beginning when administration of a treatment (or a cycle of treatment) begins and ending when administration of the treatment (or a cycle of treatment) is completed (potentially extended by a predefined buffer time interval). For example, the on-treatment period may end 30 minutes after completion of treatment administration. The on-treatment period may include the time during which the treatment (or a cycle of treatment) is infused into the subject. The on-treatment period may be (for example) at least 15 minutes, at least 30 minutes, at least 1 hour, at least 2 hours, at least 3 hours, at least 4 hours, at least 6 hours, or at least 8 hours. The on-treatment period may be (for example) less than 24 hours, less than 12 hours, less than 10 hours, less than 9 hours, less than 8 hours, 7 hours, or less than 6 hours. For example, the on-treatment period may be 3 to 5 hours long. As another example, the on-treatment period may be 7 to 9 hours long. An on-treatment period includes multiple distinct on-treatment time points, for example, a time point corresponding to the middle of a treatment administration and a time point corresponding to the end of a treatment administration.

[0103] The term "on-treatment level of a cytokine" or "on-treatment cytokine level," as used herein, refers to the level of a particular cytokine detected in a biological sample (e.g., a blood sample or tissue sample) collected during the on-treatment period. If multiple biological samples are collected from a given subject during the on-treatment period and cytokine levels are determined in each sample, the on-treatment level of a cytokine may be defined as the maximum of these cytokine levels. The on-treatment level of a cytokine may be determined (for example) using a cytokine capture and detection antibody introduced into the biological samples collected during the on-treatment period.

[0104] The term "baseline" period, as used herein, refers to a period ending with the start of the administration period. The baseline period can extend up to and may include the time when treatment is initiated. The baseline period may include the time when prior treatment is administered.

[0105] The term "baseline level of a cytokine" or "baseline cytokine level," as used herein, refers to the level of a particular cytokine detected in a biological sample (e.g., a blood sample or tissue sample) collected during a baseline period. The biological sample processed to identify the baseline level of a particular cytokine can include a sample collected at a predefined time before the start of treatment administration, or within a predefined time interval before the start of treatment administration. The baseline level of a cytokine can be determined (for example) using a cytokine capture and detection antibody introduced into a biological sample collected during the baseline period.

[0106] The term "baseline characteristics" of a subject includes characteristics of the subject detected during the baseline period, characteristics detected before the baseline period but assumed to be static, characteristics that are static, or characteristics that change in a deterministic manner. For example, if a subject was diagnosed with a disease subtype before the baseline period but the baseline period itself does not include any subtype diagnosis, it can be assumed that the subject's disease will remain the same subtype. Thus, a subtype can be a baseline characteristic. As another example, a subject's race can be recorded before the baseline period, during the baseline period, or during the on-treatment period; however, given that this type of characteristic is generally static throughout life, race can be characterized as a baseline characteristic regardless of when it was recorded. On the other hand, for more dynamic variables (e.g., age), baseline characteristics can be defined as values ​​detected during the baseline period and / or calculated based on the relative time of the baseline period. Baseline characteristics can also be based on assessment of samples collected during the baseline period. For example, baseline characteristics can be characterized by the presence or absence of malignant cells and / or the extent to which malignant cells are present within the body components (from which samples are collected). The baseline characteristics can be determined based on one or more images collected during the baseline period. For example, the baseline characteristics can characterize tumor burden or tumor spread based on computed tomography (CT) images or other medical images. The baseline characteristics can include static or variable demographic attributes and / or comorbidities (e.g., whether the subject has any comorbidities, whether the subject has a particular type of comorbidity, and / or what type of comorbidity the subject has).

[0107] The term "cytokine fold change" as used herein refers to a value calculated using at least two cytokine levels. The at least two cytokine values ​​can include a baseline level of the cytokine and any other level of the cytokine (associated with the same subject). For example, the any other level of the cytokine can include another baseline level of the cytokine, an on-treatment level of the cytokine, or a level of the cytokine determined using a sample collected from the subject after an on-treatment period. The cytokine fold change can be based on, or defined as equal to, the logarithm of the baseline level of the cytokine minus the logarithm of the other level of the cytokine. The logarithm can be any positive base (e.g., logarithm base 2 or logarithm base 10).

[0108] The term "on-treatment cytokine fold change," as used herein, refers to a cytokine fold change where the other level of the cytokine is the on-treatment level of the cytokine.

[0109] The term "cytokine release syndrome risk score," as used herein, refers to a score (i.e., typically numerical, but may be categorical) generated using one or more baseline characteristics that represents a subject's predicted risk of experiencing cytokine release syndrome. The predicted risk can be the subject's risk of experiencing any grade, at least a threshold grade (e.g., grade 2 or higher), or a specific grade of cytokine release syndrome. The predicted risk can be the subject's risk of experiencing cytokine release syndrome within a predetermined time range, for example, a time range beginning from the start or completion of treatment administration and having a duration of a predefined number of hours or days (e.g., 1 day, 2 days, 3 days, 5 days, 7 days, or 14 days).

[0110] The term "cytokine release syndrome risk," as used herein, refers to a score (typically categorical, but may be numerical) derived from one or more cytokine levels, a treatment dose or exposure, and one or more risk scores.

[0111] The term "data record," as used herein, refers to a collection of data associated with one or more indices. The one or more indices can correspond (for example) to a given subject identification, a given time, and / or a given period of time. For example, a data record can include information about a particular subject collected at a particular time point. A data record can include any collection of data that is searched by providing a query that identifies a subject (and potentially one or more other constraints, e.g., time point). For example, a data record can include a file, a row in a table, a column in a table, an element in an array, a subset of recorded data where all the subsets are associated with one or more indices, etc.

[0112] The terms "therapeutic dose," "therapeutic dose," or "at least a partial dose of a treatment," as used herein, refer to the dose of a treatment or active ingredient of a treatment. The dose may be administered during one cycle (e.g., the first cycle) of treatment, or throughout the entire treatment.

[0113] III. Exemplary Networks Stratifying Subjects for Differential Monitoring to Predict Cytokine Release Syndrome Risk FIG. 1 illustrates an exemplary network 100 for stratifying subjects for differential monitoring or treatment that predicts the risk of one or more individual subjects experiencing a cytokine risk syndrome event according to some embodiments. Network 100 includes a cytokine release syndrome prediction system 105 that receives a request from a user device 110 and predicts the risk of a particular subject subsequently experiencing cytokine release syndrome (e.g., at least one of a particular severity and / or within a predefined time period). User device 110 may be operated by (for example) a doctor, nurse, medical technician, or clinical research coordinator. The request may identify the particular subject by name and / or one or more identifiers (e.g., social security number or unique identifier). The request may identify the disease for which the particular subject has been diagnosed and / or the treatment for which the particular subject has been prescribed and / or is receiving.

[0114] III.A. Exemplary Subject Properties Certain subjects may be diagnosed with cancer, such as non-Hodgkin's lymphoma.

[0115] III.AI Non-Hodgkin's Lymphoma Non-Hodgkin lymphoma (NHL) is a tissue and molecular malignancy that is the 10th most common cancer worldwide. More than 280,000 new cases of NHL are diagnosed annually worldwide. A particular subject may reside or have been born in any geographic region. While the incidence of NHL varies by geographic region, areas with high rates of NHL include North America, Europe, and Australia, as well as some countries in Africa and South America. According to the American Cancer Society, NHL is one of the most common cancers in the United States, accounting for approximately 4% of all cancers. In 2021, approximately 81,500 people in the United States will be diagnosed with NHL, and approximately 20,720 will die from the disease.

[0116] Non-Hodgkin's lymphoma can occur at any age, so the specific target can be of any age. In fact, it is one of the most common cancers among children, teenagers, and adolescents. Overall, men have approximately 1 in 41 chance of developing non-Hodgkin's lymphoma in their lifetime. For women, the risk is approximately 1 in 53. However, each person's risk can be affected by several risk factors. Many people with non-Hodgkin's lymphoma have no clear risk factors. Some people may have multiple risk factors and never develop non-Hodgkin's lymphoma. Some factors that can increase the risk of non-Hodgkin's lymphoma include older age (most people are over 60 years old at the time of diagnosis); use of immunosuppressant medications; infections, especially those with HIV, Epstein-Barr virus, or Helicobacter pylori; and exposure to certain chemicals, such as herbicides and pesticides.

[0117] Non-Hodgkin's lymphoma is the group name for all types of lymphoma except Hodgkin's lymphoma. Non-Hodgkin's lymphoma is a diverse group of blood cancers that all arise from lymphocytes (white blood cells), which are part of the immune system. These cells are found in the lymph nodes, spleen, thymus, bone marrow, and other parts of the body. Non-Hodgkin's lymphoma is found in organs such as the skin, stomach, and intestines, and commonly arises in the lymph nodes and lymphatic tissue, with some cases involving the bone marrow and blood.

[0118] Non-Hodgkin's lymphoma develops when cells in lymph nodes or other lymphoid structures undergo mutation. The disease can begin in B lymphocytes (B cells), which produce antibodies to fight infection; T lymphocytes (T cells), which have several functions, including assisting B lymphocytes in antibody production; or natural killer (NK) cells, which attack virus-infected or tumor cells. Approximately 85 to 90 percent of non-Hodgkin's lymphoma cases begin in the affected B cells. Mutated or abnormal lymphocytes undergo uncontrolled growth and produce many abnormal cells that accumulate and form tumors. Eventually, if non-Hodgkin's lymphoma is left untreated, the abnormal (i.e., cancerous) cells crowd out normal white blood cells, preventing the immune system from effectively protecting against infection.

[0119] Non-Hodgkin's lymphoma is often asymptomatic in its early stage.Therefore, regular health checkups are important for people with known risk factors for non-Hodgkin's lymphoma (for example, HIV infection, organ transplant, autoimmune disease or previous cancer treatment).Although these people often do not develop lymphoma, their doctors are typically on the lookout for possible symptoms and signs of lymphoma.One of the most common symptoms in subjects with non-Hodgkin's lymphoma is the swelling of one or more lymph nodes in the neck, axilla or groin.Sometimes, the disease starts from a site other than lymph nodes, such as bone, lung, gastrointestinal tract or skin.In these situations, subjects may experience symptoms related to a specific site. Signs and symptoms vary, but common symptoms also include unexplained fever, night sweats, persistent fatigue, loss of appetite, unexplained weight loss, cough or chest pain, abdominal pain, abdominal distension, itchy skin, enlarged spleen or liver, and rash or skin lumps. A particular subject may have experienced or be experiencing any one or more of the above symptoms.

[0120] III.A.1.a Diagnosis of non-Hodgkin's lymphoma Certain subjects may be diagnosed with non-Hodgkin's lymphoma after the diagnosis is suspected (e.g., based on symptoms). A diagnosis can facilitate the prescription of effective treatment to manage the disease.

[0121] In addition to a physical examination, blood and urine tests are often performed to rule out infection or other diseases. For example, imaging tests such as X-rays, CT scans, MRI scans, or positron emission tomography (PET) scans can be used to detect tumors throughout the body. Biopsies of involved lymph nodes or other tumor sites are used to confirm the diagnosis and subtype of non-Hodgkin's lymphoma. Further testing can include immunophenotyping or flow cytometry to identify the specific type of cancer cells in the sample, cytogenetic analysis to look for chromosomal changes or abnormalities in the cells, and / or gene expression profiling to identify genes differentially expressed in the subject's cancerous cells.

[0122] Certain subjects may be diagnosed with any kind of non-Hodgkin's lymphoma, for example, one or more of the more than 60 subtypes of non-Hodgkin's lymphoma identified by the World Health Organization (WHO).These subtypes are categorized by the characteristics of lymphoma cells, including their appearance, the presence of specific cell surface protein and their genetic profile.Considering that the signs, symptoms and treatment of non-Hodgkin's lymphoma can vary according to subtype and the progression rate of disease, accurate diagnosis and monitoring of disease progression are important for identifying the treatment that treats certain subtypes and identifying the current progression in certain subjects.

[0123] Pathologists often describe non-Hodgkin's lymphomas by grade. High-grade lymphomas grow quickly and have cells with a different morphology than normal cells. Low-grade lymphomas resemble normal cells more closely and grow more slowly. Intermediate-grade lymphomas fall somewhere in between. The behavior of these types is also described as indolent and aggressive.

[0124] When pathologists describe high-grade or intermediate-grade lymphoma, these types of lymphoma usually grow quickly in the body, and therefore these two types of lymphoma are considered aggressive lymphomas. Low-grade non-Hodgkin's lymphoma, on the other hand, grows slowly, and these lymphomas are called indolent lymphomas. Pathologists also classify non-Hodgkin's lymphomas as follicular or diffuse lymphomas. In follicular lymphoma, cancer cells arrange themselves in spherical clusters called follicles. In diffuse non-Hodgkin's lymphoma, the cells are spread out without clustering. Generally, low-grade or indolent non-Hodgkin's lymphoma appears follicular, while intermediate-grade or high-grade non-Hodgkin's lymphoma (aggressive non-Hodgkin's lymphoma) appears diffuse on biopsy slides.

[0125] Aggressive lymphoma accounts for approximately 60 percent of all non-Hodgkin's lymphoma cases, with diffuse large B-cell lymphoma (DLBCL) being the most common aggressive non-Hodgkin's lymphoma subtype. Indolent lymphomas tend to migrate slowly, grow slowly, and have few signs and symptoms when first diagnosed. Low-grade or indolent subtypes represent approximately 40 percent of all non-Hodgkin's lymphoma cases, with follicular lymphoma (FL) being the most common subtype of indolent non-Hodgkin's lymphoma. In some cases, indolent non-Hodgkin's lymphoma can transform into aggressive non-Hodgkin's lymphoma. If a subject's disease progression rate is between indolent and aggressive, the subject is considered to have intermediate-grade disease.

[0126] Table 4 provides several diagnoses of non-Hodgkin's lymphoma categorized by cell type (B-cell, T-cell, or NK-cell) and rate of progression (aggressive or indolent) based on the WHO classification. The listed percentages reflect the frequency of diagnosed cases of the most common non-Hodgkin's lymphoma subtypes. TIFF2025169248000008.tif255167TIFF2025169248000009.tif112170

[0127] Certain subjects may have been diagnosed with and / or have any of the lymphoma subtypes shown in Table 4. Certain subjects may have been diagnosed with non-Hodgkin's lymphoma after the diagnosis was suspected (e.g., based on symptoms). A diagnosis can facilitate the prescription of an effective treatment to manage the disease.

[0128] Diagnosis can also include grading or staging non-Hodgkin's lymphoma to determine the location of the cancer, the number of affected lymph nodes, and whether the disease has spread from its original site to other parts of the body, such as the liver or lungs. The majority of lymphomas are nodal lymphomas, meaning they originate in the lymph nodes. However, lymphomas can occur anywhere in the body. When lymphoma is primarily present in the nodes, it is referred to as nodal disease. Occasionally, most lymphomas may be located in organs that are not part of the lymphatic system, such as the stomach, skin, or brain. In these cases, the lymphoma is referred to as extranodal. Nodal and extranodal refer to the primary site of disease. Lymphoma develops in the lymph nodes and subsequently involves other structures. In these cases, it is referred to as nodal lymphoma with extranodal involvement.

[0129] A particular subject may be assigned a grade of non-Hodgkin's lymphoma based on the following definitions of the various stages. Stage I: Cancer is found in a single area or organ, usually one lymph node and surrounding area. Stage II: Cancer is found in two or more lymph node areas on the same side of the diaphragm, either above or below the diaphragm. Stage III: Cancer is found in lymph nodes on both sides of the diaphragm. If cancer is also outside the lymphatic system, it is called Stage IIIE. Stage III lymphoma that is also in the spleen is Stage IIIS. Stage IIIS and has spread outside the lymphatic system is Stage IIIE+S. Stage IV: Cancer has spread to one or more tissues or organs outside the lymphatic system, such as the liver, lungs, or bones, and may be found in lymph nodes near or far from these organs. Stage V: Death.

[0130] III.A.1.b. Treatment of Non-Hodgkin's Lymphoma Certain subjects may be prescribed to receive or may already be receiving a treatment that has the potential to trigger cytokine release syndrome. The treatment may include (for example) a treatment identified in Sections II.A.1.bi or III.A.1.b.ii. below. Certain subjects may further be prescribed to receive or may already be receiving a pretreatment before the treatment is administered. The compositions and / or active agents within the pretreatment may be the same as or different from those of the treatment.

[0131] Treatment for non-Hodgkin's lymphoma may depend on the subtype, progression rate, and / or stage of the disease. Lymphoma that does not cause signs or symptoms may not require treatment for several years. In some cases, if the initial cancer is small, the tumor can be removed by biopsy and no further treatment may be required. However, if the non-Hodgkin's lymphoma is aggressive or causes signs and symptoms, treatment is often prescribed.

[0132] Treatment for indolent non-Hodgkin's lymphoma can range from a wait-and-see approach to aggressive therapies.

[0133] III.A.1.bi Indolent subtype Certain subjects may be diagnosed with indolent subtypes of non-Hodgkin's lymphoma (e.g., follicular lymphoma). The management of indolent non-Hodgkin's lymphoma depends on prognostic factors, disease stage, age, and other medical conditions. Follicular lymphoma is the most common type of indolent non-Hodgkin's lymphoma and is a very slow-growing disease. Treatment for some subjects is not recommended for several years, while others may have widespread lymph node or organ involvement, and therefore immediate treatment may be recommended. In a small percentage of subjects, follicular lymphoma can transform into more aggressive disease.

[0134] Grade 1 or grade 2 follicular lymphoma may be treated with a wait-and-see approach, involving regular examinations and imaging tests or radiation therapy. Radiation therapy is most often used to treat early-stage non-Hodgkin's lymphoma, where the cancer is in only one part of the body. Treatment is usually given in short, daily sessions, usually for up to three weeks. In some cases, early, indolent non-Hodgkin's lymphoma may be treated with chemotherapy, chemotherapy combined with radiation therapy, or chemotherapy combined with immunotherapy, such as monoclonal antibody therapy. Rituximab (Rituxan®) (Genentech, San Francisco, CA) is a monoclonal antibody used to treat many different types of B-cell non-Hodgkin's lymphoma. Rituximab works by targeting CD20 on the surface of B cells and B-cell non-Hodgkin's lymphoma. When the antibody binds to CD20 on B cells, the subject's immune system is activated, destroying some lymphoma cells or making them more susceptible to destruction by chemotherapy. Rituximab can work well on its own, but studies have shown it works even better when added to chemotherapy for most types of B-cell non-Hodgkin's lymphoma. Rituximab is also given after remission of indolent lymphoma to increase the length of remission. Other monoclonal antibodies against CD20 that are FDA-approved for use in lymphoma include obinutuzumab (Gazyva®), ofatumumab (Arzerra®), rituximab-abbs (Truxima®), rituximab-arrx (Riabni®), and rituximab-pvvr (Ruxience®).

[0135] In addition to classifying lymphoma by grade, some subjects are also classified as having recurrent or recurrent follicular lymphoma. The Follicular Lymphoma International Prognostic Index (FLIPI) is a scoring system used to predict which subjects with follicular lymphoma may be at high risk of disease recurrence. One point is assigned to each of the following risk factors (known by the acronym NoLASH): Involved nodules - 5 or more Lactate dehydrogenase (LDH) levels – higher than the upper limit of normal Age over 60 Grade 3 or grade 4 disease Hemoglobin concentration less than 12g / dL

[0136] Risk is classified as follows: low risk: 0-1 point, intermediate risk: 2 points, and high risk: 3-5 points.

[0137] In subjects with grade 2 follicular lymphoma with large lymph nodes, grade 3 follicular lymphoma, or grade 4 follicular lymphoma or advanced recurrent follicular lymphoma, treatment is based on symptoms, the subject's age and health, the extent of the disease, and the patient's selection. Other treatment options include radiation therapy to the lymph nodes causing symptoms or to large localized masses, if present, or chemotherapy (as a single chemotherapy agent or in chemotherapy combinations) in conjunction with immunotherapy (rituximab).

[0138] Chemotherapeutic agents include, but are not limited to, alkylating agents (e.g., cyclophosphamide, chlorambucil, bendamustine, ifosfamide), platinum drugs (e.g., cisplatin, carboplatin, and oxaliplatin), purine analogs (e.g., cytarabine (ara-C), gemcitabine, methotrexate, pralatrexate), anthracyclines (e.g., doxorubicin or liposomal doxorubicin), vincristine, mitoxantrone, etoposide (VP-16), and bleomycin. Often, drugs from different groups are combined. One of the most common combinations is called CHOP, which includes cyclophosphamide, doxorubicin (also known as hydroxydaunorubicin), vincristine (Oncovin®), and prednisone. Another common combination, CVP, does not include doxorubicin. CHOP or CVP may be administered in combination with rituximab (CHOP-R or CVP-R).

[0139] Some subjects with grade 2 follicular lymphoma with large lymph nodes, grade 3 follicular lymphoma, or advanced-grade recurrent follicular lymphoma can be treated by stem cell transplantation (autologous and allogeneic) or by targeted therapy with kinase inhibitors (e.g., idelalisib (Zydelig®), copanlisib (Aliquopa®), and duvelisib (Copiktra™); lenalidomide (Revlimid®); or tazemetostat (Tazverik™)).

[0140] The treatment a subject is about to receive or has received may include a bispecific antibody. Bispecific antibodies may be provided or recommended as immunotherapeutic agents for subjects with refractory or relapsed follicular lymphoma. Bispecific T-cell-engaging antibodies (BiTEs) and knob-into-hole antibodies (KIHs) are exemplary antibody-based molecular modifications that bind to two different epitopes: one targeting malignant cells and the other targeting effector cells, typically T lymphocytes that mediate tumor cell destruction. Mosunetuzumab (Genentech), a T-cell-dependent bispecific, and glofitamab (Genentech), a KIH T-cell bispecific, both specifically bind to CD20 and CD3. Both are T-cell-engaging bispecific antibodies and can be used to treat multiple types of non-Hodgkin's lymphoma, including relapsed follicular lymphoma and diffuse large B-cell lymphoma.

[0141] Glofitamab (RO7082859, also known as RG6026) and mosunetuzumab are investigational full-length CD20- and CD3-targeted T cell-specific antibodies designed to redirect T cells to engage and eliminate malignant B cells (Bacac et al., Clin. Cancer Res. doi:10.1158 / 1078-0432.CCR-18-0455; Sun et al., Science Translational Medicine 7(287):287ra70; DOI:10.1126 / scitranslmed.aaa4802). These antibodies are designed to bind to CD20, a B cell surface protein expressed on the majority of B cell malignancies, while simultaneously binding to CD3, a component of the T cell receptor on the surface of T cells. T cell-directed therapies that induce potent immune stimulation carry the risk of cytokine release syndrome, potentially limiting their dosage and usefulness. Glofitamab and mosunetuzumab contain targeted mutations in the Fc binding site to reduce unwanted lysis of attracted T cells and off-target toxicity (e.g., cytokine release syndrome).

[0142] Follicular lymphoma has a small risk of transforming into an aggressive large B-cell lymphoma, e.g., diffuse large B-cell lymphoma. Certain subjects may have been diagnosed with aggressive large B-cell lymphoma (e.g., after a previous diagnosis of follicular lymphoma).

[0143] Subjects with transformed B-cell follicular lymphoma can benefit from rituximab therapy alone or in combination with chemotherapy. Other options include axicabtagene ciloleucel (Yescarta®) and tisagenleucel (Kymriah®), both of which are CAR T-cell therapies. In a typical CAR-T cell therapy protocol, T cells are collected from the subject's blood and modified to produce a chimeric antigen receptor (CAR) on their surface. These CAR-T cells are then reinfused into the subject, where the CAR binds to a specific antigen on the subject's tumor cells and kills them. See, e.g., Lulla et al., "The Use of Chimeric Antigen Receptor T Cells in Patients with Non-Hodgkin Lymphoma," Clin. Adv. Hematol. Oncol. 16(5):375-386 (2018). As described above, bispecific antibody therapy, e.g., glofitamab or mosunetuzumab, can also be used to treat diffuse large B-cell lymphoma.

[0144] Cutaneous T-cell lymphoma (CTCL) is a group of indolent non-Hodgkin's lymphomas that make up approximately 4% of non-Hodgkin's lymphoma cases. CTCL develops primarily in the skin and can grow to involve lymph nodes, blood, and other organs. Mycosis fungoides is the most common type of CTCL and is characterized by prominent skin involvement. When malignant lymphoma invades and accumulates in the blood, the disease is called Sézary syndrome. Treatment for CTCL depends on the nature of the skin lesions and whether disease is present in the lymph nodes.

[0145] Topical therapies are often used to treat skin lesions. These include applying drugs directly to the skin and exposing the skin lesions to light via ultraviolet or electron beam therapy. Combination therapy (PUVA), which uses ultraviolet light along with psoralens (drugs that are activated when exposed to light), is also used. Chemotherapy or extracorporeal photochemotherapy can be used when there is extensive involvement of lymph nodes and other areas. Photochemotherapy is a method in which white blood cells are removed by apheresis, treated with psoralens, exposed to ultraviolet A light, and then returned to the subject's bloodstream.

[0146] Administration of histone deacetylase (HDAC) inhibitors (romidepsin (Istodax®) given by IV infusion and vorinostat (Zolinza®) given orally), and monoclonal antibodies (mogamulizumab (Poteligeo®) given IV) are indicated for the treatment of adult subjects with either relapsed or refractory disease who have received previous systemic therapy.

[0147] III.A.1.b.ii. Invasive subtype The subject with aggressive non-Hodgkin's lymphoma is often treated with chemotherapy consisting of four or more drugs.In most cases, this is the above-mentioned CHOP or R-CHOP combination therapy.This intensive multi-drug chemotherapy can be very effective for aggressive lymphoma, and has achieved cure.For example, when the large mass of non-Hodgkin's lymphoma is found during the diagnosis and stage diagnosis process, chemotherapy can be supplemented in selected cases.

[0148] There are many types of aggressive non-Hodgkin's lymphoma, but diffuse large B-cell lymphoma is the most common subtype, accounting for approximately 31 percent of all non-Hodgkin's lymphoma cases in the United States. It grows rapidly in lymph nodes and frequently involves the spleen, liver, bone marrow, or other organs. The development of diffuse large B-cell lymphoma usually begins as lymphoma in the neck or abdomen and is characterized by large B-cell masses. Additionally, diffuse large B-cell lymphoma is often accompanied by B symptoms (fever, night sweats, and weight loss of more than 10 percent over six months). In some patients, diffuse large B-cell lymphoma may be the initial diagnosis. In other patients, indolent lymphomas, such as small lymphocytic lymphoma or follicular lymphoma, transform and become diffuse large B-cell lymphoma. Treatments include CHOP, dose-adjusted EPOCH-R (dose-adjusted etoposide, prednisone, vincristine (Oncovin®), cyclophosphamide, hydroxydoxorubicin (doxorubicin) plus rituximab, and rituximab and human hyaluronidase (Rituxan Hycela™). Bispecific antibody therapy, e.g., glofitamab or mosunetuzumab, can also be used to treat diffuse large B-cell lymphoma.

[0149] Some types of aggressive non-Hodgkin's lymphoma do not respond to standard doses of chemotherapy or have a high risk of recurrence. To treat some of these cases, physicians may consider administering high-dose chemotherapy followed by stem cell transplantation. In some cases, recurrent diffuse large B-cell lymphoma can be treated with CAR-T cell therapy, such as Yescarta®, Kymriah®, or Breyanzi (lisocabtagene ciloleucel). Axicabtagene ciloleucel (Yescarta®) is a CAR T-cell therapy approved for the treatment of subjects with diffuse large B-cell lymphoma who have received at least two previous types of therapy. Tisagenleucel (Kymriah®) is another CAR T-cell therapy approved for the treatment of refractory B-cell lymphomas, including diffuse large B-cell lymphoma, after two or more previous systemic therapies. Additional CAR T-cell therapies are in development and being studied in clinical trials. Lisocabtagene malarleucel (Breyanzi®) is a CAR T-cell therapy approved for adults with relapsed or refractory large B-cell lymphoma after two or more lines of systemic therapy. It can be used to treat diffuse large B-cell lymphoma not otherwise specified, high-grade B-cell lymphoma, primary mediastinal large B-cell lymphoma, and follicular lymphoma.

[0150] Polatuzumab vedotin-piiq (Polivy®) is a monoclonal antibody that targets CD79b. Polatuzumab is used in combination with bendamustine and rituximab to treat diffuse large B-cell lymphoma that has relapsed after at least two other therapies.

[0151] Tafasitamab-cxix (Monjuvi®) is a monoclonal antibody that targets the CD19 molecule. It can be used in combination with lenalidomide to treat relapsed or refractory diffuse large B-cell lymphoma in people who cannot undergo autologous bone marrow / stem cell transplantation.

[0152] Burkitt lymphoma is an aggressive B-cell subtype that grows and spreads very quickly. It can involve the jaw, facial bones, intestine, kidneys, ovaries, bone marrow, blood, central nervous system (CNS), and other organs. Burkitt lymphoma can spread to the brain and spinal cord (part of the CNS); therefore, treatments to prevent the spread of Burkitt lymphoma are frequently included in any treatment regimen. Physicians typically use highly aggressive chemotherapy to treat this subtype of non-Hodgkin's lymphoma. Commonly used regimens include CODOX-M / IVAC (cyclophosphamide, vincristine (Oncovin®), doxorubicin, and high-dose methotrexate) alternating with IVAC (ifosfamide, etoposide, and high-dose cytarabine); high-CVAD (hyperfractionated cyclophosphamide, vincristine, doxorubicin (Adriamycin®), and dexamethasone) alternating with methotrexate and cytarabine. In small studies, rituximab has been used in combination with high-CVAD, DA-EPOCH-R (dose-adjusted etoposide, prednisone, vincristine (Oncovin®), cyclophosphamide, doxorubicin + rituximab).

[0153] Mantle cell lymphoma (MCL), which can present as aggressive or indolent non-Hodgkin's lymphoma, originates from lymphocytes in the mantle layer of lymph nodes and represents approximately 6% of non-Hodgkin's lymphoma cases. It begins in the lymph nodes and spreads to the spleen, blood, bone marrow, and occasionally the esophagus, stomach, and intestines. Some patients do not show signs or symptoms of disease, so delayed treatment may be an option for them. However, most patients require treatment after diagnosis. Standard treatment is a combination chemotherapy regimen with or without autologous stem cell transplantation. A common treatment regimen is a form of CHOP in which bendamustine plus rituximab is used, with bortezomib substituted for vincristine. The following agents are indicated for relapsed and refractory MCL: acalabrutinib (Calquence®), given orally; bortezomib (Velcade®), given by IV or subcutaneous injection; ibrutinib (Imbruvica®), given orally; zanubrutinib (Brukinsa™), given orally; and lenalidomide (Revlimid®), given orally. Allogeneic transplantation with standard or reduced-intensity conditioning may be considered for subjects with relapsed and refractory MCL who have achieved remission after second-line therapy. Brexcabtagene outrucel (Tecartus®) is approved for adults with relapsed or refractory mantle cell lymphoma.

[0154] Peripheral T-cell lymphoma (PTCL) is a group of rare, aggressive non-Hodgkin's lymphomas that develop from mature T cells and natural killer (NK) cells. They account for approximately 10% of non-Hodgkin's lymphoma cases. PTCL, not otherwise specified (PTCL NOS), is the most common subtype of PTCL, accounting for approximately 30% of PTCL cases. For most subtypes of PTCL, initial treatment is typically a combination chemotherapy regimen, such as CHOP, CHOEP (etoposide, vincristine, doxorubicin, cyclophosphamide, and prednisone), or other multidrug regimens. Because most patients with PTCL relapse, some physicians have recommended high-dose chemotherapy followed by autologous stem cell transplantation. For CD30-expressing PTCL, brentuximab vedotin (Adecetris®) is approved for initial treatment in combination with cyclophosphamide, doxorubicin, and prednisone. Brentuximab vedotin is another type of monoclonal antibody called an antibody-drug conjugate. Antibody-drug conjugates bind to targets on cancer cells and then release small amounts of chemotherapy or other toxins directly into the tumor cells. Brentuximab vedotin combined with chemotherapy is approved for the treatment of adults with certain types of peripheral T-cell lymphoma, such as peripheral T-cell lymphoma not otherwise specified, as long as the tumor expresses the CD30 protein.

[0155] III.A.1.c. Side effects of treatment for non-Hodgkin's lymphoma Each type of treatment for non-Hodgkin's lymphoma has a different set of potential side effects, which can range from mild to severe. Common side effects associated with immunotherapy, chemotherapy, radiation therapy, or a combination of these include anemia (low red blood cells), thrombocytopenia (low platelets), neutropenia (low white blood cells), risk of infection, nausea, vomiting, intestinal problems, fatigue, brain fog, alopecia, peripheral neuropathy, dry skin, oral mucositis, sleep disorders, premature menopause, and decreased fertility. Immunotherapy in particular can trigger more severe side effects, such as pulmonary inflammation, diabetes, hypophysitis (inflammation of the pituitary gland), or cytokine release syndrome. Therefore, healthcare providers typically closely monitor any subject with non-Hodgkin's lymphoma who receives immunotherapy, especially bispecific T-cell-engaging antibody or CAR-T cell therapy, for cytokine release syndrome.

[0156] III.B. Primary Sources of Exemplary Baseline Characteristics The cytokine release syndrome prediction system 105 can request and / or retrieve information about a particular subject from one or more sources (e.g., one or more data stores or one or more computing systems). For example, the cytokine release syndrome prediction system 105 can retrieve a set of baseline characteristics of a subject from a baseline characteristic data store 115. (While FIG. 1 depicts the baseline characteristic data store 115 as a single data store, it is understood that baseline characteristics may instead be stored in or retrieved from multiple separate baseline characteristic data stores 115.) Each baseline characteristic includes a characteristic of the subject detected during a baseline period, a characteristic detected before the baseline period but assumed to be static, a characteristic that is static, or a characteristic that changes in a determined manner. The baseline characteristics may be determined based on data received from the healthcare provider system 120, the imaging system 125, or the testing system 130. Each baseline characteristic may be stored in a baseline characteristic data record, which may be stored in the baseline characteristic data store 115. Each baseline characteristic data record may be associated with a particular subject. In some instances, the baseline characteristic data record relates to a particular time point at which a particular subject is characterized by the baseline characteristic.

[0157] III.B.1. Healthcare Provider Systems The healthcare provider system 120 may include one or more computational systems that detect subject data representing one or more current characteristics of a particular subject, one or more medical assessments related to the particular subject, a description of one or more treatments previously prescribed or administered to the particular subject, and one or more medical events experienced by the particular subject.

[0158] The past or current characteristics of a particular subject can identify (for example) demographic characteristics (e.g., age, race, sex), regional characteristics (e.g., city of residence), occupational characteristics (e.g., identification of current or previous occupation), current or previous symptoms, medical history information (e.g., one or more diagnoses, previous adverse events, comorbid conditions reported by the particular subject, and / or family history related to one or more disease types. The past or current medical assessment of a particular subject can identify (for example) existing or new diagnoses (e.g., identification of disease, disease state, disease subtype), results of in-clinic evaluations (e.g., how well a given task was performed, any medical abnormalities observed, vital signs, etc.). The medical assessment may include (for example) a physician or nurse associated with the same or a different healthcare provider system 120. The details of previous treatment may include identifying medications previously administered to the particular subject, an indication of when the medications were administered (e.g., identifying one or more days or one or more years), one or more doses of the medication, a route of administration, and / or a treatment schedule (e.g., identifying how many doses were received and the relative timing of the doses). Medically related events experienced by the particular subject may include symptoms, adverse events, surgical procedures, hospitalizations.

[0159] Some or all of the subject data may be detected in the healthcare provider system 120 by processing input received through an input component of the healthcare provider system 120. The input component may include a keyboard, camera, scanner, microphone, mouse, trackpad, etc. The input may correspond to (for example) a medical note from a healthcare provider, a form completed by a particular subject, a prescription order from a healthcare provider, etc. Additionally or alternatively, some or all of the subject data may be retrieved from an electronic health record.

[0160] Subject data detected during the baseline period, detected before the baseline period but presumed to be static, static, or changed in a determined manner are baseline characteristics and may be stored in baseline characteristic data store 115. Baseline characteristics may be stored in association with a particular subject identifier.

[0161] The healthcare provider system 120 may further identify one or more specifications of a treatment currently prescribed or administered to a particular subject. The one or more treatment specifications may identify a medication, a dosage, a route of administration, and / or a treatment schedule. The one or more treatment specifications may identify a prior treatment agent, a dosage of the prior treatment, or the timing of the prior treatment (relative to the initial treatment dosage). The prior treatment agent may include an agent that is not a CD3 bispecific antibody. For example, the prior treatment agent may include obinutuzumab.

[0162] If multiple different doses of a medication (e.g., a CD3 bispecific antibody) are administered over the course of treatment, the treatment schedule can specify the relative times at which the different doses are administered. For example, a set of treatment specifications can specify that 10 mg of glofitamab is administered on the first treatment day, and 16 mg of glofitamab is administered 27 days after the first treatment day. If multiple different medications are administered over the course of treatment, the treatment schedule can specify the relative times at which the different medications are administered. For example, a set of treatment specifications can specify that 10 mg of glofitamab is administered on the first treatment day, and that a combination of 10 mg of glofitamab and 1000 mg of obinutuzumab is administered 16 days and 35 days after the first treatment day, respectively. One or more treatment doses can be stored in the treatment dosage data store 135 in association with a particular subject identifier. In some examples, the treatment specifications (stored in the treatment dosage data store 135) can further specify the time at which the treatment (or corresponding prior treatment) is initiated.

[0163] III.B.2 Imaging Systems Imaging system 125 includes one or more computing systems that collect and / or assess medical images. The medical images may be (for example) computed tomography (CT) images, X-rays, magnetic resonance imaging (MRI) scans, positron emission tomography (PET) scans, or digital pathology images. Thus, medical images may be collected using (for example) a CT machine, an X-ray machine, an MRI machine, a PET machine, or a microscope. In some examples, imaging system 125 includes equipment or devices that collect medical images. In some examples, medical images are collected using a remote imaging equipment or device and transferred to imaging system 125 (e.g., in response to imaging system 125 sending a request for the image).

[0164] Medical images (e.g., CT images, X-rays, MRI scans, or PET scans) may be collected by imaging a portion of a particular subject, potentially after a contrast agent has been administered to the subject. The medical images may be two-dimensional or three-dimensional images. In some examples, multiple two-dimensional images are collected. The medical images may be processed using a computer vision algorithm (e.g., executed on the imaging system) or based on annotations by an annotator (e.g., detected by imaging system 125) to identify one or more tumor annotations. Each tumor annotation may identify a portion of the medical image that depicts a portion of a tumor. For example, imaging system 125 may provide an interface for depicting the medical image, and imaging system 125 may receive annotation data indicating which portions of the medical image have been identified by the annotator (via input) as tumor boundaries in the image displayed by imaging system 125. Imaging system 125 may identify one or more measurements of each identified tumor. Spatial measurements can include (for example) tumor volume, tumor area, the length along the longest axis of the tumor (referred to as the longest diameter), and / or the aspect ratio of the tumor.

[0165] The imaging system 125 can further automatically detect (e.g., using computer vision algorithms) and classify each depiction of an organ, or can receive input from an annotator who identifies the boundaries of each depicted organ. The imaging system 125 can then use the given tumor and organ annotations to detect what type of tumor is located.

[0166] The imaging system 125 can generate tumor characterization statistics, such as total tumor volume detected, total tumor volume (sum of tumors detected), average longest diameter of tumors, number of early stage types where at least one tumor was detected, tumor burden, and / or sum of products of overall longest tumor diameter in the tumor.

[0167] A tumor characterization statistic can be characterized as a baseline characteristic (stored in baseline characteristic data store 115) if the medical images were collected during a baseline period. In some examples, the baseline characteristic (then stored in baseline characteristic data store 115) is defined based on a numerical tumor characterization statistic. For example, the numerical tumor characterization statistic can be compared to one or more thresholds to generate a binary indicator that determines whether the statistic exceeds a single threshold. As another example, the numerical tumor characterization statistic can be compared to multiple thresholds to identify one or more ranges that include the statistic, and a categorical indicator can identify a category.

[0168] In some instances, medical images are used to detect lymph node size. Enlarged lymph nodes may indicate lymphoma. Thus, baseline statistics may be defined as the estimated volume, cross-sectional area, or longest diameter of a lymph node.

[0169] Alternatively, medical images (e.g., digital pathology images) may be collected by collecting a sample (e.g., a biopsy, tissue sample, and / or blood sample) from a particular subject, fixing the sample, potentially sectioning the sample, or dropping a liquid sample onto a slide, and staining the sample section. The imaging system 125 can then image the stained section, or a remote imaging system can image the stained section, and the imaging system 125 can access the image.

[0170] The imaging system 125 can process the images to detect the presence, location, and / or density of any biological object of a predetermined type (e.g., a particular cell type). For example, the imaging system 125 can determine the point location (or area or volume) of each tumor cell and / or each immune cell. The imaging system 125 can define baseline characteristics (and store the baseline characteristics in the baseline characteristic data store 115) to indicate the binary presence of any tumor cells, the density of tumor cells, the density of immune cells, etc.

[0171] III.B.3. Inspection Systems The testing system 130 can process the biological sample to generate one or more test results. Each test result can identify the presence, count, concentration, and / or type of one or more respective biological structures. The biological sample may be different from any sample used to collect the medical image (i.e., processed by the imaging system 125). The biological sample may include (for example) a blood sample, a urine sample, a sweat sample, or a tissue sample.

[0172] The biological structures (measured by the inspection system 130) may be cell types, cell fragments, or proteins. For example, the biological structures may include white blood cells, monocytes, platelets, hemoglobin, fibrinogen, C-reactive protein (CRP), aspartate aminotransferase (AST), and / or alkaline phosphatase (ALP). High white blood cell counts, high monocyte counts, and low platelet counts may be consistent with various types of cancer (e.g., lymphoma). Low hemoglobin levels may be consistent with certain types of cancer (e.g., non-Hodgkin's lymphoma) or advanced stages of certain types of cancer (e.g., stage III or stage IV Hodgkin's lymphoma). High levels of fibrinogen and / or C-reactive protein may indicate inflammation. High levels of AST and / or ALP may indicate that cancer (e.g., non-Hodgkin's lymphoma) has spread to the liver.

[0173] If the biological sample was collected during a baseline period, the test results may be characterized as a baseline characteristic and stored in the baseline characteristic data store 115 .

[0174] The testing system 130 includes a cytokine detection subsystem 140 that monitors the respective levels (e.g., concentrations) of one or more cytokines in a biological sample (or different biological samples). The testing system 130 stores each cytokine level in the raw cytokine level data store 145 in association with a subject identifier, a measurement time, and / or a cytokine identifier. For example, a single cytokine level data record may be generated corresponding to an individual measurement time and an individual subject and may include each level of a cytokine detected in a sample collected from the subject at that measurement time. As another example, a single cytokine level data record may be generated corresponding to an individual subject and may include each level of a cytokine detected in any sample collected from that subject. The single cytokine level data record may associate each level of a cytokine with a measurement time indicating when the sample was collected from the subject used to measure the cytokine level. Each measurement time may be an absolute time or a time relative to the start of a predetermined treatment period.

[0175] As an example, the cytokine detection subsystem 140 may detect the level of each of one or more of the following cytokines in a blood sample: IL-1β, IL-2, IL-6, IL-8, MIP1b, MCP1, IL-10, IFN-γ, TGF-β, and TNF-α.

[0176] III.C. Exemplary Cytokine Release Syndrome Prediction System The cytokine release syndrome prediction system 105 can process one or more baseline characteristics (from the baseline characteristics data store 115), one or more treatment doses (from the treatment dose data store 135), and one or more cytokine levels (from the raw cytokine level data store 145) using a machine learning model to predict a particular subject's risk of experiencing cytokine release syndrome.

[0177] III.C.1. Cytokine release syndrome Cytokine release syndrome (CRS) is an uncontrolled inflammatory response that can be triggered when treating non-Hodgkin's lymphoma, particularly with therapeutic antibodies, CAR-T cell therapy, or allogeneic transplantation. It can occur after the infusion of several antibody-based therapies, such as glofitamab, rituximab, obinutuzumab, alemtuzumab, brentuximab, dacetuzumab, or nivolumab. It has also been observed after the administration of non-antibody-based cancer drugs, such as oxaliplatin and lenalidomide. CCS is one of the most frequent and severe adverse effects following the administration of T-cell-engaged immunotherapeutic agents. T-cell-engaged immunotherapies include bispecific antibody constructs and chimeric antigen receptor (CAR) T-cell therapy, both of which have shown therapeutic efficacy in several hematological malignancies, including diffuse large B-cell lymphoma. Cytokine release syndrome can occur over several days or weeks after treatment, or immediately after treatment as immediate-onset cytokine release syndrome.Normally, cytokine signaling leads to a rapid and powerful immune response.This response usually reaches equilibrium and dissipates when malignant or infected cells are eliminated.However, in some cases, this positive feedback loop, in which activated cells continue to release more cytokines and activate more cells for cytokine release, becomes uncontrollable, resulting in cytokine syndrome, which produces excessively high levels of pro-inflammatory cytokines.

[0178] Cytokine release syndrome often manifests as a combination of fever, hypoxia, hypotension, and capillary leak syndrome, with or without organ findings. Cytokine release syndrome is a condition that occurs after immunotherapy, causing rapid release of large amounts of cytokines from immune cells, such as T cells, into the blood.

[0179] Cytokines are a large group of proteins, peptides, and glycoproteins secreted by specific cells of the immune system. Cytokines are signaling molecules that are transiently produced after cell activation and help mediate and regulate immunity, inflammation, and hematopoiesis. These molecules act as regulators that modulate the function of individual cells. Cytokines can act locally as autocrine, paracrine, or endocrine response modifiers, and these effects are exerted through specific cell surface receptors on their target cells. As used herein, "autocrine" or "autocrine action" means that a cytokine exerts its effect by binding to a receptor on the membrane of the same cell from which it is secreted. "Paracrine" or "paracrine action" means that a cytokine binds to a receptor on a target cell in close proximity to the cell that produced the cytokine. "Endocrine" or "endocrine action" means that a cytokine travels through the circulation and acts on target cells in parts of the body throughout.

[0180] Elevated levels of cytokines, such as one or more cytokines selected from the group consisting of IL-1β, IL-2, IL-6, IL-8, MIP1b, MCP1, IL-10, IFN-γ, TGF-β, and TNF-α, are often associated with cytokine release syndrome. Table 5 below lists the main cytokines associated with cytokine release syndrome and their effects (Yildizahan and Kaynar, Journal of Oncological Sciences, 4(3):134-141(2018)). TIFF2025169248000010.tif255167TIFF2025169248000011.tif49170

[0181] III.C.1.a. Mechanism Cytokine release syndrome is usually caused by the on-target effect induced by the binding of bispecific antibodies or CAR T cell receptors to their antigens, followed by the activation of bystander immune cells and non-immune cells, such as endothelial cells. Activation of bystander cells results in the massive release of a series of cytokines. Depending on the characteristics of the host, tumor, and therapeutic agent, the administration of T cell-engaging therapy can trigger an inflammatory circuit that overwhelms counterregulatory homeostatic mechanisms, resulting in cytokine syndrome that can have adverse effects on the subject.

[0182] Upon administration of immunotherapy, T cell activation or immune cell lysis induces the release of interferon gamma (IFN-γ) or tumor necrosis factor alpha (TNF-α). TNF-α induces flu-like symptoms, including fever, general malaise, and fatigue, as well as watery diarrhea, vascular leakage, cardiomyopathy, lung injury, and the synthesis of acute-phase proteins (e.g., C-anticoagulant protein). IFN-γ causes fever, chills, headache, dizziness, and fatigue. Secreted IFN-γ induces the activation of macrophages, dendritic cells, other immune cells, and endothelial cells. Activated macrophages produce excessive amounts of pro-inflammatory cytokines, such as IL-6, TNF-α, and IL-10. Importantly, macrophages and endothelial cells produce large amounts of interleukin-6 (IL-6), which activates T cells and other immune cells, leading to cytokine syndrome.

[0183] Interleukin-6 (IL-6) is a pleiotropic cytokine with both anti- and pro-inflammatory properties that plays a central role in host defense through a wide range of immune and hematopoietic activities and its ability to induce acute-phase proteins. IL-6 is a central mediator of cytokine release syndrome toxicity. IL-6 signaling requires binding to the ubiquitously expressed cell-associated gp130 (CD130) and the IL-6 receptor (IL-6R) (CD126). IL-6R is expressed on macrophages, neutrophils, hepatocytes, and some T cells and mediates classical signaling, which predominates when IL-6 levels are low. However, when IL-6 levels are elevated, soluble IL-6R can also initiate trans-signaling, which occurs in a wider range of cells. The anti-inflammatory properties of IL-6m are mediated by classical signaling, whereas pro-inflammatory responses appear to occur as a result of trans-signaling. High levels of IL-6 are present in the context of cytokine release syndrome and appear to mediate pro-inflammatory IL-6-mediated signaling.

[0184] III.C.2. Pretreatment: Generating Cytokine Fold Changes The cytokine release syndrome prediction system 105 includes a cytokine regulator 150 that aligns cytokine levels with standard time points and generates cytokine fold changes. For example, for one or more subjects, the cytokine regulator 150 can retrieve (e.g., from the treatment dosage data store 135) the time at which treatment or prior treatment was initiated for the subject. The plurality of subjects may include a set of subjects associated with the data used to train the machine learning model, or may also include a particular subject.

[0185] For each of the plurality of subjects, the cytokine regulator 150 can define a baseline period using the time a treatment or prior treatment was initiated. For example, the baseline period may define the time a treatment or prior treatment was initiated as having ended (or a predefined time prior to such initiation, e.g., one day prior to treatment initiation). In some examples, the baseline period has a predefined duration, and the cytokine regulator 150 can identify a time from the beginning of the baseline period to an end time of the baseline period based on the duration. In some examples, the baseline period is defined based solely on the end time, such that all time preceding the end time is within the baseline period.

[0186] For each of the plurality of subjects, cytokine regulator 150 retrieves (e.g., from raw cytokine level data store 145) a measurement time associated with each cytokine level stored in association with the subject's identifier. Cytokine regulator 150 can use the baseline period and the measurement times associated with the cytokine levels to detect which cytokine levels are associated with measurement times within the baseline period. Cytokine regulator 150 can characterize each cytokine level associated with a measurement time within the baseline period as a baseline cytokine level 155.

[0187] For each of the subject sets (related to the data used for training), and potentially for a particular subject, the cytokine regulator 150 can further retrieve (e.g., from the treatment dosage data store 135) the time at which the treatment or treatment cycle was completed. In some examples, the duration of the treatment or treatment cycle is known (e.g., with a predetermined degree of confidence and a predetermined degree of accuracy) or is estimated such that the time at which the treatment or treatment cycle is or will be completed can be estimated.

[0188] For each of the subject sets, and potentially for a particular subject, the cytokine regulator 150 can define an on-treatment period beginning (for example) at the time the treatment began or the cycle began (e.g., as identified in the data retrieved from the treatment dosage data store 135). It is understood that the start of the on-treatment period may differ from the end of the baseline period. In some examples, the treatment specification identifies the time the treatment administration ended or the administration of a treatment cycle ended, and the cytokine regulator 150 can define the end of the on-treatment period to be completed at that time. In some examples, the duration of the treatment or treatment cycle is known (e.g., with a predetermined degree of confidence and a predetermined degree of accuracy), and the cytokine regulator can define the end of the on-treatment period based on the duration and the start of the on-treatment period.

[0189] The cytokine regulator 150 can use the on-treatment period and the measurement times associated with the cytokine levels to determine which cytokine levels are associated with the measurement times within the on-treatment period. The cytokine regulator 150 can characterize each cytokine level associated with a measurement time within the on-treatment period as an on-treatment cytokine level 160. For each of the subject sets, the cytokine regulator 150 can further characterize each cytokine level associated with a measurement time subsequent to the on-treatment period as a post-treatment cytokine level.

[0190] The cytokine regulator 150 can generate at least one cytokine fold change using one or more baseline cytokine levels 155 and one or more on-treatment cytokine levels 160. A cytokine fold change 170 can be determined by subtracting a baseline level term from another term. For a given subject, the baseline level term is defined to be or is based on at least one baseline cytokine level 155, and the other term is defined to be or is based on at least one on-treatment cytokine level 160 or at least one other baseline cytokine level 155.

[0191] A baseline cytokine level can be defined as a cytokine level associated with a measurement time that falls within a specified portion of the baseline period. For example, a baseline cytokine level can include a baseline cytokine level 155 measured 6-8 days prior to prior treatment. In some examples, a cytokine fold change 170 is defined for each cytokine level measured in a given subject based on the baseline cytokine level and the raw cytokine level.

[0192] For each term in the baseline level term and other terms, the term can be determined using a logarithmic function. However, the logarithm of zero is undefined. Therefore, the logarithm of the sum of the corresponding cytokine level and a predefined positive value can be calculated instead of calculating the logarithm of the corresponding cytokine level. The predefined value can be, for example, a fraction, 1, 2, etc.

[0193] III.C.3. Machine Learning Model Training The cytokine release syndrome prediction system 105 includes a model training subsystem 175 that trains one or more machine learning models to predict cytokine release syndrome risk 180 based on one or more baseline characteristics 115 and treatment dose 135. The model training subsystem may additionally or alternatively predict cytokine release syndrome risk 180 based on cytokine fold change 170. The cytokine release syndrome prediction system 105 may further use one or more baseline cytokine levels 155 to predict cytokine release syndrome risk 180. The machine learning models may include (for example), a random forest model, a regression model (e.g., a linear logistic regression model), a decision tree model, and / or a neural network model.

[0194] The training data used by model training subsystem 175 to train the predictive model can be associated with a set of subjects and can include treatment dosages and an indication of whether (and if) each subject will experience cytokine release syndrome, and if so, an indication of the severity of the event. Criteria for determining the severity of cytokine release syndrome can be included in Section III.A.1.a.

[0195] The model training subsystem 175 can obtain cytokine release syndrome information from a cytokine release syndrome (CRS) report data store 182. The CRS report data store 182 can include multiple CRS report records, each identifying a subject, a cytokine release syndrome severity for each CRS event, and a time to the cytokine release syndrome. Each CRS report record can be generated based on data or input received from a healthcare provider system 120 associated with a subject (e.g., by a healthcare provider corresponding to the healthcare delivery system 120 administering a treatment to the subject, diagnosing the cytokine release syndrome, and / or treating the cytokine release syndrome). Thus, the model training subsystem 175 can query the CRS report data store 182 to determine whether cytokine release syndrome (e.g., of at least a threshold severity) has been observed in each of a set of subjects.

[0196] III.C.3.a. Train a decision tree model to convert baseline parameter, dose, and cytokine level inputs into predicted cytokine release syndrome risk In some examples, the model training subsystem 175 can define training data elements that include, in each of a set of subjects, an indication of whether cytokine release syndrome (e.g., of at least a threshold grade) was observed, baseline parameters, a treatment dose, and one or more cytokine fold changes 170 (e.g., corresponding to an on-treatment period). The training data can be composed of training data elements corresponding to the set of subjects. The indication of whether cytokine release syndrome (e.g., of at least a threshold grade) was observed can be defined as a label for the training data element.

[0197] In some examples, the model training subsystem 175 can train a model to learn a set of model parameters that facilitate converting a baseline parameter (a single baseline characteristic 115 or a baseline cytokine release syndrome risk score 184) to a cytokine release syndrome risk 180. In some examples, the model training subsystem can train a model to learn a set of model parameters that facilitate converting a baseline parameter, a treatment dose 135, and / or a cytokine fold change 170 to a cytokine release syndrome risk 180. The learning model can include (for example) a decision tree model 183, and the model parameters can include a set of decision tree thresholds.

[0198] The decision tree thresholds of the predictive model can include a dosage threshold, a baseline cytokine release syndrome risk score (CRSRS) threshold, and one or more cytokine level thresholds. Thus, the decision tree model 183 can determine whether a treatment dosage exceeds a dosage threshold, whether a baseline cytokine release syndrome risk score exceeds a CRSRS threshold, and / or whether a cytokine fold change 170 exceeds each of one or more cytokine level thresholds. For example, a different (e.g., lower) cytokine level threshold can be used when a dosage exceeds a threshold compared to when the dosage does not exceed a dosage threshold. In some examples, each training data element includes multiple cytokine fold changes 170. The decision tree model 183 can be configured to identify each "on-treatment" cytokine fold change 170 corresponding to an on-treatment period and to use the maximum on-treatment cytokine fold change for threshold comparison.

[0199] III.C.3.b. Train a feature selection model and / or risk score generation model to convert baseline characteristics into risk scores In some examples, at least some of the set of baseline characteristics (e.g., retrieved from baseline characteristic data store 115) can be used (e.g., together with treatment dose and cytokine fold change 170) to predict cytokine release syndrome risk 180. For example, risk score generation model 184 can convert at least some of the set of baseline characteristics into a cytokine release syndrome risk score, which can then be used as a single predictor by decision tree model 183 to predict cytokine syndrome risk 180.

[0200] The model training subsystem 175 can perform feature selection to identify which baseline characteristics are used by the risk score generation model 184 to predict cytokine release syndrome risk 180. In some examples, features are selected using a feature selection model 185, which can be configured to perform univariate analysis on each baseline characteristic. The univariate analysis can output a significance value that can indicate whether there is a significant relationship between the characteristics and an indication of whether cytokine release syndrome (e.g., of at least a threshold severity, such as an event of at least grade 1 severity or an event of at least grade 2 severity) will occur. An initial subset of baseline characteristics is defined as baseline characteristics with a p-value below a predefined threshold (e.g., less than 0.1 or less than 0.3). This subset is further refined by 185 by applying multivariate techniques such as floating forward / backward multiple regression or random forest analysis.

[0201] In some examples, the feature selection model 185 can perform kappa-fold cross-validation. Cross-validation can be performed multiple times, with each cross-validation run associated with a subset of baseline characteristics. In each cross-validation run and each fold, the training data can be divided into a training portion and a test portion used to assess the performance of the model. The training data used for feature selection can include a set of baseline statistics and an indication of whether cytokine release syndrome (e.g., of at least a threshold grade) was observed in each of the subject sets.

[0202] The stratification factors may be defined to include disease history and treatment dosage. Thus, the training and test datasets used for feature selection may be defined to include a distribution of disease history in the training data that is approximately the same as the test dataset, and a distribution of treatment dosage in the training data that is approximately the same as the test dataset. The training dataset may be used to train a feature selection model, and the training dataset may be used to determine performance measures. Performance statistics may be generated for each baseline feature subset based on the performance statistics for the corresponding split. The reduced feature set may be defined to be the subset associated with the best performance and stability statistics.

[0203] The model training subsystem 175 can then train the risk score generation model 184 to learn a set of model parameters (e.g., a set of weights) to convert values ​​of the reduced feature set into a cytokine release syndrome risk score. The risk score generation model 184 may include a regression model or a weighted sum of baseline parameter values. In some examples, the training set of model parameters may include learning a parameter value for each baseline characteristic represented in the reduced feature set (e.g., corresponding to a subset of the set of baseline characteristics). For example, the parameters may include a weight for each baseline characteristic represented in the reduced feature set.

[0204] The model parameters can be learned by (for example) fitting one or more functions to the training data. For example, the model training subsystem can learn weights for each baseline characteristic represented in the reduced feature set, and the weights can be identified as those with the greatest classification accuracy and stability in the training data. Each weight can be derived primarily from the log(cytokine release syndrome odds ratio) in a dose-adjusted logistic regression model that models the cytokine release syndrome odds ratio for the involved parameter value and drug dose (exposure). The weights can be further adjusted by including information about the stability of variables from random forests and floating feature selection experiments.

[0205] Additionally or alternatively, the model training subsystem 175 can use a loss function to learn the parameters of the risk score generation model 184 by iteratively using machine learning and a loss function. For example, the machine learning model can generate one or more predicted outputs (e.g., predict whether cytokine release syndrome will occur) using parameter values ​​of a reduced feature set, compare the predicted outputs to labels in the training data, calculate a loss based on the comparison and the loss function, and adjust the parameters of the risk score generation model 184 based on the loss. This process can be repeated for multiple training cycles. After the loss or a moving average of the loss falls within a predefined loss threshold and / or after several training cycles have crossed a predefined training cycle threshold, the model training subsystem 175 can fix the parameter set.

[0206] The model training subsystem 175 can be configured to train the risk score generation model 184, adjust a trained version of the risk score generation model 184, or configure a post-processing algorithm to convert a selected subset of model parameters, as well as one or more treatment specifications (e.g., treatment dose), into a predicted risk that a given subject will experience cytokine release syndrome (e.g., of at least threshold severity). For example, the parameters of the risk score generation model 184 can be learned by providing input data including values ​​of a subset of baseline characteristics to the risk score generation model 184, determining whether or to what extent a cytokine release syndrome risk score output by the model accurately predicts that cytokine release syndrome (e.g., of at least threshold severity) will be observed, and calculating a loss based on whether or to what extent the cytokine release syndrome risk score output is accurate. The set of parameters can be used regardless of the treatment dose administered to the subject, or a different set of parameters can be learned for each of multiple treatment doses. As another example, the risk score generation model 184 can be initially trained to learn a set of parameters that predicts a cytokine release syndrome risk score based on baseline characteristics. The risk score generation model 184 can then use the set of parameters, baseline characteristics and treatment dosage to determine a given subject's risk of cytokine release syndrome.

[0207] III.C.3.c. Train a decision tree to convert risk score output and cytokine level input into predicted cytokine release syndrome risk In some examples, the decision tree model 183 is configured to receive as input one or more cytokine levels and a risk score. The risk score may be a cytokine release syndrome risk score (determined using the baseline characteristics of 180). The decision tree model 183 may further receive a treatment dosage as an additional input variable.

[0208] The model training subsystem 175 can train the decision tree model 183 to learn a threshold set that can include a risk score threshold and one or more cytokine level thresholds. Thus, the decision tree model 183 can determine whether the risk score 180 exceeds the risk score threshold and / or whether the cytokine fold change 170 exceeds each of one or more cytokine level thresholds. In some examples (e.g., when the decision tree model 183 receives the cytokine release syndrome risk score 180 as input), the threshold set can further include a dosage threshold, and the decision tree model 183 can determine whether the dosage exceeds the dosage threshold.

[0209] The one or more cytokine level thresholds can include multiple thresholds. For example, a different (e.g., lower) cytokine level threshold may be used when the cytokine release syndrome risk score exceeds a risk score threshold (compared to when the risk score does not exceed the risk score threshold) and / or when the dosage exceeds a dosage threshold (compared to when the dosage does not exceed the dosage threshold). In some examples, each training data element includes multiple cytokine fold changes 170. The decision tree model 183 can be configured to identify each "on-treatment" cytokine fold change 170 corresponding to an on-treatment period and to use the maximum on-treatment cytokine fold change for threshold comparison.

[0210] III.C.4. Predicting Cytokine Release Syndrome Risk The cytokine release syndrome prediction system 105 includes a CRS risk detector 190 that uses one or more trained machine learning models to convert a subject-specific input dataset corresponding to a particular subject into a particular cytokine release syndrome risk 180. The subject-specific dataset can include one or more baseline characteristics (e.g., tumor burden, tumor spread, presence or amount of malignant cells in peripheral blood, presence or amount of malignant cells in bone marrow, demographic attributes, age, baseline LDH level, baseline WBC level, and / or comorbidities). The subject-specific dataset can further include one or more cytokine fold changes 170 associated with the particular subject and a treatment dosage associated with the particular subject (e.g., indicating the amount of treatment administered to the subject, the amount of treatment prescribed to the subject, or the amount of treatment being considered for the subject). Thus, the subject-specific dataset can include (1) one or more baseline characteristics, (2) one or more baseline characteristics and a treatment dosage, or (3) one or more baseline characteristics, a treatment dosage, and one or more cytokine fold changes 170.

[0211] The CRS risk detector 190 can combine one or more cytokine fold changes 170 associated with a particular subject and one or more other subject-specific values ​​with a decision tree model 183 to generate a cytokine release syndrome risk 180 for a particular subject (representing a predicted risk of the subject experiencing cytokine release syndrome after receiving a dose of a treatment). The one or more other subject-specific values ​​can include a treatment associated with the particular subject and / or risk score. The predicted risk can include (for example) a categorical value (e.g., representing very low risk, low risk, moderate risk, high risk, or very high risk) or a binary value (e.g., representing high risk or not high risk).

[0212] The CRS risk detector 190 can access the inpatient monitoring conditions 193 and evaluate the condition using the cytokine release syndrome risk 180 generated for a particular subject. The inpatient monitoring conditions 193 can be configured to be met when the cytokine release syndrome risk is a particular value (e.g., high risk) or exceeds a particular threshold (e.g., moderate or high risk).

[0213] The CRS risk detector 190 can select or generate an output to be directed to the user device 110 by the cytokine release syndrome prediction system 105 based on the condition evaluation. For example, a "consider inpatient monitoring" or "inpatient monitoring recommended" output may be selected if the inpatient monitoring condition 193 is met, and a "consider outpatient monitoring" or "outpatient monitoring recommended" output may be selected if the inpatient monitoring condition is met. The output can further include (for example) one or more cytokine fold changes (e.g., used to generate the cytokine release syndrome risk 180), one or more numerical risk scores, one or more raw cytokine levels, one or more baseline characteristics, and / or dosages associated with a particular subject.

[0214] The user device 110 can present the output to the user, who (or another entity) can decide whether to accept the recommendation and expedite the hospitalization or outpatient monitoring condition accordingly.

[0215] III.D. Exemplary Inpatient or Outpatient Monitoring If a particular subject is monitored as an outpatient, the particular subject may be advised (e.g., by a healthcare provider) to monitor for any one, more, or all of the symptoms identified in Section III.D.1. and to alert a healthcare provider or to go to a healthcare facility if any symptoms occur.

[0216] If a particular subject is monitored as an inpatient, a health care provider (e.g., a doctor and / or nurse) may monitor for any one, more, or all of the symptoms identified in Section III.D.1., and may be requested to monitor a particular subject for any such symptoms. Additionally, if a particular subject is monitored as an inpatient, one or more laboratory tests may be performed periodically (e.g., to detect cytokine levels) to rapidly detect any cytokine release syndrome.

[0217] III.D.1.Symptoms Cytokine release syndrome can range from mild flu-like symptoms to severe, potentially fatal symptoms. Mild symptoms of cytokine release syndrome include fever, fatigue, headache, rash, joint pain, and malaise. More severe cases are characterized by hypotension and high fever and can progress to an uncontrolled systemic inflammatory response accompanied by circulatory shock requiring vasopressors, vascular leakage, disseminated intravascular coagulation, and multiple organ system failure. Respiratory symptoms are common in subjects with cytokine release syndrome. Mild cases may present with cough and tachypnea, but can progress to acute respiratory distress syndrome (ARDS), accompanied by dyspnea, hypoxemia, and bilateral opacities on chest X-ray.

[0218] The timing of symptom onset and severity of cytokines depend on the immunotherapy agent and the magnitude of immune cell activation. + Cytokine release syndrome after rituximab for malignancies typically occurs within minutes to hours and is >50 × 10 9 Subjects with circulating lymphocytes in the 100-2000 mg / L group experience increased rates of cytokine release syndrome (Winkler et al., Blood, 94(7):2217-2224 (1999)). In contrast, symptom onset typically occurs within days (CAR T cell therapy) to weeks (cytotoxic T cell (CTL) therapy) after T cell infusion, coinciding with maximal T cell proliferation in vivo (Lee et al., Blood, 124(2):188-195 (2014)).

[0219] Symptoms and severity associated with cytokine release syndrome vary widely, and management can be complicated by concurrent conditions in these subjects. Fever is a prominent feature of cytokine release syndrome, and many features of cytokine release syndrome mimic infection. Because it is not uncommon for subjects to experience temperatures above 40.0°C, infection is considered an alternative explanation in all subjects who present with cytokine release syndrome.

[0220] Potentially fatal complications of cytokine release syndrome include cardiac dysfunction, adult respiratory distress syndrome, neurotoxicity, renal and / or hepatic dysfunction, and disseminated intravascular coagulation. Of particular concern is acute cardiac toxicity in the setting of cytokine release syndrome, which resembles sepsis-associated and stress cardiomyopathy. Neurological symptoms occurring in the context of cytokine release syndrome are diverse. Neurological symptoms may occur along with other symptoms of cytokine release syndrome or may occur when other symptoms of cytokine release syndrome have resolved.

[0221] Cytokine release syndrome may also be associated with the findings of macrophage activation syndrome / hemophagocytic lymphohistiocytosis (HLH), and the physiology of these syndromes may overlap. In subjects with cytokine release syndrome who develop HLH / MAS-like syndromes, additional cytokines, such as IL-18, IL-8, IP-10, MCP1, MIG, and MIP1β, are also elevated. These cytokines have also been reported to be elevated in traditional HLH and MAS. Some subjects may have genetic mutations that predispose them to developing HLH / MAS. Additionally, IL-6 may also promote the development of HLH / MAS in the setting of cytokine release syndrome by inducing dysfunctional cytotoxic activity in T and NK cells, a hallmark of HLH and MAS.

[0222] Tumor lysis syndrome may also occur concomitantly with cytokine release syndrome, as immune cell activation and proliferation in a subject correlates with anti-tumor efficacy.

[0223] III.D.2. Diagnosis If a particular subject is monitored as an inpatient, a healthcare provider can determine whether the particular subject has cytokine release syndrome (e.g., after one or more symptoms are observed). Similarly, if a particular subject is monitored as an outpatient but arrives at a healthcare facility later (e.g., after one or more symptoms are observed), a healthcare provider can determine whether the particular subject has cytokine release syndrome.

[0224] Cytokine release syndrome is diagnosed in the context of the underlying medical condition of a particular subject.This underlying problem may be known or may require self-diagnosis.In the context of the treatment of non-Hodgkin's lymphoma, the factors that can frequently affect treatment selection include the subtype, cycle, and type of therapy administered to the subject of the non-Hodgkin's lymphoma of a particular subject.Medical history and physical examination provide the starting point for diagnosis.

[0225] A healthcare provider can examine a subject for symptoms that may indicate the development of cytokine release syndrome, as cytokine release syndrome can affect many different systems in the body. As noted above, abnormally low blood pressure, fever, and hypoxia can indicate cytokine release syndrome.

[0226] Laboratory tests can be performed to identify abnormalities. Increased levels of one or more cytokines, decreased numbers of immune cells, markers of kidney or liver damage, elevated inflammatory markers such as C-reactive protein, abnormal markers of blood coagulation, and elevated ferritin are all consistent with the development of cytokine release syndrome.

[0227] Medical imaging can be performed. For example, a chest x-ray or CT scan can identify pulmonary involvement of cytokine release syndrome.

[0228] Based on the results of physical examinations, laboratory tests, medical imaging, etc., a healthcare provider can determine whether a subject has cytokine release syndrome, and if so, assign a cytokine release syndrome grade to the subject. The grade or stage of cytokine release syndrome guides treatment options. Before determining whether cytokine release syndrome has occurred, a healthcare provider can rule out other potential medical conditions that may be consistent with the results of physical examinations, laboratory tests, medical imaging, etc. For example, a healthcare provider can rule out certain subjects suffering from infection, neutropenic sepsis, tumor lysis syndrome, or adrenal insufficiency, because anti-cytokine therapy administered under these conditions without clear evidence of cytokine release syndrome may be harmful.

[0229] The National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE v.4.0) includes the following grading system designed for cytokine release syndrome associated with antibody therapy: Table 6 below shows the characteristic symptoms and treatment recommendations for each grade of cytokine release syndrome. TIFF2025169248000012.tif251170TIFF2025169248000013.tif21170

[0230] III.D.3. Treatment of Cytokine Release Syndrome When a subject is detected as experiencing cytokine release syndrome, a healthcare provider can administer or provide the treatments identified in this section. Management of cytokine release syndrome can follow a grade- and risk-adapted strategy of monitoring and treatment (Shimabukuro-Vornhagen et al., J. Immunother. Cancer 6:56 (2018)).

[0231] Low-grade cytokine release syndrome can be treated antipodean with antihistamines, antipyretics, analgesics, and fluid therapy. Additional diagnostic tests can be performed frequently to rule out alternative diagnoses. If infection cannot be reliably ruled out, empiric antibody therapy may be considered. Furthermore, if a particular subject exhibits early signs of cytokine release syndrome, the subject can be actively monitored (via inpatient monitoring) more frequently for signs of further deterioration.

[0232] Severe cytokine release syndrome represents a life-threatening condition requiring prompt and invasive treatment. Thus, if a particular subject experiences severe cytokine release syndrome, treatment for cytokine release syndrome can be administered promptly. Treatment for cytokine release syndrome can include anti-cytokine therapy, such as tocilizumab, with or without corticosteroids (for high-risk subjects with cytokine release syndrome of grade 3 or higher or grade 2). In some instances, depending on the type of immunotherapy, corticosteroids can be administered at grade 1 without anti-cytokine therapy, without waiting until the subject has cytokine release syndrome of grade 2 or higher, to reduce immunotherapy-associated cytokine release syndrome and neurological events. See, for example, Liu et al., Blood Cancer J. 10(2):15 (2020). As another example, blinatumomab (an immunotherapeutic drug for the treatment of acute lymphoblastic leukemia) can be administered to a particular subject in response to the detection of severe cytokine release syndrome.

[0233] Because IL-6 is elevated in the serum of subjects with cytokine release syndrome after immunotherapy, such as CAR T cell therapy or bispecific T cell engager therapy, tocilizumab (anti-IL-6 therapy) can be administered to certain subjects diagnosed with severe cytokine release syndrome. IL-6 may be a suitable target because, although not essential for T cell function, it drives many of the symptoms of cytokine release syndrome, as described above. By binding to membrane-bound and soluble IL-6 receptors, tocilizumab can interfere with both traditional and trans-signaling pathways. Research has confirmed that administering monoclonal antibodies against IL6 (siltuximab) and its receptor (tocilizumab) leads to rapid resolution of cytokine release syndrome symptoms (Shimabukuro-Vornhagen (2018)). In early clinical trials, tocilizumab demonstrated a 69% response rate in subjects with severe or fatal cytokine release syndrome. Therefore, tocilizumab is often used for the initial treatment of severe cytokine release syndrome in subjects receiving CAR T cells.

[0234] In August 2017, the FDA approved tocilizumab for the treatment of cytokine release syndrome in subjects aged 2 years and older. The approved dose of tocilizumab for cytokine release syndrome is 12 mg / kg for subjects weighing less than 30 kg and 8 mg / kg for subjects weighing 30 kg or more. Fever and hypotension generally improve within a few hours in subjects who respond to tocilizumab. However, some subjects may require supportive care for several days. Although tocilizumab has a very long half-life (11-14 days), if sufficient clinical improvement is not achieved within 48 hours, the general approach is to repeat the dose with or without corticosteroids. If the subject still does not improve and elevated IL-6 levels persist, higher doses of tocilizumab may be considered. Subjects who develop grade 3 or 4 cytokine release syndrome typically receive treatment (e.g., tocilizumab with or without corticosteroids) almost immediately. After administration of tocilizumab, C-reactive protein may no longer be used as an indicator of cytokine release syndrome severity due to blockade of IL-6 signaling, which results in a rapid decrease in C-reactive protein.

[0235] Several other IL-6-targeting monoclonal antibodies are in late-stage clinical development for the treatment of cytokine release syndrome. Siltuximab is a chimeric IGκ monoclonal antibody that binds to human IL-6 and prevents its interaction with both membrane-bound and soluble IL-6 receptors. Clazakizumab is another monoclonal antibody that targets IL-6.

[0236] Monoclonal antibodies that directly target IL-6 and thereby remove it from the circulation can be used to treat certain subjects if they experience severe cytokine release syndrome and concurrent neurotoxicity, because tocilizumab does not cross the blood-brain barrier and therefore cannot inhibit IL-6 signaling in the CNS. Corticosteroids can be used to treat certain subjects if HLHI / MAS occurs as part of cytokine release syndrome. When corticosteroids are used in subjects receiving T cell-engaged immunotherapy, the duration of treatment can be kept as short as possible to minimize any potential adverse effects on the effectiveness of immunotherapy.

[0237] When neither tocilizumab nor glucocorticoids are effective, blocking TNF-α signaling can be used. However, there are cases of severe cytokine release syndrome that do not respond to tocilizumab, etanercept (anti-TNFD antibody), and glucocorticoids. In these cases, other immunosuppressants, such as the IL-6 monoclonal antibody siltuximab, T cell depleting antibody therapy, such as alemtuzumab and ATG, IL-1R-based inhibitors (anakinra), or cyclophosphamide may be administered or provided.

[0238] Another experimental therapy for cytokine release syndrome is ibrutinib. Cytokine adsorption may also be effective in treating cytokine release syndrome. The advantage of extracorporeal cytokine adsorption over other therapeutic approaches is that it does not selectively block specific receptors or signaling cascades. Instead, this method reduces the specific elevated concentrations of various inflammatory mediators, such as cytokines with pro- and anti-inflammatory functions, such as IL-6, TNF-α, and interferon. In this method, blood is withdrawn from the subject's circulation, and cytokines are removed from the blood before the blood is returned to the circulation.

[0239] III.E. Alternative Exemplary Embodiments It is understood that various alternative embodiments to those described above or depicted in FIG. 1 are contemplated for network 100. For example, instead of or in addition to using cytokine release syndrome risk 180 to assess inpatient monitoring conditions, CRS risk detector 190 can be used to assign a particular subject to a cohort in a clinical study. Cohort assignment can be based on an algorithm or technique that optimizes or prioritizes the definition of cohort assignment so that there is high overlap between cohorts with respect to cytokine release syndrome risk. Thus, cohort assignment can be generated based on the cohort assignment and cytokine release syndrome risk 180 associated with at least one other subject. Output from cytokine release syndrome prediction system 105 can identify the cohort assignment.

[0240] As another example, instead of or in addition to using cytokine release syndrome risk 180 to evaluate inpatient monitoring conditions 193, CRS risk detector 190 can be used with cytokine release syndrome risk 180 to determine whether certain eligibility criteria for a clinical study are met. The certain eligibility criteria can require that a subject have a certain cytokine release syndrome risk and have at least a threshold level of cytokine release syndrome in order to enroll in the clinical study. Thus, the criteria can be evaluated using a particular subject's cytokine release syndrome risk 180 to determine a criteria-specific result. If the criteria are not met, the output can indicate that the particular subject is ineligible for the clinical study. If the criteria are met, it can be determined whether each remaining criterion is met, and the output can indicate whether the subject is eligible for the study.

[0241] As yet another example, instead of or in addition to using cytokine release syndrome risk 180 to assess inpatient monitoring condition 193, CRS risk detector 190 may be used with cytokine release syndrome risk 180 to determine whether to recommend, provide, and / or administer one or more medications that reduce the likelihood of cytokine release syndrome occurring. The one or more medications may include (for example) a steroid agent (e.g., a corticosteroid or methylprednisolone), or a cytokine-directed therapy (an IL-6 inhibitor, e.g., tocilizumab).

[0242] IV. Exemplary Process for Stratifying Subjects to Predict Cytokine Release Syndrome Risk IV.A. Exemplary Processes for Predicting Risk of Cytokine Release Syndrome 2A illustrates a flowchart of a process 200a for predicting a subject's risk of experiencing cytokine release syndrome. Process 200a begins at block 205, where cytokine regulator 150 detects baseline cytokine levels 155. Detecting baseline cytokine levels 155 may include processing one or more of the cytokine level data records associated with the subject (e.g., from raw cytokine level data store 145) or inputs associated with the subject to extract cytokine levels associated with a timestamp within a baseline period.

[0243] In block 210, the cytokine regulator 150 detects on-treatment cytokine levels 160. Detecting on-treatment cytokine levels 160 may include processing one or more of the cytokine level data records associated with the subject (e.g., from the raw cytokine level data store 145) or inputs associated with the subject to extract cytokine levels associated with a timestamp within the on-treatment period.

[0244] In block 215, the cytokine regulator 150 determines a cytokine fold change 170 based on the baseline cytokine level and the on-treatment cytokine level. For example, the cytokine fold change 170 may be defined by subtracting the baseline cytokine level from the on-treatment cytokine level. As another example, the cytokine fold change 170 may be defined by subtracting the logarithm of the baseline cytokine level plus a constant from the logarithm of the on-treatment cytokine level plus a constant.

[0245] In block 220, CRS risk detector 190 detects one or more baseline characteristics. Detecting the baseline characteristics may include processing one or more of the baseline characteristic data records associated with the subject (e.g., from baseline characteristic data store 115) or inputs associated with the subject to extract the baseline characteristics. In some examples, one or more timestamps associated with the one or more data records and / or associated with the one or more inputs are detected, and block 220 includes determining which of the one or more timestamps are within the baseline period and then extracting one or more corresponding data records and / or inputs.

[0246] In block 225, the CRS risk detector 190 identifies a dosage at least part of which is identified as a treatment. The dosage may be identified (for example) by querying the treatment dosage data store 135 with the subject's identifier or by detecting the dosage in input received from the user device 110. The dosage may include a dosage of an active ingredient or a treatment. The dosage may include (for example) a dosage within a cycle or a cumulative dosage of a multi-cycle treatment. The dosage may include a dosage already administered to the subject, a dosage to be administered to the subject, a dosage prescribed for the subject, or a dosage (e.g., of an active ingredient or an overall treatment) being considered as a treatment option for the subject.

[0247] In block 230, CRS risk detector 190 determines a cytokine release syndrome risk score by processing the baseline characteristics and optionally the dosage using a machine learning model. The cytokine release syndrome risk score can represent an intermediate prediction of a subject's risk of experiencing cytokine release syndrome (e.g., at least a threshold grade and / or within a predefined time range from the start of treatment). The cytokine release syndrome risk score can be determined using risk score generation model 184. The cytokine release syndrome risk score can be determined (for example) by looking up one or more learned parameters (e.g., parameters associated with each of two or more characteristics) in risk score generation model 184 and using the parameters, the baseline characteristics, and optionally the dosage to generate a risk score.

[0248] In some examples, before a risk score is generated, CRS risk detector 190 uses feature selection model 185 or results generated by feature selection model 185 to detect a subset of baseline characteristics to be used in determining the risk score. Block 230 can then selectively use the subset of baseline characteristics (potentially along with dosage) to determine the risk score.

[0249] In block 235, CRS risk detector 190 predicts the risk of the subject experiencing cytokine release syndrome (e.g., at least at a threshold grade, and / or within a predefined time period) based on the CRSRS and (potentially) dose and cytokine fold change. The predicted risk of the subject experiencing cytokine release syndrome may be cytokine release syndrome risk 180, which may be determined using a decision tree model 183.

[0250] In block 240, the cytokine release syndrome prediction system 105 outputs a result based on the predicted risk. The result may be output to the user device 110 that initiated the process 200s and / or to a healthcare provider system 120 associated with the subject. The result may identify the predicted risk. The result may additionally or alternatively identify an action (e.g., to be performed, recommended, or considered) in response to evaluating a condition (e.g., inpatient monitoring condition 193) based on the predicted risk. For example, the result may indicate that the subject has or is being considered for inpatient monitoring with a predefined monitoring period after treatment. As another example, the result may indicate that the subject has or is being considered for outpatient monitoring with a predefined monitoring period after treatment. The result may be output (for example) via transmission or display.

[0251] It is understood that modifications to process 200 are contemplated. For example, one, more, or all of blocks 205, 210, 215, and 225 may be omitted from process 200. By way of example, each of blocks 205, 210, 215, and 225 may be omitted from process 200, and the cytokine release syndrome risk score generated in block 230 is based on (e.g., and / or based only on) one or more baseline characteristics.

[0252] IV.B. Exemplary Process for Selecting Hospitalization or Outpatient Monitoring Based on Risk Prediction 2B shows a process 200b for using predicted risk to determine whether to recommend hospitalization or outpatient monitoring for cytokine release syndrome for a subject. Process 200b begins by accessing baseline characteristics 255, which may include some or all of the baseline characteristics detected in block 220. The baseline characteristics can be used to determine a baseline risk (e.g., a numerical risk score or a categorical risk) (e.g., by risk score generation model 184), which may also or alternatively be dependent on treatment dosage. One or more baseline cytokine levels can also be determined (e.g., by processing samples collected during the baseline period).

[0253] In block 260, the decision tree model 183 can determine if the baseline risk is high. For example, the decision tree model 183 can compare the risk to a threshold. If the risk is determined to be low, the process 200b proceeds to block 265, where an initial plan can be initiated for outpatient monitoring. For example, the subject can be notified of the possibility of being discharged from the hospital or discharged from the medical facility once treatment is complete. On the other hand, if the risk is determined to be high, the process 200b proceeds to block 270, where an initial plan can be initiated for inpatient monitoring. For example, the subject can be notified of the possibility of being discharged from the hospital or discharged from the medical facility once treatment is complete, hospitalization information can also be requested from the subject, and / or data can be initiated to reserve space or a room for the subject for a period of time after treatment.

[0254] At block 275, the infusion of the treatment is completed. At this point, and / or while the treatment is being infused, one or more treatment samples can be collected from the subject, and / or one or more on-treatment levels of one or more cytokines can be measured within the samples. The one or more on-treatment levels and one or more baseline levels can be used to determine a cytokine fold change.

[0255] If the subject is preliminarily defined as low risk (at block 265), process 200b proceeds from block 275 to block 280a. In block 280a, it is determined whether the cytokine fold change is below a cytokine level threshold. In some examples, the cytokine level threshold is selected based on a previous determination (at block 260) that assigned the subject to a low risk classification. If it is determined that the cytokine fold change is below the cytokine level threshold, process 200b proceeds to block 285, where the subject is monitored via outpatient monitoring. Otherwise, process 200b proceeds to block 290, where the subject is monitored via inpatient monitoring.

[0256] If the subject is preliminarily defined as high risk (at block 265), process 200b proceeds from block 275 to block 280b. In block 280b, it is determined whether the cytokine fold change exceeds a cytokine level threshold. In some instances, the cytokine level threshold is selected based on a previous determination (at block 260) that assigned the subject to a low-risk classification. Thus, the cytokine level threshold considered in block 280a can be different from the cytokine level threshold considered in block 280b. If it is determined that the cytokine fold change exceeds the cytokine level threshold, process 200b proceeds to block 290, where the subject is monitored with inpatient monitoring. Otherwise, process 200b proceeds to block 285, where the subject is monitored with outpatient monitoring.

[0257] Inpatient and / or outpatient monitoring (at blocks 290 or 285) can indicate that the subject will actually receive that type of monitoring, that the healthcare provider recommends inpatient (or, alternatively, outpatient) monitoring, that instructions are provided to the subject to prepare for that type of monitoring, and / or that a recommendation is provided to the subject for that type of monitoring. [Example]

[0258] IV Example IV.A. Example 1: Exemplary Training and Use of a Multivariate Model to Predict the Occurrence of Cytokine Release Syndrome Clinical data and laboratory values ​​were used to train multivariate models (risk score generation model and decision tree model) to predict the incidence and / or severity of cytokine release syndrome after administration of CD3-engaging bispecific cancer immunotherapy (glofitamab). The extent to which various variables predict the incidence and / or severity of cytokine release syndrome was further characterized. Still further, the trained models were used to determine the extent to which a subset of subjects could be identified as having a low (<10%) risk of grade 2+ cytokine release syndrome based on baseline observations and laboratory values.

[0259] IV.A.1. Training and Validation Data The data used to train and validate the machine learning model were from clinical study NP30179, a phase 1, multicenter, dose-escalation study. One intervention in the study involved the administration of 1000 mg of obinutuzumab via IV infusion on day 1, followed by one or more days of glofitamab (schedule-specific doses). Data corresponding to this intervention were analyzed. The study evaluated efficacy, safety, tolerability, and pharmacokinetics in subjects with relapsed / refractory B-cell non-Hodgkin's lymphoma.

[0260] The cohorts assessed in this example include: Three fixed-dose cohorts: · MQ2W (monotherapy regimen after prior treatment on day 1): 1000 mg of obinutuzumab administered on day 1 and multiple confirmed doses of glofitamab administered on each of days 8, 22, and 36 (the same dose was administered on each of days 8, 22, and 36, and the dose ranged from 0.6 mg to 25 mg); · MQ3W (monotherapy regimen after prior treatment on day 1): 1000 mg of obinutuzumab administered on day 1 and multiple confirmed doses of glofitamab administered on each of days 8, 22, and 43 (the same dose was administered on each of days 8, 22, and 36, with doses ranging from 0.6 mg to 16 mg); CQ3W (combination therapy regimen after prior treatment on day 1): 1000 mg obinutuzumab administered on day 0, multiple confirmed doses of glofitamab administered on days 8, 22, and 43 (the same dose was administered on days 22 and 36, ranging from 0.6 mg to 16 mg), and 1000 mg obinutuzumab administered on days 22 and 43. 1 split-dose cohort: 10 / 16Q3W: 1000 mg obinutuzumab administered on day 1, 10 mg glofitamab administered on day 22, and 16 mg glofitamab administered on day 43; Two step-up dose (SUD) cohorts: 2.5 / 10 / 16SUD Q3W: 1000 mg obinutuzumab administered on day 1, 2.5 mg glofitamab administered on day 8, 10 mg glofitamab administered on day 15, 16 mg glofitamab administered on day 22, and 16 mg glofitamab administered on day 43; 2.5 / 10 / 30 SUD Q3W: 1000 mg obinutuzumab administered on day 1, 2.5 mg glofitamab administered on day 8, 10 mg glofitamab administered on day 15, 30 mg glofitamab administered on day 22, and 30 mg glofitamab administered on day 43.

[0261] Figure 3 depicts the dose timing in the various cohorts. "Cycle 1" was defined as starting on day 8, "Cycle 2" was defined as starting on day 22, and "Cycle 3" was defined as starting on day 36 (Q2W regimen) or day 43 (Q3W regimen). Thus, the monotherapy fixed-dose cohorts (MQ2W and MQ3W) and the combination therapy fixed-dose cohort (CQ3W) did not differ between cohorts with respect to the type of therapy administered during cycle 1, allowing the data to be combined for analyses focused on this cycle.

[0262] The 2.5 / 10 / 30SUD Q3W cohort was used as the validation dataset.

[0263] The histology of non-Hodgkin's lymphomas (excluding mantle cell non-Hodgkin's lymphoma histology) was further accessed.

[0264] Table 7 shows the number of subjects with documented treatment completion at Cycle 1, separated by treatment regimen and non-Hodgkin's lymphoma subtype (aggressive, indolent, or unknown). The 2.5 / 10 / 30SUD Q3W dose group was used as the validation data set, and subject counts for this regimen are shown in the box. TIFF2025169248000014.tif127170

[0265] Table 8 shows how many subjects (in each treatment regimen and dose group) had documented treatment completion at Cycle 2. TIFF2025169248000015.tif39170

[0266] For each subject represented in the training / validation data, the following data were identified in this study, when available: · the dose of glofitamab administered if the subject was in a fixed-dose cohort; · Whether the dose of the administered prior treatment (obinutuzumab) was 200 g / mL; The following laboratory variables were measured / observed on Cycle 1 Day 1 (C1D1): Platelet count Monocyte levels Hemoglobin level ○White blood cell count (WBC) Fibrinogen levels Butyrate dehydrogenase (LDH) levels C-reactive protein (CRP) levels TNF-α plasma levels Interleukin-6 (IL-6) plasma levels Aspartate aminotransferase (AST) levels Alkaline phosphatase (ALP) levels Pre-Gz glofitamab (<200g / ml) The following clinical variables measured / observed on or before Day 1: Small cell non-Hodgkin's lymphoma of the aggressive subtype (aNHL; defined as including follicular lymphoma grade 1, grade 2, or grade 3A) or the indolent subtype (iNHL; defined as including diffuse large B-cell lymphoma, primary mediastinal B-cell lymphoma, Richter's transformation, transformed follicular lymphoma, and transformed marginal zone lymphoma) Whether the subject had a previous B-cell lymphocytosis Whether the subject had any comorbidities Whether the subject had any cardiac comorbidities, including any of the following: Cardiac arrhythmias (arrhythmia, supraventricular arrhythmia, atrial fibrillation, atrial flutter, atrial tachycardia, sinus bradycardia, sinus tachycardia, supraventricular extrasystoles, supraventricular tachycardia, tachycardia, paroxysmal tachycardia, ventricular extrasystoles, or ventricular tachycardia); · Heart problems, signs and symptoms NEC (heart problems or hypertensive heart disease); · Heart valve disorders (aortic stenosis or mitral valve prolapse); · Coronary artery disease (acute myocardial infarction, angina pectoris, atherosclerotic coronary arteries, myocardial infarction, or myocardial ischemia); · Heart failure (heart failure or chronic heart failure); myocardial dysfunction (cardiomyopathies, cytotoxic cardiomyopathy, diastolic dysfunction, or ischemic cardiomyopathy); or Pericardial disorders (pericarditis) The following pathology-based variables were measured / observed on Day 1: Whether bone marrow (BM) infiltration by non-Hodgkin's lymphoma was detected Whether or not peripheral blood (PB) infiltration by non-Hodgkin's lymphoma was detected Whether extranodal involvement by non-Hodgkin's lymphoma was detected Tumor burden, which determines the sum of the longest diameters (SPD) of the entire tumor, and / or tumor burden >3000 mm 2 Is it more than or equal to this? Ann Arbor lymphoma staging (and / or disease stage is at least stage III) The following demographic variables were measured / observed on Day 1: The subject's age (and / or whether the subject is at least 64 years old)

[0267] The study monitored and graded any cytokine release syndrome that occurred during or after glofitamab infusion using the grading criteria described in Table 9 in Section III.D.2. The time at which any cytokine release syndrome occurred was recorded (e.g., relative to day 1 as defined for the corresponding cohort).

[0268] IV.A.2 Data Partitioning for Model Training and Validation Figure 4 depicts what data is used to train and validate the feature selection model (which identifies a reduced feature set and threshold to convert any non-binary baseline characteristics into binary variables), the risk score generation model (which converts the binary values ​​of the reduced feature set into risk scores), and the decision tree model (which converts risk scores and cytokine fold changes into predictions of whether or not grade 2+ cytokine release syndrome will occur).

[0269] A challenge to model development posed by non-randomized or precisely stratified clinical trials (e.g., NP30179) is the size of the subject subcohorts, which are expected to exhibit multiple confounding phenomena. Predictive or prognostic factors may be imbalanced across treatment cohorts and subject subgroups. For example, the incidence of cytokine release syndrome may be confounded by glofitamab dose.

[0270] Therefore, non-overlapping training and validation datasets were used. The training dataset included data corresponding to all available regimens except for the target 2.5 / 10 / 30 SUD Q3W treatment regimen. The feature selection model used the training data (n=196) to identify which baseline characteristics in the training set significantly correlated with whether grade 2+ cytokine release syndrome occurred within 7 days of the first glofitamab infusion. A reduced feature set was defined to include each baseline characteristic significantly correlated with the incidence of grade 2+ cytokine release syndrome. For any baseline characteristics in the reduced feature set that had non-binary values ​​(e.g., real-valued values), the feature selection model further determined the weight associated with the baseline characteristic and the threshold value that most accurately separated the values ​​of the baseline characteristic that predicted the occurrence of grade 2+ cytokine release syndrome from other values ​​of the baseline characteristic that did not predict the occurrence of grade 2+ cytokine release syndrome. Separated reduced feature sets and threshold values ​​were determined for aggressive and all non-Hodgkin's lymphoma histologies.

[0271] A predictive model was defined to include a risk score generation model that converts a reduced feature set (with associated weights, and any thresholds) into a risk score, and also to include a decision tree that generates subject-specific interpretable and clinically actionable output (e.g., a recommendation whether to monitor for potential cytokine release syndrome post-treatment using inpatient or outpatient monitoring).

[0272] Data corresponding to the 2.5 / 10 / 30 step-up dose (SUD) Q3W treatment regimen were defined as the validation SUD cohort, and data corresponding to subjects receiving this treatment regimen were used to validate the weights of baseline characteristics in the reduced feature set and prediction model.

[0273] The decision tree model uses one or more thresholds (e.g., risk threshold and cytokine level threshold) to predict whether a given subject is at "high risk" or "low risk" for cytokine release syndrome. The training dataset was a combination of non-randomized or stratified multiple-dose regimens and did not include many cases of the initial 2.5 mg dose. The current 2.5 / 10 / 30 SUD CRS mitigation strategy also did not accurately match that of the initial cohort. These dataset characteristics highlight the challenges of using the training dataset to determine a classifier decision cutoff that will yield an accurate classifier for the target SUD schedule.

[0274] IV.A.3. Timing of Cytokine Release Syndrome As shown in Section IV.A.1., the NP30179 study data included cytokine release syndrome data. Each cytokine release syndrome was associated with a subject, treatment regimen, severity, and a time metric indicating when the cytokine release syndrome occurred within the treatment regimen. The time metric was then used to determine in which treatment cycle the cytokine release syndrome occurred and the time of the event within the cycle.

[0275] Figure 5 shows the timing of cytokine release syndrome in each analysis cohort. Each cytokine release syndrome is represented by a symbol. If multiple cytokine release syndromes were detected in a given subject, only the first observed event is represented in Figure 5. The position of the symbol on the (logarithmic) OY axis indicates the timing of the first cytokine release syndrome in the subject. Severity was determined using the American Society for Transplantation and Cellular Therapy (ASTCT) consensus grading recommendations (shown in Table 6 above). The severity of each event is represented both by the symbol along the O-X axis and also by the color of the symbol.

[0276] As shown in Figure 5, the majority of initial occurrences of cytokine release syndrome occurred during cycle 1 of the treatment regimen. The earliest events occurred within 1 day of the end of the first infusion of cycle 1.

[0277] Furthermore, the incidence of cytokine release syndrome increased with increasing monotherapy dose. For example, the number of cytokine release syndrome events occurred more than twice as many times in cohorts receiving glofitamab at doses of 10 to 25 mg (including both the 10 / 16 and 16 / 25 cohorts) compared with cohorts receiving glofitamab at doses of 4 to 10 mg. Similarly, the number of cytokine release syndrome events occurred more than twice as many times in cohorts receiving glofitamab at doses of 4 to 10 mg compared with cohorts receiving glofitamab at doses of 1 to 2.5 mg.

[0278] Cytokine release syndrome in the second cycle was not detected in any of the fixed-dose monotherapy cohorts. Cytokine release syndrome also rarely occurred in cycle 2 in the split-dose and step-up dose cohorts. Therefore, the subsequent analysis presented in this example focused on predicting cytokine release syndrome that occurred in cycle 1.

[0279] IV.A.4. Dose-Dependence of Cytokine Release Syndrome Figure 6 shows the percentage of subjects from the training and validation datasets within each cohort who experienced a cytokine release syndrome event during week 1 of cycle 1. Blue bars correspond to any type of cytokine release syndrome. Orange bars correspond to cytokine release syndrome grade 2 or higher. The incidence of cytokine release syndrome was dose-related.

[0280] The incidence of cytokine release syndrome in the 1.8-2.5 mg monotherapy cohort appeared to differ with the step-up dose (using a 2.5 mg initial dose). This difference may reflect differences in clinical monitoring or de-escalation practices (e.g., infusion times were extended to 8 hours in many subjects in the step-up cohort). Alternatively or additionally, the difference in cytokine release syndrome may be driven by differences between the two cohorts with respect to baseline key risk factors (lower risk in the step-up dose cohort compared with the 1.8-2.5 mg monotherapy cohort with a baseline risk profile; see also Figure 13).

[0281] IV.A.5. Predictor Training and Multivariate Model Formulation Figure 7 shows the workflow used to identify the extent to which various baseline characteristics (or "risk factors") contribute to predicting the occurrence of cytokine release syndrome and how the model parameters are trained. The training cohorts depicted in this figure included each fixed-dose cohort and the 2.5 / 10 / 16 SUD Q3W split-dose cohort. In total, 196 subjects were represented in the training cohort. The subject set included subjects diagnosed with aggressive or indolent non-Hodgkin's lymphoma.

[0282] The data were randomly stratified by three-fold cross-validation. Stratification factors included non-Hodgkin's lymphoma histology (follicular stage I-IIIA, diffuse large B-cell lymphoma, primary mediastinal B-cell lymphoma, Richter's transformation, transformed follicular lymphoma, transformation, and marginal zone lymphoma). In each iteration, data corresponding to approximately 130 subjects were used for training and approximately 65 subjects for testing.

[0283] Cross-validation (using training data) was used to select baseline characteristic predictors of cytokine release syndrome occurrence (via a feature selection model), relate selected risk factors to the predicted probability of cytokine release syndrome, adjust parameters of the regression model (via a risk score generation model), perform stability analysis (via the risk score generation model and the feature selection model), and estimate the performance of the regression model (via the risk score generation model). The risk score generation model included either a bivariate logistic regression model of CRS risk score (CRSRS) and glofitamab dose, or a multivariate logistic regression model of baseline parameter values ​​(the same parameter set as the CRSRS combined score) and glofitamab dose. The final set of baseline parameters and the weight of single baseline parameters in predicting cytokine release syndrome risk score were completed with the help of random forest and floating regression model formation. The stability of the predictors was assessed in a stratified cross-validation setting.

[0284] The 2.5 / 10 / 30 SUD Q3W split-dose cohort was used for validation and to identify one or more clinically relevant risk score thresholds. More specifically, the risk score generation model was configured to output a numerical output corresponding to the predicted probability of cytokine release syndrome (ASTCT grade 2+ CRS) occurring after the first dose of 2.5 mg glofitamab. The performance of several thresholds on the predicted probability scale was validated in the validation cohort.

[0285] Two types of analyses were performed for the "all histologies" data (corresponding to data where subjects had been diagnosed with either aggressive or indolent non-Hodgkin's lymphoma), and for the aggressive non-Hodgkin's lymphoma (aNHL) data. The first analysis, using the risk score generation model, performed multiple univariate regressions to determine the extent to which each of multiple variables independently predicted whether grade 2 or higher cytokine release syndrome was observed within 1 week after the first glofitamab dose.

[0286] IV.A.5.a. Univariate analysis assesses the degree to which individual variables predict an event Figure 8 is a plot showing the degree to which each of several baseline characteristics was predictive of the occurrence of cytokine release syndrome (grade 2+ after the first glofitamab dose). The dose-adjusted predictive strength of each baseline characteristic is provided in terms of odds ratios per unit change in factor level. Confidence intervals (not adjusted for multiple testing) aid in interpreting significance. The odds ratio represents the degree to which a given characteristic predicts whether or not cytokine release syndrome will occur. Large odds ratios indicate an increased risk of cytokine release syndrome at the indicated factor level. The odds ratio statistic considers the prediction of a single variable in isolation.

[0287] Additional input was provided by random forest and floating regression (multivariate) experiments. For any feature in the reduced feature set that corresponded to a non-binary (e.g., real number) variable, the feature was defined to be a binary value indicating whether a particular inequality was established through the use of the non-binary variable (e.g., whether a given real number was below a threshold set of features).

[0288] Figure 9A illustrates how cytokine release risk can be predicted using a multivariate logistic regression model. The figure illustrates the model output: predicted probability of grade 2+ cytokine release syndrome as a function of glofitamab dose (first predictor), tumor burden (SPD categorization), peripheral blood infiltration status, and Ann Arbor stage category (I or II vs. III or IV). For any particular combination of these parameters, the model predicts a certain risk of grade 2+ cytokine release syndrome. At 2.5 mg glofitamab and certain values ​​of other baseline parameters (indicated by the vertical dashed line and red arrow), the predicted risk was estimated to be approximately 25%.

[0289] IV.A.5.b. Combined Cytokine Release Syndrome Risk Score (CRSRS) to Predict Events The combined cytokine release syndrome risk score (CRSRS) was defined as the weighted sum of selected (reduced feature set) target characteristics at baseline. The risk score generation model defined weights with maximum classification accuracy and stability in the training data. Each weight was primarily derived from the log(odds ratio) of a univariate dose-adjusted logistic regression predicting the odds ratio of grade 2+ CRS from baseline values ​​of dose and corresponding parameters. The weights were further adjusted by including information about feature stability from random forest and floating feature selection experiments. (See Figure 7.)

[0290] In this example, a risk score generation model 184 is defined to generate a cytokine release syndrome risk score that predicts grade 2+ cytokine release syndrome occurring at a particular glofitamab dose in a particular subject. The risk of a subject experiencing cytokine release syndrome was determined based on the cytokine release syndrome risk score and the treatment dose.

[0291] In this example, the decision tree model included a classifier configured to translate a real cytokine release risk score into a binary prediction of whether cytokine release syndrome will occur based on whether the sum of the cytokine release risk score and the administered dose exceeds a risk score threshold. The cytokine release risk score was defined to have a minimum value of 0 and a maximum value of 8.5.

[0292] Figure 9B illustrates how the cytokine release risk score is calculated and used together with the dosage in the predictive model. As illustrated in Figure 9B, the slope of the plot relating the incidence of Grade 2+ cytokine release syndrome to the cytokine release risk score can be steeper than the slope of the plot relating the incidence of Grade 2+ cytokine release syndrome to the dosage.

[0293] Table 9 shows the final weights (or their binary transformations) assigned to baseline characteristics contributing to cytokine release syndrome risk. The characteristics associated with the highest weights were whether the Ann Arbor stage was at least III and whether the tumor had an overall longest diameter sum of at least 3000 mm. 2 Characteristics associated with intermediate weights indicated whether the subject was over 64 years old, whether bone marrow infiltration was observed, and whether atypical cells were detected in the peripheral blood. Characteristics associated with the lowest weights indicated whether the subject had cardiac comorbidities and whether the white blood cell count was 4.5 × 10 9 cells / l and whether lactate dehydrogenase was above 280 U / l. TIFF2025169248000016.tif78170

[0294] IV.A.6. Performance on training and validation datasets Figure 10 shows the negative predictive value (NPV) for predicted negative cases corresponding to risk scores from two versions of the risk score generation model. In one case, the risk score generation model converted binary versions of baseline characteristics represented in the reduced feature set into a combined cytokine release syndrome risk score (CRSRS, blue line; see also Figure 9B). In another case, the risk score generation model converted raw versions of baseline characteristics represented in the reduced feature set into a multivariate model output (orange line; see also Figure 9A).

[0295] Every point in Figure 10 corresponds to a unique cutoff (e.g., using a decision tree model), where values ​​above the cutoff were considered to correspond to a prediction of at least grade 2 cytokine release syndrome occurring, and values ​​below the cutoff were considered the opposite prediction. For each cutoff, the negative predictive value and predicted negative case percentage were recorded at the cutoff.

[0296] In Figure 10, the O-X coordinate indicates the negative call rate for that cutoff, which is the proportion of cases in the dataset classified as "low risk" by the decision tree model. The O-Y coordinate for each point identifies the negative predictive value at that point relative to the cutoff. The negative predictive value is the probability that a subject classified as low risk will not actually develop grade 2 or higher cytokine release syndrome. The shaded area in Figure 10 is the opportunity range, where 20% to 50% of subjects had less than a 10% chance of developing grade 2 or higher cytokine release syndrome after the first glofitamab dose.

[0297] As illustrated in Figure 10, model variability increases significantly as negative predictive value reaches 80-90%. Opportunity ranges exist for the "all histologies" and aggressive non-Hodgkin's lymphoma data.

[0298] To characterize the performance of predicting cytokine release syndrome (grade 2+ after the first glofitamab dose) in the target 2.5 / 10 / 30 SUD cohort, the first glofitamab dose was defined as 2.5 mg. Figure 11 shows the negative predictive value for predicted negative cases in the model validation dataset for the 2.5 / 10 / 30 mg step-up dose cohort. Each dot corresponds to a different threshold used by the decision tree model to convert the risk score into a binary prediction of whether cytokine release syndrome (grade 2 or higher after the first glofitamab dose) will occur. If the sum of the risk score and dose is lower than the threshold, the classifier produced a "low risk" result, corresponding to the prediction that cytokine release syndrome will occur. Thus, the percentage of predictions corresponding to a "low risk" result (corresponding to the prediction that grade 2+ cytokine release syndrome will occur after the first glofitamab dose) increases as the threshold increases.

[0299] Figure 12A shows the probability of cytokine release syndrome (grade 2 or higher after the first glofitamab dose) occurring as a function of the cytokine release syndrome risk score (CRSRS) at each of three exemplary thresholds that distinguish whether the event is predicted to occur or not. Lower thresholds corresponded to a prediction that more events would occur.

[0300] The table in Figure 12A shows how increasing the threshold from 4.0 to 6.0 reduces the predicted number of positive cases from 17 (49%) to 7 (20%). Data for the subpopulation of subjects who were low risk and who were predicted not to experience or to experience cytokine release syndrome (grade 2 or higher after the first glofitamab dose) were further examined.

[0301] When a cutoff threshold of 4.0 was used, the decision tree model predicted that 51% of subjects did not experience cytokine release syndrome (grade 2 or higher after the first glofitamab dose); such events were not observed in subjects with scores below the threshold, resulting in an observed negative predictive value of 1.0. When a cutoff threshold of 6.0 was used, the decision tree model predicted that 80% of subjects did not experience cytokine release syndrome (grade 2 or higher after the first glofitamab dose); however, such events were actually observed in 14% of subjects below the threshold (resulting in a negative predictive value of 0.86). When a cutoff threshold of 5.0 was used, the decision tree model predicted that 60% of subjects did not experience cytokine release syndrome (grade 2 or higher after the first glofitamab dose); such events were observed in only 5% of subjects. It appeared as if a cutoff close to 5.0 was optimal for distinguishing positive from negative cases. Of the subpopulation of subjects corresponding to results below the 5.0 threshold, 95% of subjects predicted to be at low risk of experiencing cytokine release syndrome (grade 2 or higher after the first glofitamab dose) did in fact not experience such an event.

[0302] The trained CRSRS model was further used to generate CRS risk predictions for the complete (model and decision cutoff) validation set. Each subject in the validation set was diagnosed with NHL and participated in the NP30179 clinical study. Each score was compared with one of two thresholds (4.0 or 5.0) to generate a binary prediction of whether the subject would experience grade 2 or higher cytokine release syndrome after the first glofitamab dose. The validation set included data from 156 subjects. The data from this analysis is shown in Figure 12B.

[0303] As shown in the plot in Figure 12B, the CRSRS threshold remains positively correlated with the percentage of predicted negative cases and negatively correlated with the negative predictive value. As shown in the table, when a cutoff threshold of 4.0 was used to assess the validation data, the trained decision tree model predicted that 42% of subjects did not experience cytokine release syndrome (grade 2 or higher after the first glofitamab dose), which was accurate for 98% of subjects with scores below the threshold (yielding a negative predictive value of 0.98). (The detection rate for subjects who did not experience cytokine release syndrome grade 2 or higher was 40%). The standard error was 0.02, with a confidence interval of 0.92–0.99.

[0304] When a cutoff threshold of 5.0 was used to assess the validation data, the trained decision tree model predicted that 52% of subjects did not experience cytokine release syndrome (grade 2 or higher after the first glofitamab dose), which was accurate in 98% of subjects with scores below the threshold (yielding a negative predictive value of 0.98).

[0305] Notably, the predicted percentage of negative cases determined using the training data (shown in Figure 12A) is very similar to that determined using the validation data (shown in Figure 12B).

[0306] Furthermore, for validation data, an association between baseline CRSRS and the severity of any observed cytokine release syndrome that occurred after the first glofitamab infusion was observed in both CART-naive and -experienced subjects, as well as in subjects pretreated with dexamethasone or other corticosteroids.

[0307] IV.A.7. Distribution and Prediction of Cytokine Release Syndrome Risk Scores Figure 13 shows the distribution of baseline cytokine release syndrome risk scores (CRSRS) corresponding to clinical study NP30179. As shown, the distribution is multimodal, with modes at 2.3, 5.6, and around 5.6.

[0308] Furthermore, baseline risk may differ across cohorts, which may explain some of the differences in cytokine release syndrome observations between subjects receiving the same treatment dose. This differential detection may explain much of the observed differences in cytokine release syndrome incidence in these cohorts. The table shown in Figure 13 summarizes the statistical distribution of cytokine release syndrome risk scores (grade 2 or higher after the first glofitamab dose) for these dose groups. (Figure 6 shows a summary of cytokine release syndrome incidence in cycle 1 after the first glofitamab infusion.)

[0309] IV.B. Example 2: Exemplary Analysis of How Early Changes in Cytokine Levels Can Predict the Incidence and Severity of Cytokine Release Syndrome In each of the subject sets in the NP30179 study (who were also in the cohort used for training, as shown in the boxes in Figure 4), cytokine data were collected and analyzed to determine cytokine dynamics and the extent to which levels of various types of cytokines indicate the incidence and / or severity of cytokine release syndrome. Each subject in the subject set was in the fixed-dose cohort of NP30179, having been diagnosed with non-Hodgkin's lymphoma and receiving a fixed dose of glofitamab on study day 8 after a C1D1 Gpt, as shown in Figure 3. Table 10 shows the breakdown of the subject set based on each dose of glofitamab and also based on the subtype of non-Hodgkin's lymphoma with which the subject was diagnosed (aggressive or indolent). TIFF2025169248000017.tif91170

[0310] For each dose range, Table 11 shows the distribution of duration of first glofitamab administration. As shown, most infusions were administered over 4 hours. TIFF2025169248000018.tif98170

[0311] IV.B.1. Exemplary Cytokine Kinetics Figures 14A and 14B show the fold change in IL-6 and TNF-α (respectively) during the first glofitamab treatment cycle. Each line corresponds to a different subject who experienced cytokine release syndrome (of any grade). The first x-position corresponds to before administration of Gpt on C1D1. The second x-position corresponds to before administration of glofitamab on C1D8. Cytokine level data for all subjects was normalized to a second time point collected before the first administration of glofitamab. The third x-position (MI) corresponds to the midpoint of the glofitamab infusion. The fourth x-position (EOI) corresponds to the end of the glofitamab infusion. The fifth, sixth, and seventh x-positions (6H EOI, 24H EOI, and 120H EOI) correspond to 6, 24, and 120 hours (respectively) after the end of the glofitamab infusion.

[0312] Peaks in both cytokines were observed after treatment initiation. For IL-6, the peak began to occur at end of infusion (EOI). For TNF-α, the peak began to occur even earlier at mid-infusion (MI).

[0313] Figure 15 contrasts the cytokine fold change in IL-6 in subjects who did not experience cytokine release syndrome (left plot) and subjects who did experience cytokine release syndrome (right plot). Note that, apart from the arrows, the right plot in Figure 15 is the same as Figure 14A.

[0314] The "on-treatment" (OT) time point was defined to include mid-infusion and end-of-infusion time points, and the "baseline" (BL) time was defined as the time at which treatment began (before C1D8 administration). As shown in Figure 15, the fold change in IL-6 at time points after treatment initiation (e.g., MI, EOI, 6H EIO, etc.) was generally substantially positive in subjects who experienced cytokine release syndrome (right graph, compare change on treatment - green arrow - variability before glofitamab administration - red arrow), whereas this relationship was not observed in subjects who did not experience cytokine release syndrome (left graph).

[0315] For each subject, on-treatment cytokine fold change was calculated with the following definition: log 2 (1+OT)-log 2 (1+BL) where OT is the maximum cytokine level (picograms per milliliter) during the on-treatment period and BL is the cytokine level (picograms per milliliter) during baseline. These on-treatment cytokine fold changes were then divided based on whether or not the subject experienced cytokine release syndrome and the severity of any observed cytokine release syndrome.

[0316] Figures 16A and 16B show box plots showing the dependence of on-treatment cytokine fold change on the presence or severity of initial cytokine release syndrome. Each point represents a subject. Each point is color-coded to indicate the dose of glofitamab the subject received.

[0317] In Figure 16A, an x ​​value of 0 indicates that no cytokine release syndrome was observed. Each non-zero x value indicates the severity of cytokine release syndrome observed. In Figure 16B, data points are divided based on whether or not a severity of cytokine release syndrome of at least 2 was observed.

[0318] As shown, on-treatment cytokine fold change increased with the severity of the first cytokine release syndrome event (left plot) and differed in groups defined based on whether at least two severity levels of cytokine release syndrome were observed (right plot). Specifically, on-treatment cytokine fold change was higher with increasing severity of cytokine release syndrome.

[0319] While the on-treatment cytokine fold change captures the difference between the logarithm of the maximum cytokine level + 1 and the logarithm of the baseline cytokine level + 1, other cytokine fold changes can be calculated to represent the difference between the cytokine level at any time point and the cytokine level at the baseline time point. That is, a cytokine fold change can be defined as: log 2 (1+T)-log 2 (1+BL) where T is the cytokine level (picograms per milliliter) during any period and BL is the cytokine level (picograms per milliliter) during baseline.

[0320] When baseline cytokine levels are associated with cytokine release syndrome, the relationship between cytokine levels and the incidence of cytokine release syndrome cannot be captured by evaluating cytokine fold change. However, this cytokine fold change measurement can facilitate characterization of intra-subject changes and reduce intra-subject variability. This cytokine fold change measurement can further facilitate capturing pharmacodynamic concepts in induction. Therefore, the subsequent cytokine level analysis in this example focuses on cytokine fold change measurement (or on-treatment cytokine fold change).

[0321] Cytokine fold changes can reflect pharmacodynamic concepts in guiding therapy. Absolute fold changes convey cytokine kinetic characteristics and can better compensate for baseline variability in subjects.

[0322] IV.B.3. Effect of Dose on Early Cytokine Changes and the Association with Cytokine Release Syndrome Figures 17A and 17B show how on-treatment levels of two cytokines (IL-6, TNF-α) change during the first cycle of glofitamab treatment. The four subplots shown in each figure correspond to four different dose ranges. Each symbol represents a subject. The color of the symbol indicates whether the subject had cytokine release syndrome and, if so, the severity of the event. For each incidence and severity of cytokine release syndrome and each dose range, the mean cytokine fold change was also calculated using the subject's cytokine levels associated with the incidence / severity and associated with the dose range. These average values ​​are shown as solid lines in Figures 17A and 17B.

[0323] These figures show a clear dependence between the dose of glofitamab administered and the level of cytokine induction: the magnitude of the peak cytokine fold change increases with the administration of higher doses of glofitamab.

[0324] Furthermore, the magnitude of the peak cytokine fold change correlates with the severity of the cytokine release syndrome: more aggressive cytokine release syndrome (represented by purple or red lines) is associated with higher peak cytokine levels.

[0325] Additionally, for IL-6, TNF-α, and IL-8, the magnitude of the peak cytokine level correlates with the timing of the peak cytokine level. More specifically, higher peak cytokine levels (and greater severity of cytokine release syndrome) are associated with earlier peak times.

[0326] For subjects who did not experience cytokine release syndrome, a dose-dependent relationship between on-treatment cytokine fold change and dose was observed for glofitamab at doses greater than 4 mg for IL-6, TNF-α, IL-8, and IL-10 cytokines. A dose-dependent relationship between on-treatment cytokine fold change and dose for MIPb was observed for glofitamab at doses greater than 2 mg. The mean subject-specific on-treatment cytokine fold changes for the 10 mg dose of glofitamab were 1.55-fold for IL-6, 2-fold for TNF-α, 1.55-fold for IL-8, 4-fold for MIPb, and 8-fold for IL-10. The mean subject-specific on-treatment cytokine fold changes for the 20 mg dose of glofitamab were 16-fold for IL-6, 8-fold for TNF-α, 4-fold for IL-8, 100-fold for MIPb, and 100-fold for IL-10.

[0327] For subjects who did not experience cytokine release syndrome, dose dependence began even at the lowest dose of glofitamab. On-treatment cytokine fold changes for IL-6 ranged in magnitude from 30 to 1000.

[0328] Figures 18A and 18B show the maximum log2-fold change in subjects for IL-6 and TNF-α, respectively. The data were divided based on whether any non-zero-grade cytokine release syndrome was observed in the first cycle of treatment. In each of Figures 18A and 18B, the lines in the left subplot correspond to subjects in whom cytokine release syndrome was not observed in the first cycle, while the lines in the right subplot correspond to subjects in whom cytokine release syndrome was observed in the first cycle. These line plots show that the highest peak in TNF-α was observed at mid-infusion (MI). Meanwhile, the highest peak in IL-6 occurred at the end of infusion (EOI) or 6 hours later.

[0329] IV.B.4. Timing of Dose Effects on Early Cytokine Changes and the Association with Cytokine Release Syndrome To explore the extent to which the dynamics of cytokine levels are related to the dynamics of cytokine release syndrome incidence, Figures 19A and 19B were generated to show exemplary time courses of cytokine fold change, while simultaneously stratifying subjects according to the time of onset of cytokine release syndrome relative to treatment initiation.

[0330] The columns correspond to different timings of any initial cytokine release syndrome. Specifically, column 1 corresponds to cases where cytokine release syndrome did not occur. Columns 2, 3, 4, and 5 correspond to cases where cytokine release syndrome occurred within 2 hours of initiating the therapeutic infusion, between 2 and 4 hours after initiating the infusion, between 4 and 10 hours after initiating the infusion, and more than 10 hours after initiating the infusion, respectively.

[0331] The different rows correspond to different glofitamab doses, with lower rows corresponding to higher doses.

[0332] Each line corresponds to a single subject and shows the cytokine fold change for a given cytokine over time (beginning of infusion). The color of the line represents the severity of cytokine release syndrome (dark green lines correspond to the absence of such an event).

[0333] The timing of cytokine release syndrome onset is indicated within the shaded region. Thus, a cytokine fold change greater than 0 within the unshaded region precedes cytokine release syndrome onset and can serve as an indicator of impending cytokine release syndrome.

[0334] For IL-6 (Figure 19A), positive fold changes were detected before symptoms of cytokine release syndrome in some, but not all, subjects. For TNF-α (Figure 19B), the majority of cases in which subjects experienced cytokine release syndrome were associated with peak cytokine fold changes in the time range preceding the onset of cytokine release syndrome, particularly with glofitamab doses above 1 mg.

[0335] To minimize the observed effect of dose on cytokine induction, a more focused analysis was performed processing only data corresponding to first glofitamab doses of 1.8 to 10 mg. Furthermore, to assess the accuracy of predictions, a "true" prediction (one of cytokine release syndrome occurrence) was recorded if the cytokine fold change was above 0 before the start of the infusion or at time point 4.0, and a "false" prediction was recorded if it was not.

[0336] For each of Figures 20A and 20B, the left subplot shows a subset of the data shown in Figures 19A and 19B (corresponding to the 1.8-10 mg dose range). As illustrated, in multiple cases, the cytokine fold change for cytokines did not cross the y=0 line at the 4 hour dosing time (and thus did not represent an increase in cytokine levels relative to baseline).

[0337] The right subplot shows boxplots comparing cytokine fold changes in cytokine levels in cases identified based on whether any type of cytokine release syndrome occurred or whether cytokine release syndrome of at least grade 2 occurred. These plots show that cytokine levels were higher in cases where cytokine release syndrome occurred (normal or at least grade 2).

[0338] The true positive, false positive, true negative and false negative statistics are further shown in Figures 20A and 20B. Predicted event occurrence was based on whether the cytokine log2 fold change exceeded zero in the x-axis range of ≦4 hours.

[0339] The data presented show that true positives outnumbered false positives and true negatives outnumbered false negatives for cytokines. Furthermore, sensitivity, specificity, positive predictive value, and negative predictive value for cytokines were almost all above chance (>0.5).

[0340] IV.B.5. Association between Cytokine Release Syndrome Risk Score and Changes in Cytokine Levels As described herein, the fold changes of various cytokines indicate the incidence and severity of cytokine release syndrome development. Furthermore, as described in the Examples (see, e.g., the results for "Risk Score" in Figure 8), the Cytokine Release Syndrome Risk Score (CRSRS) also predicts incidence.

[0341] Potentially, the prediction of fold-change cytokine levels partially or completely overlaps with the prediction of cytokine release syndrome risk scores, or potentially the combination of these variables (fold-change cytokine levels and risk scores) is more informative (and supports more accurate predictions) than either variable alone.

[0342] To investigate these issues, multidimensional plots were generated. Specifically, Figures 21A and 21B show scatter plots comparing the maximum log 2-fold change in cytokine levels across various cytokines (all doses) versus the cytokine release syndrome risk score. The cytokine release syndrome risk score was calculated for each subject according to the techniques disclosed in Section IV.A.5.b.

[0343] The CRSRS threshold was defined as 4.5 because a cytokine release syndrome risk score above 4.5 is considered to represent a higher risk of developing cytokine release syndrome of at least grade 2 compared to lower risk scores. The fold change threshold was also defined by first identifying the maximum cytokine fold change among all subjects associated with a cytokine release syndrome risk score below 4.5 and then averaging the values. The dashed lines in each of Figures 21A and 21B have a y value equal to the fold change threshold and extend to an x ​​range with a lower value equal to the CRSRS threshold.

[0344] For each cytokine, these thresholds were used to predict that at least Grade 2 cytokine release syndrome would occur if (1) the subject's cytokine release syndrome risk score was at least 4.5, and (2) the subject's maximum cytokine log2 fold change exceeded the fold change threshold. If either (or both) of these conditions were not met, the subject was predicted not to experience at least Grade 2 cytokine release syndrome. Thus, in each of Figures 20A and 20B, each data point above the dashed line was predicted to correspond to at least Grade 2 cytokine release syndrome, and each data point below or to the left of the dashed line was predicted not to correspond to at least Grade 2 cytokine release syndrome.

[0345] Each red or purple code above the dashed line (corresponding to grade 2, 3, or 4 cytokine release syndrome) is a true positive. Each red or purple code below or to the left of the dashed line is a false negative. Each green or blue code above the dashed line (corresponding to no cytokine release syndrome or grade 1 cytokine release syndrome) is a false positive. Each green or blue code below or to the left of the dashed line is a true negative.

[0346] For all five assessed cytokines, the majority of cytokine release syndromes of at least grade 2 were associated with cytokine release syndrome risk scores and maximum log2-fold changes above their respective thresholds. However, some false negatives were observed. At least some of the false negatives may be due to cytokines with kinetic profiles in which the peak fold change was not achieved during the infusion period.

[0347] Using both criteria (CRSRS and fold change thresholds) resulted in higher specificity values ​​compared to using either threshold individually, where each specificity value was defined as the number of true negatives relative to the sum of true negatives and false positives.

[0348] IV.B.6. Interpretation Both the incidence and severity of cytokine release syndrome, as well as the incidence and dose of cytokine induction, are strongly dose-dependent phenomena (see, e.g., Figures 16A, 16B, and 19A-19B). Thus, assessing the predictive value of cytokine signals for predicting the incidence and severity of cytokine release syndrome is extremely challenging in phase I nonrandomized studies without comparison groups and / or control for confounding factors.

[0349] For each of several cytokines (e.g., TNF-α, IL-8, M1P1b, IL-6, and IL-10), an association between on-therapy kinetics of the incidence and severity of cytokine release syndrome was observed.

[0350] When cytokine levels are evaluated alone to determine whether they predict the incidence or severity of cytokine release syndrome, cytokine fold changes provide reasonable predictions. The kinetics of some cytokines (e.g., IL-6) suggest that calculating cytokine fold changes using post-infusion and baseline levels may be more predictive of cytokine release syndrome than calculating cytokine fold changes using on-treatment and baseline levels of one or more cytokines. The magnitude of cytokine fold changes was relatively small for some cytokines (1.4- to 2-fold increases for TNF-α and IL-8). Given the small differences between these subject groups, it may be advantageous, or potentially even necessary, to develop an assay with high sensitivity to predict cytokine release syndrome based on cytokine levels with sufficient reliability or accuracy.

[0351] Cytokine changes alone cannot achieve reliable accuracy in predicting severe cytokine release syndrome (grade 2 or higher) in terms of both positive and negative predictive value. Improved predictive value can be achieved when early cytokine changes are combined with a baseline cytokine release syndrome risk score.

[0352] IV.C. Example 3: Exemplary Training and Use of a Multivariate Model to Predict the Occurrence of Cytokine Release Syndrome To determine the extent to which the cytokine release syndrome risk score can be used to reliably predict the incidence of cytokine release syndrome (grade 2 or higher), a score threshold for the score was determined. More specifically, a threshold was trained using training data to optimally separate cases in which at least grade 2 cytokine release syndrome was observed from cases in which no such event was observed or grade 1 cytokine release syndrome was observed. Figure 22 shows the results of a landmark analysis of how the probability of occurrence of grade 2 or higher cytokine release syndrome (adjusted for the first glofitamab dose) relates to the normalized version of the cytokine release syndrome risk score. Only events occurring after the end of the infusion are presented. Specifically, each data point represents a subject diagnosed with aggressive non-Hodgkin's lymphoma and treated with glofitamab. The color of the code represents the grade of cytokine release syndrome observed (in some cases, a dark green code represents no cytokine release syndrome observed). Jitter was introduced along the y-axis, meaning that the y-axis of the symbols does not represent any cytokine release syndrome or any characteristic of the subject.

[0353] Among the 89 subjects for whom cytokine data were available, comparison of the cytokine release syndrome risk score to the score threshold resulted in a prediction that 41 subjects (46%) were at high risk of experiencing grade 2 or higher cytokine release syndrome, and 48 subjects (54%) were at low risk of experiencing grade 2 or higher cytokine release syndrome. Of those predicted to be at high risk, 23 of these subjects experienced grade 2 or higher cytokine release syndrome, while 18 did not. Of those predicted to be at low risk, 4 of these subjects experienced grade 2 cytokine release syndrome (but none of these 4 subjects experienced grade 3 or higher cytokine release syndrome), and 44 of these subjects did not.

[0354] Figures 22 and 23 show how any observed cytokine release syndrome severity relates to both the TNF-α cytokine release syndrome risk score and cytokine fold change. Thus, each data point represented in Figure 22 has a corresponding data point represented in Figure 23 with the same x-axis value. However, the y-axis values ​​in Figure 23 represent TNF-α cytokine fold change, and Figure 23 shows the same score thresholds (corresponding to cytokine release syndrome risk scores) along the x-axis as shown in Figure 22.

[0355] Figure 23 further depicts along the y-axis two cytokine change thresholds (learned using the training data) corresponding to the cytokine fold change thresholds for TNF-α. Specifically, different cytokine change thresholds for TNF-α cytokine fold change were identified, and the cytokine change threshold selected for the cytokine release syndrome risk score was above the score threshold, whereas the cytokine change threshold for TNF-α cytokine fold change was selected for the cytokine release syndrome risk score.

[0356] In this analysis, a TNF-α cutoff was identified for each subject using a cytokine release syndrome risk score, where a threshold was selected to allow for discrimination between low- and high-risk subjects experiencing at least grade 2 cytokine release syndrome using a cytokine release syndrome score of 5. As shown in Figure 23, the majority of observed grade 2 or higher cytokine release syndrome cases were observed in subjects predicted to be at high risk (24 of 28). Furthermore, the majority of cases in which no grade 2 or higher cytokine release syndrome was observed in subjects predicted to be at low risk (54 subjects) were accurately predicted (with few false negatives). Thus, accuracy, precision, and recall generated based on both the TNF-α cytokine fold change and cytokine release syndrome risk score were better than those based on the cytokine release syndrome risk score alone.

[0357] IV.D. Example 4: Exemplary Baseline Feature Weights In Example 1, Table 9 shows weights assigned to the baseline feature set (or a binary transformation thereof), and the weights were used to generate a cytokine release syndrome risk score. However, in some instances, values ​​for all variables corresponding to the baseline features identified in Table 9 are not available. For example, tests to detect atypical cells in peripheral blood (e.g., blood smear tests) are not routinely performed. Furthermore, at the time treatment decisions are made, bone marrow samples (to determine bone marrow infiltration) may not be available, or infiltration analysis results may not be available.

[0358] Thus, a cytokine release syndrome risk score may be based on a reduced set of baseline characteristics. Table 12 identifies an exemplary reduced set of exemplary baseline characteristics. The weight of each characteristic in the reduced set of baseline characteristics was defined to be the same as the weight that would have been determined if the full set of baseline characteristics had been analyzed. TIFF2025169248000019.tif56170

[0359] The confidence in the prediction output can be reduced when a reduced set of baseline characteristics is used. Thus, the confidence cutoff for distinguishing between predicted occurrence of cytokine release syndrome and predicted non-occurrence of cytokine release syndrome was lowered from 5 to 4 based on analysis of data from the validation cohort (2.5 / 10 / 30 mg SUD).

[0360] Figure 24 shows the negative predictive value and low-risk detection rate for the cutoff score set in the SUD cohort (n=109, aNHL cases). The left panel shows data corresponding to the original cytokine release syndrome risk score (CRSRS calculated using the eight baseline characteristics identified in Table 9), and the right panel shows data corresponding to the reduced set of baseline characteristics (CRSRS.5p identified in Table 12).

[0361] Table 13 shows exemplary performance measures for predictions using each of the two adjusted confidence cutoffs (4 or 5) and each of the two sets of baseline features. Specifically, Table 13 shows performance measures when predictions are made using the baseline features identified in Table 9, and Table 13 shows performance measures when predictions are made using the baseline features identified in Table 12. Furthermore, the first row in each table corresponds to a cutoff of 4.0 (to convert real-valued outputs to binary predictions), and the second row in each table corresponds to a cutoff of 5.0. Missing values ​​were imputed with zero, thereby corresponding to a "base case" scenario that may underestimate baseline risk. Using an adjusted confidence cutoff of 4, the predictive performance of the reduced classifier was comparable to that of a classifier using eight baseline features. TIFF2025169248000020.tif255163

[0362] IV. Additional Considerations Some embodiments of the present disclosure include a system including one or more data processors. In some embodiments, the system includes a non-transitory computer-readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein. Some embodiments of the present disclosure include a computer program product tangibly embodied in a non-transitory machine-readable storage medium containing instructions configured to cause one or more data processors to perform some or all of one or more methods and / or some or all of one or more processes disclosed herein.

[0363] The terms and expressions which have been employed are used as terms of description rather than of limitation, and in the use of such terms and expressions there is no intention to exclude any equivalents of the features shown and described or portions thereof, recognizing that various modifications are possible within the scope of the claimed invention. Thus, although the claimed invention has been specifically disclosed by embodiments and optional features, it is to be understood that modifications and variations of the concepts disclosed herein may be resorted to by those skilled in the art, and that such modifications and variations are considered to be within the scope of the invention as defined by the appended claims.

[0364] This specification provides only preferred exemplary embodiments and is not intended to limit the scope, applicability, or configuration of the present disclosure. Rather, the description of preferred exemplary embodiments provides those skilled in the art with an enabling description for implementing various embodiments. It will be understood that various changes may be made in the function and arrangement of elements without departing from the spirit and scope as set forth in the appended claims.

[0365] Specific details are given in the following description to provide a thorough understanding of the embodiments. However, it will be understood that the embodiments may be practiced without these specific details. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form in order to avoid obscuring the embodiments in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail in order to avoid obscuring the embodiments.

[0366] Examples of potentially patentable subject matter include, but are not limited to, the following:

Claims

1. Identifying cytokine on-treatment levels, wherein the cytokine on-treatment levels represent the cytokine levels in on-treatment samples collected from subjects during or within one hour of the completion of treatment. Based on the on-treatment level of cytokines and the baseline level of cytokines, which indicates the level of cytokines in baseline samples collected from subjects before the initiation of treatment, the on-treatment cytokine ratio change of cytokines is determined. Using a risk score generation model, a numerical risk score for cytokine release syndrome is generated by processing changes in on-treatment cytokine ratios and dosages. To identify at least a portion of the dosage for the treatment, Based on changes in on-treatment cytokine ratios and dosage, the risk of a subject experiencing a threshold malignancy cytokine release syndrome with a threshold malignancy of 1 or greater, as defined by the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE v. 4.0), after receiving at least a portion of the treatment dosage, and To assist in determining outcomes based on predicted risk corresponding to recommendations on whether or not to monitor subjects through inpatient monitoring after completion of treatment, Outputting the results and A method that includes this.

2. Identifying a set of baseline characteristics, wherein the set of baseline characteristics, with respect to one or more baseline time points prior to the initiation of treatment, each of the baseline characteristics in the set of baseline characteristics is Tumor burden, Stage of cancer, Tumor spread, The size of one or more tumors, demographic attributes; White blood cell count, and / or Lactate dehydrogenase level This further includes identifying a set of baseline characteristics of the subject that characterize it, The method according to claim 1, wherein the predicted risk is further dependent on a set of baseline characteristics.

3. The method according to claim 2, further comprising generating a numerical cytokine release syndrome risk score by processing a set of baseline characteristics in a risk score generation model, wherein the predicted risk is based on the cytokine release syndrome risk score.

4. The method according to claim 3, wherein the risk score generation includes a regression model.

5. The method according to claim 3, wherein the risk is determined based on a linear combination of a cytokine release syndrome risk score and the dose, and the linear combination uses a set of weights.

6. The method according to claim 3, wherein predicting the risk of a subject experiencing cytokine release syndrome includes performing one or more threshold comparisons.

7. The results support the recommendation to monitor subjects through inpatient monitoring after completion of treatment, and the method is as follows: The method according to claim 1, further comprising monitoring the subject by inpatient monitoring at a medical facility for at least 24 hours after completion of treatment if the results indicate that the subject is at high risk of experiencing cytokine release syndrome.

8. The results support the recommendation to monitor subjects through outpatient monitoring after the completion of treatment, and the method is as follows: The method according to claim 1, further comprising monitoring the subject by outpatient monitoring if the results indicate that the subject is at low risk of experiencing cytokine release syndrome.

9. The method according to claim 1, wherein the subject has been diagnosed with cancer and the treatment comprises administering T-cell immunotherapy.

10. The method according to claim 1, wherein the subject has been diagnosed with cancer and the treatment comprises administering grofitamab or mosnetuzumab.

11. Determining the on-treatment cytokine ratio change of cytokines based on baseline cytokine levels is possible. Calculating the logarithm of the baseline level or its processed version of cytokines to generate baseline logarithmic values, Calculating the logarithm of the on-treatment level or the treated version of a cytokine to generate the on-treatment logarithm, Subtracting the baseline logarithm from the on-treatment logarithm and The method according to claim 1, including the method described in claim 1.

12. Determining the on-treatment cytokine ratio change of cytokines based on baseline cytokine levels is possible. Calculate the logarithm of the difference between the baseline level of cytokines and a constant to generate the baseline logarithm, The logarithm of the difference between the cytokine on-treatment level and a constant is calculated to generate the on-treatment logarithm, Subtracting the baseline logarithm from the on-treatment logarithm and The method according to claim 1, including the method described in claim 1.

13. Identifying the on-treatment level of cytokines is important. Identifying multiple preliminary on-treatment levels of cytokines, which indicate cytokine levels in multiple on-treatment samples collected from a subject during the administration of treatment or within one day after the completion of treatment, wherein each of the multiple on-treatment samples is collected at a different time point. The on-treatment level of a cytokine is defined as the maximum value among several preliminary on-treatment levels of the cytokine. The method according to claim 1, including the method described in claim 1.

14. The treatment includes administering an active ingredient, Prior to the treatment, a pre-treatment with another drug was administered. The method according to claim 1.

15. The method according to claim 14, wherein the on-treatment level is determined using a sample collected after administration of the active ingredient.

16. The method according to claim 1, wherein the cytokine comprises tumor necrosis factor alpha, interleukin 6, interleukin 8, interleukin 10, or macrophage inflammatory protein 1-beta.

17. The on-treatment level of cytokines is Collecting blood samples from the subject while treatment was being administered, and Processing blood samples using cytokine capture antibodies and detection antibodies. The method according to claim 1, as determined by [the relevant authority].

18. Determining baseline cytokine levels, which indicate the levels of cytokines in baseline samples collected from subjects before the initiation of treatment, Determining the on-treatment level of cytokines, wherein the on-treatment level of cytokines indicates the level of cytokines in on-treatment samples collected from subjects during the administration of treatment or within one hour after the completion of treatment. To identify at least a portion of the dosage for the treatment, Input the baseline level and on-treatment level of cytokines into the calculation system, To receive results that support the recommendation to monitor the subject through inpatient monitoring after the completion of treatment, The subjects will be monitored through inpatient monitoring after the completion of treatment. A method that includes this.

19. The method according to claim 18, wherein the subject is monitored by self-monitoring for at least four hours after the completion of treatment.

20. The result is, Based on the baseline level and on-treatment level of cytokines, the on-treatment cytokine ratio change of cytokines is determined, and To predict the risk of a subject experiencing cytokine release syndrome of at least threshold malignancy after receiving at least a portion of the treatment dose, based on changes in on-treatment cytokine ratios and dosage. The method according to claim 18, which is generated by a calculation system.

21. Determining baseline cytokine levels, which indicate the levels of cytokines in baseline samples collected from subjects before the initiation of treatment, Determining the on-treatment level of cytokines, wherein the on-treatment level of cytokines indicates the level of cytokines in on-treatment samples collected from subjects during the administration of treatment or within one hour after the completion of treatment. To identify at least a portion of the dosage for the treatment, Input the baseline level and on-treatment level of cytokines into the calculation system, To receive results that support the recommendation to monitor the subject through outpatient monitoring after the completion of treatment, The subjects will be monitored through outpatient monitoring after the completion of treatment. A method that includes this.

22. In response to receiving the results, The method according to claim 21, further comprising creating a queue for subjects to be discharged from a medical facility where they received treatment.

23. The result is, Based on the baseline level and on-treatment level of cytokines, the on-treatment cytokine ratio change of cytokines is determined, and To predict the risk of a subject experiencing cytokine release syndrome of at least threshold malignancy after receiving at least a portion of the treatment dose, based on changes in on-treatment cytokine ratios and dosage. The method according to claim 21, which is generated by a calculation system.

24. The use of computational prediction to assist in determining whether or not to monitor a subject for cytokine release syndrome after administration of treatment through inpatient monitoring, wherein the computational prediction is provided by a computing device that implements a risk score generation model for generating a numerical risk score for cytokine release syndrome, and the risk score generation model is Changes in cytokine on-treatment cytokine ratios, Baseline cytokine levels, which indicate the levels of cytokines in baseline samples collected from subjects before the start of treatment, and On-treatment cytokine levels indicate the levels of cytokines in on-treatment samples collected from subjects during the administration of treatment or within one hour of the completion of treatment. The decision is made based on the following: This method is used to predict the risk that a patient will experience a threshold malignancy cytokine release syndrome (THIS) with a threshold malignancy of 1 or greater, as defined by the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE v. 4.0), after administration of treatment based on changes in the on-treatment cytokine ratio.

25. One or more data processors, A non-temporary computer-readable storage medium that, when executed on one or more data processors, contains instructions causing one or more data processors to carry out the method according to any one of claims 1 to 17, A system that includes this.

26. A computer program product tangibly embodied in a non-temporary machine-readable storage medium, comprising instructions configured to cause one or more data processors to carry out the method according to any one of claims 1 to 17.

27. The method according to claim 1, 18, or 21, or the use according to claim 24, wherein the treatment comprises administering a therapy comprising an antibody or small molecule.

28. The method according to claim 1, 18, or 21, or the use according to claim 24, wherein the treatment comprises administering a therapy comprising an antibody or small molecule, and the administered therapy comprises an antibody.

29. The method according to claim 1, 18, or 21, or the use according to claim 24, wherein the treatment comprises administering a therapy comprising an antibody or small molecule, the administered therapy comprising a multispecific antibody that engages with T cells when bound to at least one of an antigen.

Images