Method for determining treatment for cancer patients

Abstract

The present invention relates to a method for determining a treatment for a cancer patient. The method includes measuring a variety of clinical pathological marker concentrations for gradient enhanced machine learning analysis.

Description

[0001] This application is a divisional application of Chinese patent application No. 201980082966.5, filed on October 18, 2019, “Method for determining treatment for cancer patients.” Technical Field

[0002] The present invention generally relates to methods of treating patients with cancer. Background Art

[0003] The development of new anticancer therapeutics, such as immuno-oncology (IO) agents and drugs targeting the DNA damage repair system, requires significant investment of time and resources to identify potentially effective drug candidates. However, before these candidates can be added to the existing armamentarium of approved anticancer therapies, they must be demonstrated to be both safe and effective in clinical trials. The initial step in this process is Phase I clinical trials, which focus on drug safety.

[0004] These Phase I clinical trials recruit patients who may potentially benefit from participation in the trial, typically because these patients have failed standard of care treatments. In many clinical trials, regardless of the type of treatment being tested or its intended indication, a subset of participants are unable to complete the study for a variety of reasons. For example, some patients with significant advanced underlying disease may enter a clinical trial or may have secondary or other pathological conditions unrelated to their cancer.

[0005] In the context of Phase I IO clinical trials (e.g., studies of PD-1 or PD-L1 (PDx) monotherapy), almost 20% of patients enrolled die within 2-3 months of study entry, even though ideally, most such trials last longer than 2-3 months. Therefore, to participate in Phase I IO clinical trials, patients should have a life expectancy of more than 24 weeks and not be at risk for early mortality (EM; defined as a life expectancy of less than 24 weeks). The EM phenomenon has been observed in numerous randomized clinical trials comparing ICIs with active comparators in patients with advanced or metastatic cancer, even with an overall benefit ultimately favoring ICI treatment (Champiat et al. 2018). While the exact etiology of this phenomenon has not been definitively determined, it is characterized by a seemingly disproportionately higher mortality rate during the early treatment period, during which survival favors the control group, followed by a subsequent benefit in overall survival (OS) favoring the ICI treatment group. This is often reflected in clinical data by a "crossover of the Kaplan-Meier curve," indicating a subpopulation of patients at higher risk for EM whose progressive rate of tumor growth may require the cytotoxic tumor reduction effect of chemotherapy.

[0006] Therefore, better prediction of a given patient's risk of early death to better inform appropriate treatment for their individual clinical status has significant therapeutic implications. Unexpected rapid disease progression or worsening of another underlying pathological condition in a significant number of patients participating in Phase I clinical trials can reduce the ability to detect early efficacy signals of the tested therapeutic agent, delay dose escalation due to premature patient withdrawal, confound analysis of predictive biomarkers of efficacy, and weaken efficacy relative to comparators in randomized trials. This can also prevent patients from receiving potentially life-prolonging therapeutic interventions outside of Phase I clinical trials. If such patients were to receive therapeutic interventions that correct or alleviate the underlying condition, they could fully benefit from trial participation. Therefore, there is a need for improved prognostic methods for predicting EM in patients undergoing clinical trials of anticancer therapeutics.

[0007] Immune checkpoint inhibitors (ICIs) are dramatically changing the treatment of many types of cancer, including melanoma, non-small cell lung cancer (NSCLC), renal cell carcinoma (RCC), squamous cell carcinoma of the head and neck (SCCHN), urothelial carcinoma (UC), and Hodgkin lymphoma, and are associated with long-term tumor responses. To aid this goal, we present a new prognostic score, the Immuno-Instant Index (3i score), which uses pre-treatment measurements of routinely collected blood-based factors to predict patients' risk of early death and optimize benefit-risk prediction for patients treated with ICIs. Summary of the Invention

[0008] As described below, in a first aspect, the present invention provides a method of treating a patient having cancer, comprising: a) determining the concentrations of a plurality of clinical pathology markers in the patient; b) providing the determined concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; c) receiving a prediction from the gradient boosting machine learning model as to whether the patient is likely to die within 12 weeks; and d) administering an anti-cancer therapeutic to the patient if the prediction indicates that the patient is unlikely to die within 12 weeks.

[0009] In one embodiment of the first aspect, the plurality of clinical pathological markers includes one or more of the following: age, albumin (g / L), alanine phosphatase (U / L), alanine aminotransferase (U / L), aspartate aminotransferase (U / L), basophils (10 3 / μL), basophils / leukocytes (%), bilirubin (μmol / L), body mass index, calcium (mmol / L), chloride (mmol / L), eosinophils (10 3 / μL), eosinophils / leukocytes (%), γ-glutamyl transferase (U / L), glucose (mmol / L), hematocrit (%), hemoglobin (g / L), potassium (mmol / L), lactate dehydrogenase (U / L), liver metastasis (yes / no), lymphocytes (103 / μL), lymphocytes / leukocytes (%), magnesium (mmol / L), monocytes (10 3 / μL), monocytes / leukocytes (%), neutrophils (10 3 / μL), neutrophils / leukocytes (%), neutrophils / lymphocytes (%), platelets (10 3 / μL), protein (g / L), gender, sodium (mmol / L), thyroid stimulating hormone (mU / L), tumor size, uric acid (mmol / L) and white blood cells (10 3 / μL).

[0010] In another embodiment of the first aspect, the plurality of clinical pathological markers includes neutrophils / leukocytes (%), neutrophils (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L).

[0011] In an embodiment of the first aspect, the method further comprises taking a sample from the patient.

[0012] In one embodiment of the first aspect, the sample comprises tissue biopsy, cerebrospinal fluid, lymph, whole blood, serum, blood cells, urine, sweat, tears, saliva and / or stool.

[0013] In one embodiment of the first aspect, the concentration of one or more of the plurality of clinical pathology markers is determined from a sample taken from the patient.

[0014] In one embodiment of the first aspect, the anti-cancer therapeutic comprises cisplatin, gemcitabine, methotrexate, vinblastine, doxorubicin, cisplatin (MVAC), carboplatin, taxanes, temozolomide, dacarbazine, vinflunine, docetaxel, paclitaxel, nab-paclitaxel, vemurafenib, erlotinib, afatinib, cetuximab, bevacizumab, gefitinib and / or pemetrexed.

[0015] In one embodiment of the first aspect, the anti-cancer therapeutic comprises an immune checkpoint inhibitor.

[0016] In one embodiment of the first aspect, the immune checkpoint inhibitor comprises an anti-CTLA-4 antibody, an anti-PD-1 antibody, and / or an anti-PD-L1 antibody.

[0017] In one embodiment of the first aspect, the anti-CTLA-4 antibody is tremelimumab or ipilimumab.

[0018] In one embodiment of the first aspect, the anti-PD-1 antibody is REGN2810, SHR1210, IBI308, PDR001, nivolumab, pembrolizumab, anti-PD-1, BGB-A317, BCD-100, or JS001.

[0019] In one embodiment of the first aspect, the anti-PD-L1 antibody comprises durvalumab, avelumab, atezolizumab, or KNO35.

[0020] In another embodiment of the first aspect, the method further comprises determining a cancer type of the patient, and further providing the determined cancer type to the gradient boosting machine learning model.

[0021] In a second aspect, the present invention provides a method of treating a patient having cancer, comprising: a) determining the concentrations of a plurality of clinical pathology markers in the patient; b) providing the determined concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; c) receiving a prediction from the gradient boosting machine learning model as to whether the patient has poor immune fitness; and d) administering an anti-cancer therapeutic to the patient if the prediction indicates that the patient does not have poor immune fitness.

[0022] In an embodiment of the second aspect, the method further comprises e) reducing at least one of cancer cell division rate, tumor growth, tumor size, tumor density, or tumor metastasis rate in the patient.

[0023] In a third aspect, the present invention provides a method for treating a patient having cancer, comprising: a) determining the concentrations of multiple clinical pathology markers in the patient; b) providing the determined concentrations of the multiple clinical pathology markers to a gradient boosting machine learning model; c) receiving a prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; d) if the prediction indicates that the patient has poor immune fitness, administering a therapeutic agent to the patient to enhance the patient's immune fitness; e) determining a second concentration of the multiple clinical pathology markers in the patient; f) providing the determined second concentrations of the multiple clinical pathology markers to the gradient boosting machine learning model; g) receiving a second prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; and h) if the prediction indicates that the patient does not have poor immune fitness, administering an anti-cancer therapeutic to the patient.

[0024] In a fourth aspect, the present invention provides a method for conducting a clinical trial of an anti-cancer therapeutic agent, comprising: a) identifying a candidate participant for the clinical trial, wherein the candidate participant has cancer; b) determining the concentrations of multiple clinical pathology markers in the patient; c) providing the determined concentrations of the multiple clinical pathology markers to a gradient boosting machine learning model; d) receiving a prediction from the gradient boosting machine learning model as to whether the patient is likely to die within 12 weeks; and e) if the prediction indicates that the patient is unlikely to die within 12 weeks, enrolling the patient in the clinical trial of the anti-cancer therapeutic agent.

[0025] In one embodiment of the fourth aspect, the anti-cancer therapeutic is an immune checkpoint inhibitor.

[0026] In a fifth aspect, the present invention provides a method for predicting the effectiveness of treating a cancer patient with an immune checkpoint inhibitor, comprising: a) determining the concentrations of multiple clinical pathology markers in the patient; b) providing the determined concentrations of multiple clinical pathology markers to a gradient boosting machine learning model; c) receiving a prediction from the gradient boosting machine learning model as to whether the patient lacks immune fitness, wherein a prediction indicating that the patient does not lack immune fitness predicts that the patient can be treated more effectively with an immune checkpoint inhibitor (ICI) than a patient predicted to lack immune fitness.

[0027] In a sixth aspect, the present invention provides a method for determining a treatment regimen for a cancer patient, comprising: a) determining the concentrations of multiple clinical pathology markers in the patient; b) providing the determined concentrations of multiple clinical pathology markers to a gradient boosting machine learning model; c) receiving a prediction from the gradient boosting machine learning model as to whether the patient lacks immune fitness, wherein the prediction indicating that the patient does not lack immune fitness suggests a treatment regimen including an immune checkpoint inhibitor (ICI), and wherein the prediction indicating that the patient lacks immune fitness suggests a treatment regimen without using an immune checkpoint inhibitor (ICI).

[0028] In a seventh aspect, a computing system includes a computer memory configured to store model training data, wherein the model training data relates to a first group of cancer patients undergoing cancer treatment, wherein the model training data correlates (i) results of laboratory tests performed on the first group of cancer patients and the tumor types of the first group of cancer patients with (ii) whether an individual from the first group of cancer patients died within a threshold number of weeks after starting cancer treatment. The computing system also includes one or more processors in communication with the computer memory and configured to execute program instructions to: (i) train a continuous series of decision trees based on the model training data until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on the residual values of its previous decision tree in the continuous series of decision trees; and (ii) generate a gradient boosting machine learning model as a summed function of the continuous series of decision trees, wherein the gradient boosting machine learning model predicts whether another cancer patient will die within the threshold number of weeks based on the summed function applied to the results of laboratory tests performed on another cancer patient and the tumor type of the other cancer patient.

[0029] In an eighth aspect, a computer-implemented method involves obtaining model training data, wherein the model training data relates to a first group of cancer patients undergoing cancer treatment, wherein the model training data associates (i) the results of laboratory tests performed on the first group of cancer patients and the tumor types of the first group of cancer patients with (ii) whether an individual from the first group of cancer patients died within a threshold number of weeks after starting cancer treatment. The computer-implemented method also involves training a continuous series of decision trees based on the model training data until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on the residual values of its previous decision tree in the continuous series of decision trees. The computer-implemented method also involves generating a gradient boosting machine learning model as a summed function of the continuous series of decision trees, wherein the gradient boosting machine learning model predicts whether another cancer patient will die within a threshold number of weeks based on the summed function applied to the results of laboratory tests performed on another cancer patient and the tumor type of the other cancer patient.

[0030] In a ninth aspect, an article of manufacture comprises a non-transitory computer-readable medium having program instructions stored thereon, the program instructions, when executed by a computing device, cause the computing device to perform the operations of the eighth aspect.

[0031] In a tenth aspect, a computing system includes a computer memory configured to store a gradient boosting machine learning model, wherein the gradient boosting machine learning model is trained on model training data, wherein the model training data includes a first group of cancer patients undergoing cancer treatment, wherein the model training data correlates (i) results of laboratory tests performed on the first group of cancer patients and the tumor type of the first group of cancer patients with (ii) whether individuals from the first group of cancer patients died within a threshold number of weeks after starting cancer treatment, wherein training the gradient boosting machine learning model involves: (i) training a continuous series of decision trees until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on residual values of its previous decision tree in the continuous series of decision trees; and (ii) determining a sum function for the continuous series of decision trees. The computing system also includes one or more processors in communication with the computer memory and configured to execute program instructions to: (i) obtain laboratory test results performed on another cancer patient; (ii) obtain a tumor type of the other cancer patient; (iii) apply the gradient boosting machine learning model to the results of the laboratory test performed on the other cancer patient and the tumor type of the other cancer patient; and (iv) receive a prediction from the gradient boosting machine learning model of whether the other cancer patient will die within a threshold number of weeks.

[0032] In an eleventh aspect, a computer-implemented method involves obtaining a gradient boosting machine learning model, wherein the gradient boosting machine learning model is trained on model training data, wherein the model training data includes a first group of cancer patients undergoing cancer treatment, wherein the model training data associates (i) the results of laboratory tests performed on the first group of cancer patients and the tumor types of the first group of cancer patients with (ii) whether an individual from the first group of cancer patients died within a threshold number of weeks after starting cancer treatment, wherein training the gradient boosting machine learning model involves: (i) training a continuous series of decision trees until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on the residual values of its previous decision tree in the continuous series of decision trees; and (ii) determining a sum function of the continuous series of decision trees. The computer-implemented method also includes obtaining the results of laboratory tests performed on another cancer patient. The computer-implemented method also includes obtaining the tumor type of the other cancer patient. The computer-implemented method also includes applying the gradient boosting machine learning model to the results of laboratory tests performed on the other cancer patient and the tumor type of the other cancer patient. The computer-implemented method also includes receiving a prediction from the gradient boosting machine learning model of whether the other cancer patient died within the threshold number of weeks.

[0033] In a twelfth aspect, an article of manufacture includes a non-transitory computer-readable medium having program instructions stored thereon, the program instructions, when executed by a computing device, cause the computing device to perform the operations of the eleventh aspect.

[0034] The present invention relates to the following technical solutions:

[0035] 1. A method of treating a patient suffering from cancer, the method comprising:

[0036] a) determining the concentrations of multiple clinical pathological markers of the patient;

[0037] b) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model;

[0038] c) receiving a prediction from the gradient boosting machine learning model as to whether the patient is likely to die within 12 weeks; and

[0039] d) administering the anti-cancer therapeutic to the patient if the prognosis indicates that the patient is unlikely to die within 12 weeks.

[0040] 2. The method of claim 1, wherein the plurality of clinical pathological markers comprises one or more of the following: age, albumin (g / L), alanine phosphatase (U / L), alanine aminotransferase (U / L), aspartate aminotransferase (U / L), basophils (10 3 / μL), basophils / leukocytes (%), bilirubin (μmol / L), body mass index, calcium (mmol / L), chloride (mmol / L), eosinophils (10 3 / μL), eosinophils / leukocytes (%), γ-glutamyl transferase (U / L), glucose (mmol / L), hematocrit (%), hemoglobin (g / L), potassium (mmol / L), lactate dehydrogenase (U / L), liver metastasis (yes / no), lymphocytes (10 3 / μL), lymphocytes / leukocytes (%), magnesium (mmol / L), monocytes (10 3 / μL), monocytes / leukocytes (%), neutrophils (10 3 / μL), neutrophils / leukocytes (%), neutrophils / lymphocytes (%), platelets (10 3 / μL), protein (g / L), gender, sodium (mmol / L), thyroid stimulating hormone (mU / L), tumor size, uric acid (mmol / L) and white blood cells (10 3 / μL).

[0041] 3. The method of scheme 2, wherein the multiple clinical pathological markers include neutrophils / leukocytes (%), neutrophils (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L).

[0042] 4. The method of Option 1, further comprising taking a sample from the patient.

[0043] 5. The method of claim 4, wherein the sample comprises tissue biopsy, cerebrospinal fluid, lymph, whole blood, serum, blood cells, urine, sweat, tears, saliva and / or feces.

[0044] 6. The method of claim 5, wherein the concentration of one or more of the plurality of clinical pathology markers is determined from the sample taken from the patient.

[0045] 7. The method of claim 1, wherein the anticancer therapeutic comprises cisplatin, gemcitabine, methotrexate, vinblastine, doxorubicin, cisplatin (MVAC), carboplatin, a taxane, temozolomide, dacarbazine, vinflunine, docetaxel, paclitaxel, nab-paclitaxel, vemurafenib, erlotinib, afatinib, cetuximab, bevacizumab, gefitinib, and / or pemetrexed.

[0046] 8. The method of Scheme 1, wherein the anti-cancer therapeutic comprises an immune checkpoint inhibitor.

[0047] 9. The method of claim 8, wherein the immune checkpoint inhibitor comprises an anti-CTLA-4 antibody, an anti-PD-1 antibody, and / or an anti-PD-L1 antibody.

[0048] 10. The method of claim 9, wherein the anti-CTLA-4 antibody is tremelimumab or ipilimumab.

[0049] 11. The method of claim 9, wherein the anti-PD-1 antibody is REGN2810, SHR1210, IBI308, PDR001, nivolumab, pembrolizumab, anti-PD-1, BGB-A317, BCD-100, or JS001.

[0050] 12. The method of claim 9, wherein the anti-PD-L1 antibody comprises durvalumab, avelumab, atezolizumab, or KNO35.

[0051] 13. The method of Option 1 further comprises determining the patient's cancer type and providing the determined cancer type to the gradient boosting machine learning model.

[0052] 14. A method of treating a patient suffering from cancer, the method comprising:

[0053] a) determining the concentrations of multiple clinical pathological markers of the patient;

[0054] b) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model;

[0055] c) receiving a prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; and

[0056] d) administering an anti-cancer therapeutic to the patient if the prediction indicates that the patient does not have poor immune fitness.

[0057] 15. The method of claim 14, further comprising e) reducing at least one of cancer cell division rate, tumor growth, tumor size, tumor density, or tumor metastasis rate in the patient.

[0058] 16. A method of treating a patient suffering from cancer, the method comprising:

[0059] a) determining the concentrations of multiple clinical pathological markers of the patient;

[0060] b) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model;

[0061] c) receiving a prediction from the gradient boosting machine learning model whether the patient has poor immune fitness;

[0062] d) if the prediction indicates that the patient has poor immune fitness, administering a therapeutic agent to the patient to enhance the patient's immune fitness;

[0063] e) determining a second concentration of a plurality of clinical pathology markers in the patient;

[0064] f) providing the determined second concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model;

[0065] g) receiving a second prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; and

[0066] h) administering an anti-cancer therapeutic to the patient if the prognosis indicates that the patient does not have poor immune fitness.

[0067] 17. A method for conducting a clinical trial of an anticancer therapeutic agent, the method comprising:

[0068] a) identifying a candidate participant for a clinical trial, wherein the candidate participant has cancer;

[0069] b) determining the concentrations of multiple clinical pathological markers in the patient;

[0070] c) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model;

[0071] d) receiving a prediction from the gradient boosting machine learning model as to whether the patient is likely to die within 12 weeks; and

[0072] e) enrolling the patient in a clinical trial of the anti-cancer therapeutic agent if the prognosis indicates that the patient is unlikely to die within 12 weeks.

[0073] 18. The method of claim 17, wherein the anti-cancer therapeutic agent is an immune checkpoint inhibitor.

[0074] 19. A computing system comprising:

[0075] a computer memory configured to store model training data, wherein the model training data relates to a first group of cancer patients undergoing cancer treatment, wherein the model training data correlates (i) results of laboratory tests performed on the first group of cancer patients and tumor types of the first group of cancer patients with (ii) whether an individual from the first group of cancer patients died within a threshold number of weeks after starting cancer treatment; and

[0076] One or more processors in communication with the computer memory and configured to execute program instructions to:

[0077] training a continuous series of decision trees based on the model training data until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on residual values of its previous decision tree in the continuous series of decision trees; and

[0078] A gradient boosting machine learning model is generated as a summed function of the continuous series of decision trees, wherein the gradient boosting machine learning model predicts whether the other cancer patient will die within the threshold number of weeks based on the summed function applied to a result of a laboratory test performed on the other cancer patient and the tumor type of the other cancer patient.

[0079] 20. The computing system of claim 19, wherein the threshold number of weeks is 12 weeks.

[0080] 21. The computing system of claim 19, wherein the results of the laboratory tests provide one or more of: neutrophil to lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration, or gamma-glutamyl transferase concentration.

[0081] 22. The computing system of claim 21, wherein the neutrophil-to-lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration, or gamma-glutamyl transferase concentration is selected from 20 or more laboratory tests performed on the first group of cancer patients, wherein the selection comprises:

[0082] training the gradient boosting machine learning model on the 20 or more laboratory tests;

[0083] For each corresponding laboratory test, training a corresponding variation of the gradient boosting machine learning model on the 20 or more laboratory tests with the corresponding laboratory test removed; and

[0084] The predictive value of each of the 20 or more laboratory tests is determined by comparing the accuracy of the variations of the gradient boosting machine learning model.

[0085] 23. The computing system of claim 19, wherein the tumor type is one of: non-small cell lung cancer, urothelial carcinoma, head and neck squamous cell carcinoma, gastric cancer, renal cell carcinoma, or Hodgkin lymphoma.

[0086] 24. A computing system as described in claim 19, wherein the sum function applies a multiplicative learning rate to the results of each of the decision trees from the continuous series.

[0087] 25. The computing system of claim 19, wherein predicting whether another cancer patient will die within a threshold number of weeks comprises:

[0088] The probability that the other cancer patient will die within a threshold number of weeks is determined by using a summation function.

[0089] 26. The computing system of claim 25, wherein predicting whether the other cancer patient will die within a threshold number of weeks further comprises:

[0090] Based on the probability and the cutoff value, the other cancer patient is placed in a high risk of death category or a low risk of death category.

[0091] 27. A computing system as described in embodiment 26, wherein the cutoff value is selected so that the false positive rate does not exceed 5%-20% when predicting whether the first group of cancer patients will die within a threshold number of weeks.

[0092] 28. A computing system as described in embodiment 26, wherein cancer patients in a high risk of death category are excluded from clinical trials for treating cancer, and wherein cancer patients in a low risk of death category are included in clinical trials for treating cancer.

[0093] 29. The computing system of claim 19, wherein the stopping condition is that the continuous series of decision trees includes a threshold number of decision trees.

[0094] 30. A computing system as described in Option 19, wherein the stopping condition is that the residual values of a specific decision tree in the continuous series of decision trees are all lower than a threshold.

[0095] 31. The computing system of claim 19, wherein the computer memory is further configured to store model adjustment data, wherein the model adjustment data relates to a second group of cancer patients undergoing cancer treatment, wherein the model adjustment data relates (i) results of laboratory tests performed on the second group of cancer patients and the tumor type of the second group of cancer patients to (ii) whether the second group of cancer patients died within a threshold number of weeks after starting cancer treatment; and

[0096] The one or more processors are further configured to execute other program instructions to:

[0097] generating two or more variations of a gradient boosting machine learning model by training a sequential series of decision trees for each variation with a different set of hyperparameters;

[0098] Apply these changes to the model training data and the model tuning data; and

[0099] A specific variation is selected from these variations as a gradient boosting machine learning model, which provides a prediction of whether the first group of cancer patients and the second group of cancer patients die within a threshold number of weeks after starting cancer treatment within a threshold accuracy.

[0100] 32. A computing system as described in embodiment 31, wherein the threshold accuracy is based on a true positive rate or a true negative rate of predicting whether the first group of cancer patients will die within a threshold number of weeks after the start of cancer treatment.

[0101] 33. A computing system as described in embodiment 31, wherein the threshold accuracy is based on a p-value of the interaction between: (i) the prediction of whether the second group of cancer patients dies within a threshold number of weeks after starting cancer treatment, and (ii) PD-L1 positive patients in the second group of cancer patients.

[0102] 34. A computing system as described in claim 31, wherein the hyperparameters include one or more of the following: the maximum depth of the decision tree is set to 2, 3, 4, 5 or 6, the minimum loss reduction required to split a node in the decision tree is set to 0, 1, 2, 3 or 4, the minimum sum of the weights of all observations required in child nodes in the decision tree is set between 3.0 and 7.0, the fraction of input parameters randomly sampled for each decision tree is set between 0.7 and 1.0, the fraction of observations randomly sampled for each decision tree is set between 0.4 and 0.8, and the number of decision trees in a continuous series of decision trees is set between 10 and 25.

[0103] 35. A computer-implemented method, the computer-implemented method comprising:

[0104] obtaining model training data, wherein the model training data relates to a first group of cancer patients undergoing cancer treatment, wherein the model training data correlates (i) results of laboratory tests performed on the first group of cancer patients and tumor types of the first group of cancer patients with (ii) whether individuals from the first group of cancer patients died within a threshold number of weeks after starting the cancer treatment;

[0105] training a continuous series of decision trees based on the model training data until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on residual values of its previous decision tree in the continuous series of decision trees; and

[0106] A gradient boosting machine learning model is generated as a summed function of the continuous series of decision trees, wherein the gradient boosting machine learning model predicts whether the other cancer patient will die within the threshold number of weeks based on the summed function applied to a result of a laboratory test performed on the other cancer patient and the tumor type of the other cancer patient.

[0107] 36. The computer-implemented method of claim 35, wherein the threshold number of weeks is 12 weeks.

[0108] 37. The computer-implemented method of claim 35, wherein the results of the laboratory tests provide one or more of: neutrophil-to-lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration, or gamma-glutamyl transferase concentration.

[0109] 38. The computer-implemented method of claim 37, wherein the neutrophil-to-lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration, or gamma-glutamyl transferase concentration is selected from 20 or more laboratory tests performed on the first group of cancer patients, wherein the selection comprises:

[0110] training the gradient boosting machine learning model on the 20 or more laboratory tests;

[0111] For each corresponding laboratory test, training a corresponding variation of the gradient boosting machine learning model on the 20 or more laboratory tests with the corresponding laboratory test removed; and

[0112] The predictive value of each of the 20 or more laboratory tests is determined by comparing the accuracy of the variations of the gradient boosting machine learning model.

[0113] 39. The computer-implemented method of claim 35, wherein the tumor type is one of: non-small cell lung cancer, urothelial carcinoma, head and neck squamous cell carcinoma, gastric cancer, renal cell carcinoma, or Hodgkin lymphoma.

[0114] 40. A computer-implemented method as described in claim 35, wherein the sum function applies a multiplicative learning rate to the results of each of the decision trees from the continuous series.

[0115] 41. The computer-implemented method of claim 35, wherein predicting whether another cancer patient will die within a threshold number of weeks comprises:

[0116] The probability that the other cancer patient will die within a threshold number of weeks is determined by using a summation function.

[0117] 42. The computer-implemented method of claim 41, wherein predicting whether the other cancer patient will die within a threshold number of weeks further comprises:

[0118] Based on the probability and the cutoff value, the other cancer patient is placed in a high risk of death category or a low risk of death category.

[0119] 43. A computer-implemented method as described in embodiment 42, wherein the cutoff value is selected so that the false positive rate does not exceed 5%-20% when predicting whether the first group of cancer patients will die within a threshold number of weeks.

[0120] 44. The computer-implemented method of claim 42, wherein cancer patients in a high risk of death category are excluded from clinical trials for treating cancer, and wherein cancer patients in a low risk of death category are included in clinical trials for treating cancer.

[0121] 45. A computer-implemented method as described in claim 35, wherein the stopping condition is that the continuous series of decision trees includes a threshold number of decision trees.

[0122] 46. A computer-implemented method as described in Option 35, wherein the stopping condition is that the residual values of a specific decision tree in the continuous series of decision trees are all lower than a threshold.

[0123] 47. The computer-implemented method of claim 35, further comprising:

[0124] obtaining model-adjusted data, wherein the model-adjusted data relates to a second group of cancer patients undergoing cancer treatment, wherein the model-adjusted data relates (i) results of laboratory tests performed on the second group of cancer patients and tumor types of the second group of cancer patients to (ii) whether the second group of cancer patients died within a threshold number of weeks after starting cancer treatment;

[0125] generating two or more variations of a gradient boosting machine learning model by training a sequential series of decision trees for each variation with a different set of hyperparameters;

[0126] applying the changes to the model training data and the model tuning data; and selecting a specific change from the changes as a gradient boosting machine learning model that provides a prediction of whether the first group of cancer patients and the second group of cancer patients died within a threshold number of weeks after starting cancer treatment within a threshold accuracy.

[0127] 48. The computer-implemented method of claim 47, wherein the threshold accuracy is based on a true positive rate or a true negative rate of predicting whether the first group of cancer patients will die within a threshold number of weeks after initiation of cancer treatment.

[0128] 49. The computer-implemented method of claim 47, wherein the threshold accuracy is based on a p-value for an interaction between: (i) the prediction of whether the second group of cancer patients dies within a threshold number of weeks after starting cancer treatment, and (ii) the number of PD-L1-positive patients in the second group of cancer patients.

[0129] 50. A computer-implemented method as described in claim 47, wherein the hyperparameters include one or more of the following: the maximum depth of the decision tree is set to 2, 3, 4, 5 or 6, the minimum loss reduction required to split a node in the decision tree is set to 0, 1, 2, 3 or 4, the minimum sum of the weights of all observations required in child nodes in the decision tree is set between 3.0 and 7.0, the fraction of input parameters randomly sampled for each decision tree is set between 0.7 and 1.0, the fraction of observations randomly sampled for each decision tree is set between 0.4 and 0.8, and the number of decision trees in a continuous series of decision trees is set between 10 and 25.

[0130] 51. An article of manufacture comprising a non-transitory computer-readable medium having program instructions stored thereon, the program instructions, when executed by a computing device, causing the computing device to perform the operations of any one of schemes 35-50.

[0131] 52. A computing system, the computing system comprising:

[0132] a computer memory configured to store a gradient boosting machine learning model, wherein the gradient boosting machine learning model is trained on model training data,

[0133] wherein the model training data comprises a first group of cancer patients undergoing cancer treatment, wherein the model training data correlates (i) results of laboratory tests performed on the first group of cancer patients and tumor types of the first group of cancer patients with (ii) whether individuals from the first group of cancer patients died within a threshold number of weeks after starting cancer treatment,

[0134] wherein training the gradient boosting machine learning model involves: (i) training a continuous series of decision trees until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on the residual values of its previous decision tree in the continuous series of decision trees; and (ii) determining a sum function of the continuous series of decision trees; and

[0135] One or more processors in communication with the computer memory and configured to execute program instructions to:

[0136] Obtaining the results of laboratory tests performed on another cancer patient;

[0137] obtaining the tumor type of the other cancer patient;

[0138] applying the gradient boosting machine learning model to the results of the laboratory test performed on the other cancer patient and the tumor type of the other cancer patient; and

[0139] Receive a prediction from the gradient boosting machine learning model whether the other cancer patient died within the threshold number of weeks.

[0140] 53. The computing system of claim 52, wherein the threshold number of weeks is 12 weeks.

[0141] 54. The computing system of claim 52, wherein the results of the laboratory tests provide one or more of: neutrophil to lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration, or gamma-glutamyl transferase concentration.

[0142] 55. The computing system of claim 52, wherein the tumor type is one of: non-small cell lung cancer, urothelial carcinoma, head and neck squamous cell carcinoma, gastric cancer, renal cell carcinoma, or Hodgkin lymphoma.

[0143] 56. A computing system as described in claim 52, wherein the sum function applies a multiplicative learning rate to the results of each of the decision trees from the continuous series.

[0144] 57. The computing system of claim 52, wherein predicting whether another cancer patient will die within a threshold number of weeks comprises:

[0145] The probability that the other cancer patient will die within a threshold number of weeks is determined by using a summation function.

[0146] 58. The computing system of claim 57, wherein predicting whether the other cancer patient will die within a threshold number of weeks further comprises:

[0147] Based on the probability and the cutoff value, the other cancer patient is placed in a high risk of death category or a low risk of death category.

[0148] 59. The computing system of claim 58, wherein the cutoff value is selected so that the false positive rate does not exceed 5%-20% when predicting whether the first group of cancer patients will die within a threshold number of weeks.

[0149] 60. The computing system of claim 58, wherein cancer patients in a high risk of death category are excluded from clinical trials for treating cancer, and wherein cancer patients in a low risk of death category are included in clinical trials for treating cancer.

[0150] 61. A computing system as described in embodiment 52, wherein the stopping condition is that the continuous series of decision trees includes a threshold number of decision trees.

[0151] 62. A computing system as described in Option 52, wherein the stopping condition is that the residual values of a specific decision tree in the continuous series of decision trees are all lower than a threshold.

[0152] 63. The computing system of claim 52, wherein the gradient boosting machine learning model is also trained using model adjustment data,

[0153] wherein the model-adjusted data relates to a second group of cancer patients who underwent cancer treatment, wherein the model-adjusted data relates (i) results of laboratory tests performed on the second group of cancer patients and the tumor type of the second group of cancer patients to (ii) whether the second group of cancer patients died within a threshold number of weeks after starting cancer treatment; and

[0154] The use of the model to adjust data and train the gradient boosting machine learning model includes:

[0155] generating two or more variations of a gradient boosting machine learning model by training a sequential series of decision trees for each variation with a different set of hyperparameters;

[0156] applying the changes to the model training data and the model tuning data; and selecting a specific change from the changes as a gradient boosting machine learning model that provides a prediction of whether the first group of cancer patients and the second group of cancer patients died within a threshold number of weeks after starting cancer treatment within a threshold accuracy.

[0157] 64. The computing system of claim 63, wherein the threshold accuracy is based on a true positive rate or a true negative rate of predicting whether the first group of cancer patients will die within a threshold number of weeks after the start of cancer treatment.

[0158] 65. The computing system of claim 63, wherein the threshold accuracy is based on a p-value for an interaction between: (i) the prediction of whether the second group of cancer patients dies within a threshold number of weeks after starting cancer treatment, and (ii) the number of PD-L1-positive patients in the second group of cancer patients.

[0159] 66. A computing system as described in embodiment 63, wherein the hyperparameters include one or more of the following: the maximum depth of the decision tree is set to 2, 3, 4, 5 or 6, the minimum loss reduction required to split a node in the decision tree is set to 0, 1, 2, 3 or 4, the minimum sum of the weights of all observations required in child nodes in the decision tree is set between 3.0 and 7.0, the fraction of input parameters randomly sampled for each decision tree is set between 0.7 and 1.0, the fraction of observations randomly sampled for each decision tree is set between 0.4 and 0.8, and the number of decision trees in a continuous series of decision trees is set between 10 and 25.

[0160] 67. A computer-implemented method, the computer-implemented method comprising:

[0161] Obtaining a gradient boosting machine learning model, wherein the gradient boosting machine learning model is trained on model training data, wherein the model training data comprises a first group of cancer patients undergoing cancer treatment, wherein the model training data correlates (i) results of laboratory tests performed on the first group of cancer patients and tumor types of the first group of cancer patients with (ii) whether individuals from the first group of cancer patients died within a threshold number of weeks after initiating cancer treatment, wherein training the gradient boosting machine learning model involves: (i) training a continuous series of decision trees until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on residual values of a previous decision tree in the continuous series of decision trees; and (ii) determining a sum function for the continuous series of decision trees;

[0162] Obtain results from a laboratory test performed on another cancer patient;

[0163] obtaining the tumor type of the other cancer patient;

[0164] applying the gradient boosting machine learning model to results of laboratory tests performed on another cancer patient and the tumor type of the other cancer patient; and

[0165] A prediction is received from the gradient boosting machine learning model as to whether the other cancer patient will die within a threshold number of weeks.

[0166] 68. The computer-implemented method of claim 67, wherein the threshold number of weeks is 12 weeks.

[0167] 69. The computer-implemented method of claim 67, wherein the results of the laboratory tests provide one or more of: neutrophil-to-lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration, or gamma-glutamyl transferase concentration.

[0168] 70. The computer-implemented method of claim 67, wherein the tumor type is one of: non-small cell lung cancer, urothelial carcinoma, head and neck squamous cell carcinoma, gastric cancer, renal cell carcinoma, or Hodgkin lymphoma.

[0169] 71. A computer-implemented method as described in claim 67, wherein the sum function applies a multiplicative learning rate to the results of each of the decision trees from the continuous series.

[0170] 72. The computer-implemented method of claim 67, wherein predicting whether another cancer patient will die within a threshold number of weeks comprises:

[0171] The probability that the other cancer patient will die within a threshold number of weeks is determined by using a summation function.

[0172] 73. The computer-implemented method of embodiment 72, wherein predicting whether the other cancer patient will die within a threshold number of weeks further comprises:

[0173] Based on the probability and the cutoff value, the other cancer patient is placed in a high risk of death category or a low risk of death category.

[0174] 74. A computer-implemented method as described in embodiment 73, wherein the cutoff value is selected so that the false positive rate does not exceed 5%-20% when predicting whether the first group of cancer patients will die within a threshold number of weeks.

[0175] 75. The computer-implemented method of claim 73, wherein cancer patients in a high risk of death category are excluded from clinical trials for treating cancer, and wherein cancer patients in a low risk of death category are included in clinical trials for treating cancer.

[0176] 76. A computer-implemented method as described in embodiment 67, wherein the stopping condition is that the continuous series of decision trees includes a threshold number of decision trees.

[0177] 77. A computer-implemented method as described in Option 67, wherein the stopping condition is that the residual values of a specific decision tree in a continuous series of decision trees are all lower than a threshold.

[0178] 78. The computer-implemented method of claim 67, wherein the gradient boosting machine learning model is also trained using model adjustment data, wherein the model adjustment data relates to a second group of cancer patients undergoing cancer treatment, wherein the model adjustment data correlates (i) results of laboratory tests performed on the second group of cancer patients and the tumor type of the second group of cancer patients with (ii) whether the second group of cancer patients died within a threshold number of weeks after starting the cancer treatment, wherein training the gradient boosting machine learning model using the model adjustment data comprises:

[0179] generating two or more variations of a gradient boosting machine learning model by training a sequential series of decision trees for each variation with a different set of hyperparameters;

[0180] applying the changes to the model training data and the model tuning data; and selecting a specific change from the changes as a gradient boosting machine learning model that provides a prediction of whether the first group of cancer patients and the second group of cancer patients died within a threshold number of weeks after starting cancer treatment within a threshold accuracy.

[0181] 79. The computer-implemented method of claim 78, wherein the threshold accuracy is based on a true positive rate or a true negative rate of predicting whether the first group of cancer patients will die within a threshold number of weeks after initiation of cancer treatment.

[0182] 80. The computer-implemented method of claim 78, wherein the threshold accuracy is based on a p-value for an interaction between: (i) the prediction of whether the second group of cancer patients dies within a threshold number of weeks after starting cancer treatment, and (ii) the number of PD-L1-positive patients in the second group of cancer patients.

[0183] 81. A computer-implemented method as described in claim 78, wherein the hyperparameters include one or more of the following: the maximum depth of the decision tree is set to 2, 3, 4, 5 or 6, the minimum loss reduction required to split a node in the decision tree is set to 0, 1, 2, 3 or 4, the minimum sum of the weights of all observations required in child nodes in the decision tree is set between 3.0 and 7.0, the fraction of input parameters randomly sampled for each decision tree is set between 0.7 and 1.0, the fraction of observations randomly sampled for each decision tree is set between 0.4 and 0.8, and the number of decision trees in a continuous series of decision trees is set between 10 and 25.

[0184] 82. An article of manufacture comprising a non-transitory computer-readable medium having program instructions stored thereon, the program instructions, when executed by a computing device, causing the computing device to perform the operations of any one of schemes 67-81.

[0185] Other features and advantages of the invention will be apparent from the detailed description and from the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0186] Figure 1A Shown is the overall survival of patients enrolled in a Phase I immuno-oncology trial, durvalumab study 1108 (2L+UC cohort). The chart shows that approximately 20% of patients died early in the Phase II / O clinical trial despite meeting eligibility criteria. Figure 1B Prognostic scores for estimated life expectancy based on published retrospective studies are described. These scores were developed to identify patients who are likely to have a short survival. dNLR, derived neutrophil / (white blood cell minus neutrophil) ratio; LDH, lactate dehydrogenase; LLN, lower limit of normal; NLR, neutrophil-to-lymphocyte ratio; ULN, upper limit of normal. Arkenau, J Clin Oncol, 2009; Bigot, Eur J Cancer, 2017; Mezquita, JAMA Oncol, 2018.

[0187] Figure 2 The protocol for developing a high-risk early mortality (HREM) prognostic score is described. Briefly, data from all patients enrolled in three Phase I trials (NCT01693562, NCT02000947, and NCT02261220) were divided into two pools: training data and test data. The training data was used to construct a model with 10-fold cross-validation. The test data was then applied to the model, and the results were compared with those obtained using the RMH, GRIM, and LIPI prognostic scores (see Figure 1).

[0188] Figure 3A The proportion of patients with an overall survival (OS) score of 8 weeks or less (ie, patients at high risk of early death) is shown. Figure 3B The proportion of patients with an overall survival (OS) score of 12 weeks or less (i.e., patients at high risk of early death) is shown. Study sites that enrolled 7 or more patients (n=55 sites) were included. The selection of study sites can be improved by determining whether the site tends to enroll patients who are more or less likely to have EM to identify patients at high risk of early death.

[0189] Figure 4 HREM scores are shown ( Figure 4 A) and the associated predictive performance for identifying patients with overall survival (OS) ≤ 12 weeks ( Figure 4 B) The average true positive rate (TPR) of the training and test groups is about 75%, and the false positive rate (FPR) is controlled at about 10%.

[0190] Figure 5 The time-dependent TPR of HREM was shown to be significantly higher than that of RMH, GRIM and LIPI prognostic scores ( Figure 5 A) and FPR( Figure 5 B). HREM better predicted patients at high risk of early mortality (TPR): OS ≤ 4 weeks (88.4%); OS ≤ 8 weeks (81.7%); and OS ≤ 12 weeks (74.7%). HREM also had a better control of FPR. P values: *** < 0.001, ** < 0.01, * < 0.05.

[0191] Figure 6 The OS of patients with high risk of early death identified by HREM and patients without high risk of early death is shown. The predicted HREM survival rate was significantly reduced compared with the non-HREM survival rate. The median survival time in weeks and the 95% confidence interval (CI) are shown.

[0192] Figure 7 Comparison of OS of patients with high risk of early death predicted by HREM with RMH, GRIM and LIPI prognostic scores is shown. Median survival time in weeks and 95% confidence interval (CI) are shown according to the score.

[0193] Figure 8 The effect of PD-L1 on OS was shown in patients predicted to have a high risk of early death. Figure 8 A shows a composite graph of the survival probability of tested patients based on HREM, RMH, GRIM and LIPI prognostic scores for PD-L1 low / negative (solid line) and PD-L1 high (dashed line). Figure 8 B, 8C, 8D, and 8E show the prognostic scores for PD-L1 low / negative (solid line) and PD-L1 high (dashed line) for each of the HREM, RMH, GRIM, and LIPI prognostic scores, respectively. A PD-L1 effect was observed in patients predicted by HREM to have a high risk of early death, but not by the other three scores. HREM-predicted patients at high risk of early death with low / negative PD-L1 progressed even faster than the other scores. The median survival time in weeks and 95% confidence interval (CI) are shown. PD-L1 status was determined in tumor cells (TC) and immune cells (IC). *For NSCLC, high: TC ≥ 25%, low / negative: TC < 25%; for UC, high: TC ≥ 25% or IC ≥ 25%, low / negative: TC < 25% and IC < 25%.

[0194] Figure 9 To show the impact of the presence of liver metastases (mets) on OS in patients predicted to be at high risk of early death. Figure 9 A shows a composite graph of the survival probability based on the HREM, RMH, GRIM and LIPI prognostic scores for the test patients with (solid line) and without (dashed line) liver metastases. Figure 9 B and 9C show the prognostic scores of the presence (solid line) and absence (dashed line) of liver metastases for each of the HREM, RMH, GRIM and LIPI prognostic scores, respectively. Patients predicted to have a high risk of early death according to the four scores had similar survival in the liver metastasis-positive group. Compared with the other three scores, HREM can better capture patients with faster progression in the liver metastasis-negative group.

[0195] Figure 10 Shown by treatment ( Figure 10 A and 10C = durvalumab; Figure 10 B and 10D = durvalumab + tremelimumab) and cancer type ( Figure 10 A and 10B = non-small cell lung cancer (NSCLC); Figure 10 C and 10D = PD-L1 and HREM patient stratification for bladder cancer (UBC).

[0196] Figure 11 The FastProgIO prognostic score was shown to be associated with a high inflammatory proteomic signature.

[0197] Figure 12 A shows the F1 model prognostic score. Figure 12 B highlights a comparison of HREM rates (boxes) for the same tumor type from different trials. This comparison can help understand unexpected efficacy results. Boxes represent the same tumor type.

[0198] Figure 13 The variation in rapid progression rates between study sites used in Example 4 is shown. HREM rates varied widely between study sites. Approximately 44% of study sites had HREM rates ≥ 20%. Only study sites with ≥ 7 patients enrolled were included (n = 79 study sites).

[0199] Figure 14 A schematic diagram for developing the F1 model is shown.

[0200] Figure 15 The predictive performance of the F1 model using the training and test data is shown. Similar predictive performance can be seen for the training and test data. For the 12-week OS cutoff, the FPR was controlled at 10%.

[0201] Figure 16 Shows the overall survival of the F1 model prediction group.

[0202] Figure 17Comparison of OS in patients at high risk of early death using the F1 model and the RMH, GRIM, and LIPI prognostic scores is shown. OS in patients predicted to be at high risk of early death by the F1 model was inferior compared to existing predictors. The F1 model was the only model that predicted 12-week life expectancy (i.e., the 95% CI did not overlap with the 12-week scale).

[0203] Figure 18 The results showed the effect of PD-L1 on OS in patients predicted by the F1 model to be at high risk of early death. Most patients predicted to be at high risk of early death had worse OS with PD-L1 low / negative compared to PD-L1 high. Patients predicted to be at high risk of early death by the F1 model had worse OS compared to published scores in both the PD-L1 low / negative and high populations.

[0204] Figure 19 The results showed the effect of liver metastasis on OS in patients predicted by the F1 model to be at high risk of early death. Patients with liver metastasis who were predicted to be at high risk of early death had worse OS compared with patients without liver metastasis. Patients predicted to be at high risk of early death by the F1 model had a worse OS prediction compared with published scores in both liver metastasis-negative and -positive populations.

[0205] Figure 20 A shows the results of study 1108, in which durvalumab alone had promising efficacy with a median OS of 4.9 months (2.6-9.1). Despite the crossover trial, the results seemed poorer and may be explained by F1, but in study 21, durvalumab alone (D) showed worse results than 1108, and durvalumab + tremelimumab (D+T) in the INFy+ cohort was also associated with worse results. Figure 20 Figure B shows that applying F1 to Study 21 revealed a substantial imbalance between cohorts in patients at high risk for early mortality. An imbalance in patients with a "poor prognosis" predicted by the F1 model (i.e., not mitigated by randomization) was observed between cohorts. Both the D and D+T (IFNy+) cohorts enrolled a higher proportion of high-risk patients, providing a potential explanation for the poor outcomes in these cohorts. Other potential explanations, such as differences in baseline clinical prognostic parameters, baseline biomarker differences in PDL1, or IFNy+ status, did not explain the differences in results.

[0206] Figure 21Results from the ARCTIC study, which examined patients with NSCLC, were presented. Durvalumab plus tremelimumab (D+T) showed greater benefit in patients with negative PD-L1 levels (<1%). For patients with low PD-L1 levels (≥1%, <25%), durvalumab monotherapy and chemotherapy showed similar benefit, outperforming D+T and tremelimumab alone.

[0207] Figure 22 This analysis shows an imbalance in the F1 model-predicted patients at high risk of early death between cohorts. OS excludes patients predicted to have a high risk of early death. Predicted rapid progression rates were similar between cohorts, and the difference in OS was not due to an imbalance among patients with a "poor prognosis." Even after removing patients predicted to have a high risk of early death, D+T still exhibited a better OS association than chemotherapy in PD-L1-negative patients.

[0208] Figure 23 Showing correlation of F1 model results with inflammatory proteomic signatures (Study 1108, UBC).

[0209] Figure 24 Shown are the intention-to-treat (ITT) survival probabilities for the three treatment regimens in the EAGLE study (chemotherapy, durvalumab, and durvalumab plus tremelimumab) over 144 weeks (top) and a table of patients at risk over the same period (bottom).

[0210] Figure 25 Figure 3. PD-L1 high (≥25%) survival probability analysis over 144 weeks for the three treatment regimens in the EAGLE study (chemotherapy, durvalumab, and durvalumab plus tremelimumab) (top) and at-risk patients over the same period (bottom).

[0211] Figure 26 Figure 3. PD-L1 low / negative (≤25%) survival probability analysis over 144 weeks for the three treatment regimens in the EAGLE study (chemotherapy, durvalumab, and durvalumab plus tremelimumab) (top) and a table of patients at risk over the same period (bottom).

[0212] Figure 27 Shown are the probability of survival for all patients over 144 weeks for the three treatment regimens in the EAGLE study (chemotherapy, durvalumab, and durvalumab plus tremelimumab) (top) and a table of patients at risk over the same period (bottom). HR, hazard ratio. CI, confidence interval. Also shown are the hazard ratios (HRs) and p-values for the durvalumab and durvalumab plus tremelimumab arms of the EAGLE study for head and neck squamous cell carcinoma.

[0213] Figure 28Figure 3. Intention-to-treat (ITT) survival probability analysis (top) for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0214] Figure 29 Shown are the intention-to-treat (ITT) survival probabilities for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients at high risk of early death (top) and a table of at-risk patients over the same period (bottom). Also shown are the hazard ratios (HRs) and p-values for the F1 analysis.

[0215] Figure 30 Figure 3: Analysis of the probability of survival for PD-L1-high patients (≥25%) across the two treatment regimens in the EAGLE study (chemotherapy and durvalumab) over 144 weeks (top) and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0216] Figure 31 Figure 2: Analysis of the probability of survival for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients with PD-L1 high risk (≥25%) at high risk of early death (top) and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0217] Figure 32 Figure 3: Survival probability analysis of PD-L1 low patients (≥1, <25%) for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks (top) and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0218] Figure 33 Figure 3: Analysis of the probability of survival for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients with low PD-L1 levels (≥1, <25%) at high risk of early death (top) and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0219] Figure 34 Figure 3: Analysis of the probability of survival for PD-L1-negative patients (<1%) across the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks (top) and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0220] Figure 35Figure 3. Survival probability analysis (top) of the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding PD-L1-negative patients at high risk of early death (<1%), and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0221] Figure 36 Figure 3: Analysis of the probability of survival for patients at high risk of early death for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks (top) and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0222] Figure 37 Shown are the intention-to-treat (ITT) survival probabilities for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients at high risk of early death (top) and a table of at-risk patients over the same period (bottom). Also shown are the hazard ratios (HRs) and p-values for the F1 analysis.

[0223] Figure 38 Figure 3: Analysis of the probability of survival for PD-L1 high patients (≥25%) over 144 weeks in the EAGLE study for both treatment regimens (chemotherapy and durvalumab), excluding patients at high risk of early death (top) and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0224] Figure 39 Figure 3: Analysis of the probability of survival for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients with low PD-L1 levels (≥1, <25%) at high risk of early death (top) and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0225] Figure 40 Figure 3: Analysis of the probability of survival of PD-L1-negative patients (<1%) in the EAGLE study for both treatment regimens (chemotherapy and durvalumab) over 144 weeks, excluding patients at high risk of early death (top) and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0226] Figure 41Figure 3: Survival probability analysis (top) of patients at high risk of early death as measured by NLR, NEUT, ALB, LDH, GGT, and AST for the three treatment regimens in the EAGLE study (chemotherapy, durvalumab, and durvalumab + tremelimumab) over 144 weeks and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0227] Figure 42 Shown are the intention-to-treat (ITT) survival probabilities for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients at high risk of early death (as measured by NLR, NEUT, ALB, LDH, GGT, and AST) and a table of patients at risk over the same period (bottom). Also shown are the hazard ratios (HRs) and p-values for the F1 analysis.

[0228] Figure 43 Figure 2: Analysis of the probability of survival of PD-L1 high patients (≥25%) for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients at high risk of early death (as measured by NLR, NEUT, ALB, LDH, GGT, and AST) (top) and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0229] Figure 44 Figure 2: Analysis of the probability of survival of patients with low PD-L1 (≥1, <25%) for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients at high risk of early death (as measured by NLR, NEUT, ALB, LDH, GGT, and AST) (top) and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0230] Figure 45 Figure 2: Analysis of the probability of survival of PD-L1-negative patients (<1%) for the two treatment regimens (chemotherapy and durvalumab) in the EAGLE study over 144 weeks, excluding patients at high risk of early death (as measured by NLR, NEUT, ALB, LDH, GGT, and AST) (top) and a table of at-risk patients over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0231] Figure 46Figure 3: Survival probability analysis (top) of patients at high risk of early death as measured by NLR, NEUT, ALB, LDH, GGT, and AST for the three treatment regimens in the MYSTIC study (chemotherapy, durvalumab, and durvalumab + tremelimumab) over 144 weeks and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0232] Figure 47 Figure 3. Survival probability analysis of PD-L1 high patients (≥25%) for the two treatment regimens (chemotherapy and durvalumab) in the MYSTIC study over 144 weeks (top) and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown. Arrows indicate the intersection of Kaplan-Meier (KM) curves.

[0233] Figure 48 Figure 2: Analysis of the probability of survival for the two treatment regimens (chemotherapy and durvalumab) in the MYSTIC study over 144 weeks, excluding patients with high PD-L1 (≥25%) at high risk of early death (measured using NLR, NEUT, ALB, LDH, GGT, and AST) and a table of patients at risk over the same period (bottom). Hazard ratios (HRs) and p-values for the F1 analysis are also shown.

[0234] Figure 49 Overall survival (OS) among treatment arms is shown for Studies 1108, HAWK, CONDOR, and EAGLE.

[0235] Figure 50 Shown are F1-predicted patients at high risk for early death (FastProg) and patients not at high risk for early death (Non-FastProg) by treatment arm for Studies 1108, HAWK, CONDOR, and EAGLE.

[0236] Figure 51 Shown are the overall Kaplan–Meier estimates of PD-L1-positive patients who were alive at 12 weeks.

[0237] Figure 52 The Immunity Immediate Index (3i) score development workflow is shown. Model requirements (1) were defined to address current challenges in predicting early mortality. The design (2) selected the model requirements in terms of dataset, feature, and method selection. A test plan (3) was prepared to evaluate model performance according to globally accepted machine learning practices and mitigate any risks associated with the use of open source packages. The development (4) followed globally accepted principles of dataset separation, k-fold cross-validation during model training, hyperparameter optimization, and independent dataset testing (5).

[0238] Figure 53 Figure 4. Quantification of variable importance in feature selection. A gradient boosting model was fitted using a set of variables to predict 12-week life expectancy. The top 20 variables with the highest importance values are plotted. Variables were clustered into more and less important groups using the K-means clustering method. Abbreviations: ALB = albumin; ALP = alkaline phosphatase; AST = aspartate aminotransferase; BMI = body mass index; CL = chloride; EOSLE = eosinophil / leukocyte ratio; GGT = gamma-glutamyltransferase; GLUC = glucose; HCT = hematocrit; LDH = lactate dehydrogenase; LYMLE = lymphocyte / leukocyte ratio; MG = magnesium; NEUT = neutrophils; NEUTLE = neutrophil / leukocyte ratio; NLR = neutrophil / lymphocyte ratio; PROT = protein; TSH = thyroid-stimulating hormone; TUMSZINV = tumor size; URATE = uric acid; WBC = white blood cells.

[0239] Figure 54 Figure 3. Quantification of the importance of the Immuno-Instant Index (3i) score variables in the final 3i score model. Variables were clustered into more and less important groups using the K-means clustering method. Abbreviations: ALB = albumin; AST = aspartate aminotransferase; GGT = gamma glutamyltransferase; LDH = lactate dehydrogenase; NEUT = neutrophil; NLR = neutrophil / lymphocyte ratio; NSCLC = non-small cell lung cancer; SCCHN = head and neck squamous cell carcinoma; UBC = urothelial bladder cancer.

[0240] Figure 55 The overall survival according to the 3i score in the training set is shown. The total n=2213; including study 1108 (all solid tumors, n=923), study 06 (non-small cell lung cancer, n=353), study 10 (urothelial bladder cancer, n=164), study 21 (gastric cancer, n=98), ATLANTIC (non-small cell lung cancer, n=371), CONDOR (head and neck squamous cell carcinoma, n=193) and HAWK (head and neck squamous cell carcinoma, n=111).

[0241] Figure 56A and 56B show the overall survival of the 3i score, GRIM and LIPI models in the entire training set. Total n = 2213; including study 1108 (all solid tumors, n = 923), study 06 (non-small cell lung cancer, n = 353), study 10 (urothelial bladder cancer, n = 164), study 21 (gastric cancer, n = 98), ATLANTIC (non-small cell lung cancer, n = 371), CONDOR (head and neck squamous cell carcinoma, n = 193) and HAWK (head and neck squamous cell carcinoma, n = 111). Left: Kaplan-Meier curves for patients identified as high risk using the 3i score, GRIM or LIPI. Right: Kaplan-Meier curves for patients identified as low risk using the 3i score, GRIM or LIPI.

[0242] Figure 57 Figures A and B show overall survival using the 3i score, GRIM, and LIPI models for MYSTIC ITT patients randomized to durvalumab. N = 374 patients with non-small cell lung cancer. Kaplan-Meier curves for patients identified as high risk (left panel) or high risk (right panel) by 3i, GRIM, or LIPI scores.

[0243] Figure 58 A-58C shows Kaplan-Meier estimates of overall survival in the MYSTIC ITT population. Figure 58 A shows all patients, Figure 58 B shows 3i score low risk, Figure 58 C shows the high-risk group with 3i score.

[0244] Figure 59 A-59C shows Kaplan Meier estimates of overall survival in the MYSTIC PD-L1 ≥ 25% subgroup. Figure 59 A shows all patients, Figure 59 B shows 3i score low risk, and Figure 59 C shows the 3i high-risk group. N.missing refers to the number of patients with at least one missing laboratory value for whom a 3i score could not be generated. N.3i refers to the number of patients with a 3i high-risk score.

[0245] Figure 60 A and 60B show overall survival of EAGLE ITT patients randomized to durvalumab using the 3i score, GRIM, and LIPI models. N = 240 patients with head and neck squamous cell carcinoma. Kaplan-Meier curves for patients identified as high risk (60A) or low risk (60B) by 3i, GRIM, or LIPI scores.

[0246] Figure 61A-61C shows the Kaplan Meier estimates of overall survival in the EAGLE ITT population. Figure 61 A shows all patients, Figure 61 B shows 3i score low risk, Figure 61 C shows the high-risk patient group with 3i score.

[0247] Figure 62 A-62C shows Kaplan Meier estimates of overall survival in the EAGLE PD-L1 ≥ 25% subgroup. Figure 62 A shows all patients, Figure 62 B shows 3i score low risk, Figure 62 C shows the high-risk patient group with 3i score.

[0248] Figure 63 Figure 3. Boxplots showing differences in input variables between high- and low-risk patients with 3i scores in MYSTIC. Abbreviations: ALB = albumin; AST = aspartate aminotransferase; GGT = gamma-glutamyltransferase; LDH = lactate dehydrogenase; NEUT = neutrophil; NLR = neutrophil / lymphocyte ratio.

[0249] Figure 64 Figure 3. Box plots showing differences in input variables between high- and low-risk patients with 3i scores in EAGLE. Abbreviations: ALB = albumin; AST = aspartate aminotransferase; GGT = gamma-glutamyltransferase; LDH = lactate dehydrogenase; NEUT = neutrophil; NLR = neutrophil / lymphocyte ratio.

[0250] Figure 65 Figure 3: 3i-high and 3i-low NSCLC patients with lower expression of T effectors and higher immunosuppressive gene signatures compared to 3i-low NSCLC, HNSCC, and UBC patients. Across solid tumors and studies, 3i-high patients showed 1.9-fold lower (FC) expression of T effectors and 1.5-fold higher expression of immunosuppressive gene signatures. The number of patients (n) in each 3i-high and 3i-low group and the Wilcox test p-value (****P<0.0001) are shown. NSCLC data are from the MYSTIC, ATLANTIC, CP1108, and 006 studies. HNSCC data are from the HAWK, CONDOR, and CP1108 studies. UBC data are from Study 10 and CP1108.

[0251] Figure 66 Figure 3: 3i high vs. 3i low NSCLC patients with lower expression of T effectors and higher immunosuppressive gene signatures, regardless of PD-L1 status. The number of patients (n) in each 3i high and low group and the Wilcox test p-value (****P<0.0001) are shown.

[0252] Figure 67 Compared with 3i low NSCLC patients, 3i high-grade NSCLC patients had lower absolute counts of T, NK, and B cells and higher absolute counts of MNC MDSCs. Compared with 3i low NSCLC patients, 3i high-grade NSCLC patients had at least 24% lower absolute counts of T, NK, and B cells (data from CP1108 and 006 studies). The absolute counts of MNC MDSCs (CD14+HLA-DRlo / -) and monocytes (CD14+HLA-DR+) were significantly higher by 138.2% in 3i high-grade NSCLC patients compared with 3i low (Mystic), while no difference was found in the absolute counts of monocytes. The number of patients (n) in each 3i high and low group and the Wilcox test p value are shown (***P<0.001; ****p<0.0001).

[0253] Figure 68 Shown are lower absolute counts of naive, activated, effector, and central memory T cells in 3i-high vs. 3i-low NSCLC patients. Data are from studies CP1108 and 006. Shown are the number of patients (n) in each 3i-high and 3i-low group and Wilcox test p-values (***P < 0.001; ****P < 0.0001).

[0254] Figure 69 Figure 3 shows lower serum expression concentrations of CRP, IL6, and IL8 in 3i-high compared with 3i-low NSCLC, HNSCC, and UBC patients. The number of patients (n) in each 3i-high and 3i-low group and the Wilcox test p-value (****P < 0.0001) are shown. NSCLC data are derived from the CP1108 study. HNSCC data are derived from the HAWK and CONDOR study and CP1108. UBC data are derived from CP1108.

[0255] Figure 70 Depicted are example computing devices that can be configured to perform training and / or execution of gradient boosting machine learning models.

[0256] Figure 71 Depicted are example computing clusters that can be configured to perform training and / or execution of gradient boosting machine learning models.

[0257] Figure 72 A flowchart representation depicting an example training of a gradient boosting machine learning model.

[0258] Figure 73 A flowchart representation depicting an example execution of a gradient boosting machine learning model. DETAILED DESCRIPTION

[0259] Unless otherwise defined, all technical and scientific terms used herein have the meanings commonly understood by those skilled in the art to which the present invention belongs. The following references provide a general definition of a plurality of terms used in the present invention for technical personnel: Singleton et al., Dictionary of Microbiology and Molecular Biology [microbiology and molecular biology dictionary] (2nd edition 1994); The Cambridge Dictionary of Science and Technology [Cambridge dictionary of science and technology] (Walker writes, 1988); The Glossary of Genetics [genetics vocabulary], 5th edition, R. Rieger et al. (eds.), Springer Verlag (Springer Verlag) (1991); and Hale and Marham, The HarperCollins Dictionary of Biology [HarperCollins biological dictionary] (1991). Unless otherwise indicated, the following terms as used herein have the following meanings given to them.

[0260] "Anti-PD-L1 antibody" means an antibody that selectively binds to a PD-L1 polypeptide. Exemplary anti-PD-L1 antibodies are described, for example, in U.S. Patent Nos. 8,779,108; 9,493,565; and 10,400,039, which are incorporated herein by reference. Durvalumab (MEDI4736) or "Durva" is an exemplary anti-PD-L1 antibody suitable for use in the methods described herein.

[0261] "Anti-CTLA4 antibody" means an antibody that selectively binds to a CTLA4 polypeptide. Exemplary anti-CTLA4 antibodies are described, for example, in U.S. Patent Nos. 6,682,736; 7,109,003; 7,123,281; 7,411,057; 7,824,679; 8,143,379; 7,807,797; and 8,491,895 (in which tremelimumab is 11.2.1), which are incorporated herein by reference. Tremelimumab or "Treme" is an exemplary anti-CTLA4 antibody.

[0262] As used herein, the terms "biomarker," "marker," or "clinical pathology marker" (which are used interchangeably) generally refer to a protein, nucleic acid molecule, clinical indicator, or other analyte associated with a disease. In one embodiment, a clinical pathology marker can be differentially expressed (or present) in a biological sample obtained from a subject having a disease (e.g., bladder cancer) relative to the concentration present in a control sample or reference.

[0263] In another embodiment, clinical pathology markers are indicators of disease extent. For example, when the concentration of a clinical pathology marker in a patient with a disease (e.g., cancer) is increased compared to a control (e.g., a subject without cancer), the increased concentration may indicate disease progression in the patient. Similarly, when the concentration of a clinical pathology marker in a patient with a disease (e.g., cancer) is increased compared to an earlier measured value of a marker in the patient themselves, the increased concentration may indicate disease progression in the patient. In addition, in some embodiments, the concentration of a clinical pathology marker may indicate the relative health (e.g., body mass index) of the patient and / or the overall physiological resilience (e.g., age) of the patient.

[0264] In some embodiments, the concentration of one or more clinical pathology markers may be indicative of a patient's immune fitness, e.g., the relative ability of a patient's immune system to fight a particular disease on its own or to be enhanced or modified to fight a particular disease, such as cancer, by treatment with a therapeutic agent, such as an immune checkpoint inhibitor (ICI).

[0265] In this disclosure, “comprises,” “comprising,” “containing,” and “having,” etc. may have the meanings ascribed to them by U.S. patent law and may mean “includes,” “including,” etc.; “consisting essentially of or consists essentially of” also has the meaning ascribed to it by U.S. patent law and the term is open-ended, allowing for existence beyond what is recited as long as the basic or novel features recited are not altered by existence beyond what is recited, but excluding prior art embodiments.

[0266] As used herein, the terms "determine," "evaluate," "determine," "measure," and "detect," as well as "identify," refer to both quantitative and qualitative determinations, and thus, the terms "determine" and "determine," "measure," and the like are used interchangeably herein. Where quantitative determination is the goal, the phrase "determine the amount of an analyte, substance, protein, etc." is used. Where qualitative and / or quantitative determination is the goal, the phrase "determine the concentration of an analyte" or "detect" an analyte is used.

[0267] "Disease" means any condition or disorder that damages, interferes with, or abnormally regulates the normal function of a cell, tissue, or organ. In diseases such as cancer (e.g., bladder cancer), the normal function of a cell, tissue, or organ is disrupted, enabling immune evasion and / or escape.

[0268] "Responsive" in the context of therapy means susceptible to the effects of treatment. "Subject" means a mammal, including but not limited to a human (e.g., a human patient), a non-human primate, or a non-human mammal (e.g., a bovine, equine, canine, ovine, or feline).

[0269] As used herein, the terms "treat," "treating," "treatment," and the like refer to reducing, diminishing, alleviating, alleviating, eliminating, or ameliorating a disorder and / or the symptoms associated therewith. It will be understood that treating a disorder or condition does not require complete elimination of the disorder, condition, or symptoms associated therewith, although it cannot be eliminated.

[0270] Unless specifically stated or obvious from its context, the term "or" as used herein is understood to be inclusive. Unless specifically stated or obvious from its context, the terms "a," "an," and "the" as used herein are understood to be singular or plural. Similarly, a particular term, when expressed in the singular, also encompasses the same term expressed in the plural, and vice versa. For example, the term "drug" also includes "drugs," and vice versa.

[0271] Unless expressly stated or obvious from the context, the term "about" as used herein is understood to be within the normal tolerance range in the art, for example, within 2 standard deviations of the mean. The term "about" is understood to mean 5%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, 0.5%, 0.1%, 0.05% or 0.01% of the stated value. Unless obvious from the context, all numerical values provided herein are modified by the term about.

[0272] The recitation of a list of chemical groups in any definition of a variable herein includes definitions of that variable as any single group or combination of listed groups. The recitation of an embodiment of a variable or aspect herein includes that embodiment as any single embodiment or in combination with any other embodiment or portion thereof.

[0273] Early death

[0274] As described herein, in one embodiment, the invention features a method of predicting whether a cancer patient is a rapid progressor or at risk for early mortality (EM; patients with an overall survival (OS) of 24 weeks or less) by collecting a blood sample or other sample and performing physiological measurements as described elsewhere herein, measuring the concentrations of clinicopathological markers, and generating a composite score of the marker concentrations, wherein the clinicopathological markers can be, for example, neutrophils, hemoglobin, aspartate aminotransferase, and alkaline phosphatase.

[0275] In certain embodiments, EM is defined as patients whose overall survival (OS) is 12 weeks or less.

[0276] In one embodiment, any number of clinicopathological markers may be considered, such as those markers typically measured to determine the relative disease state of cancer patients and those markers typically measured to determine the relative health of an individual (whether the individual is healthy or diseased).

[0277] In certain embodiments, commercially available test can be used for measuring the concentration of clinical pathology markers, such as mensuration based on antibodies, wherein the target marker has a specific antibody for targeting this type of marker. The combination of the antibody and the specific marker can be detected, and the concentration and / or the concentration of the target marker can be measured by the relative amount of the antibody binding. Other tests can be used, such as those measuring the relative abundance of specific cell types, such as flow cytometry. Other tests that can be used for this paper include quantitative PCR and similar techniques that provide information about relative gene expression concentrations. As will be appreciated by the skilled person, other tests for measuring the relative health of people or the various clinical pathology markers of specific disease states are also contemplated herein.

[0278] In some embodiments, the sample used may include tissue biopsy, cerebrospinal fluid, lymph, whole blood, blood fractions (e.g., serum or specific blood cell populations), urine, sweat, tears, saliva, and / or stool.

[0279] In one embodiment, the patient can be assessed upon arrival at the clinic for participation in the clinical trial. In this case, blood can be drawn using standard techniques, and the patient's clinical pathology markers (e.g., neutrophils, hemoglobin, aspartate aminotransferase, alkaline phosphatase, etc.) can be assessed, and the patient's overall survival over a given time period, e.g., 4, 8, 12, or 24 weeks, 4, 6, 8, or 10 months, 1, 2, 3, 4, or 5 years, can be predicted.

[0280] In another embodiment, patients can be pre-approved, remotely enrolled, or flagged for accelerated treatment through remote review of laboratory reports, where a centralized software platform can assess the patient's clinicopathological markers (e.g., neutrophils, hemoglobin, aspartate aminotransferase, alkaline phosphatase, etc.) and determine overall survival.

[0281] In some embodiments, the present invention provides a method for conducting a clinical trial of an anti-cancer therapeutic agent, comprising the steps of identifying a candidate participant for the clinical trial, wherein the candidate participant has cancer, determining the candidate participant's blood concentration of i) neutrophils, ii) hemoglobin, iii) aspartate dehydrogenase, and iv) alkaline phosphatase, calculating the candidate participant's predicted overall survival, administering a first anti-cancer therapeutic agent to a candidate participant with a predicted overall survival of 12 weeks or longer, administering a second anti-cancer therapeutic agent to a candidate participant with a predicted overall survival of 12 weeks or less, and reducing at least one of cancer cell division rate, tumor growth, tumor size, tumor density, or metastasis rate in the candidate participant.

[0282] In other embodiments, the present invention provides a method of treating cancer in a subject in need thereof, comprising the steps of collecting a sample from the subject, determining the concentration of i) neutrophils, ii) hemoglobin, iii) aspartate dehydrogenase, and iv) alkaline phosphatase in the sample, calculating a predicted overall survival score for the subject, administering a first anti-cancer therapeutic to the subject if the subject's predicted overall survival is 12 weeks or longer, or administering a second anti-cancer therapeutic to the subject if the subject's predicted overall survival is less than 12 weeks, and reducing at least one of cancer cell division rate, tumor growth, tumor size, tumor density, or metastasis rate in the subject.

[0283] In some embodiments, the present invention provides a method for conducting a clinical trial of an anti-cancer therapeutic agent, comprising the steps of identifying a candidate participant for the clinical trial, wherein the candidate participant has cancer, determining the candidate participant's blood concentration of i) a neutrophil / lymphocyte ratio, ii) neutrophils, iii) albumin, iv) lactate dehydrogenase, v) gamma-glutamyl transferase, and vi) aspartate dehydrogenase, calculating a predicted overall survival score for the candidate participant, administering a first anti-cancer therapeutic agent to a candidate participant with a predicted overall survival of 12 weeks or longer, administering a second anti-cancer therapeutic agent to a candidate participant with a predicted overall survival of 12 weeks or less, and reducing at least one of cancer cell division rate, tumor growth, tumor size, tumor density, or metastasis rate in the candidate participant.

[0284] In other embodiments, the present invention provides a method of treating cancer in a subject in need thereof, comprising the steps of collecting a sample from the subject, determining the concentration of i) neutrophil / lymphocyte ratio (NLR) protein, ii) neutrophils, iii) albumin, iv) lactate dehydrogenase, v) gamma-glutamyl transferase, and vi) aspartate dehydrogenase in the sample, calculating a predicted overall survival for the subject, administering a first anti-cancer therapeutic to the subject if the predicted overall survival of the subject is 12 weeks or longer, or administering a second anti-cancer therapeutic to the subject if the predicted overall survival of the subject is less than 12 weeks, and reducing at least one of cancer cell division rate, tumor growth, tumor size, tumor density, or metastasis rate in the subject.

[0285] In some embodiments where a patient is predicted to have an overall survival of less than 12 weeks from the date of assessment, an accelerated and / or more aggressive treatment regimen may be prescribed that is different from the treatment regimen contemplated by the particular clinical trial for which the patient was assessed when overall survival was assessed.

[0286] Immune checkpoint inhibitors and early death

[0287] At the forefront of the ICI class are antibodies targeting programmed cell death ligand-1 (PD-L1), which exhibit therapeutic activity by restoring anti-tumor T cell responses. An example of the profound impact demonstrated by ICIs is the 23% 5-year overall survival observed in patients with PD-L1 high advanced or metastatic NSCLC treated with pembrolizumab (Garon et al. 2019). However, during the development of these agents, EM phenomena were frequently observed when ICIs were delivered as monotherapy in the metastatic disease setting, with delivery occurring in the early months of treatment in clinical trials across multiple tumor types including NSCLC (Borghaei et al. 2015, Rizvi et al. 2018, Socinski et al. 2018, Mok et al. 2019), UC (Bellmunt et al. 2017, Powles et al. 2018), HNSCC (Ferris et al. 2016), and gastric cancer (Shitara et al. 2018). Early mortality has been observed regardless of the specificity of antibodies targeting PD-1 or PD-L1 (Winquist et al. 2018, Mulkey et al. 2019), suggesting a class effect. Although this phenomenon has been widely described, there remains a critical need for an objective measure or tool to better predict a patient's risk of EM and help direct treatment to those who are most likely to have a favorable benefit-risk profile when treated with ICI therapy.

[0288] Prevalence of expected early death in clinical trials investigating ICIs

[0289] To ensure comprehensive analysis across tumor types, the prevalence of death within 4, 8, and 12 weeks after the first study treatment was first assessed in recent Phase I-III studies. These studies included:

[0290] Study 1108 (CD-ON-MEDI4736-1108), a Phase 1 study evaluating the safety, tolerability, and pharmacokinetics of MEDI4736 in subjects with advanced solid tumors (n=1002)

[0291] Study 06 (D4190C0006), a Phase 1b open-label study evaluating the safety and tolerability of MEDI4736 in combination with tremelimumab in subjects with advanced non-small cell lung cancer (n=379 patients)

[0292] Study 10 (D4190C00010), a Phase 1b / 2 multicenter, open-label study evaluating the safety, tolerability, and efficacy of MEDI4736 in combination with tremelimumab in subjects with advanced malignancies (n=168 patients)

[0293] Study 21 (D4190C00021), a Phase 1b / 2 study of MEDI4736 in combination with tremelimumab, MEDI4736 monotherapy and tremelimumab monotherapy in subjects with metastatic or recurrent gastric or gastroesophageal junction adenocarcinoma (n=113 patients).

[0294] CONDOR (D4193C00003), a Phase II, multicenter, single-arm, global study of MEDI4736 monotherapy, tremelimumab monotherapy, and MEDI4736 in combination with tremelimumab in patients with recurrent or metastatic head and neck squamous cell carcinoma (N=263).

[0295] • HAWK (D4193C00001), a Phase II, multicenter, single-arm, global study of MEDI4736 monotherapy in patients with recurrent or metastatic head and neck squamous cell carcinoma (N=112).

[0296] ATLANTIC (D4191C00003), a Phase II, non-comparative, open-label, multicenter, international study (N=371) of MEDI4736 in patients with locally advanced or metastatic non-small cell lung cancer (stage IIIB-IV) who have received at least 2 prior systemic therapy regimens, including 1 platinum-based chemotherapy regimen.

[0297] MYSTIC (D419AC00001), a phase III randomized, open-label, multicenter, global study of MEDI4736 in combination with tremelimumab or MEDI4736 monotherapy versus standard of care platinum-based chemotherapy in the first-line treatment of patients with advanced or metastatic non-small cell lung cancer (N=1118)

[0298] EAGLE (D4193C00002), a Phase III randomized, open-label, multicenter, global study of MEDI4736 monotherapy and MEDI4736 in combination with tremelimumab versus standard of care in patients with recurrent or metastatic head and neck squamous cell carcinoma (N=736)

[0299] By combining these studies together, it was observed that 17.84% to 30.36% of patients treated with ICI died ≤ 12 weeks after the first day of study treatment (Table 9).

[0300] Table 9. Prevalence of EM in clinical trials

[0301]

[0302] #All solid tumors, including NSCLC, SCCHN, UBC, gastric cancer, HPV-positive cancer, advanced malignant melanoma, uveal melanoma, pancreatic cancer, advanced cutaneous melanoma, hepatocellular carcinoma, nasopharyngeal carcinoma, ovarian cancer, SCLC, soft tissue sarcoma, MSI-high cancer, triple-negative breast cancer, colorectal cancer, and renal cell carcinoma. ICI represents patients treated with immune checkpoint inhibitors; SOC represents patients treated with chemotherapy; and NT represents patients not treated.

[0303] In conclusion, the potential for higher EM in patients treated with ICIs compared with active comparators suggests the need for improved prognostic methods to better identify those patients who are unlikely to survive beyond 12 weeks and therefore benefit from ICI treatment.

[0304] Existing prognostic scores for predicting early death

[0305] Many studies have been conducted to improve the accuracy of physician prognosis for patients with advanced or metastatic cancer (Ploquin et al., Crit Rev Oncol Hematol. [Tumor Hematology Review] 2012) and to contribute to the development of clinical prognostic scores. Most of these studies identified specific clinical features and laboratory values associated with adverse outcomes in clinical studies. Several issues limit the application of these studies in addressing the phenomenon of early mortality. First, several studies relied on subjective and investigator-dependent parameters such as performance status (i.e., Eastern Cooperative Oncology Group ("ECOG"), Karnfosky index) (Krishnan M; J Support Oncol [Supportive Oncology Journal] 2013). Another limitation of these clinical scores is the lack of validation in large independent studies. Finally, these scores have barely evolved in the ICI era, so their relevance in this setting remains unknown. These scores are summarized below.

[0306] Royal Marsden Hospital (RMH) score

[0307] The RMH score is the first score in oncology to predict early life expectancy. It was originally developed in the context of phase 1 trials investigating cytotoxic and targeted therapies. The RMH score is based on the following variables: lactate dehydrogenase (LDH, > upper limit of normal [ULN]), albumin (<35 g / L), and number of metastatic sites (>2) (Arkenau et al., 2009). Each variable is assigned 1 point. If the sum of these variables exceeds 1, the patient is classified as having a poor prognosis by RMH. Application of the RMH score in a large retrospective cohort (N > 1000 patients) of patients with various tumor types (Wheler et al., 2012) confirmed the poor prognosis associated with an RMH score > 1. In retrospective patient series, high-risk RMH has been associated with a poor prognosis in the setting of ICI (Bigot et al., 2017; Minami et al., 2019).

[0308] Gustave Roussy Immunity (GRIm) score

[0309] The GRIm score is a recent attempt to improve patient life expectancy estimates in the context of phase 1 oncology trials investigating ICIs. It is based on albumin (<35 g / L), LDH (>ULN), and the neutrophil-to-lymphocyte ratio (NLR; >6) (Bigot et al., 2017). Similar to the RMH score, each variable is assigned 1 point. In retrospective patient series, a sum of these variables >1 was associated with poor overall survival in ICI-treated patients (Bigot et al., 2017; Minami et al., 2019).

[0310] Lung Immunoprognostic Index (LIPI) score

[0311] The LIPI score was developed to predict outcomes in NSCLC patients treated with ICIs in a variety of settings, including routine clinical care, expanded access, compassionate use programs, and clinical trials. It is based on the following variables: derived neutrophil / (leukocyte minus neutrophil) ratio (dNLR) > 3 and LDH > ULN. Each variable is assigned 1 point. According to the LIPI, NSCLC patients whose sum of the variables is > 1 are associated with a poor prognosis. In the original publication, the LIPI score was retrospectively applied in two independent cohorts of NSCLC patients treated with ICIs (N = 305 patients) or chemotherapy (N = 162 patients) (Mezquita et al., 2018). The authors found that a LIPI score > 1 was associated with poor overall survival in patients treated with ICIs, but not in those treated with chemotherapy. However, a subsequent FDA analysis of the LIPI, which included pooled clinical trial data from studies evaluating 1,368 patients with second-line metastatic NSCLC who received ICIs and 1,072 patients who received chemotherapy, did not confirm these results (Kazandjian et al., 2018). This study showed that LIPI has a prognostic impact in second-line metastatic NSCLC, regardless of treatment modality (ICI or chemotherapy). Finally, another study investigated the LIPI score in patients with metastatic disease from various solid tumors participating in a phase 1 trial. Although this analysis was performed on a very heterogeneous cohort, a LIPI score >1 was also associated with poor overall survival, suggesting that its prognostic role is not limited to patients with NSCLC (Varga et al. 2019).

[0312] Immune fitness

[0313] All six 3i blood test variables (NLR, neutrophils, albumin, LDH, GGT, AST), assessed by standard laboratory measurements and identified by feature selection, have previously been reported to be prognostic in various tumor types. For example, a high NLR is associated with poor prognosis in multiple tumor types, including gastrointestinal (Bowen et al. 2017), melanoma (Ding et al. 2018), NSCLC (Fukui et al. 2019), UC (Sacdalan et al. 2018), and HNSCC (Yu et al. 2018, Tham et al. 2018). This ratio may reflect a measure of immune fitness that may influence tumor immune responses, and the association with poor prognosis was judged to be independent of ICI type. Fukui et al. observed a significant correlation between high NLR and poor prognosis in NSCLC patients treated with nivolumab. In a meta-analysis of ICI studies in melanoma, NSCLC, and UC, Sacdalan et al. observed an association between high NLR and worse outcomes, suggesting its potential as a prognostic marker. It has also been speculated that tumor burden and chronic inflammation lead to high NLR, which in turn leads to poor prognosis (Bigot et al., 2017). Absolute neutrophil count (ANC) is a surrogate marker of physiological stress including inflammation, and in NSCLC patients, high ANC after nivolumab treatment has been observed to be independently associated with poor overall survival (Khunger et al., 2018). Albumin (ALB) is a well-known marker of malnutrition and cachexia in cancer patients, and low ALB has been observed to be associated with poor overall survival in many malignancies such as SCCHN, breast cancer, lung cancer and gastrointestinal cancer (Gupta et al. 2010). In the case of chronic inflammation, the production of ALB is also reduced.

[0314] LDH is an established marker that reflects increased cell turnover, primarily in malignant diseases, but also in non-malignant diseases (e.g., hemolytic anemia). In a recent FDA analysis of early deaths in ICI studies, elevated LDH was reported as a significant risk factor, primarily in melanoma patients (Mulkey et al. 2019). In a systematic review of 76 studies, high LDH was associated with poor prognosis in renal cancer, melanoma, gastric cancer, prostate cancer, nasopharyngeal carcinoma, and lung cancer (Petrelli et al. 2015). In advanced cancers, increased LDH concentrations in the blood reflect increased metabolic activity and rapid cell proliferation, which leads to increased leakage of the intracellular enzyme LDH into the circulation.

[0315] Abnormal liver function tests, such as AST and GGT, are carefully monitored in cancer patients. Abnormal liver function tests reflect not only pre-existing disease but also cancer progression / liver metastasis and drug toxicity, which are known causes of death in cancer patients. Therefore, AST and GGT have been reported as prognostic markers in cancer (Freis et al. 2017, Luo et al. 2017), and GGT has also been proposed to play a role in tumor initiation, invasion, and drug resistance.

[0316] Methods for identifying immunocompromised patients

[0317] The 3i score can be used to identify patients with a high likelihood of EM and, therefore, likely lacking immune fitness. Lack of immune fitness suggests that ICI therapy should be postponed until these patients have been effectively treated to improve their immune fitness. Therefore, therapeutic interventions that enhance immune system fitness in 3i high-risk patients could allow these patients to receive ICI therapy, reduce their risk of EM, and, in addition, increase their chances of deriving therapeutic benefit from ICI therapy.

[0318] Similarly, this article considers methods for predicting the effectiveness of ICI treatment. For example, patients identified as 3i high-risk patients will be predicted to be unable to be effectively treated with ICI such as durvalumab.

[0319] Cancer type

[0320] Cancer types contemplated for treatment herein include, for example, NSCLC, advanced solid malignancies, biliary tract tumors, bladder cancer, colorectal cancer, diffuse large B-cell lymphoma, esophageal tumors, esophageal squamous cell carcinoma, extensive-stage small cell lung cancer, gastric adenocarcinoma, gastric cancer, gastroesophageal junction cancer, head and neck cancer, head and neck squamous cell carcinoma, hepatocellular carcinoma, Hodgkin's lymphoma, lung cancer, melanoma, mesothelioma, metastatic clear cell renal carcinoma, metastatic melanoma, metastatic non-cutaneous melanoma, multiple myeloma, nasopharyngeal tumors, non-Hodgkin's lymphoma, ovarian cancer, fallopian tube cancer, peritoneal tumors, pleural mesothelioma, prostate tumors, recurrent or metastatic PD-L1 positive or negative SCCHN, recurrent squamous cell lung cancer, renal cell cancer (renal cell cancer / renal cell carcinoma), SCCHN, hypopharyngeal squamous cell carcinoma, laryngeal squamous cell carcinoma, small cell lung cancer, squamous cell carcinoma of the head and neck, squamous cell lung cancer, TNBC, transitional cell carcinoma, unresectable or metastatic melanoma, urothelial cancer / urothelial carcinoma. In a particular embodiment, the intended treatment regimen may include a biological component, such as an antibody and a chemotherapeutic component. Contemplated antibodies include anti-PD-L1 antibodies (e.g., durvalumab (MEDI4736), nivolumab, pembrolizumab, avelumab, atezolizumab, KNO35), anti-PD-1 antibodies (e.g., REGN2810, SHR1210, IBI308, PDR001, anti-PD-1, BGB-A317, BCD-100, and JS001), and anti-CTLA4 antibodies (e.g., tremelimumab or ipilimumab). Additional antibodies are also contemplated herein. Any therapeutically effective antibody sub-portion is also contemplated herein.

[0321] Treatment

[0322] In one embodiment, the present invention relates to effectively treating cancer patients with novel anti-cancer therapeutics and providing cancer patients with greater access to existing therapeutics. For example, the present invention contemplates treating cancer patients with immune checkpoint inhibitors (ICIs) such as those targeting the PD-1 / PD-L1 axis (PDX) and other IO therapies such as immune system agonists.

[0323] Anticancer therapeutics

[0324] Information about durvalumab (or a fragment thereof) for use in the methods provided herein can be found in U.S. Patent Nos. 8,779,108; 9,493,565; and 10,400,039, the disclosures of which are incorporated herein by reference in their entirety. In a specific aspect, durvalumab or an antigen-binding fragment thereof for use in the methods provided herein comprises the variable heavy chain and variable light chain CDR sequences of the 2.14H90PT antibody as disclosed in the aforementioned U.S. Patents.

[0325] Information regarding tremelimumab (or an antigen-binding fragment thereof) for use in the methods provided herein can be found in U.S. Patent No. 6,682,736 (where tremelimumab is referred to as 11.2.1), the disclosure of which is incorporated herein by reference in its entirety.

[0326] The amount of anti-cancer therapeutic agent, such as an antibody or antigen-binding fragment thereof, administered to a patient will depend on various parameters, such as the patient's age, weight, clinical assessment, tumor burden, and / or other factors (including the judgment of the attending physician). Any acceptable route of administration is contemplated, such as, but not limited to, intravenous administration (e.g., intravenous infusion), parenteral, or subcutaneous administration.

[0327] Additional therapeutic agents (chemotherapeutic agents or biologics) contemplated herein include, but are not limited to, cisplatin / gemcitabine or methotrexate, vinblastine, ADRIAMYCIN TM (doxorubicin), cisplatin (MVAC), carboplatin-based regimens, or single-agent taxanes or gemcitabine, temozolomide, or dacarbazine, vinflunine, docetaxel, paclitaxel, nab-paclitaxel, vemurafenib, erlotinib, afatinib, cetuximab, bevacizumab, erlotinib, gefitinib, and / or pemetrexed. Additional examples include drugs that target the DNA damage repair system, such as poly (ADP-ribose) polymerase 1 (PARP1) inhibitors and therapeutic agents that inhibit WEE1 protein kinase activity, ATR protein kinase activity, ATM protein kinase activity, Aurora protein kinase B activity, and DNA-PK activity.

[0328] Any therapeutic composition or method contemplated herein can be combined with one or more of any other therapeutic compositions and methods provided herein.

[0329] Companion diagnostics

[0330] In one embodiment, the present invention contemplates a device comprising one or more laboratory tests as described elsewhere herein for measuring the concentration of a clinical pathology marker in a patient sample applied to the device. The device may also include a computing system comprising a gradient boosting machine learning model to predict whether a patient is likely to die within a given time period based on the measured concentrations of the clinical pathology markers from the one or more laboratory tests.

[0331] In another embodiment, a system is contemplated that includes one or more testing devices directly or indirectly linked to one or more computing devices, such that when the concentration of one or more clinical pathology markers is determined, the data is sent to the computing device to be provided to a gradient boosting machine learning model within the computing device. The gradient boosting machine learning model can, in turn, output a prediction of the patient's EM and / or immune fitness or the effectiveness of an ICI therapeutic agent (such as durvalumab) to another component of the system.

[0332] In some embodiments, commercially available tests can be used to measure the concentration of clinical pathology markers, such as antibody-based assays, in which antibodies specific for a marker of interest are used to target such markers. Binding of the antibody to the specific marker can be detected, and the concentration and / or concentration of the marker of interest can be determined by the relative amount of antibody binding.

[0333] In some embodiments, samples used include tissue biopsy, cerebrospinal fluid, lymph, whole blood, blood fractions (eg, serum or specific blood cell populations), urine, sweat, tears, saliva, and / or stool.

[0334] In one embodiment, the patient can be assessed upon arrival at the clinic for participation in the clinical trial. In this case, blood can be drawn using standard techniques, and the patient's clinical pathology markers (e.g., neutrophils, hemoglobin, aspartate aminotransferase, alkaline phosphatase, etc.) can be assessed, and the patient's overall survival over a given time period, e.g., 4, 8, 12, or 24 weeks, 4, 6, 8, or 10 months, 1, 2, 3, 4, or 5 years, can be predicted.

[0335] In another embodiment, patients can be pre-approved, remotely enrolled, or flagged for accelerated treatment through remote review of laboratory reports, where a centralized software platform can assess the patient's clinicopathological markers (e.g., neutrophils, hemoglobin, aspartate aminotransferase, alkaline phosphatase, etc.) and determine an overall score.

[0336] The following examples are put forth so as to provide those skilled in the art with a complete disclosure and description of how to carry out the invention and are not intended to limit the scope of what the inventors regard as their invention.

[0337] Example 1 - Validation of a Prognostic Score to Improve Patient Selection in Immuno-Oncology Trials (High Risk of Early Mortality; HREM)

[0338] This example provides an overview of a new prognostic scoring method for immuno-oncology clinical trials. As used herein, "high risk of early mortality," "HREM," and "FastProgIO" are interchangeable.

[0339] background

[0340] Life expectancy estimates are critical for selecting patients for oncology clinical trials, particularly in the phase 1 setting. However, most study protocols use subjective eligibility criteria to meet this goal (e.g., 12-week life expectancy). Although eligibility criteria can help determine which patients can be enrolled, they do not provide a reliable measure for determining overall survival (see Figure 1A Several prognostic scores have been developed from retrospective studies to help predict EM and OS in patients ( Figure 1B ), but their utility in the immuno-oncology (IO) setting remains unknown. The high percentage of EM patients (overall survival [OS] ≤ 12 weeks) observed in IO trials further emphasizes the need for more objective prognostic criteria.

[0341] Purpose

[0342] The aim of this study was to develop a new prognostic score using basic laboratory tests at baseline to identify patients at high risk of early death (OS ≤ 12 weeks) using data from three phase I IO trials including patients with non-small cell lung cancer (NSCLC) and urothelial carcinoma (UC). In addition, this study sought to compare the prevalence of patients at high risk of early death across clinical sites in these studies. The new prognostic score (HREM) was benchmarked against three published prognostic scores (RMH, GRIM, and LIPI). Finally, the impact of programmed cell death ligand-1 (PD-L1) expression on tumor / immune cells and liver metastasis on predicting patients at high risk of early death was evaluated.

[0343] method

[0344] Methods: Patients with NSCLC and UC (N = 972) who were prospectively enrolled in three phase 1 trials (NCT01693562, NCT02000947, and NCT02261220) investigating durvalumab (anti-PD-L1) plus or minus tremelimumab (anti-cytotoxic T lymphocyte antigen [CTLA]-4) at 160 centers were analyzed. Differences in 12-week life expectancy rates were also assessed across centers.

[0345] A composite score of clinicopathological variables predicting 12-week life expectancy (HREM) was developed in 648 patients and validated in 324 patients using a multivariate regression approach ( Figure 2 Neutrophil, hemoglobin aspartate aminotransferase, and alkaline phosphate concentrations were measured in blood samples taken from patients using commercially available assays.

[0346] The performance of HREM (predicting patients at high risk of early death with an overall survival less than or equal to 12 weeks) was compared with existing published scores (see Figure 1) in terms of time-dependent true positive rate (TPR) and false positive rate (FPR).

[0347] result

[0348] The rates of rapid progression in patients analyzed in each trial are shown in Table 1 (see also Figure 3A and 3B ).

[0349] Table 1. Rapid progression rates in phase I immuno-oncology trials

[0350]

[0351] Baseline characteristics (shown in Table 2) were similar between patients with NSCLC and UC.

[0352] Table 2. Patient characteristics

[0353]

[0354]

[0355] *For NSCLC, high: TC ≥ 25%, low / negative: TC < 25%; for UC, high: TC ≥ 25% or IC ≥ 25%, low / negative: TC < 25% and IC < 25%.

[0356] HREM score

[0357] The HREM score includes neutrophil count, aspartate aminotransferase, alkaline phosphatase, and hemoglobin as predictive markers. The true positive rate (TPR) for identifying patients with OS ≤ 12 weeks was 74.7% (74.1% in the training dataset and 75.9% in the test dataset), while the false positive rate (FPR) was controlled at around 10% ( Figure 4 HREM had a time-dependent TPR superior to or similar to RMH, GRIM, and LIPI in the OS range of 4-12 weeks ( Figure 5 The time-dependent FPR of HREM is better than that of GRIM, slightly lower than that of RMH and higher than that of LIPI ( Figure 5 B) Patients at high risk of early death identified by HREM and those not at high risk of early death showed different survival profiles ( Figure 6 Patients at high risk of early death identified by HREM had shorter OS than those identified by three published scores ( Figure 7 ).

[0358] Impact of PD-L1 expression and liver metastasis on patients with high risk of early death

[0359] The PD-L1 effect was observed in patients at high risk of early death predicted by HREM but not in those predicted by other scores: patients at high risk of early death predicted by HREM with low / negative PD-L1 expression progressed faster than those with high PD-L1 expression ( Figure 8 ).

[0360] The effect of liver metastases was observed in patients at high risk of early death predicted by all four scores: patients with liver metastases (mets) progressed faster than those without ( Figure 9 Among patients without liver metastases, those at high risk of early death predicted by HREM had worse survival than those predicted by the other three scores.

[0361] in conclusion

[0362] The significant differences between study sites in the prevalence of patients at high risk for early death suggest that the use of subjective criteria to estimate 12-week life expectancy at study recruitment is clearly suboptimal. In this large cohort of patients with NSCLC and UC, HREM significantly outperformed the RMH, GRIM, and LIPI prognostic scores in predicting patients at high risk for early death. An effect of PD-L1 was observed in patients at high risk for early death predicted by HREM but not by the three published scores, suggesting that the HREM approach has a higher sensitivity for evaluating patients.

[0363] The effect of liver metastases was observed in patients predicted to be at high risk of early death by all four scores. However, among patients without liver metastases, those predicted to be at high risk of early death by HREM had shorter OS than those predicted by published scores. The success of HREM prognostic assessment in patients with NSCLC and UC suggests that this approach could be effective in other cancer types.

[0364] Example 2 - Evaluation of HREM in various tumor types

[0365] This example demonstrates the applicability of the HREM prognostic scoring method to scoring other cancer types.

[0366] method

[0367] The prevalence of patients predicted to be at high risk for early death in patients with various cancer types from a Phase I study (Study 1108) was analyzed.

[0368] result

[0369] Table 3 shows the rapid progression rate of the patients analyzed (see also Figure 10 ).

[0370] Table 3. Rapid progression rates in various cancer types

[0371]

[0372]

[0373] Table 4 shows the comparison of HREM with other published scores.

[0374] Table 4. Comparison of prediction scores in other cancer types

[0375]

[0376] in conclusion

[0377] HREM can be used to predict patients at high risk of early death across multiple cancer types. Therefore, this prognostic score can be used to improve patient intervention and phase I outcomes across multiple cancer types. The strength of the predictions can be increased by increasing the sample size.

[0378] Example 3 - Correlation between HREM and inflammatory proteins

[0379] This example investigates the potential correlation between HREM findings and inflammatory proteins.

[0380] method

[0381] Using the UC cohort, the concentrations of inflammatory proteins in patient serum samples were tested for correlation with HREM scores using Spearman correlation.

[0382] result

[0383] HREM scores were significantly correlated with 10 different inflammatory proteins (see Figure 11 ).

[0384] in conclusion

[0385] HREM can be used to draw therapeutically relevant biological inferences to help inform patient care and / or Phase I study selection.

[0386] Example 4 - Development of a pan-tumor prognostic model (F1 model)

[0387] This example attempts to develop an improved prognostic model (F1 model) based on the successful construction of the HREM model.

[0388] method

[0389] Patients from a combination of 4 studies with several cancer types were used to develop a new prognostic model with improved true-positive and false-positive rates (Table 5) (see also Figure 12 and 13 ).

[0390] Table 5. Studies used for the F1 patient pool

[0391]

[0392]

[0393] F1 model development Figure 14 Several standard laboratory tests, such as ALB, LAP, AST, LDH, NEUT, and WBC, were included in the model. The model was developed using predictive modeling methods such as ensemble tree models and random forest models.

[0394] result

[0395] The prediction performance of the F1 model in the training and test data is as follows Figure 15 The total survival of the F1 model is shown as Figure 16 shown. Figure 17 Comparison of the F1 model with published prognostic scores is shown. The impact of PD-L1 on liver metastasis is shown in Figure 18 and 19 Taken together, these results indicate that the F1 model is a much better predictor of OS than previously available prognostic scores such as RMH, GRIM, and LIPI.

[0396] Example 5 - Retrospective Use of the F1 Model in Study 21 (Gastroesophageal Cancer) and ARCTIC (NSCLC)

[0397] To demonstrate its usefulness in retrospective analyses of previous trials, the F1 model was used to review 2 phase I trials in gastroesophageal cancer using durvalumab alone (Study 1108) or durvalumab alone (D), tremelimumab alone (T), durvalumab plus tremelimumab (D+T), and durvalumab plus tremelimumab in an IFNγ+ cohort (Study 21). Results are compared in Figure 20 A and 20B.

[0398] The ARCTIC study results ( Figure 21 The results of applying the F1 model to the ARCTIC study are as follows. Figure 22 shown.

[0399] in conclusion

[0400] The F1 model is a more powerful OS predictor than the HREM model. It is superior to published scores (poor OS, generalizable to all tumor types, PD-L1 influence). The F1 model can be used to evaluate ongoing or previously conducted clinical trials to assess any imbalance in rapid progression rates between groups that may have a potential impact on the results. In addition, it is believed that the F1 model may be incorporated into the inclusion / exclusion criteria in future Phase I trials (incremental phase, randomized expansion) as a more objective way to exclude patients with poor risk. In addition, the selection of study sites is improved by excluding sites with a history of enrollment of high-risk patients with a high ratio of early death. Excluding patients at high risk of early death in Phase 1 studies (incremental and randomized expansion parts) will likely increase the ability to detect efficacy signals in Phase I trials.

[0401] Example 6-Correlation between F1 model and inflammatory proteins

[0402] This example investigates potential correlations between F1 model results and inflammatory proteins.

[0403] method

[0404] Using Study 1108, the UC cohort, patient serum samples were tested for concentrations of inflammatory proteins (C-reactive protein, IL-6, TIMP-1, and IL-8) that correlated with F1 model scores.

[0405] result

[0406] HREM scores were significantly correlated with 10 different inflammatory proteins (see Figure 23Other relevant biomarkers (r>0.5) included ferritin (FRTN), tenascin-C (TN-C), 70-kilodalton heat shock protein (HSP70), necrosis factor receptor 1 (TNFR1), interleukin-2 receptor α (IL-2Rα), tumor necrosis factor receptor 2 (TNFR2), insulin-like growth factor binding protein 2 (IGFBP2), and myeloid progenitor inhibitory factor 1 (MPIF1).

[0407] in conclusion

[0408] The F1 model can be used to make therapeutically relevant biological inferences to help inform patient care and / or Phase I study selection.

[0409] Example 7 - Training and testing an updated F1 model using the EAGLE test and neutrophils (NEUT), lymphocyte percentage (LYM%), albumin (ALB), lactate dehydrogenase (LDH), and URATE as clinical pathology markers

[0410] This example sought to test an improved prognostic model (F1 model) that was successfully constructed based on the first F1 model.

[0411] method

[0412] Patients from the combined 3 studies with several cancer types were used to develop a new prognostic model with improved true-positive and false-positive rates (Table 6).

[0413] Table 6. Studies used to train F1 models

[0414]

[0415] Model development such as Figure 14 shown.

[0416] To test this model, the EAGLE study was used. (736 patients with head and neck squamous cell carcinoma were treated with three treatment regimens: chemotherapy, durvalumab monotherapy, and durvalumab plus tremelimumab.) Overall survival for each regimen in the intention-to-treat trial is shown in Table 1. Figures 24-26 Hazard ratios (HRs) were calculated using proportional cox regression, including PD-L1 (≥25% vs <25%), tumor location / HPV status, and smoking status as strata. P values were calculated using the log-rank test, stratified by PD-L1 (≥25% vs <25%), tumor location / HPV status, and smoking status.

[0417] result

[0418] The updated F1 model predicted patients at high risk of early death ( Figure 27 ). Figure 28 Shown are the overall survival of patients in chemotherapy and durvalumab regimens tested with the updated F1 model. Figure 29 Overall survival is shown for patients on chemotherapy and durvalumab regimens excluding those with a predicted high risk of early death. Figure 30 Shows the overall survival of all patients with PD-L1 high, Figure 31 Data for patients who were not predicted to be at high risk for early death are shown. Figure 32 The overall survival of all patients with low PD-L1 is shown. Figure 33 Data for patients who were not predicted to be at high risk for early death are shown. Figure 34 The overall survival of all PD-L1 negative patients is shown. Figure 35 Data are shown for patients without a prediction of a high risk of early death.

[0419] Example 8 - Training and testing an updated F1 model using the EAGLE test and neutrophils (NEUT), lymphocyte percentage (LYM%), albumin (ALB), and lactate dehydrogenase (LDH) as clinical pathology markers

[0420] This example attempts to improve the F1 model by removing URATE from the clinical pathology markers used.

[0421] method

[0422] Patients from the combined cohort of 3 studies with several cancer types were used to develop a new prognostic model with improved true-positive and false-positive rates (Table 7). Hazard ratios (HRs) were calculated by proportional cox regression, including PD-L1 (≥25% vs. <25%), tumor location / HPV status, and smoking status as strata. P values were calculated by log-rank test stratified by PD-L1 (≥25% vs. <25%), tumor location / HPV status, and smoking status.

[0423] Table 7. Studies used to train F1 models

[0424]

[0425]

[0426] Model development such as Figure 14 shown.

[0427] To test the model, the EAGLE study (736 patients with head and neck squamous cell carcinoma treated with three treatment regimens: chemotherapy, durvalumab monotherapy, and durvalumab plus tremelimumab) was used.

[0428] result

[0429] The F1 model using neutrophil count (NEUT), lymphocyte percentage (LYM%), albumin (ALB) and lactate dehydrogenase (LDH) was used to predict patients at high risk of early death ( Figure 36 ). Figure 37 Shown are the overall survival of patients in chemotherapy and durvalumab regimens tested with the updated F1 model, excluding those patients identified as at high risk for early death. Figure 38 Shown is the overall survival of all PD-L1 high patients excluding patients predicted to be at high risk of early death. Figure 39 showed that there was no overall survival in patients with low PD-L1 who were predicted to have a high risk of early death. Figure 40 Overall survival of PD-L1-negative patients was shown, with no prediction of patients at high risk of early death.

[0430] Example 9 - Training and testing an updated F1 model using the EAGLE and MYSTIC assays and neutrophil / lymphocyte ratio (NLR); neutrophils; albumin; lactate dehydrogenase; gamma-glutamyl transferase; and aspartate aminotransferase as clinicopathological markers for non-small cell lung cancer and squamous cell head and neck cancer

[0431] This example attempts to use the discovery of predictors of non-small cell lung cancer and squamous cell carcinoma of the head and neck from patients at high risk of early death to develop a single model for predicting both cancers. The non-small cell lung cancer markers evaluated were neutrophil / lymphocyte ratio (NLR); albumin (ALB), lactate dehydrogenase (LDH), gamma-glutamyl transferase (GGT), and aspartate aminotransferase (AST). The squamous cell carcinoma of the head and neck predictors of patients at high risk of early death were neutrophil (NEUT), lymphocyte percentage (LYM%), albumin (ALB), lactate dehydrogenase (LDH), and URATE. Six markers were selected: neutrophil / lymphocyte ratio (NLR); neutrophil (NEUT), albumin (ALB), lactate dehydrogenase (LDH), gamma-glutamyl transferase (GGT), and aspartate aminotransferase (AST).

[0432] method

[0433] Patients from a combination of 7 studies with several cancer types were used to develop a new prognostic model with improved true-positive and false-positive rates (Table 8). Hazard ratios (HRs) were calculated by proportional cox regression, including PD-L1 (≥25% vs. <25%), tumor location / HPV status, and smoking status as strata. P values were calculated by log-rank test stratified by PD-L1 (≥25% vs. <25%), tumor location / HPV status, and smoking status.

[0434] Table 8. Studies used to train F1 models

[0435]

[0436]

[0437] Model development such as Figure 14 shown.

[0438] To test the model, the ARCTIC, EAGLE (736 participants with head and neck squamous cell carcinoma treated with three treatment regimens: chemotherapy, durvalumab monotherapy, and durvalumab plus tremelimumab), and MYSTIC studies (1,118 participants with PD-L1-high non-small cell lung cancer treated with multiple treatment regimens: chemotherapy, durvalumab monotherapy, and durvalumab plus tremelimumab) were used.

[0439] result

[0440] The F1 model used neutrophil / lymphocyte ratio (NLR); neutrophils; albumin; lactate dehydrogenase; gamma-glutamyl transferase; and aspartate aminotransferase to predict patients at high risk of early death in the EAGLE trial ( Figure 41 ). Figure 42 Shown are the overall survival of patients in the chemotherapy and durvalumab regimens of the EAGLE trial tested with this F1 model, excluding those patients identified as having a high risk of early death. Figure 43 shows the overall survival of all PD-L1 high patients in the EAGLE trial excluding patients predicted to be at high risk of early death, Figure 44 showed that there was no overall survival in patients with low PD-L1 who were predicted to have a high risk of early death. Figure 45 The overall survival of PD-L1 negative patients was not predicted for patients at high risk of early death. The F1 model predicted patients at high risk of early death in the MYSTIC trial ( Figure 46 ). Figure 47 shows the overall survival of all patients with PD-L1 high in the MYSTIC trial, Figure 48 As shown in , PD-L1 high patients in the MYSTIC trial who were predicted to have a high risk of early death were excluded. Figure 49 The percentages of overall survival observed in several cohorts that were longer than 12 weeks or equal to or shorter than 12 weeks are shown in Figure 50 Patients with and without high risk of early death predicted by F1 in these cohorts are shown.

[0441] in conclusion

[0442] The F1 model using neutrophil / lymphocyte ratio (NLR); neutrophils; albumin; lactate dehydrogenase; gamma-glutamyl transferase; and aspartate aminotransferase as clinical pathological markers is a powerful predictor of OS. The F1 model can be used to evaluate ongoing or previously conducted clinical trials to assess any imbalance in the rapid progression rate between groups that may have a potential impact on the results. In addition, it is believed that the F1 model may be incorporated into the inclusion / exclusion criteria in future Phase 1 trials (incremental phase, randomized expansion) as a more objective way to exclude patients with poor risk. In addition, the selection of study sites is improved by excluding sites with a high ratio of high-risk patients with early death history. Excluding high-risk patients with early death in Phase 1 studies (incremental and randomized expansion parts) will likely increase the ability to detect the efficacy signal of Phase I trials.

[0443] Example 10 - Development of an Immunological Immediate Index (3i Score) for Identifying Patients at Risk of Early Mortality

[0444] summary

[0445] New approach: identifying patients at high risk of early death in the setting of ICIs

[0446] While the RMH, GRIM, and LIPI scores demonstrate prognostic impact, the recent availability of a large cohort of patients treated with ICIs in a prospective trial by AstraZeneca represents an opportunity to improve the identification of patients at high risk of early mortality in the setting of ICIs relative to active chemotherapy comparators.

[0447] In randomized trials of ICIs versus active comparators in the advanced disease setting, a crossing of the Kaplan-Meier (KM) curve for OS has been primarily observed, with higher mortality in the ICI group typically observed within the first 12 weeks after randomization. This was illustrated in the MYSTIC study of durvalumab versus chemotherapy in first-line NSCLC. Figure 47 Shown are the OS curves (crossing of the curves is indicated by arrows) in the primary analysis population of PD-L1-positive patients (defined as those whose tumors expressed PD-L1 in ≥25% of cells). The percentage of deaths in the first 12 weeks was 17.2% in the experimental group and 8.0% in the control group. Between weeks 12 and 24, these rates were 12.7% and 15.0%, respectively. Over time, the mortality rate continued to improve in favor of durvalumab. Figure 51Shown are KM curves for OS after excluding patients who died within the first 12 weeks. In the subset of patients who survived 12 weeks or longer, no crossing of the OS curves was observed. Similar observations have been made in other studies of ICIs versus chemotherapy (Borghaei et al. 2015, Mok et al. 2019). Therefore, based on this empirical observation in multiple trials, to predict the risk of EM with ICIs, EM was defined as any death occurring within the first 12 weeks after randomization (in randomized trials) or start of study treatment (in single-arm trials).

[0448] Development of the 3i score

[0449] Considering the data collected in studies investigating ICIs, the challenges in objectively predicting EM, and the limitations of previously established scores, we proposed using machine learning to develop a prognostic model (hereafter referred to as the Immunity Immediate Index [3i score]) for identifying patients at risk for EM. The development workflow, including definition of model requirements, design, development, and testing, is described below.

[0450] Model requirements

[0451] like Figure 52 As shown, the key requirement of the 3i score is to use machine learning to predict EM (death within 12 weeks after randomization or treatment initiation), ideally using continuous input values obtained from some or all routinely performed baseline laboratory tests. Machine learning models are increasingly used in healthcare and clinics to identify subtle but predictive patterns in raw data that may be lost due to discrete laboratory parameters. Deploying machine learning models in real-world environments is complex due to the need for robust performance and interface software. Therefore, acceptable performance metrics throughout training, tuning, and testing, as well as flexible deployment options, influence the final 3i score.

[0452] The identification of datasets for development, separation of datasets for training, tuning, and testing, and design choices for feature and method selection are driven by model requirements and cross-functional expertise in machine learning and clinical research.

[0453] Develop machine learning models through training, tuning, and testing

[0454] The core activities in developing supervised machine learning models are most commonly broken down into three key steps: training, tuning, and testing.

[0455] During training, model parameters such as the tree depth are optimized to encode the relationship between input and output. In an iterative manner, the model accepts input from the training dataset and produces output, which is compared to the real data, and the error guides the update of the model parameters. Over time, the model learns input representations that lead to the desired target output using the training dataset.

[0456] The same machine learning technique can produce different models in terms of architecture, weights, and performance depending on the training constraints chosen. The optimal training constraints, called hyperparameters, are different in each machine learning case and are set before training, in contrast to other model parameters, which are developed during training.

[0457] Hyperparameter optimization (also known as tuning) is a step in machine learning development that is often found to have a profound impact on a model's ability to generalize successfully. The choice of hyperparameters can lead to models that overfit or underfit on the same training set. A common approach to exploring the space of hyperparameter ranges is grid search, where each time a set of hyperparameters is sampled, the model is trained on the training dataset and then evaluated on an independent dataset (the tuning dataset). Performance evaluation on the tuning set guides the selection of optimal training constraints, which produce a model that is robust enough to function outside of the training set.

[0458] Model testing is performed using the chosen hyperparameters to evaluate the final model fit on the training dataset. Testing is performed on an independent dataset (that was not used for training or tuning) to assess how well the model generalizes and derives patterns beyond the data it has previously encountered.

[0459] Dataset Identification

[0460] A comprehensive set of ICI trials was used together with individual patient concentration data (Table 9) to develop the 3i scoring model. The trials were selected so that the training and adjustment sets would include data from patients with a range of different tumor types and different lines of treatment.

[0461] Separation of datasets for training, tuning, and testing

[0462] The data for training, tuning, and testing are separated as follows:

[0463] Training set (for final input feature selection, model training, and hyperparameter optimization):

[0464] - Study 1108 (N=923), Study 06 (N=353), Study 10 (N=164), Study 21 (N=98)

[0465] ATLANTIC (D4191C00003), a phase II, non-comparative, open-label, multicenter, international study of MEDI4736 in patients with locally advanced or metastatic non-small cell lung cancer (stage IIIB-IV) who have received at least 2 prior systemic therapy regimens, including 1 platinum-based chemotherapy regimen (N=371)

[0466] CONDOR (D4193C00003), a Phase II, multicenter, single-arm, global study of MEDI4736 monotherapy, tremelimumab monotherapy, and MEDI4736 in combination with tremelimumab in patients with recurrent or metastatic head and neck squamous cell carcinoma (N=193)

[0467] -HAWK (D4193C00001), a Phase II, multicenter, single-arm, global study of MEDI4736 monotherapy in patients with recurrent or metastatic head and neck squamous cell carcinoma (N=111)

[0468] Note: Patients censored before 12 weeks were not included in the training set.

[0469] Tuning set (for model hyperparameter optimization):

[0470] -MYSTIC (D419AC00001), a phase III randomized, open-label, multicenter, global study of MEDI4736 in combination with tremelimumab or MEDI4736 monotherapy versus standard of care platinum-based chemotherapy in the first-line treatment of patients with advanced or metastatic non-small cell lung cancer (N=1118)

[0471] Test set (for model testing):

[0472] -EAGLE (D4193C00002), a Phase III randomized, open-label, multicenter, global study of MEDI4736 monotherapy and MEDI4736 in combination with tremelimumab versus standard of care in patients with recurrent or metastatic head and neck squamous cell carcinoma (N=736)

[0473] The total training dataset includes 2213 patients, mainly including patients diagnosed with NSCLC, SCCHN, UBC, and gastric cancer, which are summarized in Table 10.

[0474] Table 10 Number of patients with different tumor types in the training set

[0475]

[0476] Abbreviations: NSCLC = non-small cell lung cancer; SCCHN = squamous cell carcinoma of the head and neck; UBC = urothelial bladder cancer; other = HPV-positive cancer, advanced malignant melanoma, uveal melanoma, pancreatic cancer, advanced cutaneous melanoma, hepatocellular carcinoma, nasopharyngeal carcinoma, ovarian cancer, SCLC, soft tissue sarcoma, MSI-high cancer, triple-negative breast cancer, colorectal cancer, and renal cell carcinoma.

[0477] Feature Selection

[0478] One advantage of using certain machine learning algorithms is their ability to determine feature importance and weight predictive features relative to non-predictive features. Given that model complexity increases with the number of features or dimensions, and that one of the requirements was the use of routinely available laboratory measurements (i.e., covering laboratory measurements of hematology, chemistry, and thyroid hormones available in the clinical trials selected here), a comprehensive starting list of 37 candidate predictors was collated and summarized in Table 11. It was planned to remove unimportant or highly correlated features during the development phase.

[0479] Table 11. List of candidate predictors considered for constructing a prognostic early death model

[0480]

[0481]

[0482] Machine Learning Method Selection (Gradient Boosting Trees)

[0483] A gradient boosting modeling approach was selected to predict the 12-week EM. The specific version of the gradient boosting method used is called XGBoost (extreme gradient boosting). This machine learning method has gradually become the preferred choice in the statistical community due to its robust predictive performance, its speed (i.e., scalability on parallel processes), and the regulation of missing values (Friedman et al. 2000, Friedman et al. 2001, Chen and Guestrin 2016). Decision trees are used as weak learners in gradient boosting. Each time a variable is selected to split a node into two subnodes, until no split is available to further improve the model fit. A separate tree is built on the residuals of the previous tree. Variables can be selected multiple times in one tree and in different trees. Decision trees are constructed sequentially until the stopping criteria are met. By cumulatively assembling decision trees, it is possible to apply a weighted learning rate to each to build the final gradient boosting model. By selecting appropriate model hyperparameters, the XGBoost model can be trained to minimize the prediction error while the model complexity is controlled, resulting in a good balance of bias-variance trade-off. The XGBoost model handles missing values by learning the default direction. Internally, XGBoost learns the direction of travel when values are missing.

[0484] Test Plan

[0485] The testing plan was designed to evaluate the 3i score by retrospective application to the ICI clinical trial test set (EAGLE). The EAGLE studies were excluded from training and tuning to represent an independent test set, where the model was evaluated based on: (1) ability to generalize beyond the training set as defined by true positive rate (TPR), false positive rate (FPR), and fractional area under the curve (AUC) in the specificity region between 0.7 and 0.9; (2) comparison with other previously developed risk scores; and (3) clinical utility, defined as the increase in treatment effect after excluding the high-risk group of the 3i score.

[0486] Code testing is not suitable for the test plan because no new functionality is introduced, nor is any existing functionality extended in the open source R packages used in development.

[0487] Development

[0488] Development Environment

[0489] All development was performed in R v.3.5.1 on an x86_64pc Linux GNU (64-bit) operating system. Table 12 lists all open source packages used for development.

[0490] Table 12. Open-source software packages used to develop the 3i scoring model

[0491]

[0492]

[0493] Dataset separation and labeling

[0494] The collected datasets were preprocessed internally according to internal data management standards. Patients who died within 12 weeks (i.e., the time from randomization / first dose to death was ≤12 weeks) were labeled as EM. Patients who survived beyond 12 weeks (i.e., the time from randomization / first dose to death was >12 weeks) were not labeled as EM. During model development, patients censored within 12 weeks (i.e., those lost to follow-up for ≤12 weeks) were not included in the training set. No other data processing was performed.

[0495] Feature Selection

[0496] Feature selection was performed to examine a wide range of candidate predictors for the 3i scoring model and to return a subset of predictors from which an efficient model could be fitted. First, a gradient boosting fitting procedure was used on the training data ( Figure 52 ). Model fitting was performed using 10-fold cross validation. A grid search of hyperparameters was performed, and the model that gave the highest partial AUC was used to guide feature selection.

[0497] A gradient boosting model was fitted to the training data to calculate variable importance. Variable importance was quantified by how much improvement in accuracy resulted from including the variable. All available pre-treatment blood test variables (n=31) covering hematology, clinical chemistry, and thyroid hormones were initially included in this first step (i.e., all laboratory variables). Clinical variables recognized to be associated with oncological prognosis, such as age ("AGE"), sex ("SEX"), baseline tumor size (estimated by RECIST, "TUMSZINV"), body mass index ("BMI"), tumor type ("typ"), and the presence of liver metastases ("LIVERBL"), were also introduced into the model. Figure 53 As shown, albumin, LDH, neutrophils, neutrophil / leukocyte ratio (NEUTLE), NLR, gamma-glutamyl transferase (GGT) and aspartate aminotransferase (AST) were found to be the most important variables. NEUTLE was not selected when constructing the final model because neutrophils and NEUTLE were highly correlated (r=0.98, p<0.001). Based on this feature importance analysis and subject matter expert input, the final set of input features for training and adjusting the 3i scoring model was albumin, LDH, neutrophils, NLR, GGT and AST.

[0498] Biological relevance of blood test variables selected for the 3i scoring model

[0499] All six blood test variables, assessed by standard laboratory measurements and identified by feature selection, have previously been reported to be prognostic in various tumor types.

[0500] A high NLR has been associated with poor prognosis in multiple tumor types, including gastrointestinal (Bowen et al. 2017), melanoma (Ding et al. 2018), NSCLC (Fukui et al. 2019), UC (Sacdalan et al. 2018), and HNSCC (Yu et al. 2018, Tham et al. 2018). This ratio may reflect a measure of immune fitness that may influence tumor immune responses, and the association with poor prognosis was judged to be independent of the type of ICI. Fukui et al. observed a significant correlation between high NLR and poor prognosis in NSCLC patients treated with nivolumab. In a meta-analysis of ICI studies in melanoma, NSCLC, and UC, Sacdalan et al. observed an association between high NLR and worse outcome, suggesting its potential as a prognostic marker. It has also been hypothesized that tumor burden and chronic inflammation lead to high NLR, which in turn leads to poor prognosis (Bigot et al. 2017).

[0501] Absolute neutrophil count (ANC) is a surrogate marker of physiological stress including inflammation, and in NSCLC patients, high ANC has been observed to be independently associated with poor overall survival after nivolumab treatment (Khunger et al., 2018).

[0502] ALB is a well-known marker of malnutrition and cachexia in cancer patients, and low ALB has been observed to be associated with poor overall survival in many malignancies such as SCCHN, breast cancer, lung cancer, and gastrointestinal cancer (Gupta et al. 2010).

[0503] LDH is an established marker that reflects increased cell turnover, primarily in malignant diseases, but also in non-malignant diseases (e.g., hemolytic anemia). In a recent FDA analysis of early deaths in ICI studies, elevated LDH was reported as a significant risk factor, primarily in melanoma patients (Mulkey et al. 2019). In a systematic review of 76 studies, high LDH was associated with poor prognosis in renal cancer, melanoma, gastric cancer, prostate cancer, nasopharyngeal carcinoma, and lung cancer (Petrelli et al. 2015). In advanced cancers, increased LDH concentrations in the blood reflect increased metabolic activity and rapid cell proliferation, which leads to increased leakage of the intracellular enzyme LDH into the circulation.

[0504] Abnormal liver function tests, such as AST and GGT, are carefully monitored in cancer patients. Abnormal liver function tests reflect not only pre-existing disease but also cancer progression / liver metastasis and drug toxicity, which are known causes of death in cancer patients. Therefore, AST and GGT have been reported as prognostic markers in cancer (Freis et al. 2017, Luo et al. 2017), and GGT has also been proposed to play a role in tumor initiation, invasion, and drug resistance.

[0505] 3i scoring model training and hyperparameter selection

[0506] Only the most predictive biological variables described previously in the feature selection step (NLR, neutrophils, albumin, LDH, AST, GGT) were retained ( Figure 53 ) and tumor types were used to train the final 3i scoring model ( Figure 52). There are 12 tree booster hyperparameters in XGBoost that set training constraints, such as the type of base learner, learning rate, maximum tree depth allowed, weight regularization. Since each hyperparameter set will produce a different XGBoost model during training, we use the MYSTIC data as an adjustment set to guide the selection of the best hyperparameter set. A grid search was performed for max.depth, gamma, min.child.weight, colsample.bytree, and subsample hyperparameters. A pool of XGBoost models was generated by fitting the models on the training dataset using different hyperparameter sets. The model performance in the training set was quantified as the true positive rate and partial AUC (specificity between 0.7 and 0.9). The hazard ratio and p-value (Cox regression) for the interaction between 3i score status and treatment in the PD-L1 population were calculated and used as a quantification of model performance in the adjustment set (MYSTIC). The model with the smallest interaction p-value (MYSTIC) in the adjustment set and a high partial AUC in the training set was selected as the final model.

[0507] The final 3i scoring model selected from the pool of 432 models has a pAUC of 0.83 (0.7, 0.9) on the training dataset (ranked 8th among all 432 models) and was trained with the following non-default XGBoost hyperparameters:

[0508] Maximum tree depth (max.depth): 3

[0509] Minimum loss reduction (γ) required to make a further partition on a leaf node of the tree: 1

[0510] Minimum required sum of instance weights (hessian) of children (min.child.weight): 5

[0511] Column subsample ratio when building each tree (colsample.bytree): 1

[0512] Subsample ratio of training instances (subsample): 0.6

[0513] scale_pos_weight: negative cases / positive cases

[0514] Number of rounds: 22

[0515] To describe the nature of the final model, the importance of each variable selected in the final model was calculated. Figure 54 As shown in Figure 3, NLR, NEUT, ALB, LDH, and GGT were more important for classification than AST and tumor type.

[0516] Final 3i scoring model

[0517] The following six key predictors were retained in the final model: NLR, NEUT, ALB, LDH, GGT, AST, and tumor type. The model generates a score (i.e., a value between 0 and 1) representing the probability of death within ≤12 weeks for each patient. The 3i score was then converted to a status that assigned the patient to a high or low prognostic or risk category. A cutoff value was determined to allow a 10% false positive rate for predicting EM in the training dataset and was calculated to be 0.649 on July 2, 2019. Patients with scores above the 0.649 cutoff were identified as high risk for EM, and patients at or below the cutoff (0.649) were identified as low risk for EM. The 3i score would not be calculated for patients with missing any of the six laboratory values. This decision was supported by the expectation that data for these variables would be available for patients globally as they are standard laboratory measurements collected routinely.

[0518] 3i relative to GRIm and LIPI in training and tuning datasets

[0519] When used in the training set (N = 2213 patients), the TPR at 12 weeks was 67% and the FPR was 10%. The performance, defined by pAUC (0.7, 0.9), was 0.83. The median OS for patients identified as high-risk 3i score status (486 / 2213 patients) was 9.29 weeks (95% CI: 8.29 weeks, 9.86 weeks), and the median OS for patients with low-risk 3i score status was 61.43 weeks (95% CI: 57.71, 66.14 weeks) (see Figure 55 ).

[0520] Patients identified as high risk by the 3i score (see Table 1) had a significantly higher OS than those identified as high risk using other prognostic models developed to predict early death in patients with advanced or metastatic cancer: GRIm (624 patients, median OS 14.29 weeks [95% CI: 12.71, 15.71]); and LIPI score (363 patients, median OS 14.00 weeks [95% CI: 11.43, 16.00]). Figure 56 A and 56B) had a shorter median OS. ULN = 240 was used to calculate LDH for GRIm and LIPI scores.

[0521] When used in the adjustment set (MYSTIC) (N=1118) (see Figure 57A and 57B), durvalumab-treated patients had a TPR of 52% and a FPR of 11% at week 12. Performance, as defined by pAUC (0.7, 0.9), was 0.63. Patients identified as high-risk 3i score status (69 of 374 patients) had a median OS of 12.00 weeks (95% CI: 9.57, 22.00), while those identified as low-risk 3i score status (289 of 374 patients) had a median OS of 68.71 weeks (95% CI: 60.29, 81.00). Patients identified as high risk by the 3i score had a shorter median OS compared with patients identified as high risk using other prognostic models: GRIm (115 patients, median OS 20.43 [95% CI: 14.29, 28.71]); and LIPI score (71 patients, median OS 18.14 [95% CI: 10.00, 33.29]).

[0522] In the MYSTIC intention-to-treat population (ITT) (see Figure 58 A-58C), the median OS of the durvalumab and SoC groups were 53.29 weeks and 51.14 weeks, respectively. In the MYSTIC ITT population (N=746), 120 patients (16%) were identified as high risk using the 3i score, 51 patients in the SoC group and 69 patients in the durvalumab group. Compared with the chemotherapy group (median OS: 29.14 weeks, HR=1.39 [95% CI: 0.947, 2.053]), the median OS of patients (120 / 746, 16% patients) who were identified as high-risk 3i score status in the durvalumab group (median OS: 12.00 weeks) appeared to show worse results. In low-risk patients, the treatment effect was still only moderately improved (HR=0.89 [95% CI: 0.735, 1.079]), and the crossing of the OS curve was not eliminated. This could be attributed to a possible lack of benefit of monotherapy ICIs in PD-L1 low / negative patients relative to active chemotherapy comparators in the advanced disease setting.

[0523] In the MYSTIC PD-L1 ≥25% subgroup (see Figure 59 A-59C), a total of 56 patients (17%) were identified as high 3i score risk, 24 in the SoC group and 32 in the durvalumab group. As in the ITT, for high-risk 3i score patients in the PD-L1 ≥ 25% subgroup, the median OS in the durvalumab group was shorter than that in the chemotherapy group (chemotherapy was 35.79 weeks, durvalumab was 11.36 weeks, HR = 1.533 [95% CI: 0.873, 2.726], Figure 59C). Importantly, exclusion of patients with high-risk status by 3i score from the MYSTIC PD-L1 ≥ 25% subgroup (56 / 325 patients, 17% excluded by 3i score) reduced the previously observed crossing of the OS curves ( Figure 59 B). The treatment effect of durvalumab was greater relative to chemotherapy (HR: 0.626 [95% CI: 0.459, 0.849]) compared with the original primary analysis PD-L1 ≥ 25% population (HR: 0.76 [95% CI: 0.59, 0.98]).

[0524] 3i Rating Test

[0525] Application of 3i scoring model in EAGLE (test set)

[0526] When used on the test set (EAGLE) (N=736) (see Figure 60 A and 60B), durvalumab-treated patients had a TPR of 57% and a FPR of 16% at week 12. Performance, as defined by pAUC (0.7, 0.9), was 0.64. Patients identified as high-risk 3i score status (59 / 240 patients) had a median OS of 12.14 weeks (95% CI: 8.57, 19.14), while those identified as low-risk 3i score status (172 / 240 patients) had a median OS of 48.00 weeks (95% CI: 41.29, 58.71). Patients identified as high risk by the 3i score had a shorter OS compared with patients identified as high risk using other prognostic models: GRIm (67 patients, median OS 15.71 weeks [95% CI: 12.14, 21.86]); and LIPI score (38 patients, median OS 19.36 weeks [95% CI: 13.71, 35.43]).

[0527] In EAGLE ITT (see Figure 61 A-61C), the median OS of the durvalumab (N=240) and SoC (N=249) groups were 33.14 and 36.00 weeks, respectively. Using the 3i score in the EAGLE ITT, a total of 121 patients (25%) were identified as high risk, 62 in the SoC group and 59 in the durvalumab group. The median OS of patients identified as high risk by the 3i score in the durvalumab group (median OS=12.14 weeks) appeared to be worse than that in the chemotherapy group (median OS=16.00 weeks, HR 1.30 [95% CI: 0.86, 1.99]). Excluding those patients identified as high risk by the 3i score from the EAGLE ITT (i.e., 25% of patients) reduced the crossing of the overall survival curve ( Figure 61B) The magnitude of the efficacy of durvalumab was increased (HR: 0.75 [95% CI: 0.58, 0.96]) compared with the original ITT population (HR: 0.88 [95% CI: 0.72, 1.08]).

[0528] In the PD-L1 ≥ 25% subgroup (N = 140) (see Figure 62 A-62C), a total of 43 patients (31%) were identified as high risk by the 3i score, 20 in the SoC group and 23 in the durvalumab group ( Figure 62 C). Similar to what was seen in the MYSTIC dataset (adjustment set), excluding these patients partially eliminated the crossing of the overall survival curves ( Figure 62 B). The magnitude of efficacy of durvalumab was increased (HR: 0.66 [95% CI: 0.38, 1.12]) compared with the original primary analysis PD-L1 ≥ 25% population (HR: 0.93 [95% CI: 0.63, 1.39]). Distribution of laboratory tests in patients with 3i low-risk and 3i high-risk scores

[0529] In the MYSTIC and EAGLE studies, the Wilcoxon test was used to compare the six laboratory values used in the 3i score between patients identified as high risk (predicted life expectancy ≤ 12 weeks) or low risk (predicted life expectancy > 12 weeks) by the 3i score. For all six parameters, the values for patients with high-risk and low-risk 3i score status were significantly different (p < 0.05) (see Figure 63 and 64 ).

[0530] Example 11 - Determination of Patient Immunological Fitness

[0531] summary

[0532] The variables included in the 3i score are associated with host immune fitness and cancer-associated inflammation. In addition, additional translational analyses of samples obtained from clinical trials showed that 3i high-risk patients were characterized by higher baseline concentrations of peripheral immune suppressor cells (MDSCs) in the blood and higher expression of immunosuppressive gene markers and higher proinflammatory protein concentrations (CRP, IL6, and IL8) in the serum. Interestingly, unlike immune checkpoint blockade, chemotherapy is recognized to reduce IL6, IL8, and CRP concentrations. This finding may explain the earlier death observed in patients with advanced / metastatic cancer treated with ICB compared with chemotherapy.

[0533] To investigate whether the 3i score could be used as an indicator of patient immune fitness, baseline molecular profiles of 3i high-risk patients were assessed by blood gene expression, peripheral blood immune profiling by flow cytometry, and serum protein expression.

[0534] result

[0535] Peripheral blood gene expression

[0536] Compared with 3i low in NSCLC, HNSCC, and UBC patients, 3i high (poor prognosis) was associated with lower baseline T effector and higher immunosuppressive cell gene expression signatures, regardless of patient PD-L1 status (see Figure 65 and 66 51 / 80 genes measured by qPCR were differentially expressed between 3i high and 3i low in the MYSTIC, ATLANTIC, CP1108, HAWK, CONDOR, 006, and 10 studies with a cutoff difference of 1.5-fold and a p-value < 0.01% (Wilcox test).

[0537] Peripheral immune profiling by flow cytometry

[0538] 3i high was associated with lower baseline T, NK, and B cell and higher absolute counts of monocytic myeloid-derived suppressor cells (MNC MDSC) compared with 3i low in NSCLC patients (see Figure 67 and 68 ). 11 / 22 immunophenotypes assessed in CP1108,006 and Mystic NSCLC patients and from 5 flow cytometry panels were found to be differentially expressed between the 3i high and 3i low patient groups (>20% CV, Wilcox test p value <0.05%).

[0539] Serum protein expression

[0540] Compared with 3i low NSCLC, HNSCC, and UBC patients, 3i high (poor prognosis) was highly associated with higher baseline proinflammatory protein concentrations, particularly C-reactive protein (CRP) and interleukins IL6 and IL8 (see Figure 69 ). 16-22 / 66 serum protein concentrations assessed by Luminex immunoassay (Myriad RBM) in the CP1108, HAWK, and CONDOR studies were differentially expressed in 3i high and 3i low NSCLC, HNSCC, and UBC patients (% difference compared with 3i low > 20% CV, Wilcox test p value < 0.05%). In addition to CRP, IL6, and IL8, pro-angiogenic and angiogenic proteins (e.g., ANGPT2, VEGF, ICAM1, and vWF) were found to be higher in the 3i high patient group than in the 3i low patient group.

[0541] in conclusion

[0542] The 3i index appears to have clinical utility in identifying patients who may lack immune fitness. Lack of immune fitness suggests that ICI therapy should be delayed until these patients have been effectively treated to improve their immune fitness or that they should be treated with a non-ICI therapy. Therefore, therapeutic interventions that enhance immune system fitness in patients with a high 3i index could allow these patients to receive ICI therapy, reduce their risk of EM, and further increase their chances of deriving therapeutic benefit from ICI therapy.

[0543] Example computing system

[0544] Figure 70 is a simplified block diagram of an exemplary computing device 7000, which illustrates some of the components that may be included in a computing device arranged to operate according to embodiments herein. Computing device 7000 may be a client device (e.g., a device actively operated by a user), a server device (e.g., a device that provides computing services to client devices), or some other type of computing platform. Some server devices may sometimes operate as client devices in order to perform certain operations, and some client devices may incorporate server features.

[0545] In this example, computing device 7000 includes a processor 7002, memory 7004, a network interface 7006, and an input / output unit 7008, all of which may be coupled via a system bus 7010 or similar mechanism. In some embodiments, computing device 7000 may include other components and / or peripherals (e.g., removable storage, a printer, etc.).

[0546] The processor 7002 may be one or more of any type of computer processing element, such as a central processing unit (CPU), a coprocessor (e.g., a math, graphics, or encryption coprocessor), a digital signal processor (DSP), a network processor, and / or an integrated circuit or controller that performs processor operations. In some cases, the processor 7002 may be one or more single-core processors. In other cases, the processor 7002 may be one or more multi-core processors having multiple independent processing units. The processor 7002 may also include register memory for temporarily storing instructions being executed and associated data, and a cache memory for temporarily storing recently used instructions and data.

[0547] Memory 7004 can be any form of computer-usable memory, including but not limited to random access memory (RAM), read-only memory (ROM), and non-volatile memory (e.g., flash memory, hard drives, solid-state drives, compact disks (CDs), digital video disks (DVDs), and / or magnetic tape storage). Thus, memory 7004 represents both primary memory and long-term storage. Other types of memory can include biological memory.

[0548] The memory 7004 may store program instructions and / or data on which the program instructions may operate. For example, the memory 7004 may store these program instructions on a non-transitory computer-readable medium so that these instructions may be executed by the processor 7002 to perform any method, process, or operation disclosed in this specification or the accompanying drawings.

[0549] like Figure 70 As shown, memory 7004 may include firmware 7004A, kernel 7004B, and / or application programs 7004C. Firmware 7004A may be program code used to boot or otherwise start some or all of the computing device 7000. Kernel 7004B may be an operating system, including modules for memory management, process scheduling and management, input / output, and communication. Kernel 7004B may also include device drivers that allow the operating system to communicate with hardware modules (e.g., memory units, network interfaces, ports, and buses) of computing device 7000. Application programs 7004C may be one or more user-space software programs, such as a web browser or email client, and any software libraries used by these programs. Memory 7004 may also store data used by these and other programs and applications.

[0550] The network interface 7006 may take the form of one or more wired interfaces, such as Ethernet (e.g., Fast Ethernet, Gigabit Ethernet, etc.). The network interface 7006 may also support communication over one or more non-Ethernet media, such as coaxial cable or power line, or over wide area media, such as Synchronous Optical Network (SONET) or Digital Subscriber Line (DSL) technology. The network interface 7006 may also take the form of one or more wireless interfaces, such as IEEE 802.11 (Wifi), Global Positioning System (GPS) or wide area wireless interface. However, other forms of physical layer interfaces and other types of standard or proprietary communication protocols may be used on the network interface 7006. In addition, the network interface 7006 may include multiple physical interfaces. For example, some embodiments of the computing device 7000 may include Ethernet, and Wi-Fi interface.

[0551] The input / output unit 7008 can facilitate interaction between the user and peripheral devices and the computing device 7000. The input / output unit 7008 can include one or more types of input devices, such as a keyboard, a mouse, a touch screen, etc. Similarly, the input / output unit 7008 can include one or more types of output devices, such as a screen, a monitor, a printer, and / or one or more light-emitting diodes (LEDs). Additionally or alternatively, the computing device 7000 can communicate with other devices using, for example, a Universal Serial Bus (USB) or High-Definition Multimedia Interface (HDMI) port interface.

[0552] In some embodiments, one or more computing devices, such as computing device 7000, may be deployed to support a machine learning framework. The exact physical location, connectivity, and configuration of these computing devices may be unknown and / or unimportant to the client device. Thus, the computing devices may be referred to as "cloud-based" devices, which may be located in various remote data center locations.

[0553] Figure 71 Depicted is a cloud-based server cluster 7100 according to an example embodiment. Figure 71 In the embodiment of the present invention, the operation of a computing device (e.g., computing device 7000) can be distributed among server devices 7102, data storage 7104, and routers 7106, all of which can be connected by a local cluster network 7108. The number of server devices 7102, data storage 7104, and routers 7106 in server cluster 7100 can depend on the computing tasks and / or applications assigned to server cluster 7100.

[0554] For example, the server device 7102 can be configured to perform various computing tasks of the computing device 7000. Therefore, the computing tasks can be distributed among one or more server devices 7102. In terms of being able to perform these computing tasks in parallel, such task distribution can reduce the total time to complete these tasks and return results. For simplicity, the server cluster 7100 and each server device 7102 can be referred to as a "server device." This term should be understood to imply that one or more different server devices, data storage devices, and cluster routers may be involved in the operation of the server device.

[0555] The data storage 7104 may be a data storage array including a drive array controller configured to manage read and write access to a bank of hard disk drives and / or solid-state drives. The drive array controller, alone or in conjunction with the server device 7102, may also be configured to manage backup or redundant copies of data stored in the data storage 7104 to protect against drive failure or other types of failures that prevent one or more server devices 7102 from accessing units of the data storage 7104. Other types of storage besides drives may be used.

[0556] The router 7106 may include a networking device configured to provide internal and external communications for the server cluster 7100. For example, the router 7106 may include one or more packet switching and / or routing devices (including switches and / or gateways) configured to provide (i) network communications between the server devices 7102 and the data storage 7104 via the local cluster network 7108, and / or (ii) network communications between the server cluster 7100 and other devices via communication links 7110 to the network 7112.

[0557] In addition, the configuration of the router 7106 can be based at least in part on the data communication requirements of the server device 7102 and the data storage 7104, the latency and throughput of the local cluster network 7108, the latency, throughput and cost of the communication link 7110, and / or other factors that may contribute to the cost, speed, fault tolerance, resiliency, efficiency and / or other design goals of the system architecture.

[0558] As a possible example, data storage 7104 may include any form of database, such as a Structured Query Language (SQL) database. Various types of data structures may store information in such a database, including but not limited to tables, arrays, lists, trees, and tuples. Furthermore, any database in data storage 7104 may be monolithic or distributed across multiple physical devices.

[0559] The server device 7102 may be configured to send data to and receive data from the data storage 7104. Such sending and retrieval may take the form of SQL queries or other types of database queries, and the output of such queries, respectively. Other text, images, video, and / or audio may also be included. In addition, the server device 7102 may organize the received data into a web page or web application presentation. Such presentation may take the form of a markup language, such as Hypertext Markup Language (HTML), Extensible Markup Language (XML), or some other standardized or proprietary format. In addition, the server device 7102 may have the ability to execute various types of computerized scripting languages, such as, but not limited to, Perl, Python, PHP Hypertext Preprocessor (PHP), Active Server Pages (ASP), Etc. Computer program codes written in these languages can facilitate providing web pages to client devices, and interaction between client devices and web pages. Alternatively or additionally, It can be used to facilitate the generation of web pages and / or provide web application functionality.

[0560] Example Operation

[0561] Figure 72 and 73 is a flowchart illustrating an example embodiment. Figure 72 and 73 The illustrated processes may be performed by a computing device such as computing device 7000 and / or a cluster of computing devices such as server cluster 7100. However, the processes may be performed by other types of devices or device subsystems. Figure 72 and 73 The embodiments may be simplified by removing any one or more of the features shown therein. Furthermore, the embodiments may be combined with each other and / or any features, aspects and / or implementations described in other ways in the previous figures or text.

[0562] Figure 72 Block 7200 of the present invention may involve obtaining model training data, wherein the model training data relates to a first group of cancer patients undergoing cancer treatment, wherein the model training data relates to (i) results of laboratory tests performed on the first group of cancer patients and tumor types of the first group of cancer patients with (ii) whether individuals from the first group of cancer patients died within a threshold number of weeks after starting cancer treatment. The laboratory tests may include tissue biopsy, cerebrospinal fluid, lymph, whole blood, serum, blood cells, urine, sweat, tears, saliva, and / or stool tests, or any type of test that indicates a clinical pathology marker.

[0563] Block 7202 may involve training a continuous series of decision trees based on the model training data until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on residual values of its previous decision tree in the continuous series of decision trees.

[0564] Block 7204 may involve generating a gradient boosting machine learning model as a summed function of a continuous series of decision trees, wherein the gradient boosting machine learning model predicts whether the other cancer patient will die within a threshold number of weeks based on the summed function applied to results of laboratory tests performed on the other cancer patient and the tumor type of the other cancer patient.

[0565] In some embodiments, the threshold number of weeks is 12 weeks, but other values are possible (e.g., 8, 10, 15, or 20 weeks).

[0566] In some embodiments, the results from the laboratory tests provide one or more of the following: neutrophil to lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration, or gamma-glutamyl transferase concentration. Some or all of these results may be used.

[0567] In some embodiments, the neutrophil to lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration or gamma-glutamyl transferase concentration are selected from 20 or more laboratory tests performed on the first group of cancer patients. The selection may involve (i) training a gradient boosting machine learning model on 20 or more laboratory tests; (ii) for each corresponding laboratory test, training a corresponding change of the gradient boosting machine learning model on 20 or more laboratory tests with the corresponding laboratory test removed; and (iii) determining the predicted value of each of the 20 or more laboratory tests by comparing the accuracy of the changes in the gradient boosting machine learning model. The 20 or more laboratory tests may be selected from the predictors of Table 11 and may include non-test predictors, such as sex and gender.

[0568] In some embodiments, the tumor type is one of: non-small cell lung cancer, urothelial carcinoma, head and neck squamous cell carcinoma, gastric cancer, renal cell carcinoma, or Hodgkin lymphoma. Other types are possible.

[0569] In some embodiments, the sum function applies a multiplicative learning rate to the results from each of the consecutive series of decision trees.

[0570] In some embodiments, predicting whether the other cancer patient will die within the threshold number of weeks includes determining a probability that the other cancer patient will die within the threshold number of weeks by using a sum function.

[0571] In some embodiments, predicting whether the other cancer patient will die within the threshold number of weeks further comprises placing the other cancer patient in a high risk of death category or a low risk of death category based on the probability and the cutoff value. In some embodiments, the cutoff value is selected so that the false positive rate does not exceed 5%-20% when predicting whether the first group of cancer patients will die within the threshold number of weeks. In some cases, the cutoff value can be selected so that the false positive rate is 10%.

[0572] In some embodiments, cancer patients in a high risk category for death are excluded from a clinical trial for treating cancer, and wherein cancer patients in a low risk category for death are included in a clinical trial for treating cancer.

[0573] In some embodiments, the stopping condition is that the continuous series of decision trees includes a threshold number of decision trees. Alternatively or additionally, the stopping condition is that the residual values of a particular decision tree in the continuous series of decision trees are all below a threshold.

[0574] Some embodiments may further involve obtaining model tuning data, wherein the model tuning data relates to a second group of cancer patients undergoing cancer treatment, wherein the model tuning data correlates (i) results of laboratory tests performed on the second group of cancer patients and the tumor type of the second group of cancer patients with (ii) whether the second group of cancer patients died within a threshold number of weeks after starting cancer treatment; generating two or more variations of the gradient boosting machine learning model by training a sequential series of decision trees for each variation with a different set of hyperparameters; applying the variations to the model training data and the model tuning data; and selecting a particular variation from the variations as the gradient boosting machine learning model, the particular variation providing predictions of whether the first group of cancer patients and the second group of cancer patients died within a threshold number of weeks after starting cancer treatment within a threshold accuracy.

[0575] The threshold accuracy can be based on the true positive rate or true negative rate of the prediction of whether the first group of cancer patients died within a threshold number of weeks after the start of cancer treatment. Alternatively or additionally, the threshold accuracy can be based on the p-value of the interaction between: (i) the prediction of whether the second group of cancer patients died within a threshold number of weeks after the start of cancer treatment, and (ii) PD-L1 positive patients in the second group of cancer patients.

[0576] In some embodiments, the hyperparameters include one or more of the following: the maximum depth of the decision tree is set to 2, 3, 4, 5, or 6, the minimum loss reduction required to split a node in a decision tree is set to 0, 1, 2, 3, or 4, the minimum sum of weights of all observations required in child nodes in a decision tree is set between 3.0 and 7.0, the fraction of input parameters randomly sampled for each decision tree is set between 0.7 and 1.0, the fraction of observations randomly sampled for each decision tree is set between 0.4 and 0.8, and the number of decision trees in a continuous series of decision trees is set between 10 and 25. Other ranges may also be used.

[0577] Figure 73 Block 7300 of can involve obtaining a gradient boosting machine learning model, wherein the gradient boosting machine learning model is trained on model training data, wherein the model training data relates to a first group of cancer patients undergoing cancer treatment, wherein the model training data relates (i) results of laboratory tests performed on the first group of cancer patients and tumor types of the first group of cancer patients to (ii) whether an individual from the first group of cancer patients died within a threshold number of weeks after starting cancer treatment, wherein training the gradient boosting machine learning model involves: (i) training a continuous series of decision trees until a stopping condition is reached, wherein each subsequent decision tree in the continuous series of decision trees is constructed based on residual values of its previous decision tree in the continuous series of decision trees; and (ii) determining a sum function of the continuous series of decision trees. The laboratory test can include a tissue biopsy, cerebrospinal fluid, lymph, whole blood, serum, blood cells, urine, sweat, tears, saliva, and / or stool test, or any type of test indicative of a clinical pathology marker.

[0578] Block 7302 may involve obtaining results from a laboratory test performed on another cancer patient.

[0579] Block 7304 may involve obtaining a tumor type for another cancer patient.

[0580] Block 7306 may involve applying the gradient boosting machine learning model to the results of laboratory tests performed on another cancer patient and the tumor type of the other cancer patient.

[0581] Block 7308 may involve receiving a prediction from the gradient boosting machine learning model as to whether another cancer patient will die within a threshold number of weeks.

[0582] In some embodiments, the threshold number of weeks is 12 weeks, but other values are possible (e.g., 8, 10, 15, or 20 weeks).

[0583] In some embodiments, the results from the laboratory tests provide one or more of the following: neutrophil to lymphocyte ratio, neutrophil concentration, albumin concentration, lactate dehydrogenase concentration, aspartate aminotransferase concentration, or gamma-glutamyl transferase concentration. Some or all of these results may be used.

[0584] In some embodiments, the tumor type is one of: non-small cell lung cancer, urothelial carcinoma, head and neck squamous cell carcinoma, gastric cancer, renal cell carcinoma, or Hodgkin lymphoma. Other types are possible.

[0585] In some embodiments, the sum function applies a multiplicative learning rate to the results from each of the consecutive series of decision trees.

[0586] In some embodiments, predicting whether the other cancer patient will die within the threshold number of weeks includes determining a probability that the other cancer patient will die within the threshold number of weeks by using a sum function.

[0587] In some embodiments, predicting whether the other cancer patient will die within the threshold number of weeks further comprises placing the other cancer patient in a high risk of death category or a low risk of death category based on the probability and the cutoff value. In some embodiments, the cutoff value is selected so that the false positive rate does not exceed 5%-20% when predicting whether the first group of cancer patients will die within the threshold number of weeks. In some cases, the cutoff value can be selected so that the false positive rate is 10%.

[0588] In some embodiments, cancer patients in a high risk category for death are excluded from a clinical trial for treating cancer, and wherein cancer patients in a low risk category for death are included in a clinical trial for treating cancer.

[0589] In some embodiments, the stopping condition is that the continuous series of decision trees includes a threshold number of decision trees. Alternatively or additionally, the stopping condition is that the residual values of a particular decision tree in the continuous series of decision trees are all below a threshold.

[0590] In some embodiments, the gradient boosting machine learning model is also trained using model adjustment data, wherein the model adjustment data relates to a second group of cancer patients undergoing cancer treatment, wherein the model adjustment data associates (i) results of laboratory tests performed on the second group of cancer patients and the tumor type of the second group of cancer patients with (ii) whether the second group of cancer patients died within a threshold number of weeks after starting cancer treatment, wherein training the gradient boosting machine learning model using the model adjustment data comprises: (i) generating two or more variations of the gradient boosting machine learning model by training a sequential series of decision trees for each variation with a different set of hyperparameters; (ii) applying the variations to the model training data and the model adjustment data; and (iii) selecting a particular variation from the variations as the gradient boosting machine learning model, the particular variation providing a prediction of whether the first group of cancer patients and the second group of cancer patients died within the threshold number of weeks after starting cancer treatment within a threshold accuracy.

[0591] The threshold accuracy can be based on the true positive rate or true negative rate of the prediction of whether the first group of cancer patients died within a threshold number of weeks after the start of cancer treatment. Alternatively or additionally, the threshold accuracy can be based on the p-value of the interaction between: (i) the prediction of whether the second group of cancer patients died within a threshold number of weeks after the start of cancer treatment, and (ii) PD-L1 positive patients in the second group of cancer patients.

[0592] In some embodiments, the hyperparameters include one or more of the following: the maximum depth of the decision tree is set to 2, 3, 4, 5, or 6, the minimum loss reduction required to split a node in a decision tree is set to 0, 1, 2, 3, or 4, the minimum sum of weights of all observations required in child nodes in a decision tree is set between 3.0 and 7.0, the fraction of input parameters randomly sampled for each decision tree is set between 0.7 and 1.0, the fraction of observations randomly sampled for each decision tree is set between 0.4 and 0.8, and the number of decision trees in a continuous series of decision trees is set between 10 and 25. Other ranges may also be used.

[0593] in conclusion

[0594] A new model has been developed that relies on baseline measurements, as opposed to subjective clinical measures such as the Eastern Cooperative Oncology Group (ECOG) score. Based on its evaluation in the MYSTIC and EAGLE studies, this index appears to have clinical utility in identifying patients who may have a relatively more favorable benefit-risk profile compared with the ITT population.

[0595] All patents and publications mentioned in this specification are herein incorporated by reference to the same extent as if each individual patent and publication was specifically and individually indicated to be incorporated by reference. Citation or identification of any reference in any part of this application shall not be construed as an admission that such reference is available as prior art with respect to the present invention.

[0596] References

[0597] Arkenau HT, Barriuso J, Olmos D, Ang JE, de Bono J, Judson I, et al. Prospective validation of a prognostic score to improve patient selection for oncology phase I trials. J Clin Oncol. 2009; 27(16): 2692-6.

[0598] Bellmunt J, de Wit R, Vaughn DJ, Fradet Y, Lee JL, Fong L et al. Pembrolizumab as Second-Line Therapy for Advanced Urothelial Carcinoma. NEngl J Med. 2017; 376(11):1015-26.

[0599] Bigot F, Castanon E, Baldini C, Hollebecque A, Carmona A, Postel-Vinay S, et al. Prospective validation of a prognostic score for patients in immunotherapy phase I trials: The Gustave Roussy Immune Score (GRIm-Score). EurJ Cancer. 2017;84:212-8.

[0600] Borghaei H,Paz-Ares L,Horn L,Spigel DR,Steins M,Ready NE,etal.Nivolumab versus Docetaxel in Advanced Nonsquamous Non-Small-Cell LungCancer.N Engl J Med.2015;373(17):1627-39.

[0601] Bowen RC,Little NAB,Harmer JR,Ma J,Mirabelli LG,Roller KD,etal.Neutrophil-to-lymphocyte ratio as prognostic indicator in gastrointestinalcancers:a systematic review and meta-analysis.Oncotarget.2017;8(19):32171-89.

[0602] Caliskan,A,Bryson JJ,and Narayanan A.Semantics derived automaticallyfrom language corpora contain human-like biases.Science.356(6334):183-186,2017.

[0603] Champiat S,Ferrara R,Massard C,Besse B,Marabelle A,Soria JC,etal.Hyperprogressive disease:recognizing a novel pattern to improve patientmanagement.Nat Rev Clin Oncol.2018;15(12):748-62.

[0604] Chen T,Guestrin C.XGBoost:A Scalable Tree Boosting System.Proceedingsof the 22nd ACM SIGKDDInternational Conference on Knowledge Discovery andData Mining(KDD).2016;785-794.DOI:http: / / dx.doi.org / 10.1145 / 2939672.2939785.

[0605] Chen T,He T,Benesty M,Khotilovich V,Tang Y,Cho H,Chen K,Mitchell R,Cano I,Shou T,Li M,Xie J,Lin M,Geng Y,Li Y.R Package‘xgboost’,March 12,2019.https: / / cran.r-project.org / web / packages / xgboost / xgboost.pdf.

[0606] Ding Y,Zhang S,Qiao J.Prognostic value of neutrophil-to-lymphocyteratio in melanoma:Evidence from a PRISMA-compliant meta-analysis.Medicine(Baltimore).2018;97(30):e11446.

[0607] Ferris RL,Blumenschein G,Jr.,Fayette J,Guigay J,Colevas AD,Licitra L,et al.Nivolumab for Recurrent Squamous-Cell Carcinoma of the Head and Neck.NEngl J Med.2016;375(19):1856-67.

[0608] Friedman J,Hastie T,Tibshirani R.(2000).Additive logistic regression:a statistical view ofboosting(with discussion and a rejoinder by theauthors).The annals ofstatistics.2000;28(2),337-407.

[0609] Friedman JH.Greedy function approximation:a gradient boostingmachine.Annals ofStatistics,pp.2001;1189-1232.

[0610] Freis P,Graillot E,Rousset P,Hervieu V,Chardon L,Lombard-Bohas C,etal.Prognostic factors in neuroendocrine carcinoma:biological markers are moreuseful than histomorphological markers.Sci Rep.2017;7:40609.

[0611] Fukui T,Okuma Y,Nakahara Y,Otani S,Igawa S,Katagiri M,et al.ActivityofNivolumab and Utility ofNeutrophil-to-Lymphocyte Ratio as a PredictiveBiomarker for Advanced Non-Small-Cell Lung Cancer:A Prospective ObservationalStudy.Clin Lung Cancer.2019;20(3):208-14e2.

[0612] Garon EB,Hellmann MD,Rizvi NA,Carcereny E,Leighl NB,Ahn MJ,etal.Five-Year Overall Survival for Patients With Advanced NonSmall-Cell LungCancer Treated With Pembrolizumab:Results From the Phase I KEYNOTE-001Study.J Clin Oncol.2019;JCO1900934.

[0613] Gupta D,Lis CG.Pretreatment serum albumin as a predictor of cancersurvival:a systematic review ofthe epidemiological literature.Nutr J.2010;9:69.

[0614] Kazandjian DG,Gong Y,Kazandjian H,Pazdur R,Blumenthal GM.Explorationof baseline derived neutrophil to lymphocyte ratio(dNLR)and lactatedehydrogenase(LDH)in patients(pts)with metastatic non-small cell lung cancer(mNSCLC)treated with immune checkpoint inhibitors(ICI)or cytotoxicchemotherapy(CCT).2018;36(15_suppl):3035.

[0615] Khunger M,Patil PD,Khunger A,Li M,Hu B,Rakshit S,et al.Post-treatmentchanges in hematological parameters predict response to nivolumab monotherapyin non-small cell lung cancer patients.PLoS One.2018;13(10):e0197743.

[0616] Luo M,Sun W,Wu C,Zhang L,Liu D,Li W,et al.High pretreatment serumgamma-glutamyl transpeptidase predicts an inferior outcome in nasopharyngealcarcinoma.Oncotarget.2017;8(40):67651-62.

[0617] Mezquita L,Auclin E,Ferrara R,Charrier M,Remon J,et al.Association ofthe Lung Immune Prognostic Index With Immune Checkpoint Inhibitor Outcomes inPatients With Advanced Non-Small Cell Lung Cancer.JAMA Oncol.2018 Mar 1;4(3):351-357.

[0618] Minami S,Ihara S,Ikuta S,Komuta K.Gustave Roussy Immune Score andRoyal Marsden Hospital Prognostic Score Are Biomarkers of Immune-CheckpointInhibitor for Non-Small Cell Lung Cancer.World J Oncol.2019 Apr;10(2):90-100.

[0619] Mok TSK,Wu YL,Kudaba I,Kowalski DM,Cho BC,Turna HZ,etal.Pembrolizumab versus chemotherapy for previously untreated,PD-L1-expressing,locally advanced or metastatic non-small-cell lung cancer(KEYNOTE-042):a randomised,open-label,controlled,phase 3trial.Lancet.2019;393(10183):1819-30.

[0620] Mulkey F,By K,Theoret MR,Maher VE,Pazdur R,Sridhara R.Analysis ofearly mortality in randomized clinical trials evaluating anti-PD-1 / PD-L1antibodies:A systematic analysis by the United States Food and DrugAdministration(FDA).2019;37(15_suppl):2516-.

[0621] Petrelli F,Cabiddu M,Coinu A,Borgonovo K,Ghilardi M,Lonati V,etal.Prognostic role of lactate dehydrogenase in solid tumors:a systematicreview and meta-analysis of 76 studies.Acta Oncol.2015;54(7):961-70.

[0622] Powles T,Duran I,van der Heijden MS,Loriot Y,Vogelzang NJ,De Giorgi Uet al.Atezolizumab versus chemotherapy in patients with platinum-treatedlocally advanced or metastatic urothelial carcinoma (IMvigor211):amulticentre,open-label,phase 3 randomised controlled trial.Lancet.2018;391(10122):748-57.

[0623] Rizvi N,Cho BC,Reinmuth N,Lee KH,Ahn MJ,Luft A et al.Durvalumab withor without tremelimumab vs platinum-based chemotherapy as first-linetreatment for metastatic non-small cell lung cancer:MYSTIC.ESMO Immuno-Oncology Congress;2018 December 13-16;Geneva,Switzerland.Abstract LBA6[oralpresentation].

[0624] Sacdalan DB,Lucero JA,Sacdalan DL.Prognostic utility of baselineneutrophil-to-lymphocyte ratio in patients receiving immune checkpointinhibitors:a review and meta-analysis.Onco Targets Ther.2018;11:955-65.

[0625] Socinski MA,Jotte RM,Cappuzzo F,Orlandi F,Stroyakovskiy D,Nogami N,etal.Atezolizumab for First-Line Treatment of Metastatic Nonsquamous NSCLC.NEngl J Med.2018;378(24):2288-301.

[0626] Shitara K,Ozguroglu M,Bang YJ,Di Bartolomeo M,Mandala M,Ryu MH,etal.Pembrolizumab versus paclitaxel for previously treated,advanced gastric orgastro-oesophagealjunction cancer(KEYNOTE-061):a randomised,open-label,controlled,phase 3 trial.Lancet.2018;392(10142):123-33.

[0627] Tham T,Bardash Y,Herman SW,Costantino PD.Neutrophil-to-lymphocyteratio as a prognostic indicator in head and neck cancer:A systematic reviewand meta-analysis.Head Neck.2018;40(11):2546-57.

[0628] Varga A,Bernard-Tessier A,Auclin E,Mezquita Pérez L.Baldini C,etal.Applicability of the Lung Immune Prognostic Index(LIPI)in patients withmetastatic solid tumors when treated with Immune Checkpoint Inhibitors(ICI)inearly clinical trials.Ann Oncol.2019 Feb 1;30(Supplement 1).

[0629] Wheler J,Tsimberidou AM,Hong D,Naing A,Falchook G,Piha-Paul S,etal.Survival of 1,181 patients in a phase I clinic:the MD Anderson ClinicalCenter for targeted therapy experience.Clin Cancer Res.2012;18(10):2922-9.

[0630] Winquist E KS,Nichols AC,Lenehan JG,Macneil SD,Palma DA;.Earlymortality with immune checkpoint inhibitors(IOs)in solid tumors:Aninconvenient truth?J Clin Oncol.2018;36(suppl;abstr 12121).

[0631] Yu Y,Wang H,Yan A,Wang H,Li X,Liu J,et al.Pretreatment neutrophil tolymphocyte ratio in determining the prognosis of head and neck cancer:a meta-analysis.BMC Cancer.2018;18(1):383.

Claims

1. Use of a reagent for determining the concentration of multiple clinical pathological markers in a patient in the preparation of a system for predicting whether a patient is at risk of developing non-small cell lung cancer (NSCLC) by a method comprising the following steps: a) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; The multiple clinical pathological markers include neutrophil / leukocyte (%), neutrophil (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or gamma-glutamyl transferase (U / L); and b) receiving a prediction from the gradient boosting machine learning model as to whether the patient has a likelihood of developing NSCLC.

2. Use of a reagent for determining the concentration of multiple clinical pathological markers in a patient in the preparation of a system for predicting whether a patient is at risk of developing bladder cancer by a method comprising the following steps: a) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; The multiple clinical pathological markers include neutrophil / leukocyte (%), neutrophil (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or gamma-glutamyl transferase (U / L); and b) receiving a prediction from the gradient boosting machine learning model as to whether the patient has a likelihood of developing bladder cancer.

3. The use according to claim 1 or claim 2, further comprising taking a sample from the patient.

4. The method of claim 3, wherein the sample comprises tissue biopsy, cerebrospinal fluid, lymph, whole blood, serum, blood cells, urine, sweat, tears, saliva and / or feces.

5. The method of claim 3, wherein the concentration of one or more of the plurality of clinical pathological markers is determined from the sample taken from the patient.

6. The use of claim 1 or claim 2, wherein the patient has previously received an anti-cancer therapeutic agent.

7. The method of claim 1 or claim 2, wherein the anticancer therapeutic comprises cisplatin, gemcitabine, methotrexate, vinblastine, doxorubicin, cisplatin (MVAC), carboplatin, a taxane, temozolomide, dacarbazine, vinflunine, docetaxel, paclitaxel, nab-paclitaxel, vemurafenib, erlotinib, afatinib, cetuximab, bevacizumab, gefitinib, and / or pemetrexed.

8. The method of claim 1 or claim 2, wherein the anti-cancer therapeutic comprises an immune checkpoint inhibitor.

9. The use according to claim 8, wherein the immune checkpoint inhibitor comprises an anti-CTLA-4 antibody, an anti-PD-1 antibody, and / or an anti-PD-L1 antibody.

10. The use according to claim 9, wherein the anti-CTLA-4 antibody is tremelimumab or ipilimumab.

11. The use according to claim 9, wherein the anti-PD-1 antibody is REGN2810, SHR1210, IBI308, PDR001, nivolumab, pembrolizumab, anti-PD-1, BGB-A317, BCD-100 or JS001.

12. The use according to claim 9, wherein the anti-PD-L1 antibody comprises durvalumab, avelumab, atezolizumab or KNO35.

13. The use of claim 1 or claim 2, further comprising determining the patient's cancer type, and further providing the determined cancer type to the gradient boosting machine learning model.

14. Use of a reagent for determining the concentration of a plurality of clinical pathological markers in a patient in the preparation of a system for treating a patient suffering from non-small cell lung cancer (NSCLC) by a method comprising the steps of: a) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; wherein the plurality of clinical pathology markers include neutrophils / leukocytes (%), neutrophils (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L); b) receiving a prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; and c) administering an anti-cancer therapeutic to the patient if the prediction indicates that the patient does not have poor immune fitness.

15. Use of a reagent for determining the concentration of a plurality of clinical pathology markers in a patient in the preparation of a system for treating a patient suffering from bladder cancer by a method comprising the steps of: a) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; wherein the plurality of clinical pathology markers include neutrophils / leukocytes (%), neutrophils (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L); b) receiving a prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; and c) administering an anti-cancer therapeutic to the patient if the prediction indicates that the patient does not have poor immune fitness.

16. The use of claim 14 or claim 15, further comprising d) reducing at least one of cancer cell division rate, tumor growth, tumor size, tumor density, or tumor metastasis rate in the patient.

17. Use of a reagent for determining the concentration of a plurality of clinical pathological markers in a patient in the preparation of a system for treating a patient suffering from non-small cell lung cancer (NSCLC) by a method comprising the steps of: a) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; wherein the plurality of clinical pathology markers include neutrophils / leukocytes (%), neutrophils (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L); b) receiving a prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; c) if the prediction indicates that the patient has poor immune fitness, administering a therapeutic agent to the patient to enhance the patient's immune fitness; d) determining a second concentration of a plurality of clinical pathology markers in the patient; The multiple clinical pathological markers include neutrophil / leukocyte (%), neutrophil (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L); e) providing the determined second concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; f) receiving a second prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; as well as g) administering an anti-cancer therapeutic to the patient if the prognosis indicates that the patient does not have poor immune fitness.

18. Use of a reagent for determining the concentration of a plurality of clinical pathology markers in a patient in the preparation of a system for treating a patient suffering from bladder cancer by a method comprising the steps of: a) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; wherein the plurality of clinical pathology markers include neutrophils / leukocytes (%), neutrophils (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L); b) receiving a prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; c) if the prediction indicates that the patient has poor immune fitness, administering a therapeutic agent to the patient to enhance the patient's immune fitness; d) determining a second concentration of a plurality of clinical pathology markers in the patient; The multiple clinical pathological markers include neutrophil / leukocyte (%), neutrophil (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L); e) providing the determined second concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; f) receiving a second prediction from the gradient boosting machine learning model whether the patient has poor immune fitness; as well as g) administering an anti-cancer therapeutic to the patient if the prognosis indicates that the patient does not have poor immune fitness.

19. Use of an agent for identifying a candidate participant for a clinical trial by a method comprising the steps of: a) determining the concentrations of multiple clinical pathological markers of the patient, wherein the multiple clinical pathological markers include neutrophil / leukocyte (%), neutrophil (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L); b) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; c) receiving a prediction from the gradient boosting machine learning model of the likelihood that the patient will die from NSCLC within 12 weeks; and d) if the prediction indicates that the patient is unlikely to die within 12 weeks, enrolling the patient in a clinical trial of the anti-cancer therapeutic agent.

20. Use of an agent for identifying a candidate participant for a clinical trial by a method comprising the steps of: a) determining the concentrations of multiple clinical pathological markers of the patient, wherein the multiple clinical pathological markers include neutrophil / leukocyte (%), neutrophil (10 3 / μL), albumin (g / L), lactate dehydrogenase (U / L), aspartate aminotransferase (U / L), and / or γ-glutamyltransferase (U / L); b) providing the measured concentrations of the plurality of clinical pathology markers to a gradient boosting machine learning model; c) receiving a prediction from the gradient boosting machine learning model of a likelihood that the patient is likely to die from bladder cancer within 12 weeks; and d) if the prediction indicates that the patient is unlikely to die within 12 weeks, enrolling the patient in a clinical trial of the anti-cancer therapeutic agent.

21. The use of any one of claims 14-20, wherein the anti-cancer therapeutic agent is an immune checkpoint inhibitor.

Images