Systems and methods for predicting efficacy of checkpoint inhibitor therapies

Abstract

Predicting treatment benefits of immune checkpoint inhibitor drugs (ICIs) without resorting to advanced genomic or immunologic assays is a major unmet clinical need. This disclosure provides a predictive model using machine-learning approaches based on routine laboratory test results in clinical practice.

Description

084284.00347SYSTEMS AND METHODS FOR PREDICTING EFFICACY OF CHECKPOINT INHIBITOR THERAPIESCROSS-REFERENCE

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 742,161 filed January 6, 2025, which is incorporated by reference herein in its entirety.BACKGROUND

[0002] Immune checkpoint inhibitors (ICIs) such as anti-cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) or anti -programmed death 1 (PD-l)Zprogrammed death ligand 1 (PD-L1) agents can induce durable responses in patients with advanced-stage cancers. However, most patients incur treatment costs, and the potential for serious immunotherapy-related adverse events without having durable clinical benefit. Thus, a predictive model of ICI efficacy across different cancer types would have important ramifications in precision medicine by helping physicians identify patients more likely to benefit while potentially prioritizing other therapies in patients less likely to respond to ICI.

[0003] Tumor mutational burden (TMB) and PD-L1 expression are biomarkers approved by the U.S. Food and Drug Administration (FDA) for this purpose. However, these biomarkers have limited accuracy and practical constraints that have precluded their widespread clinical use, such as the need for sufficient tumor tissue to sequence DNA in the case of TMB and the lack of standardized antibody clones and scoring systems for PD-L1 immunohistochemistry. Thus, there remains a clinical need for a quantitative predictive marker that can be easily obtained at a low cost and quick turn-around time in diverse geographical regions, including developing countries.SUMMARY OF THE INVENTION

[0004] This disclosure addresses the need mentioned above in a number of aspects. In one aspect, this disclosure provides a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the method comprises: (a) obtaining a set of individual characteristic variables of the subject; (b) assaying a sample (e.g., blood sample) obtained or derived from the subject to obtain a set of laboratory measurements therefrom; (c) computer processing at least the set of individual characteristic variables and the set of laboratory measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (d)1180755440.1084284.00347determining, based at least in part on the computer processing in (c), a predicted clinical outcome of the subject upon receiving the ICI.

[0005] In some embodiments, the cancer is selected from the group consisting of bladder cancer (BLCA), hepatobiliary cancer, melanoma, non-small cell lung cancer (NSCLC), renal cell carcinoma (RCC), small cell lung cancer, and a combination thereof.

[0006] In some embodiments, the predicted clinical outcome comprises a clinical benefit of the ICI or an overall survival of the subject. In some embodiments, the predicted clinical outcome comprises the clinical benefit of the ICI. In some embodiments, the clinical benefit of the ICI comprises a complete response to the ICI, a partial response to the ICI, a stable disease without progression for at least six months after initially receiving the ICI, a progression of the cancer, or a regression of the cancer.

[0007] In some embodiments, the method comprises determining a likelihood of the clinical benefit of the ICI. In some embodiments, the likelihood comprises a probability of the clinical benefit of the ICI. In some embodiments, the predicted clinical outcome comprises the overall survival of the subject.

[0008] In some embodiments, the overall survival of the subject comprises at least about 1 month, 2 months, 3 months, 4 months, 5 months, 6 months, 7 months, 8 months, 9 months, 10 months, 11 months, 12 months, 2 years, 3 years, 4 years, 5 years, or more than 5 years.

[0009] In some embodiments, the method further comprises determining a risk status indicative of a likelihood of poor outcome.

[0010] In some embodiments, the trained machine learning algorithm, also known as artificial intelligence, comprises a member selected from the group consisting of a logistic regression, a Cox regression, a support vector machine, a random forest, and a combination thereof. In some embodiments, the logistic regression comprises a ridge logistic regression. In some embodiments, the Cox regression comprises a ridge Cox regression. In some embodiments, the support vector machine comprises a fast survival support vector machine. In some embodiments, the random forest comprises a random survival forest. In some embodiments, the method further comprises training a machine learning model. In some embodiments, the machine learning model comprises a neural network.

[0011] In some embodiments, step (d) further comprises determining the predicted clinical outcome with an area under receiver operating characteristic curve (AUC) of at least about 0.50, 0.55, 0.60, 0.65, 0.70, 0.75, 0.80. 0.85, 0.90, or 0.95. In some embodiments, step (d) further comprises determining the predicted clinical outcome with an accuracy of at least about 2180755440.1084284.0034750%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%. In some embodiments, step (d) further comprises determining the predicted clinical outcome with a sensitivity of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%. In some embodiments, step (d) further comprises determining the predicted clinical outcome with a specificity of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%. In some embodiments, step (d) further comprises determining the predicted clinical outcome with a positive predictive value of at least about 50%, 55%. 60%. 65%, 70%, 75%. 80%, 85%, 90%. or 95%. In some embodiments, step (d) further comprises determining the predicted clinical outcome with a negative predictive value of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.

[0012] In some embodiments, the method further comprises administering the ICI to the subject, based at least in part on the predicted clinical outcome determined in step (d). In some embodiments, the method further comprises selecting the subject to not receive the ICI and administering an alternative therapy to the subject, based at least in part on the predicted clinical outcome determined in step (d).

[0013] In some embodiments, the ICI compnses a combination therapy, a first-line therapy, a second-line therapy, or a third-line therapy. In some embodiments, the ICI is selected from the group consisting of an anti-cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) agent, an anti-programmed death 1 (PD-l) / programmed death ligand 1 (PD-L1) agent, and a combination thereof.

[0014] In some embodiments, the set of individual characteristic variables comprises a member selected from the group consisting of: a demographic characteristic, a clinical characteristic, a risk group stratification of the subject, and a combination thereof. In some embodiments, the demographic characteristic comprises age or sex of the subject.

[0015] In some embodiments, the clinical characteristic is selected from the group consisting of: body mass index (BMI), drug class (DrugClass), chemotherapy during immunotherapy (DuringChemo), systemic therapy history (PreChemo), Eastern Cooperative Oncology Group performance status (ECOG-PS), smoking history (Smoking), tumor stage (Stage), viral infection (Virus), and a combination thereof.

[0016] In some embodiments, the risk group stratification comprises a low-risk group, a moderate-risk group, or a high-risk group.

[0017] In some embodiments, the blood sample comprises a whole blood sample, a serum sample, or a plasma sample.3180755440.1084284.00347

[0018] In some embodiments, the set of laboratory measurements comprises a member selected from the group consisting of: a comprehensive metabolic panel (CMP) measurement, a complete blood count (CBC) measurement, a coagulation panel measurement, conjugated bilirubin (CB), direct bilirubin (DB), glucose-6-phosphate dehydrogenase (G6PD), ionized calcium (iCA), lactate dehydrogenase (LDH), lipase (LPS), and a combination thereof.

[0019] In some embodiments, the CMP measurement is selected from the group consisting of: albumin (ALB), alkaline phosphatase (ALK), alanine aminotransferase (ALT), anion gap (AGAP), aspartate aminotransferase (AST), blood urea nitrogen (BUN), calcium (CA), chloride (CL), carbon dioxide (CO2), creatine (CREAT), estimated glomerular filtration rate (eGFR), glucose (GLU), potassium (K), bilirubin (BILI), total protein (PROT), magnesium (MG), phosphorus (P). and a combination thereof.

[0020] In some embodiments, the CBC measurement is selected from the group consisting of: white blood cell count (WBC), basophil count (BASO), eosinophil count (EOS), granulocytes count (GRAN), lymphocyte count (LYM), monocyte count (MONO), neutrophil count (NEUT), basophil proportion among WBC (BASO%), eosinophil proportion among WBC (EOS%), granulocytes proportion among WBC (GRAN%), lymphocyte proportion among WBC (LYM%), monocyte proportion among WBC (M0N0%), neutrophil proportion among WBC (NEUT%), hematocrit (HCT), hemoglobin count (HGB), mean corpuscular hemoglobin concentration (MCHC), mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV), platelet count (PLT). red blood cell count (RBC). red blood cell distribution width (RDW), basophil-to-lymphocyte rate (BLR), eosinophil -to-lymphocyte ratio (ELR), monocyte-to-lymphocyte ratio (MLR), and neutrophil-to-lymphocyte ratio (NLR), and a combination thereof.

[0021] In some embodiments, the coagulation panel measurement is selected from the group consisting of: activated partial thromboplastin time (APTT), international normalized ratio (INR), prothrombin time (PT), and a combination thereof.

[0022] In some embodiments, the computer processing in (c) further comprises computer processing measurements or values relating to one or more biomarkers associated with activity of the ICE In some embodiments, the one or more biomarkers associated with activity of the ICI comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers comprise one or more of: tumor mutational burden 4180755440.1084284.00347(TMB), PD-L1 expression, microsatellite instability (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some cases, protein-based biomarkers comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles comprise IL-6, TNF-a, CXCL9, CXCL10, or any combination thereof. In some cases, H&E staining feature biomarkers comprise tumor-infiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers comprise exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0023] In another aspect, this disclosure also provides a non-transitory computer-readable medium comprising machine-executable code that, upon execution by one or more computer processors, implements a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the method comprises: (a) obtaining a set of individual characteristic variables of the subject; (b) assaying a sample (e.g., blood sample) obtained or derived from the subject to obtain a set of laboratory measurements therefrom; (c) computer processing at least the set of individual characteristic variables and the set of laboratory measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (d) determining, based at least in part on the computer processing in (c), a predicted clinical outcome of the subject upon receiving the ICI. In some embodiments, the computer processing in (c) further comprises computer processing measurements or values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity' of the ICI comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some 5180755440.1084284.00347cases, protein-based biomarkers comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profdes comprise IL-6. TNF-a, CXCL9. CXCL10. or any combination thereof. In some cases, H&E staining feature biomarkers comprise tumorinfiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers comprise exosomal RNA, exosomal proteins, microbiome profdes, epigenetic modifications, or any combination thereof.

[0024] In another aspect, this disclosure further provides a computer system for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the system comprises: a database that is configured to store a set of individual characteristic variables of the subject; and one or more computer processors operatively coupled to the database, wherein the one or more computer processors are individually or collectively programmed to: (i) assay a sample (e.g, blood sample) obtained or derived from the subject to obtain a set of laboratory measurements therefrom; (ii) process at least the set of individual characteristic variables and the set of laboratory measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (iii) determine, based at least in part on the computer processing in (ii), a predicted clinical outcome of the subject upon receiving the ICI. In some embodiments, the computer processing in (ii) further comprises computer processing measurements or values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity of the ICI comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some cases, protein-based biomarkers comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles comprise IL-6, TNF-a, CXCL9, CXCL10, or any combination thereof. In some cases, H&E staining feature biomarkers comprise tumor-infiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers comprise 6180755440.1084284.00347exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0025] In yet another aspect, this disclosure further provides a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, comprising: (a) assaying a blood sample obtained or derived from said subject to obtain a set of laboratory measurements therefrom; (b) computer processing at least said set of laboratory measurements against (i) a reference set of laboratory measurements or (ii) with a trained machine learning algorithm, or both (i) and (ii); and (c) determining, based at least in part on said computer processing in (b), a predicted clinical outcome of said subject upon receiving said ICI.

[0026] In some embodiments, the method does not comprise obtaining a set of individual characteristic variables of said subject. In some embodiments, the computer processing in (b) further comprises computer processing measurements or values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity of the ICI comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability7(MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some cases, protein-based biomarkers comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles comprise IL-6. TNF-a, CXCL9. CXCL10. or any combination thereof. In some cases, H&E staining feature biomarkers comprise tumorinfiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers comprise exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0027] In yet another aspect, this disclosure further provides a non-transitory computer-readable medium comprising machine-executable code that, upon execution by one or more computer processors, implements a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, said method comprising: (a) assaying a blood sample obtained or derived from said subject to obtain a set of laboratory measurements 7180755440.1084284.00347therefrom; (b) computer processing at least said set of laboratory' measurements (i) against a reference set of laboratory measurements or (ii) with a trained machine learning algorithm, or both (i) and (ii); and (c) determining, based at least in part on said computer processing in (c), a predicted clinical outcome of said subject upon receiving said ICI.

[0028] In some embodiments, the method implemented by the machine-executable code upon execution by one or more processors does not comprise obtaining a set of individual characteristic variables of said subject. In some embodiments, the computer processing in (b) further comprises computer processing measurements or values relating to one or more biomarkers associated with activity7of the ICI. In some embodiments, the one or more biomarkers associated with activity' of the ICI comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability' (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some cases, protein-based biomarkers comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles comprise IL-6. TNF-a, CXCL9. CXCL10. or any combination thereof. In some cases, H&E staining feature biomarkers comprise tumorinfiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers comprise exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0029] In yet another aspect, this disclosure further provides a computer system for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, comprising one or more computer processors operatively coupled to said database, wherein said one or more computer processors are individually or collectively programmed to (a) assay a blood sample obtained or derived from said subject to obtain a set of laboratory measurements therefrom; (b) process at least said set of laboratory measurements (i) against a reference set of laboratory7measurements or (ii) with a trained machine learning algorithm, or both (i) and (ii); and (c) determine, based at least in part on said computer processing in (ii), a predicted clinical outcome of said subject upon receiving said ICI.8180755440.1084284.00347

[0030] In some embodiments, the computer system does not comprise a database that is configured to store a set of individual characteristic variables of said subject. In some embodiments, the processing in (b) further comprises computer processing measurements or values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity7of the ICI comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory7gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some cases, protein-based biomarkers comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles comprise IL-6, TNF-a, CXCL9, CXCL10, or any combination thereof. In some cases, H&E staining feature biomarkers comprise tumor-infiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers comprise exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.BRIEF DESCRIPTION OF THE DRAWINGS

[0031] FIGs. 1A-1C show a real-world cohort (MSK-I) from Memorial Sloan Kettering Cancer Center (MSKCC), for developing a machine learning model. Mid panel, two real-world cohorts from MSKCC (MSK-II) and Mount Sinai Health System (MSHS). Bottom panel, ten global phase 3 clinical trials. ITT stands for intention-to-treat. FIG. 1A shows detailed patient characteristics of each clinical trial. FIG. IB shows feature selection analysis. In the top panel of FIG. IB, number of features collected in the MSK-I cohort for model development are shown. In the bottom panel of FIG. IB, 47 features were tested for the association with the clinical benefit by' Cochran-Mantel-Haenszel test or overall survival by Cox proportionalhazards regression. Systemic therapy history7was adjusted as a confounding factor in both tests.FIG. 1C shows machine learning analysis. The top panel of FIG. 1C shows a non-limiting example of model construction. One model was trained to predict overall survival, and the 9180755440.1084284.00347other was trained to predict clinical benefit. Mid panel, model performance comparison. ROC and AUC denote the receiver operating characteristic and area under the receiver operating characteristic curve, respectively. The bottom panel of FIG. 1C illustrates a non-limiting example of model performance evaluation. Among the two machine learning models, the one that performed the best on the hold-out test set was subjected to the analyses.

[0032] FIG. 2 illustrates summarizing the model performance in predicting overall survival at 6-, 12-, 18-, 24-, and 30-month and predicting clinical benefit in the three real-world cohorts and 12 experimental arms from 10 phase 3 clinical trials. RWD, RCT, NSCLC, and SCLC denote real-world data, randomized clinical trials, and non-small and small cell lung cancer, respectively.

[0033] FIGs. 3A and 3B show performance of SCORPIO on the MSK-II cohort. FIG. 3A shows Kaplan-Meier plots for overall survival of the three risk groups predicted by SCORPIO. Tick marks indicate censored data. Black vertical and horizontal dotted lines indicate the median survival time of each risk group. HR, CI, NSCLC, and SCLC denote hazard ratio, confidence interval, and non-small and small cell lung cancer, respectively. FIG.3B shows bar charts for clinical benefit rates of the three risk groups predicted by SCORPIO.

[0034] FIGs. 4A-4G show performance of SCORPIO on the ten global phase 3 clinical trial cohorts in the form of a dot plot for the aggregated SHAP values of the features in SCORPIO.FIG. 4A illustrates a higher value in a feature with a negative aggregated SHAP value, shown in yellow, lowers the risk status value. On the contrary, a higher value in a feature with a positive aggregated SHAP value, shown in purple, increases the risk status value. Features were sorted based on the absolute aggregated SHAP value. CL: chloride; ALB: albumin; HGB: hemoglobin; ECOG-PS: Eastern Cooperative Oncology Group performance status; EOS%: eosinophil proportion among white blood cells (WBCs); RBC: red blood cell; AGAP: anion gap; PROT: total protein; LYM%: lymphocyte proportion among WBCs; NEUT%: neutrophil proportion among WBCs; BMI: body mass index; Smoking: smoking history: NEUT: neutrophil count; CREAT: creatinine; LYM: lymphocyte count; HCT: hematocrit; GLU: glucose; MONO: monocyte count; ALT: alanine aminotransferase; Age: age at ICI; NLR: neutrophil-to-lymphocyte ratio; AST: aspartate aminotransferase; MCHC: mean corpuscular hemoglobin concentration; Stage: tumor stage at ICI; MLR: monocyte-to-lymphocyte ratio; RDW: red blood cell distribution width; ALK: alkaline phosphatase; BASO%: basophil proportion among WBCs; eGFR: estimated glomerular filtration rate; PLT : platelet; BILL total bilirubin; BLR: basophil-to-lymphocyte ratio. FIG. 4B and FIG. 4C show two representative 10180755440.1084284.00347cases with complete response (CR) to ipilimumab / nivolumab and atezolizumab, respectively, and FIG. 4D and FIG. 4E show two representative cases with progressive disease (PD) to atezolizumab and pembrolizumab, respectively. Each case is depicted with a bar chart in the left panel, displaying the aggregated SHAP values that indicate the magnitude and the direction of each feature’s impact on the predicted risk status. The right panel shows pre- and posttreatment radiographic images. Feature values of the corresponding features in a given patient are provided in the bar charts. The best overall tumor response and survival of each patient are also shown. Density plots show the distribution of the risk statuses in the training set, and black dotted lines indicate the predicted risk status of each patient. TMB and MSS denote tumor mutational burden and microsatellite stable, respectively. Heatmaps display the association between 14 immune cell types and the top five features, along with the predicted risk status from SCORPIO in FIG. 4F shows patients with NSCLC and FIG. 4G shows patients with head and neck (H&N) cancer. The number in each cell denotes Spearman’s p. * false discovery rate (FDR) adjusted P < 0.05. ** FDR adjusted P < 0.01. FIGs.5A and 5B show performance of SCORPIO on the MSHS cohort. Kaplan-Meier plots for overall survival of the three risk groups predicted by SCORPIO. Tick marks indicate censored data. Black vertical and horizontal dotted lines indicate the median survival time of each risk group. FIG. 5A shows Kaplan-Meier plots for overall survival of the three risk groups stratified by SCORPIO in the 12 experimental arms from the 10 clinical trial cohorts. Tick marks indicate censored data. Black vertical and horizontal dotted lines indicate the median survival time of each risk group. HCC denotes hepatocellular carcinoma. ACNP denotes atezolizumab plus carboplatin and nanoparticle albumin-bound paclitaxel. ACP denotes atezolizumab plus carboplatin and paclitaxel. ABCP denotes atezolizumab plus bevacizumab, carboplatin, and paclitaxel. FIG.5B shows bar charts for clinical benefit rates of the three risk groups stratified by SCORPIO in the 12 experimental arms from the 10 clinical trial cohorts.

[0035] FIG. 6 shows Kaplan-Meier plots for overall survival of the three risk groups stratified by SCORPIO. Tick marks indicate censored data. Black vertical and horizontal dotted lines indicate the median survival time of each risk group.

[0036] FIGs. 7A, 7B, 7C, 7D, and 7E show bar charts for the number of patients in each cancer type. FIG. 7A shows a MSK-I cohort. FIG. 7B shows a training set. FIG. 7C shows a Hold-out test set. FIG. 7D shows a MSK-II cohort. FIG. 7E shows a MSHS cohort.11180755440.1084284.00347

[0037] FIGs. 8A-8C show flow charts for the method performance for various groups. FIG.8A shows a MSK-I flow diagram, FIG.8B shows a MSK-II flow diagram, and FIG.8C shows a MSHS flow diagram.

[0038] FIGs. 9A-9J show flow charts for the method performance for various groups. FIG.9A shows IMbravel50. HCC denotes hepatocellular carcinoma. FIG. 9B shows IMspirel50.FIG. 9C shows IMmotionl51. FIG. 9D shows IMvigor211. FIG. 9E shows IMpowerl33. SCLC denotes small cell lung cancer. FIG.9F shows IMpowerl30. NSCLC denotes non-small cell lung cancer. FIG. 9G shows IMpowerl31. ACNP and ACP denote atezolizumab plus carboplatin and nanoparticle albumin-bound paclitaxel, and atezolizumab plus carboplatin and paclitaxel, respectively. FIG. 9H shows IMpowerl32. FIG. 91 shows IMpowerl50. ABCP denotes atezolizumab plus bevacizumab, carboplatin, and paclitaxel. FIG. 9J shows OAK.

[0039] FIG. 10A-10B show feature selection analysis. FIG. 10A shows a Cox proportionalhazards regression. FIG. 10B shows a Cochran-Mantel-Haenszel test.

[0040] FIG. 11 shows density plots showing prediction of clinical outcomes. Patients were stratified into high-risk, moderate-risk, and low-risk groups based on the first and third quartile of the risk statuses observed in the training set.

[0041] FIGs. 12A-12B show performance comparisons of SCORPIO, SCORPIO-CB, and TMB on the hold out test set. FIG. 12A shows the performance comparisons for predicting overall survival after administration with ICI was evaluated using time-dependent AUC (AUC[f|) values. FIG. 12B shows the performance comparisons for predicting clinical benefit to ICI were evaluated using AUC values.

[0042] FIGs. 13A-13B show performance comparisons with an others group. FIG. 13A shows performance comparisons for predicting overall survival after administration with ICI were evaluated using AUC(t) values. FIG. 13B shows performance comparisons for predicting clinical benefit to ICI were evaluated using AUC values.

[0043] FIGs. 14A-14B show- performance comparison. FIG. 14A show-s performance comparisons for predicting overall survival after administration with ICI were evaluated using AUC(t) values. FIG. 14B shows performance comparisons for predicting clinical benefit to ICI were evaluated using AUC values.

[0044] FIGs. 15A-15B show performance comparison. FIG. 15A show-s a performance comparison for predicting overall survival after administration with ICI evaluated using AUC(t) values FIG. 15B shows a performance comparison for predicting clinical benefit to ICI was evaluated using AUC values.12180755440.1084284.00347

[0045] FIGs. 16A-16B show survival predictions. FIG. 16A shows Kaplan-Meier plots for overall survival of the three risk groups predicted by SCORPIO. FIG. 16B shows charts for clinical benefit rates of the three risk groups predicted by SCORPIO.

[0046] FIGs. 17A-17B show performance comparisons for MSK-II cohort. FIG. 17A shows the performance comparisons for predicting overall survival after administration with ICI w ere evaluated using AUC(t) values. FIG. 17B shows the performance comparisons for predicting clinical benefit to ICI were evaluated using AUC values.

[0047] FIGs. 18A-18B show performance comparisons for SCORPIO with “other”. FIG.18A show s performance comparisons for predicting overall survival after administration with ICI evaluated using AUC(t) values. FIG. 18B shows performance comparisons for predicting clinical benefit to ICI were evaluated using AUC values.

[0048] FIGs. 19A-19B show performance comparisons for SCORPIO within the others group on the MSK-II cohort. FIG. 19A show-s Kaplan-Meier plots for overall survival corresponding to each of the three risk groups predicted by SCORPIO. FIG. 19B show s bar charts for clinical benefit rates of the three risk groups predicted by SCORPIO.

[0049] FIGs. 20A-20B show subgroup analysis on the hold out test set. FIG. 20A shows Kaplan-Meier plots for overall survival of the three risk groups predicted by SCORPIO in each subgroup. FIG.20B shows bar charts for clinical benefit rates of the three risk groups predicted by SCORPIO in each subgroup.

[0050] FIGs. 21A-21B show subgroup analysis on the MSK-II test set. FIG. 21A shows Kaplan-Meier plots for overall survival of the three risk groups predicted by SCORPIO in each subgroup. FIG.21B shows bar charts for clinical benefit rates of the three risk groups predicted by SCORPIO in each subgroup.

[0051] FIGs. 22A-22B show performance of SCORPIO on MSK non-ICI cohort. FIG. 22A shows Kaplan-Meier plots for overall survival of the three risk groups predicted by SCORPIO.FIG. 22B also shows Kaplan-Meier plots for overall survival of the three risk groups predicted by SCORPIO.

[0052] FIG. 23 shows an association between TMB and top five features along with risk status from SCORPIO.

[0053] FIGs. 24A-24X show the results of subgroup analyses on the 10 phase 3 global clinical trials. FIG.24A shows Kaplan-Meier plots for overall survival of the three risk groups predicted by SCORPIO in each subgroup in IMbravel50. FIG. 24B shows bar charts for13180755440.1084284.00347clinical benefit rates of the three risk groups predicted by SCORPIO in each subgroup in IMbravelSO.

[0054] FIGs. 25A-25X show a performance comparison of SCORPIO and PD-L1 staining.

[0055] FIG. 26 shows a subgroup analysis on the MSHS cohort.

[0056] FIG. 27 shows an association between systemic therapy history and features.

[0057] FIGs. 28A-28B show model interoperability. FIG.28A shows summary’ plots for the distribution of feature impact on the predicted risk status from SCORPIO. FIG. 28B shows feature importance with the direction of impact per feature.

[0058] FIG. 29 shows data representing feature impact on risk score in various models, both universal and specific to analysis of various tissue types.DETAILED DESCRIPTION OF THE INVENTION

[0059] Predicting treatment benefits of immune checkpoint inhibitor drugs (ICIs) without resorting to advanced genomic or immunologic assays is a major unmet clinical need. This disclosure provides a predictive model using machine-learning approaches based on routine laboratory test results in clinical practice. The disclosed methods can effectively predict treatment benefit to an immune checkpoint therapy.Methods for Predicting Efficacy of Immune Checkpoint Therapy

[0060] In one aspect, this disclosure provides a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the method comprises: (a) obtaining a set of individual characteristic variables of the subject; (b) assaying a sample (<?.g, blood sample) obtained or derived from the subject to obtain a set of laboratory' measurements therefrom; (c) computer processing at least the set of individual characteristic variables and the set of laboratory measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (d) determining, based at least in part on the computer processing in (c), a predicted clinical outcome of the subject upon receiving the ICI.

[0061] In another aspect, this disclosure also provides a non-transitory computer-readable medium comprising machine-executable code that, upon execution by one or more computer processors, implements a method for determining an efficacy’ of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the method comprises: (a) obtaining a set of individual characteristic variables of the subject; (b) assaying a sample (e.g, blood sample) obtained or derived from the subject to obtain a set of laboratory measurements 14180755440.1084284.00347therefrom; (c) computer processing at least the set of individual characteristic variables and the set of laboratory measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (d) determining, based at least in part on the computer processing in (c), a predicted clinical outcome of the subj ect upon receiving the ICI.

[0062] In another aspect, this disclosure further provides a computer system for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the system comprises: a database that is configured to store a set of individual characteristic variables of the subject; and one or more computer processors operatively coupled to the database, wherein the one or more computer processors are individually or collectively programmed to: (i) assay a sample (e.g, blood sample) obtained or derived from the subject to obtain a set of laboratory measurements therefrom; (ii) process at least the set of individual characteristic variables and the set of laboratory' measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (iii) determine, based at least in part on the computer processing in (ii), a predicted clinical outcome of the subject upon receiving the ICI.

[0063] The terms “sample” or “biological sample,” as used herein, include any biological specimen obtained (isolated, removed) from a subject. Samples may include, without limitation, organ tissue (e.g., primary or metastatic tumor tissue), whole blood, plasma, serum, whole blood cells, red blood cells, white blood cells (e.g, peripheral blood mononuclear cells), saliva, urine, stool (feces), tears, sweat, sebum, nipple aspirate, ductal lavage, tumor exudates, synovial fluid, cerebrospinal fluid, lymph, fine needle aspirate, amniotic fluid, any other bodily fluid, exudate or secretory fluid, cell lysates, cellular secretion products, inflammation fluid, semen, and vaginal secretions. In some embodiments, a sample may be readily obtainable by non-invasive or minimally invasive methods, such as blood collection (“liquid biopsy”), urine collection, feces collection, tissue (e.g., tumor tissue) biopsy or fine-needle aspiration, allowing the provision / removal / isolation of the sample from a subject. The term “tissue,” as used herein, encompasses all types of cells of the body, including cells of organs but also including blood and other body fluids recited above. The tissue may be healthy or affected by pathological alterations, e.g., tumor tissue. The tissue may be from a living subject or may be cadaveric tissue. In some embodiments, useful samples are those known to comprise, expected, or predicted to comprise, known to potentially comprise, or expected or predicted to potentially comprise tumor cells.15180755440.1084284.00347

[0064] In some embodiments, the disclosed methods may be used for prognosis or predicting responsiveness to a therapy (e.g, anti -tumor treatment).

[0065] The terms “determining responsiveness,” “predicting responsiveness,” and “assessing a likelihood of a therapeutic response” may be used interchangeably herein.

[0066] The term “prognosis,” as used herein, refers to anticipation of progression of a disease (e.g, cancer) or condition and prospect (e.g. the probability, duration, and / or extent) of recovery. A good prognosis of the diseases or conditions may generally encompass anticipation of a satisfactory partial or complete recovery from the diseases or conditions, such as within an acceptable time period. A good prognosis of such may more commonly encompass anticipation of not further worsening or aggravating within a given time period. A poor prognosis of the diseases or conditions as taught herein may generally encompass anticipation of a substandard recovery and / or unsatisfactorily slow recovery, or to substantially no recovery or even further worsening of such.

[0067] In some embodiments, a therapeutic response may include an anti-tumor response when referring to a cancer patient treated with a cancer therapy. For example, an anti-tumor response may include at least one positive therapeutic effect, such as a reduced number of cancer cells, reduced tumor size, reduced rate of cancer cell infiltration into peripheral organs, reduced rate of tumor metastasis or tumor growth, or progression-free survival. Positive therapeutic effects in cancer can be measured in a number of ways (see, e.g., W. A. Weber, J. Null. Med. 50TS-10S (2009); Eisenhauer et al., 2009 European Journal of Cancer. 45: 228-247). In some embodiments, an anti-tumor response to a cancer therapy is assessed using RECIST 1.1 criteria, bidimensional irRC, or unidimensional irRC. In some embodiments, an anti-tumor response is any of stable disease (SD), partial response (PR), complete response (CR), progression-free survival (PFS), and disease-free survival (DFS). In some embodiments, one or more biomarkers of this disclosure predict whether a subject with a solid tumor is likely to achieve a complete response or a partial response.

[0068] The terms “predicting,” “prediction,” or “predictive,” as used herein, refer to an advance declaration, indication, or foretelling of a response or reaction to a therapy (e.g., chemotherapy, immunotherapy, immunochemotherapy) in a subject not (yet) having been treated with the therapy. For example, a prediction of responsiveness (or sensitivity or susceptibility) to a cancer therapy in a subject may indicate that the subject will respond or react to the cancer therapy, for example, within a certain time period, e.g, so that the subject will have a clinical benefit from the cancer therapy. A prediction of unresponsiveness (or 16180755440.1084284.00347insensitivity or insusceptibility) to a cancer therapy in a subject may indicate that the subject will minimally or not respond or react to the cancer therapy, for example, within a certain time period, e.g, so that the subject will have no clinical benefit from the cancer therapy.

[0069] In some embodiments, the predicted clinical outcome comprises a clinical benefit of the ICI or an overall survival of the subject. In some embodiments, the predicted clinical outcome comprises the clinical benefit of the ICI. In some embodiments, the clinical benefit of the ICI comprises a complete response to the ICI, a partial response to the ICI, a stable disease without progression for at least six months after initially receiving the ICI, a progression of the cancer, or a regression of the cancer.

[0070] In some embodiments, the method comprises determining a likelihood of the clinical benefit of the ICI. In some embodiments, the likelihood comprises a probability of the clinical benefit of the ICI. In some embodiments, the predicted clinical outcome comprises the overall survival of the subject.

[0071] In some embodiments, the overall survival of the subject comprises at least about 1 month, 2 months, 3 months, 4 months. 5 months, 6 months, 7 months, 8 months, 9 months, 10 months, 11 months, 12 months, 2 years, 3 years, 4 years, 5 years, or more than 5 years.

[0072] In some embodiments, the method further comprises determining a risk status indicative of a likelihood of poor outcome.

[0073] In some embodiments, step (d) (e.g., determining a predicted clinical outcome of the subject upon receiving the ICI) further comprises determining the predicted clinical outcome with an area under receiver operating characteristic curve (AUC) of at least about 0.50, 0.55, 0.60, 0.65, 0.70, 0.75, 0.80, 0.85, 0.90, or 0.95. In some embodiments, step (d) further comprises determining the predicted clinical outcome with an accuracy of at least about 50%, 55%, 60%. 65%. 70%. 75%. 80%. 85%. 90%. or 95%. In some embodiments, step (d) further comprises determining the predicted clinical outcome with a sensitivity’ of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%. In some embodiments, step (d) further comprises determining the predicted clinical outcome with a specificity of at least about 50%, 55%, 60%, 65%, 70%, 75%. 80%. 85%. 90%. or 95%. In some embodiments, step (d) further comprises determining the predicted clinical outcome with a positive predictive value of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%. In some embodiments, step (d) further comprises determining the predicted clinical outcome with a negative predictive value of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.17180755440.1084284.00347

[0074] In some embodiments, the method further comprises making, changing, or retracting a clinical practice decision based on the predicted clinical outcome. In some embodiments, the clinical practice decision can be a treatment decision. In some embodiments, the treatment decision can be an ICI. In some embodiments, the treatment decision can comprise applying a different treatment, stopping a treatment, starting a treatment, delaying a treatment, giving a treatment along with another treatment, removing a treatment from a treatment plan, extending a treatment, or any other changes to a treatment plan. In some embodiments, treatment can comprise treatment for cancer or immunotherapy. In some embodiments, treatment can comprise combination therapies. In some embodiments, the clinical practice decision can comprise allocating resources differently. In some embodiments, the clinical practice decision can comprise modifying patient diagnostics and testing protocols. In some embodiments, the clinical practice decision can comprise administering diagnostic tests to the patient more frequently, less frequently, or changing test type.

[0075] In some embodiments, the method further comprises administering the ICI to the subject, based at least in part on the predicted clinical outcome determined in step (d). In some embodiments, the method further comprises selecting the subject to not receive the ICI and administering an alternative therapy to the subject, based at least in part on the predicted clinical outcome determined in step (d).

[0076] In some embodiments, the ICI comprises a combination therapy, a first-line therapy, a second-line therapy, or a third-line therapy. In some embodiments, the ICI is selected from the group consisting of an anti-cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) agent, an anti-programmed death 1 (PD-l) / programmed death ligand 1 (PD-L1) agent, and a combination thereof.

[0077] In some embodiments, the method further comprises predicting results of one or more clinical trials. In some embodiments, the results can be treatment effectiveness. In some embodiments, the predictions can be predictions of ICI results.

[0078] In some embodiments, the method further comprises predicting an appropriate second medical use of treatments such as ICIs. In some embodiments, ICIs can be applied with one or more biomarkers as a companion diagnostic. In some embodiments, the one or more biomarkers can be standard biomarkers.

[0079] In some embodiments, the set of individual characteristic variables comprises a member selected from the group consisting of: a demographic characteristic, a clinical characteristic, a risk group stratification of the subject, and a combination thereof. In some 18180755440.1084284.00347embodiments, the set of individual characteristic variables comprises one or more characteristics described in Table 1 or Table 2.

[0080] In some embodiments, the demographic characteristic comprises age or sex of the subject.

[0081] In some embodiments, the clinical characteristic is selected from the group consisting of: body mass index (BMI), drug class (DrugClass), chemotherapy during immunotherapy (DuringChemo), systemic therapy history (PreChemo), Eastern Cooperative Oncology Group performance status (ECOG-PS), smoking history (Smoking), tumor stage (Stage), viral infection (Virus), and a combination thereof.

[0082] In some embodiments, the risk group stratification comprises a low-risk group, a moderate-risk group, or a high-risk group.

[0083] In some embodiments, the blood sample comprises a whole blood sample, a serum sample, or a plasma sample.

[0084] In some embodiments, the set of laboratory measurements comprises a member selected from the group consisting of: a comprehensive metabolic panel (CMP) measurement, a complete blood count (CBC) measurement, a coagulation panel measurement, conjugated bilirubin (CB), direct bilirubin (DB), glucose-6-phosphate dehydrogenase (G6PD), ionized calcium (iCA), lactate dehydrogenase (LDH), lipase (LPS), and a combination thereof.

[0085] In some embodiments, the CMP measurement is selected from the group consisting of: albumin (ALB), alkaline phosphatase (ALK). alanine aminotransferase (ALT), anion gap (AGAP), aspartate aminotransferase (AST), blood urea nitrogen (BUN), calcium (CA), chloride (CL), carbon dioxide (CO2), creatine (CREAT), estimated glomerular filtration rate (eGFR), glucose (GLU), potassium (K), bilirubin (BILI), total protein (PROT), magnesium (MG), phosphorus (P). and a combination thereof.

[0086] In some embodiments, the CBC measurement is selected from the group consisting of: white blood cell count (WBC), basophil count (BASO), eosinophil count (EOS), granulocytes count (GRAN), lymphocyte count (LYM), monocyte count (MONO), neutrophil count (NEUT), basophil proportion among WBC (BASO%), eosinophil proportion among WBC (EOS%), granulocytes proportion among WBC (GRAN%), lymphocyte proportion among WBC (LYM%), monocyte proportion among WBC (M0N0%), neutrophil proportion among WBC (NEUT%), hematocrit (HCT), hemoglobin count (HGB), mean corpuscular hemoglobin concentration (MCHC), mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV), platelet count (PLT). red blood cell count (RBC). red blood cell distribution 19180755440.1084284.00347width (RDW), basophil-to-lymphocyte rate (BLR), eosinophil-to-lymphocyte ratio (ELR), monocyte-to-lymphocyte ratio (MLR), and neutrophil-to-lymphocyte ratio (NLR). and a combination thereof.

[0087] In some embodiments, the coagulation panel measurement is selected from the group consisting of: activated partial thromboplastin time (APTT), international normalized ratio (INR), prothrombin time (PT), and a combination thereof.

[0088] In some embodiments, a subject is a human with cancer. In some embodiments, the cancer is selected from adrenal gland tumors, biliary cancer, bladder cancer, brain cancer, breast cancer, carcinoma, central or peripheral nervous system tissue cancer, cervical cancer, colon cancer, endocrine or neuroendocrine cancer or hematopoietic cancer, esophageal cancer, fibroma, gastrointestinal cancer, glioma, head and neck cancer, Li-Fraumeni tumors, liver cancer, lung cancer, lymphoma, melanoma, meningioma, multiple neuroendocrine type I and type II tumors, nasopharyngeal cancer, oral cancer, oropharyngeal cancer, osteogenic sarcoma tumors, ovarian cancer, pancreatic cancer, pancreatic islet cell cancer, parathyroid cancer, pheochromocytoma, pituitary tumors, prostate cancer, rectal cancer, renal cancer, respiratory cancer, sarcoma, skin cancer, stomach cancer, testicular cancer, thyroid cancer, tracheal cancer, urogenital cancer, and uterine cancer. In some embodiments, the cancer is selected from colorectal cancer, bile duct cancer, bone cancer, fallopian tube cancer, gallbladder cancer, kidney cancer, laryngeal cancer, leukemia, lip cancer, mesothelioma, mouth cancer, myeloma, oesophageal cancer, omental cancer, penile cancer, peritoneal cancer, salivary gland cancer, sinus cancer, small intestine cancer, spinal cancer, throat cancer, tonsil cancer, vaginal cancer, vulvar cancer, and Waldenstrom's macroglobulinemia.

[0089] In some embodiments, the cancer is selected from the group consisting of bladder cancer (BLCA), hepatobiliary cancer, melanoma, non-small cell lung cancer (NSCLC), renal cell carcinoma (RCC), small cell lung cancer, and a combination thereof.

[0090] In some embodiments, the computer processing in can further comprise computer processing measurements or values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity of the ICI can comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers can comprise one or more of: tumor mutational burden (TMB). PD-L1 expression, microsatellite instability (MSI) status. DNA modifications,20180755440.1084284.00347copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers can comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some cases, protein-based biomarkers can comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles can comprise IL-6. TNF-a, CXCL9, CXCL10, or any combination thereof. In some cases, H&E staining feature biomarkers can comprise tumor-infiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers can comprise exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0091] In another aspect, this disclosure also provides a non-transitory computer-readable medium which can comprise machine-executable code that, upon execution by one or more computer processors, can implement a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the method can comprise obtaining a set of individual characteristic variables of the subject. In some embodiments, the method can comprise assaying a sample (e.g., blood sample) obtained or derived from the subject to obtain a set of laboratory measurements therefrom. In some embodiments, the method can comprise computer processing at least the set of individual characteristic variables and the set of laboratory measurements against a reference set of individual characteristic variables and laboratory measurements or with a trained machine learning algorithm, or both. In some embodiments, the method can comprise determining, based at least in part on the computer processing, a predicted clinical outcome of the subject upon receiving the ICI.

[0092] In another aspect, this disclosure further provides a computer system for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the system can comprise a database that is configured to store a set of individual characteristic variables of the subject. In some embodiments, the system can further comprise one or more computer processors operatively coupled to the database. In some embodiments, the one or more computer processors can be individually or collectively programmed to assay a sample (e.g., blood sample) obtained or derived from the subject. In some embodiments, the one or more computer processors can be individually or collectively programmed to assay a sample (e.g., blood sample) obtained or derived from the subject to obtain a set of laboratory measurements therefrom. In some embodiments, the one or more computer processors can be 21180755440.1084284.00347individually or collectively programmed to process at least the set of individual characteristic variables and the set of laboratory measurements against a reference set of individual characteristic variables and laboratory measurements. In some embodiments, the one or more computer processors can be individually or collectively programmed to process at least the set of individual characteristic variables and the set of laboratory measurements with a trained machine learning algorithm. In some embodiments, the system can determine, based at least in part on the computer processing, a predicted clinical outcome of the subject upon receiving the ICI. In some embodiments, the computer processing can further comprise computer processing measurements or values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity of the ICI can comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers can comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers can comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS). or any combination thereof. In some cases, protein-based biomarkers can comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles can comprise IL-6, TNF-a, CXCL9, CXCL10, or any combination thereof. In some cases, H&E staining feature biomarkers can comprise tumor-infiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers can comprise exosomal RNA. exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0093] In yet another aspect, this disclosure further provides a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer can comprise: assaying a blood sample obtained or derived from said subject to obtain a set of laboratory measurements therefrom. In some embodiments, the method can further comprise computer processing at least said set of laboratory measurements against a reference set of laboratory' measurements, or with a trained machine learning algorithm, or both. In some embodiments, the method can further comprise determining, based 22180755440.1084284.00347at least in part on said computer processing, a predicted clinical outcome of said subject upon receiving said ICI.

[0094] In some embodiments, the method does not comprise obtaining a set of individual characteristic variables of said subject. In some embodiments, the computer processing can further comprise computer processing measurements or values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity of the ICI can comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multimodal biomarkers, or any combination thereof. In some cases, the DNA biomarkers can comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers can comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some cases, protein-based biomarkers can comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles can comprise IL-6, TNF-a, CXCL9, CXCL10, or any combination thereof. In some cases, H&E staining feature biomarkers can comprise tumor-infiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers can comprise exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0095] In yet another aspect, this disclosure further provides a non-transitory computer-readable medium which can comprise machine-executable code that, upon execution by one or more computer processors, implements a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer. In some embodiments, the method can comprise assaying a blood sample obtained or derived from said subject to obtain a set of laboratory measurements therefrom. In some embodiments, the method can further comprise computer processing at least said set of laboratory measurements against a reference set of laboratory measurements, or with a trained machine learning algorithm, or both. In some embodiments, the method can further comprise determining, based at least in part on said computer processing, a predicted clinical outcome of said subject upon receiving said ICI.

[0096] In some embodiments, the method implemented by the machine-executable code upon execution by one or more processors does not comprise obtaining a set of individual 23180755440.1084284.00347characteristic variables of said subject. In some embodiments, the computer processing can further comprise computer processing measurements or values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity of the ICI can comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multimodal biomarkers, or any combination thereof. In some cases, the DNA biomarkers can comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers can comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS), or any combination thereof. In some cases, protein-based biomarkers can comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles can comprise IL-6, TNF-a, CXCL9, CXCL10, or any combination thereof. In some cases, H&E staining feature biomarkers can comprise tumor-infiltration lymphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers can comprise exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0097] In yet another aspect, this disclosure further provides a computer system for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, which can comprise one or more computer processors operatively coupled to said database. In some embodiments, said one or more computer processors can be individually or collectively programmed to assay a blood sample obtained or derived from said subject to obtain a set of laboratory measurements therefrom. In some embodiments, said one or more computer processors can be individually or collectively programmed to process at least said set of laboratory measurements against a reference set of laboratory measurements, or with a trained machine learning algorithm, or both. In some embodiments, said one or more computer processors can be individually or collectively programmed to determine, based at least in part on said computer processing, a predicted clinical outcome of said subject upon receiving said ICI.

[0098] In some embodiments, the computer system does not comprise a database that is configured to store a set of individual characteristic variables of said subject. In some embodiments, the processing can further comprise computer processing measurements or 24180755440.1084284.00347values relating to one or more biomarkers associated with activity of the ICI. In some embodiments, the one or more biomarkers associated with activity of the ICI can comprise one or more of: DNA biomarkers, RNA biomarkers, protein biomarkers, biomarkers associated with Hematoxylin and Eosin (H&E) staining, exosomal biomarkers, microbiome biomarkers, epigenetic biomarkers, multi-modal biomarkers, or any combination thereof. In some cases, the DNA biomarkers can comprise one or more of: tumor mutational burden (TMB), PD-L1 expression, microsatellite instability (MSI) status, DNA modifications, copy number variations, circulating tumor DNA (ctDNA), or any combination thereof. In some embodiments, the RNA-based biomarkers can comprise one or more of: immune cell infiltration signatures, interferon-gamma (IFN-y) signatures, inflammatory gene expression, or tumor inflammation signature (TIS). or any combination thereof. In some cases, protein-based biomarkers can comprise cytokine and chemokine profiles. In some cases, cytokine and chemokine profiles can comprise IL-6, TNF-a, CXCL9, CXCL10, or any combination thereof. In some cases, H&E staining feature biomarkers can comprise tumor-infiltration ly mphocytes (TILs), stromal architecture, necrosis, or any combination thereof. In some cases, emerging markers can comprise exosomal RNA, exosomal proteins, microbiome profiles, epigenetic modifications, or any combination thereof.

[0099] A sample can be obtained from a subject in any way ty pically used in clinical settings for obtaining a sample comprising the required cells or nucleic acid, including RNA, genomic DNA, mitochondrial DNA, and protein-associated nucleic acids. For example, the sample can be obtained from fresh, frozen, or paraffin-embedded surgical samples or biopsies of an organ or tissue comprising the suitable cells or nucleic acid to be tested. If desired, the sample can be mixed with a fluid, purified, amplified, or otherwise treated. For examples, samples may be treated in one or more purification steps to increase the purity of the desired cells or nucleic acid in the sample, or they may be examined without any purification steps. Any nucleic acid specimen in purified or non-purified form obtained from such sample can be utilized in the methods as taught herein.

[0100] As used herein, a ‘“machine learning model,” a “model,” or a “classifier” refers to a set of algorithmic routines and parameters that can predict an output(s) for a process input based on a set of input features, w ith or without being explicitly programmed. A structure of the softw are routines (e.g., number of subroutines and relation betw een them) and / or the values of the parameters can be determined in a training process, which can use actual results of the process that is being modeled. Such systems or models are understood to be necessarily rooted 25180755440.1084284.00347in computer technology, and in fact, cannot be implemented or even exist in the absence of computing technology. While machine learning systems utilize various types of statistical analyses, machine learning systems are distinguished from statistical analyses by virtue of the ability to leam without explicit programming and being rooted in computer technology. A neural network or an artificial neural network is one set of algorithms used in machine learning for modeling the data using graphs of neurons. Any network structure may be used. Any number of layers, nodes within layers, types of nodes (activations), types of layers, interconnections, learnable parameters, and / or other network architectures may be used. Machine training uses the defined architecture, training data, and optimization to leam values of the learnable parameters of the architecture based on the samples and ground truth of training data.

[0101] A typical machine learning pipeline may include building a machine learning model from a sample dataset (referred to as a “training set”), evaluating the model against one or more additional sample datasets (referred to as a “validation set” and / or a “test set”) to decide whether to keep the model and to benchmark how good the model is, and using the model in “production” to make predictions or decisions against live input data captured by an application service. For training the model to be applied as a machine-learned model, training data is acquired and stored in a database or memory. The training data is acquired by aggregation, mining, loading from a publicly or privately formed collection, transfer, and / or access. Ten, hundreds, or thousands of samples of training data are acquired. The samples are from scans of different patients and / or phantoms. Simulation may be used to form the training data. The training data includes the desired output (ground truth), such as segmentation, and the input, such as protocol data and imaging data.

[0102] In some embodiments, the training set will be used to create a single classifier using any now or hereafter known methods. In other embodiments, a plurality of training sets will be created to generate a plurality of corresponding classifiers. Each of the plurality of classifiers can be generated based on the same or different learning algorithm that utilizes the same or different features in the corresponding one of the pluralities of training sets. For example, each of the plurality of neural network models can be trained on a training set classified on sequence type, view type, anatomy type and / or other image classifying data as discussed in conjunction with the disclosure.

[0103] Once trained, the machine-learned or trained model is stored for later application. The training determines the values of the learnable parameters of the network. The network 26180755440.1084284.00347architecture, values of non-leamable parameters, and values of the learnable parameters are stored as the machine-learned network. Once stored, the machine-learned network may be fixed. The same machine-learned network may be applied to different patients, different scanners, and / or with different imaging protocols for the scanning. The machine-learned network may be updated. As additional training data is acquired, such as through application of the network for patients and corrections by experts to that output, the additional training data may be used to re-train or update the training.

[0104] The model may include a supervised learning model. Supervised learning models may include different approaches and algorithms including analytical learning, artificial neural network, backpropagation, boosting (meta-algorithm), Bayesian statistics, case-based reasoning, decision tree learning, inductive logic programming, Gaussian process regression, genetic programming, group method of data handling, kernel estimators, learning automata, learning classifier systems, minimum message length (decision trees, decision graphs, etc.), multilinear subspace learning, naive Bayes classifier, maximum entropy classifier, conditional random field, Nearest Neighbor Algorithm, probably approximately correct learning (PAC) learning, ripple down rules, a knowledge acquisition methodology, symbolic machine learning algorithms, subsymbolic machine learning algorithms, support vector machines, Minimum Complexity Machines (MCM) , random forests, ensembles of classifiers, ordinal classification, data pre-processing, handling imbalanced datasets, statistical relational learning, or Proaftn, a multicriteria classification algorithm The model may linear regression, logistic regression, deep recurrent neural network (e.g, long short term memory, LSTM), Bayes classifier, hidden Markov model (HMM) , linear discriminant analysis (LDA) , k-means clustering, densitybased spatial clustering of applications with noise (DBSCAN) , random forest algorithm, support vector machine (SVM), or any model described herein.

[0105] In some embodiments, the trained machine learning algorithm comprises a member selected from the group consisting of a logistic regression, a Cox regression, a support vector machine, a random forest, and a combination thereof. In some embodiments, the logistic regression comprises a ridge logistic regression. In some embodiments, the Cox regression comprises a ridge Cox regression. In some embodiments, the support vector machine comprises a fast survival support vector machine. In some embodiments, the random forest comprises a random survival forest.

[0106] In some embodiments, statistical models that are not machine learning models may be used for prediction methods.27180755440.1084284.00347

[0107] In some embodiments, the method comprises treating cancer comprising a) determining the likelihood of ICI efficacy with the biomarker; b) administering the ICI to subjects with a high likelihood of response; and / or c) excluding or offering alternative therapies to subjects with a low likelihood of response.Additional Definitions

[0108] To aid in understanding the detailed description of the compositions and methods according to the disclosure, a few express definitions are provided to facilitate an unambiguous disclosure of the various aspects of the disclosure. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

[0109] Unless defined otherwise, all technical and scientific terms used herein have the meaning commonly understood by a person skilled in the art to which this invention belongs. The following references provide one of skill with a general definition of many of the terms used in this invention: Singleton et al., Dictionary of Microbiology and Molecular Biology (2nd ed. 1994); The Cambridge Dictionary of Science and Technology’ (Walker ed., 1988); The Glossary of Genetics, 5th Ed., R. Rieger et al. (eds.). Springer Verlag (1991); and Hale & Marham, The Harper Collins Dictionary of Biology (1991). As used herein, the following terms have the meanings ascribed to them below, unless specified otherwise.

[0110] Aspects of the present disclosure are described herein ith reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. In some embodiments, the flow chart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, a segment, or a portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flow chart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that28180755440.1084284.00347perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

[0111] These computer readable program instructions may be provided to a processor of a general-purpose computer, a special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0112] The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0113] Unless specifically stated otherwise, as apparent from the above discussion, it is appreciated that throughout the description, discussions utilizing terms such as ‘'processing,” “performing,” “receiving,” “computing,” “calculating,” “determining,” “identifying,” “displaying,” “providing,” “merging,” “combining,” “running,” “transmitting,"’ or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (or electronic) quantities within the computer system memories or registers or other such information storage, transmission or display devices.

[0114] As used herein, the term “if may be construed to mean “when"’ or “upon” or “in response to determining” or “in response to detecting,” depending on the context. Similarly, the phrase “if it is determined” or “if (a stated condition or event) is detected” may be construed to mean “upon determining” or”in response to determining” or “upon detecting (the stated condition or event)” or “in response to detecting (the stated condition or event),” depending on the context.29180755440.1084284.00347

[0115] As used herein, "in vitro” refers to events that occur in an artificial environment, e.g. , in a test tube or reaction vessel, in cell culture, etc.. rather than within a multi-cellular organism.

[0116] As used herein, “in vivo” refers to events that occur within a multi-cellular organism, such as anon-human animal.

[0117] It is noted here that, as used in this specification and the appended claims, the singular forms “a,” “an,” and ’the" include plural reference unless the context clearly dictates otherwise.

[0118] As used herein, "plurality ’ means two or more. As used herein, a “set” of items may include one or more of such items.

[0119] As used herein, “including,” “comprising,” “containing,” or “having” and variations thereof are meant to encompass the items listed thereafter and equivalents thereof as well as additional subject matter unless otherwise noted.

[0120] As used herein, the phrases “in one embodiment,” “in various embodiments,” “in some embodiments,” and the like do not necessarily refer to the same embodiment, but may unless the context dictates otherwise.

[0121] As used herein, the terms “and / or” or “I” means any one of the items, any combination of the items, or all of the items with which this term is associated.

[0122] It will be understood that, although the terms “first,” “second,” etc., may be used herein to describe various elements, components, regions, layers and / or sections. These elements, components, regions, layers and / or sections should not be limited by these terms. These terms are only used to distinguish one element, component, region, layer or section from another element, component, region, layer or section. Thus, a first element, component, region, layer or section discussed below could be termed a second element, component, region, layer or section without departing from the teachings of example embodiments.

[0123] As used herein, the term “substantially” does not exclude “completely,” e.g., a composition which is “substantially free” from Y may be completely free from Y. Where necessary, the word “substantially” may be omitted from the definition of the present disclosure.

[0124] As used herein, the term “approximately” or “about,” as applied to one or more values of interest, refers to a value that is similar to a stated reference value. In some embodiments, the term “approximately” or “about” refers to a range of values that fall within 25%, 20%, 19%, 18%, 17%, 16%, 15%, 14%, 13%, 12%, 11%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, or less in either direction (greater than or less than) of the stated reference value unless otherwise stated or otherwise evident from the context (except where such number would 30180755440.1084284.00347exceed 100% of a possible value). Unless indicated otherwise herein, the term “about” is intended to include values, e.g, weight percents, proximate to the recited range that are equivalent in terms of the functionality of the individual ingredient, the composition, or the embodiment.

[0125] As used herein, the term “each,” when used in reference to a collection of items, is intended to identify an individual item in the collection but does not necessarily refer to every item in the collection. Exceptions can occur if explicit disclosure or context clearly dictates otherwise.

[0126] As disclosed herein, a number of ranges of values are provided. It is understood that each intervening value, to the tenth of the unit of the lower limit, unless the context clearly dictates otherwise, between the upper and lower limits of that range is also specifically disclosed. Each smaller range between any stated value or intervening value in a stated range and any other stated or intervening value in that stated range is encompassed within the present disclosure. The upper and lower limits of these smaller ranges may independently be included or excluded in the range, and each range where either, neither, or both limits are included in the smaller ranges is also encompassed within the present disclosure, subj ect to any specifically excluded limit in the stated range. Where the stated range includes one or both of the limits, ranges excluding either or both of those included limits are also included in the present disclosure.

[0127] The use of any and all examples, or exemplary language (e.g.. “such as”) provided herein, is intended merely to better illuminate the invention and does not pose a limitation on the scope of the present disclosure unless otherwise claimed. No language in the specification should be construed as indicating any non-claimed element as essential to the practice of the present disclosure.

[0128] All methods described herein are performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. In regard to any of the methods provided, the steps of the method may occur simultaneously or sequentially. When the steps of the method occur sequentially, the steps may occur in any order, unless noted otherwise. In cases in which a method comprises a combination of steps, each and every combination or subcombination of the steps is encompassed within the scope of the disclosure, unless otherwise noted herein.

[0129] Each publication, patent application, patent, and other reference cited herein is incorporated by reference in its entirety to the extent that it is not inconsistent with the present 31180755440.1084284.00347disclosure. Publications disclosed herein are provided solely for their disclosure prior to the filing date of the present disclosure. Nothing herein is to be construed as an admission that the present disclosure is not entitled to antedate such publication by virtue of prior disclosure. Further, the dates of publication provided may be different from the actual publication dates, which may need to be independently confirmed.

[0130] It is understood that the examples and embodiments described herein are for illustrative purposes only and that various modifications or changes in light thereof will be suggested to persons skilled in the art and are to be included within the spirit and purview of this application and scope of the appended claims.ExamplesEXAMPLE 1Study Design

[0131] To develop the model, data from 2,035 patients across 17 cancer types treated with ICIs between 2014 and 2019 at Memorial Sloan Kettering Cancer Center (MSK; hereafter referred to as MSK-I’) were first retrospectively collected as illustrated in FIGs. 1A-2. A machine-learning model was trained and tested using demographic, clinical, and routine laboratory blood test data from this cohort. The model was then further tested on an independent cohort of additional 2,104 ICI-treated patients from MSK (hereafter referred to as ‘MSK-IF). The MSK-II cohort was collected after initial model development and used identical inclusion and exclusion criteria as the MSK-I cohort but expanded the years of eligibility to patients treated between 2011 and 2020.

[0132] The model was then trained and tested on 4,447 ICI-treated patients in ten global phase 3 clinical trials. Further external testing of the predictive model was performed on a real-world cohort of 1,159 patients treated with ICIs between 2011 and 2019 at Mount Sinai Health System (MSHS). The study protocol was approved by the institutional review boards at Icahn School of Medicine at Mount Sinai and MSK. The use of clinical trial cohorts was approved by data contributor Roche, and data were accessed through Vivli, Inc.Study Patients

[0133] This study included 9,745 patients across 21 cancer types treated with ICIs from MSK, MSHS, and ten global phase 3 clinical trials as illustrated in Table 1, Table 2, and FIGs.1A-2. The ten clinical trial cohorts included patients from 12 experimental arms treated with atezolizumab (anti-PD-Ll): IMbravel50 (Cheng AL, et al. J Hepatol 2022;76(4):862-873), IMspirel50 (Gutzmer R, et al. Lancet 2020;395(10240):1835-1844), IMmotionl51 (Rini BI,32180755440.1084284.00347et al. Lancet 2019;393(10189):2404-2415), IMvigor211 (van der Heijden MS, et al. Eur Urol 2021;80(l):7-ll), IMpowerl33 (Liu SV, et al. J Clin Oncol 2021;39(6):619-630), IMpowerl30 (West H, et al. Lancet Oncol 2019;20(7):924-937), IMpowerl31 (atezolizumab plus carboplatin and nanoparticle albumin-bound paclitaxel (ACNP)) (Jotte R, et al. J Thorac Oncol 2020;15(8):1351-1360), IMpowerl31 (atezolizumab plus carboplatin and paclitaxel (ACP)) (Jotte R, et al. J Thorac Oncol 2020;15(8):1351-1360), IMpowerl32 (Nishio M, et al. J Thorac Oncol 2021;16(4):653-664), IMpowerl50 (atezolizumab plus bevacizumab, carboplatin, and paclitaxel (ABCP)) (Socinski MA, et al. N Engl J Med 2018;378(24):2288-2301), IMpowerl50 (ACP) (Socinski MA, et al. N Engl J Med 2018;378(24):2288-2301), and OAK (Mazieres J, et al. J Thorac Oncol 2021 ; 16(1): 140-150).Clinical Features

[0134] Clinical variables and standardized measurements from routine laboratory blood tests performed on the date of, or no more than 30 days before, the first ICI infusion was retrospectively collected as illustrated in Table 4 and FIG. 7B. In the MSK cohorts, TMB (calculated as the number of nonsynonymous mutations per megabase (mut / Mb)) was collected from patients’ tumors based on the FDA-authorized MSK-IMPACT next-generation sequencing platform (Cheng DT, et al. J Mol Diagn 2015; 17(3):251 -64). In the clinical trial cohorts, PD-L1 immunostaining data based on the SP142 or SP263 clones (Ventana Medical Systems, Tucson, AZ) were collected..Clinical Outcomes

[0135] The primary outcomes were clinical benefit and overall survival. Clinical benefit was defined as a complete response (CR), partial response (PR), or stable disease (SD) without progression for at least six months after the first infusion of ICI. Patients whose tumors experienced progression of disease (PD) or SD for less than six months after the first ICI infusion were classified as having no clinical benefit. CR, PR, SD, and PD were based on RECISTvl.l criteria (Eisenhauer EA, et al. Eur J Cancer 2009;45(2):228-47). Overall survival was calculated from the first ICI infusion to death from any cause. The first line was used for patients who received multiple ICI lines. Overall survival was calculated from randomization to death from any cause for the clinical trial cohorts. Patients alive at the time of review were censored at the last contact. Both clinical benefit and overall survival data were available in the MSK and clinical trial cohorts, but only overall survival data was available in the MSHS cohort.Development of the Machine-Learning Model33180755440.1084284.00347

[0136] The MSK-I cohort was randomly split with 80:20 ratio for the training set (N=l,628) and the hold-out test set (N=407). Two machine-learning models were developed to predict ICI efficacy - one trained on clinical benefit and one trained on overall survival - and selected the one that performed the best in the hold-out test set as illustrated in FIG. 7C. Each predictive model consisted of an ensemble of three algorithms with soft-voting. The predictive model trained on clinical benefit integrated a ridge logistic regression (Le Cessie S, et al. Journal of the Royal Statistical Society Series C (Applied Statistics) 1992:41(1 ): 191-201), a support vector machine (Cortes C, et al. Machine Learning 1995;20(3):273-297), and a random forest (Breiman L. Random Forests. Machine Learning 2001;45(l):5-32). The predictive model trained on overall survival integrated a ridge cox regression (Verweij PJM, et al. Statistics in Medicine 1994; 13 (23 -24): 2427-2436), a fast survival support vector machine (Polsterl S. etal. LectNotes Artif Int 2015;9285:243-259), and arandom survival forest (Ishwaran H, etal. The Annals of Applied Statistics 2008;2(3):841-860). A five-fold cross-validation was used to optimize each algorithm’s hyperparameters during training.

[0137] The model trained on clinical benefit generates a probability of clinical benefit with a higher status indicating a higher probability of having clinical benefit. The model trained on the overall survival calculates a risk status (range from 0 to 1), where a higher status indicates a higher probability of poor outcome (no efficacy or early death) after administration with ICI. The performance of the two models to predict clinical benefit and overall survival was assessed using the area under the receiver operating characteristic curve (AUC) and time-dependent AUC, respectively.Statistical Analysis

[0138] For the primary analysis of predicting clinical outcomes, patients were stratified into high-risk, moderate-risk, and low-risk groups based on the first and third quartile of the risk statuses observed in the training set as illustrated in FIG. 11. Cox proportional -hazards regression tested the association of risk statuses with overall survival. Analyses with overall survival are presented as hazard ratios (HRs) with 95% confidence inter als (Cis) for low -risk and moderate-risk groups relative to the high-risk group as a reference group. P-values for comparing survival probabilities among the three risk groups were computed using the logrank test. Fisher’s exact test compared clinical benefit rates across the three risk groups. The SHapley Additive exPlanations (SHAP) method (Lundberg SM, Lee S-I. Proceedings of the 31st International Conference on Neural Information Processing Sy stems 2017:4768-4777)34180755440.1084284.00347was used to understand the relative feature importance of the model. (See also Example 3 below).EXAMPLE 2Characteristics of the Patient Data

[0139] 9,745 patients diagnosed with 21 different cancer types from MSK, MSHS, and ten clinical trials were identified (Table 3). Patients were treated with inhibitors of PD-1 / PD-L1 (N=9,080), CTLA-4 (N=72), or a combination of both (N=593). Clinical data and routine clinical laboratory test measurements from peripheral blood were obtained in each patient. The median follow-up duration for each cohort was: 25.38 months (interquartile range (IQR), 13.50-45.01) for the training set (Table 3), 27.37 months (IQR, 13.68-49.58) for the hold-out test set. 9.42 months (IQR. 3.10-20.67) for the MSK-II cohort, 8.84 months (IQR, 2.75-28.47) for the MSHS cohort, and 13.64 months (IQR, 6.72-19.86) across clinical trials. The median follow-up duration of each clinical trial cohort is provided in Table SI in the Supplementary Appendix. The MSK data was used to train and internally test the machine-learning model, while the MSHS and clinical trial cohorts were used to test the model performance externally. Additionally, bladder cancer (BLCA), hepatobiliary cancer, melanoma, non-small cell lung cancer (NSCLC), renal cell carcinoma (RCC), and small cell lung cancer were analyzed as separate cancer types as they were collected in all available cohorts. The remaining cancer types were grouped as ‘Others’ in each cohort.Prediction Performance in the Internal Test Datasets

[0140] In the hold-out test data, a machine-learning model trained on overall survival, SCORPIO (Blood-based Estimate from Routine Laboratories for Immuno-Oncology optimization), predicted clinical benefit from ICI with a pan-cancer AUC of 0.714. In contrast, a similar model trained on clinical benefit (denoted SCORPIO-CB) predicted clinical benefit with a pan-cancer AUC of 0.701 as shown in FIG. 12A. TMB had a pan-cancer AUC of 0.546. Moreover, SCORPIO consistently outperformed SCORPIO-CB and TMB in each cancer ty pe, as indicated by the corresponding AUCs. In line with these results, SCORPIO also outperformed SCORPIO-CB and TMB at predicting overall survival, according to the timedependent AUCs as illustrated in FIG. 12B.

[0141] It was then sought to investigate the association between risk statuses calculated by SCORPIO and ICI outcomes. Three risk groups for ICI efficacy (low-risk, moderate-risk, and high-risk) were defined based on the first and third quartiles (0.24 and 0.47, respectively) of the risk statuses that were observed in the training set as illustrated in FIG. 11. In the hold-out 35180755440.1084284.00347test data, the low-risk, moderate-risk, and high-risk groups experienced significantly different overall survival as illustrated in FIG. 13 A. Across tumor types, HRs for death in the low-risk and moderate-risk groups compared to the high-risk group was 0.25 (95% Confidence Interval (CI), 0.18 to 0.34) and 0.48 (95% CI, 0.37 to 0.63), respectively. Furthermore, the clinical benefit rates significantly differed in each risk group across tumor types - low-risk, 55.96%; moderate-risk, 28.64%; high-risk, 12.12% (P<0.001) as illustrated in FIG. 13B. Importantly, the association between nsk groups and clinical outcomes was independent of line of therapy in which ICI was administered, sex, age, Eastern Cooperative Oncology Group performance status (ECOG-PS), microsatellite instability (MSI) status, and TMB as illustrated in FIGs. 14A and 14B.

[0142] It was then sought to test SCORPIO on the independent real-world MSK-II cohort. In line with the results from the hold-out test data, SCORPIO outperformed TMB based on both AUC and time-dependent AUC as shown in FIGs. 15A and 15B. The low-risk, moderaterisk, and high-risk groups had significantly different overall survival as illustrated in FIG. 7A and FIG. 16A. Across tumor types. HRs for death in the low-risk and moderate-risk groups compared to the high-risk group were 0.15 (95% CI, 0.12 to 0.17) and 0.35 (95% CI, 0.31 to 0.40), respectively. Furthermore, the clinical benefit rates significantly differed in each risk group across tumor types - low-risk, 64.44%; moderate-risk, 52.03%; high-risk, 31.25% (P<0.001) as illustrated in FIG. 7A and FIG. 16B. The association between risk groups and clinical outcomes was independent of the line of therapy in which ICI was administered, sex, age, ECOG-PS, MSI status, and TMB as illustrated in FIGs. 17A and 17B.Prediction Performance in the External Test Datasets

[0143] In each clinical trial cohort, the low-risk, moderate-risk, and high-risk groups had significantly different overall survival (P<0.001 for each trial) as illustrated in FIG.4A. In line with this, the clinical benefit rates were significantly different across the risk groups (P=0.04 for IMpowerl33; P=0.003 for IMpowerl31 (ACNP); P=0.001 for IMpowerl31 (ACP); and P<0.001 in the remaining eight trials) as illustrated in FIG. 4B. Notably, these results were independent of the sex, age, and PD-L1 expression as illustrated in FIGs. 18A and 18B. It is important to mention that the patients in the IMvigor211 and OAK trials received ICI in subsequent (2nd or 3rd) lines of therapy, indicating that the model can predict ICI efficacy in patients receiving these drugs in either first or subsequent lines. In these trials, SCORPIO outperformed PD-L1 staining in predicting clinical benefit and overall survival as indicated by different performance metrics as illustrated in FIGs. 19A and 19B.36180755440.1084284.00347

[0144] In the real-world MSHS cohort, the low-risk, moderate-risk, and high-risk groups had significantly different overall survival after ICI administration FIGs. 5A-5B. Across tumor types, HRs for death in the low-risk and moderate-risk groups compared to the high-risk group were 0.25 (95% CI, 0.18 to 0.34) and 0.41 (95% CI, 0.33 to 0.50), respectively. Notably, all these results were independent of the line of therapy in which ICI was administered, sex, age, and ECOG-PS as shown in FIGs. 20A and 20B.Model Interpretability

[0145] The relative effect of all features of SCORPIO in the training set was analyzed based on the SHAP approach, which assesses the direction and relative magnitude of the effect of each feature on the model (FIG. 7 A). The top five features that contributed the most were chloride (CL), albumin (ALB), hemoglobin (HGB), ECOG-PS, and eosinophil proportion among white blood cells (EOS%). A higher ECOG-PS was associated with higher risk, while higher values of the other features were associated with lower risk.

[0146] Representative patients from the hold-out test set with differing risk statuses and corresponding clinical responses are shown in FIGs. 7B-7E. In each case, the direction and magnitude of each feature’s contribution differed according to its own value and that of the other features, demonstrating the complexity of SCORPIO in predicting ICI efficacy in each patient.Discussion

[0147] There is a significant clinical need to develop better and more universally accessible biomarkers to predict which patients with cancer are more or less likely to respond to checkpoint inhibitor drugs; however, currently, available genomic or immunological assays are not widely accessible in global settings. Here, a machine-learning model was described, which relies only on routinely available results from laboratory blood tests and basic clinical data that can predict ICI efficacy better than existing biomarkers (TMB and PD-L1 immunohistochemistry). The data used in the study were collected from two centers and ten global phase 3 clinical trials, resulting in a total of 9,745 patients, which represents the largest predictive modeling analysis in cancer immunotherapy to date.

[0148] The machine-learning system successfully predicted how patients with cancer would respond to ICI across many cancer types. Notably, in most clinical trials, patients in the low-risk and moderate-risk groups identified retrospectively by the machine-learning model had a higher percentage of patients with clinical benefit than those in the immunotherapy arm from the original clinical trial study. These results thus indicate that the model has important 37180755440.1084284.00347ramifications for improving patient outcomes in future clinical trials. Importantly, risk group stratification was based on generalized cutoffs that predicted patients' outcomes to ICI across cancer types.

[0149] The MSHS cohort consists of patients with diverse backgrounds recruited from outpatient centers all over New Y ork City7. Compared to the MSK and clinical trial cohorts, the MSHS cohort is expected to have a more heterogeneous patient population in terms of ethnicity, socioeconomic status, comorbidity, and health literacy. Even with this heterogeneity, consistent results were found compared to the findings from the MSK and clinical trial cohorts.

[0150] Notably, the training set was retrospectively collected over several years from MSK. As a result, NSCLC, melanoma, BLCA, head and neck cancer, and RCC were this cohort’s most prevalent cancer types. Despite this, the model was able to predict ICI efficacy7regardless of the cancer type in multiple external datasets. It was also demonstrated that SCORPIO outperformed TMB and PD-L1 staining in predicting ICI efficacy when feasible.

[0151] Importantly , all features in the machine-learning model are routinely collected in hospitals and clinics worldwide and can be accessed via patient clinical records, which makes the approach more cost-effective and more globally accessible than genomic or other advanced molecular assays.Table 1. Patient characteristics.38180755440.1084284.0034739180755440.1084284.00347IQR: interquartile range; ECOG-PS: Eastern Cooperative Oncology Group performance status; CNS: central nervous system; NSCLC: non-small cell lung cancer; SCLC: small cell lung cancer; ICE immune checkpoint inhibitor; PD-1: programmed death 1; PD-L1: programmed death ligand 1; CTLA-4: cytotoxic T-lymphocyte-associated antigen 4; 1) anti-CTLA-4 with anti-PD-1, 2) anti-CTLA-4 with anti-PD-Ll, 3) anti-CTLA-4 with anti-PD-1 and anti-PD-Ll, and 4) anti-PD-1 and anti-PD-Ll. #Detailed patient characteristics of each clinical trial are described in Table S 1 in the Supplementary Appendix. *Stage at first dose of ICI for the MSK-I, MSHS, and clinical trial cohorts, and stage at diagnosis for the MSK-II cohort.Table 2: Patient characteristics: phase 3 clinical trial cohorts180755440.1084284.00347180755440.1084284.00347Table 2: Patient characteristics: phase 3 clinical trial cohorts (cont.)42180755440.1084284.00347ACNP: atezolizumab plus carboplatin and nanoparticle albumin-bound paclitaxel; ACP: atezolizumab plus carboplatin and paclitaxel; ABCP: atezolizumab plus bevacizumab, carboplatin, and paclitaxel; IQR: interquartile range; ECOG-PS: Eastern Cooperative Oncology Group performance status; ICE immune checkpoint inhibitor; HCC: hepatocellular carcinoma; NSCLC: non-small cell lung cancer; SCLC: small cell lung cancer; PD-L1: programmed death ligand 1.Table 3: Patient characteristics: MSK non-ICI cohort.43180755440.1084284.00347CNS: central nervous system; NSCLC: non-small cell lung cancer; SCLC: small cell lung cancer.Table 4: Feature description.44180755440.1084284.00347<>45180755440.1084284.0034746180755440.1084284.00347ICI: immune checkpoint inhibitor; CTLA-4: cytotoxic T-lymphocyte-associated antigen 4; PD-1: programmed death 1; PD-L1: programmed death ligand 1.Table 5: Features included in models.47180755440.1084284.00347&CNS: central nervous system; H&N: head and neck; NSCLC: non-small cell lung cancer; SCLC: small cell lung cancer. *No significant feature was identified in the feature selection analysis.180755440.1084284.00347Table 6: Hyperparameters."""""49180755440.1084284.00347&50180755440.1084284.0034751180755440.1084284.00347H&N: head and neck; NSCLC: non-small cell lung cancer; SCLC: small cell lung cancer.52180755440.1084284.00347EXAMPLE 3Efficacy Checkpoint Inhibitor Immunotherapy for CancerStudy cohorts

[0152] This study included 9,745 patients across 21 cancer types treated with ICIs from Memorial Sloan Kettering Cancer Center (MSKCC), Mount Sinai Health System (MSHS), and 10 global phase 3 clinical trials as illustrated in FIG. 1A, Table 2, and FIGs. 7A-7E.

[0153] To develop the model, of which steps are shown in FIGs. 8A-8C, data was first retrospectively collected from 2,035 patients across 17 cancer types treated with ICIs between 2014 and 2019 at MSKCC (hereafter referred to as ‘MSK-I’), which were randomly divided into a training set (n = 1,628) and hold-out test set (n = 407) with 80:20 ratio as illustrated in FIG. 8A-8B. Machine learning models were developed using the training set from this cohort and then tested in the hold-out test sets as illustrated in FIG. 8A. The model was further tested on an independent cohort of additional 2,104 ICI-treated patients from MSKCC (hereafter referred to as ‘MSK-II’). The MSK-II cohort was collected after initial model development, and identical inclusion and exclusion criteria were used as the MSK-I cohort, but the years of eligibility’ were expanded to patients treated between 2011 and 2020 as illustrated in FIG. 8B . The model was trained and tested on 4,447 ICI-treated patients in 10 global phase 3 clinical trials. Further external testing of the model was performed on a real-world cohort of 1,159 patients treated with ICIs between 2011 and 2019 at MSHS, a large comprehensive health system serving a diverse patient population across the New York metropolitan region as illustrated in FIG. 8C. A cohort of 6,629 patients were treated for cancer at MSKCC who did not receive ICI (hereafter referred to as ‘MSK non-ICT).Characteristics of the patient data

[0154] Patients were treated with inhibitors of PD-1 (n = 3,793), PD-L1 (n = 5.253), CTLA-4 (n = 72), or combinations of more than one drug (n = 627), including 1) anti-CTLA-4 with anti-PD-1; 2) anti-CTLA-4 with anti-PD-Ll; 3) anti-CTLA-4 with anti-PD-1 and anti-PD-Ll; and 4) anti-PD-1 and anti-PD-Ll. The median follow-up duration for each cohort was: 25.38 months (interquartile range [IQR] 13.50-45.01) forthe training set, 27.37 months (IQR 13.68-49.58) for the hold-out test set, 9.42 months (IQR. 3.10-20.67) for the MSK-II cohort, 8.84 months (IQR 2.75-28.47) fortheMSHS cohort, and 13.64 months (IQR 6.72-19.86) across the clinical trials. As shown in FIG. 1A, the 10 clinical trial cohorts included patients from 12 experimental arms treated with atezolizumab (anti-PD-Ll): IMbravel50, IMspirel50, IMmotionl51, IMvigor211. IMpowerl33, IMpowerl30, IMpowerl31 (atezolizumab plus 53180755440.1084284.00347carboplatin and nanoparticle albumin-bound paclitaxel [ACNP]), IMpowerl31 (atezolizumab plus carboplatin and paclitaxel [ACP]), IMpowerl32, IMpowerl50 (atezolizumab plus bevacizumab, carboplatin, and paclitaxel [ABCP]), IMpowerl50 (ACP), and OAK. Methods of analyzing the cohorts are illustrated in FIGs. 9A-9J. The cancer groups comprised bladder cancer (BLCA), hepatobiliary cancer, melanoma, non-small cell lung cancer (NSCLC), renal cell carcinoma (RCC), and small cell lung cancer (SCLC) as separate cancer types as they were collected in all available cohorts. The remaining cancer types were grouped as ‘Others’ in each cohort.Clinical features and outcomes

[0155] Clinical variables and standardized measurements from routine laboratory blood tests were performed on the date of, or no more than 30 days before, the first ICI infusion. In the MSKCC cohorts, TMB was collected from patients’ tumors based on the FDA-authorized MSK-IMPACT platform. In the clinical trial cohorts, PD-L1 immunostaining data using the SP142 or SP263 clones (Ventana Medical Systems, Tucson, AZ) were collected. (For a description of all features, see Methods). The two primary outcomes were overall survival and a treatment effect outcome, measured as clinical benefit. Overall survival was measured from the first ICI infusion to death from any cause, with the first line used for patients who received multiple ICI treatments. For clinical trial cohorts, overall survival was measured from randomization to death from any cause. Patients alive at the time of review were censored at their last contact. Clinical benefit was defined as a patient’s tumor undergoing either a complete response (CR), partial response (PR), or stable disease without progression for at least six months after the first infusion of ICI (SD>6m), as in prior studies. Patients whose tumors experienced progression of disease (PD) or SD for less than six months after the first ICI infusion (SD<6m) were classified as having no clinical benefit. CR, PR, SD, and PD were based on RECIST vl.l criteria. Both primary outcomes were available in the MSKCC and clinical trial cohorts, but only overall survival data was available in the MSHS cohort (For a description of clinical features and outcomes, see Methods).Development of the machine learning model

[0156] Prior to model training, feature selection analyses were performed on the training set to identify features associated with the target outcomes of ICI treatment as shown in FIG. IB and FIGs. 10A-10B. Two machine learning models were developed using demographic, clinical, and routine laboratory blood test data and inflammatory blood test data, for example C-Reactive Protein (CRP) to predict outcomes after ICI administration- one trained to predict 54180755440.1084284.00347overall survival and the other trained to predict clinical benefit (CR, PR, and SD>6m)- and selected the one that performed the best in the hold-out test set. Each model consisted of an ensemble of three algorithms with soft-voting. A five-fold cross-validation (CV) was used to optimize each algorithm’s hyperparameters during training.

[0157] During model training, the training set was divided into five equal-sized folds, each containing the same proportion of data. The algorithm then underwent five iterations of training and evaluation. In each iteration, four folds were used for training, and one fold was used for validation. Model performance was assessed using the concordance index (C-index) for overall survival and the area under the receiver operating characteristic curve (AUC) for clinical benefit. The performance metrics from the five iterations were averaged to obtain a single performance measurement. This process was repeated for all possible hyperparameter combinations, and the hyperparameter with the highest performance metric was selected as the optimal hyperparameter as shown in FIG. 1C.

[0158] The model trained to predict overall survival, SCORPIO (Standard Clinical and laboratory featuRes for Prognostication of Immunotherapy Outcomes), calculates a risk status ranging from 0 to 1, where a higher status indicates a higher probability of a poor outcome (i.e., lack of efficacy or early death) after ICI administration. This model was trained using 33 features significantly associated with overall survival, identified through feature selection analysis as shown in FIGs. 10A-10B. Similarly, SCORPIO-CB, trained to predict clinical benefit, generates a probability status from 0 to 1, with a higher status indicating a higher likelihood of clinical benefit. This model was trained with 22 features significantly associated with clinical benefit (Table 2), as identified in the feature selection analysis as shown in FIG.2. The performance of the tw o models w as assessed using time-dependent AUC (AUC[t]) for overall survival and AUC for clinical benefit as show in FIG. 2.

[0159] For the primary analysis of predicting clinical outcomes, patients were stratified into high-risk, moderate-risk, and low-risk groups according to the first and third quartile of the risk statuses that were observed in the training set as shown in FIG. 11. Cox proportional -hazards regression tested the association of risk statuses with overall survival and Fisher's exact test compared clinical benefit rates across the three risk groups.

[0160] As shown in FIGs. 10A-10B, Cox proportional-hazards regression was used for testing the association between each feature and overall survival. CL: chloride; ALB: albumin; ALK: alkaline phosphatase; AST: aspartate aminotransferase; AGAP: anion gap; PROT: total protein; ALT: alanine aminotransferase: GLU: glucose; CREAT: creatinine; BILL total 55180755440.1084284.00347bilirubin; eGFR: estimated glomerular filtration rate; CO2: carbon dioxide; BUN: blood urea nitrogen; CA: calcium; K: potassium; ECOG-PS: Eastern Cooperative Oncology Group performance status; BMI: body mass index; Age: age at ICI; Smoking: smoking history; DrugClass: type of ICI drug; Virus: HPV or EBV infection; DuringChemo: systemic therapy during ICI; NEUT: neutrophil count; NLR: neutrophil-to-lymphocyte ratio; WBC: white blood cell; LYM%: lymphocyte proportion among WBCs; NEUT%: neutrophil proportion among WBCs; MLR: monocyte-to-lymphocyte ratio; HGB: hemoglobin; HCT: hematocrit; MONO: monocyte count; RBC: red blood cell; RDW: red blood cell distribution width; EOS%: eosinophil proportion among WBCs; LYM: lymphocyte count; PLT: platelet; BASO%; basophil proportion among WBCs; BLR: basophil-to-lymphocyte ratio; MCHC; mean corpuscular hemoglobin concentration; MCH: mean corpuscular hemoglobin; EOS: eosinophil count; MCV: mean corpuscular volume; ELR: eosinophil -to-lymphocyte ratio; BASO: basophil count; MONO%: monocyte proportion among WBCs. b, Cochran-Mantel-Haenszel test was used to test the association between each feature and clinical benefit. These analyses were performed using the training set. Features were displayed in order of significance in each modality. * False discovery rate (FDR) adjusted P < 0.05. ** FDR adjusted P < 0.01. *** FDR adjusted P < 0.001. **** FDR adjusted P < 0.0001.Model performance in the internal test datasets

[0161] In the hold-out test data, SCORPIO, the machine learning model trained to predict overall survival using time-dependent AUC (AUC[f|) values as shown in FIGs. 12A-12B, predicted overall survival at 6-, 12-, 18-, 24-, and 30-month following ICI with a median pancancer AUC(t) of 0.763 as shown in FIGs. 12A-12B. SCORPIO outperformed SCORPIO-CB and TMB in predicting overall survival, as shown by AUC(t) values. SCORPIO also predicted clinical benefit with a pan-cancer AUC 0.714, surpassing SCORPIO-CB CB (pan-cancer AUC of 0.701) and TMB (pan-cancer AUC 0.546). SCORPIO consistently outperformed both SCORPIO-CB and TMB across all cancer types.

[0162] To determine whether cancer-type-specific models provide better predictive value than SCORPIO, a pan-cancer model, models were trained on data specific to each cancer type. First, feature selection analyses were conducted and model training separately for each cancer type. Among the 17 cancer types in the training set, 10 were identified with features significantly associated with overall survival (Table 5) as shown in FIGs. 12A-13B.Performance comparisons for predicting overall survival after administration with ICI were evaluated using AUC(t) values as shown in FIG. 13A. Performance comparisons for predicting 56180755440.1084284.00347clinical benefit to ICI were evaluated using AUC values as shown in FIG. 13B. Pancreatic cancer was excluded from the analysis because all patients showed no clinical benefit. Then 10 models were trained, and their performance was compared to SCORPIO in the hold-out test set. SCORPIO outperformed most of the cancer-type-specific models in predicting both overall survival (Table 5) and clinical benefit as shown in FIGs. 14A-14B. Performance comparisons for predicting overall survival after administration with ICI were evaluated using AUC(t) values as illustrated in FIG. 14A. Performance comparisons for predicting clinical benefit to ICI were evaluated using AUC values as illustrated in FIG. 14B. This indicates that SCORPIO trained on the large pan-cancer data successfully learned relevant relationships across cancer types.

[0163] Next. SCORPIO’S performance was compared to nine machine learning models from Vanguri et al., which predict ICI efficacy in patients with NSCLC using uni-, bi-, or multimodal data (radiology, pathology, tumor genetics, and PD-L1 scoring). SCORPIO, trained on more accessible pan-cancer data, outperformed these models in predicting overall survival as shown in FIG. 15A and showed comparable performance in predicting clinical benefit, even though it used simpler, more accessible data as shown in FIG. 15B. Performance comparison for predicting overall survival after administration with ICI was evaluated using AUC(t) values as shown in FIG. 15A. Performance comparison for predicting clinical benefit to ICI was evaluated using AUC values as shown in FIG. 15B. Dy AM multi-modal with TPS refers to the model from Vanguri et al. based on multi-modal data (radiology, pathology, genomics, and PD-L1 tumor proportion status).

[0164] In the hold-out test data, the three risk groups (low-risk, moderate-risk, and high-risk) showed significantly different overall survival as shown in FIG. 16A. Across tumor types, the hazard ratios (HRs) for death compared to the high-risk group were 0.25 (95% CI. 0.18-0.34) for the low-risk group and 0.48 (95% CI, 0.37-0.63) for the moderate-risk group. Furthermore, the clinical benefit rates significantly differed in each risk group across tumor ty pes - low-risk, 55.96%; moderate-risk, 28.64%; high-risk, 12.12% (P = 3.22 x 10-11) as shown in FIG. 16B.Kaplan-Meier plots were made for overall survival of the three risk groups predicted by SCORPIO. Tick marks indicate censored data. Black vertical and horizontal dotted lines indicate the median survival times of each risk group as shown in FIG. 16A. As shown in FIG.16B, bar charts w ere made for clinical benefit rates of the three risk groups predicted by SCORPIO.57180755440.1084284.00347

[0165] SCORPIO was then tested on the independent real-world MSK-II cohort. In this cohort, SCORPIO predicted overall survival at 6-, 12-, 18-, 24-, and 30-month following ICI with a median pan-cancer AUC(t) of 0.759 as shown in FIG.2. It also predicted clinical benefit from ICI with a pan-cancer AUC of 0.641. In accordance with the results from the hold-out test data, SCORPIO outperformed TMB based on both AUC(t) and AUC as shown in FIGs.17A-18B. Performance comparisons for predicting overall survival after administration with ICI were evaluated using AUC(t) values as shown in FIG. 18A. Performance comparisons for predicting clinical benefit to ICI were evaluated using AUC values as shown in FIG. 18B.Central nervous system cancer (CNS) was excluded from this analysis since TMB data was not available for patients with CNS in the MSK-II cohort. The three risk groups had significantly different overall survival as shown in FIG. 3A and FIGs. 19A-19B. Kaplan-Meier plots for overall survival corresponding to each of the three risk groups predicted by SCORPIO. Tick marks indicate censored data. Black vertical and horizontal dotted lines indicate the median survival times of each risk group as shown in FIG. 19A. Thirteen cancer types from the ‘Others’ group were evaluated separately. Bar charts were plotted for clinical benefit rates of the three risk groups predicted by SCORPIO as shown in FIG. 19B. Twelve cancer types from the ‘Others’ group were evaluated separately. Across tumor types, HRs for death in the low-risk and moderate-risk groups compared to the high-risk group were 0.16 (95% CI, 0.14-0.19) and 0.38 (95% CI, 0.34-0.43), respectively. Furthermore, the clinical benefit rates significantly differed in each risk group across tumor types - low-risk, 65.09%; moderate-risk. 52.20%; high-risk, 32.89% (P = 2.35 x 10-11) as shown in FIG. 3B. In both internal test datasets, the association between risk groups and clinical outcomes was independent of the line of therapy in which ICI was administered, sex, age, Eastern Cooperative Oncology Group performance status (ECOG-PS). microsatellite instability (MSI) status, and TMB as shown in FIGs. 20A-21B

[0166] To determine whether SCORPIO is specifically prognostic for ICI efficacy or generally prognostic for patients with cancer regardless of treatment, a cohort of non-ICI-treated patients was analyzed from MSKCC. SCORPIO was able to predict overall survival for non-ICI patients across various cancer types as illustrated in FIGs. 22A-22B. A Kaplan-Meier plot was generated for overall survival of the three risk groups predicted by SCORPIO in each subgroup as shown in FIG. 22A. Patients older than or equal to 63.42 years old (median age from the training set) and younger than 63.42 years old were classified as old and young age groups, respectively. Tick marks indicate censored data. Black vertical and horizontal dotted 58180755440.1084284.00347lines indicate the median survival times of each risk group. Bar charts for clinical benefit rates of the three risk groups predicted by SCORPIO in each subgroup were made in accordance with FIG. 22B. However, in contrast to the ICI treatment context, its prognostic accuracy decreased for specific cancers such as bladder cancer (P = 0.2713), hepatobiliary cancer (P = 0.1038), esophageal cancer (P = 0.8886), and ovarian cancer (P = 0.4305 as shown in FIG.3A, FIG. 16A, FIG. 19A, and FIG.22A-22B. Kaplan-Meier plots were created for overall survival of the three nsk groups predicted by SCORPIO. Tick marks indicate censored data. Black vertical and horizontal dotted lines indicate the median survival times of each risk group. These findings indicate that SCORPIO is more effective at predicting survival in the context of ICI treatment.Model interpretability

[0167] To understand how each feature contributes to SCORPIO’S risk status prediction, the relative effect of its 33 features in the training set were analyzed using the SHapley Additive exPlanations (SHAP) approach as shown in FIG. 4A. SHAP quantified the contribution of each feature to patient-to-patient variation in ICI efficacy as shown in FIG. 4A. The top five features contributing the most were chloride (CL), albumin (ALB), hemoglobin (HGB), ECOG performance status (ECOG-PS), and eosinophil proportion among white blood cells (EOS%).

[0168] As shown in FIGs. 4A-4E, data was gathered for representative patients from the hold-out test set with different risk statuses and clinical responses. Each feature's contribution varied in direction and magnitude based on its value and the values of other features, demonstrating the model's complexity in predicting ICI efficacy for each patient as shown in FIGs. 4A-4E

[0169] The top five features were analyzed, and the predicted risk status reflects characteristics of the tumor microenvironment (TME) as shown in FIG. 4F. An additional cohort of 264 patients with NSCLC was analyzed, with available bulk RNA-sequencing (RNA-seq), blood test values (performed on the date of, or no more than 30 days before, the tumor biopsy), and clinical data. Using the Danaher signature, which was validated as the most accurate immune cell deconvolution method for NSCLC, 14 immune cell types were deconvoluted as shown in FIGs. 4F-4G. The correlations between their abundances and the levels of the top five features w ere analyzed, as well as the predicted risk status as shown in FIG. 4F. Higher ALB levels w ere associated with an increased abundance of mast cells, T-cells, B-cells, CD45 cells, and regulatory T-cells. Conversely, lower ECOG-PS was linked to a greater abundance of T-cells, B-cells, CD45 cells, exhausted CD8 cells, and cytotoxic cells.59180755440.1084284.00347Additionally, a lower predicted risk status (indicating better-predicted response to immunotherapy) corresponded with higher abundances of mast cells, T-cells, B-cells, CD45 cells, regulatory T-cells, natural killer CD56 dim cells, and Thl cells. The association between the abundances of the 14 immune cell types and the levels of the top five features were analyzed, as well as the predicted risk status, in patients with head and neck (H&N) cancer (n = 32) from the MSK-I cohort as shown in FIG. 4G. Compared to the NSCLC cohort, there were fewer significant associations, likely due to the smaller sample size. However, very similar relationships were observed - various immune cell types were positively correlated with ALB levels, while ECOG-PS and predicted risk status were negatively correlated with many immune cell types. These results indicate that some features in SCORPIO reflect the TME status, and a low predicted risk status corresponds to an immune-infl amed phenotype in patients.

[0170] Also assessed was whether the top five features correlated with TMB. Using patients from multiple MSKCC cohorts (n = 2,969), it w as found that TMB w as generally not associated with the top five features or the risk statuses, except in a few cancer types as shown in FIG.23. For FIG. 23, a number in each circle denotes Spearman’s p. * FDR adjusted P < 0.05. ** FDR adjusted P < 0.01. CL: chloride; ALB: albumin; HGB: hemoglobin; ECOG-PS: Eastern Cooperative Oncology' Group performance status; EOS%: eosinophil proportion among white blood cellsModel performance in the external test datasetsa) Among the clinical trial cohorts, SCORPIO achieved its highest performance in predicting overall survival at 6-, 12-, 18-, 24-, and 30-months in the IMvigor211 trial (bladder cancer) with a median AUC(t) of 0.782, and in predicting clinical benefit in IMspirel50 trial (melanoma), with an AUC of 0.684 as shown in FIG. 2. In each clinical trial cohort, the three risk groups showed significantly different overall survival rates (P < 0.0001 for each trial) as shown in FIG. 5A. Similarly, clinical benefit rates varied significantly across risk groups (P-values: 0.043 for IMpowerl33, 0.0027 for IMpowerl31 [ACNP], 0.001 for IMpowerl31 [ACP], 0.0004 for IMpowerl50 [ABCP], 0.0003 for IMspirel50 and OAK, 0.0001 for IMpowerl50 [ACP], and <0.0001 for the remaining trials as shown in FIG. 5B. Importantly, these results were independent of sex, age, and PD-L1 expression as shown in FIGs. 24A-24X. In clinical trials, SCORPIO outperformed PD-L1 staining in predicting clinical benefit and overall survival, as indicated by various performance metrics as shown in FIGs.25A-25X.60180755440.1084284.00347

[0171] To further test model generalizability, a real-world cohort of patients treated at a large, comprehensive health system (MSHS) was analyzed, encompassing a diverse patient population. In this cohort, SCORPIO predicted overall survival at 6-, 12-, 18-, 24-, and 30-month following ICI with a median pan-cancer AUC(t) of 0.725 as shown in FIG. 2, and the three risk groups had significantly different overall survival after ICI administration as shown in FIG. 6. Across tumor types, HRs for death in the low-risk and moderate-risk groups compared to the high-risk group were 0.25 (95% CI, 0.18-0.34) and 0.41 (95% CI, 0.33-0.50), respectively. Importantly, all these results were independent of the line of therapy in which ICI was administered, sex, age, and ECOG-PS as illustrated in FIG. 26. Kaplan-Meier plots were generated for overall survival of the three risk groups predicted by SCORPIO in each subgroup as in FIG.26. Patients older than or equal to 63.42 years old (median age from the training set) and younger than 63.42 years old were classified as old and young age groups, respectively. Tick marks indicate censored data. Black vertical and horizontal dotted lines indicate the median survival times of each risk group.Model performance comparison across cohorts and tumor types

[0172] SCORPIO performed better in predicting overall survival in real-world cohorts compared to phase 3 clinical trials across most cancer types as shown in FIG. 2. For example, in bladder cancer, the median AUC(t) in real-world cohorts was 0.809 (across all time points and cohorts), higher than the median AUC(t) of 0.782 observed in the IMvigor211 trial. Similarly, for hepatobiliary cancer, the median AUC(t) in real-world data was 0.746 compared to the median AUC(t) of 0.704 in the IMbravel 50 trial as shown in FIG.2. In RCC, SCORPIO showed the most robust performance in real-world cohorts, with a median AUC(t) of 0.829 compared to the median AUC(t) of 0.668 in the IMmotionl51 trial as shown in FIG. 2.

[0173] The model performed better in real-world cohorts likely due to the broader range of patient characteristics, cancer types, and treatment environments in the training data. This also indicates that the model effectively captures the complexities and variations found in everyday clinical practice, enhancing its applicability for predicting the efficacy of ICIs in diverse patient populations. Notably, the model performed better at predicting overall survival than predicting clinical benefit across most cancer types and cohorts, likely reflecting the robustness of overall survival as a reliable clinical endpoint, which is often prioritized in oncology for its clear and objective outcomes to improve the prediction of clinical benefit.

[0174] Furthermore, the analysis revealed that SCORPIO’S performance in predicting clinical benefit is not uniform across different cancer types (FIG. 2). To understand the 61180755440.1084284.00347variability in performance across cancer types, SHAP values between cancer-type-specific models and SCORPIO were compared. This analysis revealed key features that SCORPIO'S pan-cancer modeling approach may have overlooked. The findings showed some variability in the importance of specific features in different cancer types within the SCORPIO model FIG.27. For example, while SHAP analyses indicated that ALB and HGB were important in SCORPIO, their importance was reduced in cancer-type-specific models, particularly in bladder cancer, ovarian cancer, H&N cancer, and NSCLC. Additionally, features like viral infection, relevant in H&N cancer due to human papillomavirus status, and platelet, influential in both H&N cancer and melanoma, highlight the unique biological characteristics of each cancer type. Notably, SCORPIO outperformed cancer-type-specific models in predicting overall survival and clinical benefit, demonstrating its robustness and generalizability FIG. 14.Discussion

[0175] SCORPIO, a machine learning model that relies on routine blood tests and basic clinical data, predicts clinical outcomes after ICI administration more effectively than existing FDA-approved biomarkers like TMB and PD-L1 immunohistochemistry.

[0176] Data were collected from two centers and 10 global phase 3 clinical trials, totaling 9,745 patients, representing the largest dataset in cancer immunotherapy to date. Importantly, risk group stratification was based on generalized cutoffs that predicted patient outcomes across cancer types. The MSHS cohort consists of patients with diverse backgrounds from outpatient centers across New York City. Compared to the MSKCC cohort and clinical trial cohorts, the MSHS cohort is more heterogeneous regarding ethnicity, socioeconomic status, comorbidity, and health literacy. Despite this heterogeneity’, consistent results were found across the MSHS, MSKCC, and clinical trial cohorts.b) SCORPIO outperformed TMB and PD-L1 staining in predicting ICI efficacy. In addition, PD-L1 immunohistochemistry is not universally available and is performed using various platforms, antibodies, and quality assurance practices. TMB estimation requires resource-intensive genomic profiling, and measured TMB varies across genomic panels due to differences in panel size, gene content, and bioinformatics pipelines.

[0177] The model can predict clinical benefits and survival across multiple external datasets from other medical centers and global clinical trials. The diversity’ of these external datasets introduces heterogeneity but also confirms the model's generalizability

[0178] SCORPIO is a highly accessible model for ICI efficacy prediction. It can support clinical decision-making in several scenarios, such as: 1) prioritizing treatment options when 62180755440.1084284.00347considering ICI, cytotoxic, and / or targeted therapies, 2) weighing the risk versus benefit probabilities of ICI drugs for patients at risk of immune-related adverse events, 3) informing clinical trial design to select or enrich patients more or less likely to benefit from ICI therapies.

[0179] Another advantage of SCORPIO is its accessibility in all practice settings, including low-resource healthcare environments. All features in SCORPIO are routinely collected in hospitals and clinics worldwide and can be accessed via patient clinical records, making the approach non-invasive, cost-effective, and globally accessible.MethodsCohort description: MSK-I cohort

[0180] A real-world cohort was assembled comprising 3,278 patients who were treated with at least one dose of ICI from 2014 through 2019 from MSKCC. Excluded were 818 patients with a history of more than one cancer, 26 patients who were enrolled in blinded trials, 115 patients with cancer types with few er than 25 cases, 184 patients with inadequate clinical or laboratory7data. Also excluded were 100 patients w ho received ICI in a neoadjuvant or adjuvant setting. As a result, the MSK-I cohort consisted of 2,035 patients across 17 cancer types (Table 1). Of 2,035 patients, the median age was 63.50 years (IQR 54.77-70.92 years), and 1,164 (57.20%) were male. Of the total, 1,332 patients (56.11%) were treated with ICI as the first line of therapy. The most abundant cancer types were: NSCLC (n = 666, 32.73%); RCC (n = 229, 11.25%); melanoma (n = 210, 10.32%); H&N cancer (n = 168, 8.26%), and BLCA (n = 111, 5.45%) as seen in FIG.7A.Clinical features: MSK-I cohort

[0181] Two features were collected for demographic data (Table 4) (Age and Sex), and eight features were collected for clinical data (BMI, drug class [DrugClass], chemotherapy during immunotherapy [DuringChemo], systemic therapy history [PreChemo], ECOG-PS, smoking history [Smoking], tumor stage [Stage], and viral infection [Virus]). From blood tests, 47 features were initially collected: 17 features from comprehensive metabolic panel (CMP: ALB, alkaline phosphatase [ALK], alanine aminotransferase [ALT], anion gap [AGAP], aspartate aminotransferase [AST], blood urea nitrogen [BUN], calcium [CA], chloride [CL], carbon dioxide [CO2], creatinine [CREAT], estimated glomerular filtration rate [eGFR], glucose [GLU], potassium [K], bilirubin [BILI], total protein [PROT], magnesium [MG], and phosphorus [P]), 21 features from complete blood count (CBC: white blood cell count [WBC], basophil count [BASO], eosinophil count [EOS], granulocytes count [GRAN], lymphocyte count [LYM], monocyte count [MONO], neutrophil count [NEUT], basophil proportion 63180755440.1084284.00347among WBC [BASO%], eosinophil proportion among WBC (EOS%), granulocytes proportion among WBC [GRAN%], lymphocyte proportion among WBC [LYM%], monocyte proportion among WBC [MONO%], neutrophil proportion among WBC [NEUT%], hematocrit [HCT], HGB, mean corpuscular hemoglobin concentration [MCHC], mean corpuscular hemoglobin [MCH], mean corpuscular volume [MCV], platelet [PLT], red blood cell [RBC], and red blood cell distribution width [RDW]), 3 features from coagulation panel (activated partial thromboplastin time |APTT|. international normalized ratio [INR], and prothrombin time [PT]), conjugated bilirubin (CB), direct bilirubin (DB), glucose-6-phosphate dehydrogenase (G6PD), ionized calcium (iCA), lactate dehydrogenase (LDH), and lipase (LPS). Among these, 13 features, which had more than or equal to 70% missing values across the patients in the cohort, were removed from the subsequent analyses: 2 features from CMP (MG and P), 2 features from CBC (GRAN and GRAN%), and all features from coagulation panel (APTT, INR, and PT), CB, DB, G6PD, iCA, LDH, and LPS.

[0182] Then, four immune cell-to-lymphocyte ratios were manually calculated as the absolute count of each immune cell type divided by the absolute count of lymphocytes: basophil-to-lymphocyte rate (BLR), eosinophil-to-lymphocyte ratio (ELR). monocyte-to-lymphocyte ratio (MLR), and neutrophil -to-lymphocyte ration (NLR). The above immune cell-to-lymphocyte ratios were considered as part of the CBC.

[0183] In total, there were 48 features from four types of data modalities: demographic (n = 2), clinical (n = 8). CMP (n = 15), and CBC (n = 23). All clinical features were collected prior to the first ICI infusion (performed on the date of, or no more than 30 days before, the first ICT infusion). For eGFR, results are reported without race adjustment. Tumors were staged at the time of ICI administration following the guidelines from the American Joint Committee on Cancer, 8th edition (with the exception of primary central nervous system [CNS] malignancies, which were not staged).

[0184] TMB data from MSK-IMPACT next-generation sequencing assay, approved by the FDA as a tumor profiling method, was available. TMB was defined as the total number of somatic nonsynonymous mutations per megabase (mut / Mb). For the subgroup analyses, patients with TMB>10 and TMB<10 were defined as TMB-High and TMB-Low groups, respectively. MSI status was evaluated using MSIsensor with the following criteria: stable (0 < MSI status < 3), indeterminate (3 < MSI status < 10), and unstable (MSI status > 10).

[0185] Prior to performing feature selection and training the machine learning algorithms, missing values were imputed in the MSK-I cohort using MissForest from the missingpy 64180755440.1084284.00347package (v.0.2.0) with default parameters (max_iter=10, decreasing=False, missing_values=np.nan, copy=True. n_estimators=100, criterion=('mse', 'gini'), max_depth=None, min_samples_split=2, min_samples_leaf=l, in_weight_fraction_leaf=0.0, max_features='auto', max_leaf_nodes=None, rnin_impurity_decrease=O.O. bootstrap=True, oob_status=False, njobs=-l, random_state=None, verbose=0, warm_start=False, class_weight=None) using the Python 3.8.8 (https: / / www.python.org / ). The average number of missing values across the 48 features was 0.70 per patient. After missing value imputation, the MSK-I cohort was randomly split with 80:20 ratio for the training set (n = 1,628) as shown in FIG. 7A and FIG. 7B and the hold-out test set (n = 407) as shown in FIG. 7. To avoid any potential bias of model performance between the training and hold-out test sets, the same distribution of tumor response, cancer types, and systemic therapy history between the training and hold-out test sets were maintained as show n in FIG.7D. Splitting the training and the holdout test sets was performed with the group_by and sample_frac functions from the dplyr (v.1.1.4) and tidy verse (v.2.0.0) packages using the R programming language version 4.1.1 (https : / / www. r-proj ect. org / ).Cohort description: MSK-II cohort

[0186] An additional real -wo rid cohort with 3,159 patients who were treated with at least one dose of ICI from 2011 through 2020 at MSKCC was retroactively detected to further test the model internally, drawn from patients captured under a broader time period than the MSK-I cohort and contemporaneous patients not undergoing tumor genomic sequencing as shown in FIG. 7B. Excluded were 660 patients with a history of more than one cancer, 14 patients who were enrolled in blinded trials, 65 patients with cancer types with few er than 10 cases, 184 patients with inadequate clinical or laboratory data. Also excluded were 132 patients who received ICI in a neoadjuvant or adjuvant setting. As a result, the MSK-II cohort consisted of 2,104 patients across 19 cancer types (Table 1) as shown in FIG. 7D. Of the 2,104 patients, the median age was 67.13 years (IQR 58.59-74.33 years), and 1,180 (56.08%) were male. Of the total, 1,189 patients (56.51%) were treated with ICI as the first line of therapy. The most abundant cancer types were: NSCLC (n = 755, 35.88%); BLCA (n = 156, 7.41%); RCC (n = 154, 7.32%); melanoma (n = 151, 7.18%), and SCLC (n = 137, 6.51%) as shown in FIG. 7B.Clinical features: MSK-II cohort

[0187] For the MSK-II cohort, four types of data modalities required for SCORPIO were collected: demographic (n = 1), clinical (n = 4), CMP (n = 11), and CBC (n = 17). All clinical features except for tumor stage were retrieved (performed on the date of, or no more than 3065180755440.1084284.00347days before, the first ICI infusion). Tumors were staged at diagnosis following the guidelines from the American Joint Committee on Cancer, 8th edition (except for primary CNS malignancies, which were not staged). For eGFR, results are reported without race adjustment. In the MSK-II cohort, 934 patients (44.39%) underwent MSK-IMPACT sequencing. Hence, TMB data was only available for this subset of patients. For subgroup analyses, patients with TMB>10 and TMB<10 were defined as TMB-High and TMB-Low groups, respectively. MSI status was evaluated using MSIsensor with the following criteria: stable (0 < MSI status < 3), indeterminate (3 < MSI status < 10), and unstable (MSI status > 10).

[0188] Missing values were imputed using MissForest from the missingpy package with default parameters after combining this cohort with the training set into a single data frame. The average number of missing variables across the 33 features was 0.57 per patient.Cohort description: MSK non-ICI cohort

[0189] For the MSK non-ICI cohort, 6,629 patients not treated with ICI were derived from a previous study. The median age in this cohort was 61.15 years (IQR 50.83-69.55 years), and 2,912 (43.93%) were male. The most abundant cancer types wereNSCLC (n= 1,160, 17.50%), colorectal cancer (n = 1,124, 16.96%), breast cancer (n = 820, 12.37%), pancreatic cancer (n = 753, 11.36%), and sarcoma (n = 541, 8.16%). For eGFR, results are reported without race adjustment.Clinical features: MSK non-ICI cohort

[0190] For the MSK non-ICI cohort, four types of data modalities required for SCORPIO were collected: demographic (n = 1), clinical (n = 4), CMP (n = 11), and CBC (n = 17). All clinical features were collected at the time of diagnosis. Tumors were also staged at diagnosis following the guidelines from the American Joint Committee on Cancer, 8th edition (with the exception of primary CNS malignancies, which were not staged).

[0191] Missing values were imputed using MissForest from the missingpy package with default parameters after combining this cohort with the training set into a single data frame.Cohort description: MSHS cohort

[0192] An additional retrospective real-world cohort from MSHS was collected to test whether the prognostic power of the disclosed framework was generalizable to a different healthcare setting as shown in FIG.7E. In the MSHS cohort, 1,230 patients who were treated with at least one dose of ICI from 2011 through 2019 w ere identified. Excluded were one patient who was enrolled in a blinded trial, 26 patients treated with ICI for hematologic malignancies, 16 patients with cancer types of fewer than 10 cases, and 28 patients with 66180755440.1084284.00347inadequate clinical or laboratory data. As a result, the MSHS cohort consisted of 1,159 patients across 18 cancer types (Table 1). Of 1,159 patients, the median age was 66.84 years (IQR 58.92-74.38 years), and 691 (59.62%) were male. Of the total, 551 patients (47.52%) were treated with ICI as the first line of therapy. The most abundant cancer types were: hepatobiliary cancer (n = 304, 26.23%); NSCLC (n = 281, 24.25%); melanoma (n = 128, 11.04%); H&N cancer (n = 94, 8.11%), and RCC (n = 62, 5.35%) as shown in FIG. 7E.Clinical features: MSHS cohort

[0193] For the MSHS cohort, four types of data modalities that were required for SCORPIO were collected: demographic (n = 1), clinical (n = 4), CMP (n = 11), and CBC (n = 17). All clinical features were retrieved (performed on the date of, or no more than 30 days before, the first ICI infusion). For eGFR, results are reported without race adjustment. The following records of the 1,175 patients were manually reviewed for clinical data verification: ECOG-PS, cancer type, tumor stage, smoking history, drug type, and systemic therapy history. Tumors were staged at the time of ICI administration following the guidelines from the American Joint Committee on Cancer. 8th edition (with the exception of primary CNS malignancies, which were not staged). Missing values were imputed using MissForest from the missingpy package with default parameters after combining this cohort with the training set into a single data frame. The average number of missing variables across 33 features was 1.94 per patient as shown in FIG.7C.Outcomes: real-world cohorts

[0194] Overall survival was calculated from the first ICI infusion to death from any cause, with patients alive at the time of review censored at their last contact. For patients who received multiple ICI treatments, the start date of the first treatment was used in the analysis. In the MSK-I cohort, both clinical benefit and overall survival data were available for all patients. In the MSK.-II cohort, overall survival data were available for all patients, but only 934 patients (44.39%) had clinical benefit data. For the MSK non-ICI and MSHS cohorts, only overall survival data were available. The primary clinical outcomes were clinical benefit to ICI and overall survival after ICI. Clinical benefit was classified based on RECIST vl.l. If formal RECIST reads were unavailable, the physician notes and imaging studies were manually reviewed by physician investigators to categorize the overall best response for each patient using the same criteria based on the change in the sum of diameters of target lesions. CR, PR, and SD>6m were classified as clinical benefit whereas SD<6m and PD were classified as no clinical benefit. The rationale for using clinical benefit as a treatment effect outcome is derived 67180755440.1084284.00347from systematic reviews in the cancer immunotherapy context, which indicate that patients with SD>6m have more similar overall survival outcomes to patients with tumor response categorized as minor PR, in contrast to patients with SD<6m, who have overall survival outcomes more similar to patients experiencing PD.Cohort description: clinical trial cohorts

[0195] Performed were ten phase 3 clinical trials for a further external testing: IMbravel 50 (n = 279), IMspirel50 (n = 256), IMmotionl51 (n = 445), IMvigor211 (n = 444), IMpowerl33 (n = 197), IMpowerl30 (n = 467), IMpowerl31 (n = 680), IMpowerl32 (n = 288), IMpowerl50 (n = 793), and OAK (n = 595) (Supplementary Table 1) and FIGs. 9A-9J. There were six different cancer types across the 10 cohorts: hepatocellular carcinoma (IMbravel50), BRAFV600E-positive melanoma (IMspirelSO), RCC (IMmotionlSl), bladder cancer (IMvigor211), SCLC (IMpowerl33), and NSCLC (IMpowerl30, IMpowerl31, IMpowerl32, IMpowerl50, and OAK).

[0196] For eight clinical trials, at least one additional drug was treated in addition to atezolizumab (anti-programmed death ligand 1 [PD-L1]): 1) atezolizumab plus bevacizumab for IMbravel50, 2) atezolizumab plus vemurafenib and cobimetinib for IMspirel50, 3) atezolizumab plus bevacizumab for IMmotionl51, 4) atezolizumab plus carboplatin and etoposide for IMpowerl33, 5) atezolizumab plus carboplatin and nanoparticle albumin-bound paclitaxel (nab-paclitaxel) for IMpowerl30, 6) atezolizumab plus carboplatin and nab-paclitaxel or paclitaxel (ACNP or ACP) for IMpowerl31. 7) atezolizumab plus pemetrexed and carboplatin or cisplatin for IMpowerl32, and 8) atezolizumab plus carboplatin and paclitaxel with or without bevacizumab (ABCP or ACP) for IMpowerl50. For two clinical trials, only atezolizumab was administrated: IMvigor211 and OAK. All analyses were performed based on the intention-to-treat principle. Therefore, 12 experimental arms were subjected to the external testing analysis.Clinical features: clinical trial cohorts

[0197] All patients with baseline laboratory test results were analyzed (results with “Y” flag in the LBBLFL column from the laboratory test file shared by Roche). For the clinical trial data, missing values were imputed using the MissForest from the missingpy package with default parameters after combining each cohort with the training set into a single data frame. Among the 33 features used in SCORPIO, eGFR, MCHC, and RDW were unavailable for all clinical trial cohorts. Smoking history was unavailable for the IMspireI50 and IMmotionl51. In addition, PROT was unavailable for the IMspirel50. The average number of missing values 68180755440.1084284.00347across the 33 features per patient in each clinical trial was as follows: 3.50 for IMbravel50, 5.30 for IMspirelSO. 4.84 for IMmotionl51, 3.48 for IMvigor211, 3.54 for IMpowerl33, 3.45 for IMpowerl30, 3.31 forIMpowerl31, 3.44 forIMpowerl32, 3.28 for IMpowerl50, and 3.35 for OAK.

[0198] In the clinical trial cohorts, PD-L1 immunostaining results using the SP142 or SP263 clones (Ventana Medical Systems, Tucson. AZ) were available. The SP263 clone data was available for IMbravel50 and IMpowerl33, and the rest of the clinical trials had the SP142 clone data available. The raw PD-L1 immunostaining values from the immune cell (IC) or tumor cell (TC) were available for IMbravel50, IMmotionI51 (only raw IC value was available), IMpowerl33, IMpowerl30, IMpowerl31, IMpowerl32, IMpowerl50, and OAK. The raw PD-L1 staining values for IC and TC were unavailable for IMspirel50 and IMvigor211, but categorical group information based on the PD-L1 staining levels was available (ICO / 1 / 2 / 3 and TCO / 1 / 2 / 3). To categorize patients based on the PD-L1 expression level, the FDA-approved cutoffs were applied on the clinical trials with NSCLC (IMpowerl 30, IMpowerl31, IMpowerl32, IMpowerl50, and OAK; PD-L1 expression in > 50% TC or> 10% IC [PD-Ll-High group] and < 50% TC and < 10% IC [PD-Ll-Low group] and BLCA (IMvigor211; PD-L1 expression in > 5% IC [PD-Ll-High group] and < 5% IC [PD-Ll-Low group]). In cancer ty pes without the FDA-approved cutoff, the same criteria from the original publications were used: IMbravel50 (PD-L1 expression in> 1% TC or> 1% IC [PD-Ll-High group] and < 1% TC and < 1% IC [PD-Ll-Low group]), IMspirel50 (PD-L1 expression in > 1% IC [PD-Ll-High group] and < 1% IC [PD-Ll-Low group]), IMmotionl51 (PD-L1 expression in > 1% IC [PD-Ll-High group] and < 1% IC [PD-Ll-Low group]), and IMpowerl33 (PD-L1 expression in > 5% TC or > 5% IC [PD-Ll-High group], > 1% TC or > 1% IC [PD-Ll-Mid group], and < 1% TC and < 1% IC [PD-Ll-Low group]).Outcomes: clinical trial cohorts

[0199] Overall survival was defined as the time from randomization to death from any cause. Patients alive at the time of the last follow-up were censored. CR, PR, and SD>6m were classified as clinical benefit whereas SD<6m and PD were classified as no clinical benefit. In the clinical trial protocols, patients who did not have post-baseline imaging for RECIST vl.l evaluation (data missing, not available [NA], or not evaluated [NE]) were classified as nonresponders, and, therefore in this analysis were categorized in the no clinical benefit group.Machine learning model constructionFeature selection analysis69180755440.1084284.00347

[0200] In the MSK-I cohort, 64.47% of the patients (n = 1,312) received systemic therapy as a first-line treatment prior to ICI. Since medications used for systemic therapy can influence the measurement of blood cell counts, metabolic compositions, or BML the impact of the systemic therapy history was investigated first. Using the training set, it was first tested if there was any bias in the collected data toward the systemic therapy history (PreChemo). Of the total of 47 features with a missing value of less than 30% across patients in the MSK-I cohort, 7 features including Age, Sex, DuringChemo, Virus, DrugClass, Smoking, and Stage were excluded from this analysis as they were not affected by the systemic therapy history (Table 4). Therefore, the association of 40 features (15 CMP, 23 CBC, and two clinical features) was tested with PreChemo, the methods of which are illustrated in FIGs. 9A-9J. It was found that 75.00% (30 out of 40) of the features exhibited significantly different values with respect to PreChemo in the training set as illustrated in FIG. 27. Analysis performed on the training set comprises supplementary information comprising ALB: albumin; ALK: alkaline phosphatase; ALT: alanine aminotransferase; AGAP: anion gap; AST: aspartate aminotransferase; BUN: blood urea nitrogen; CA: calcium; CL: chloride; CO2: carbon dioxide; CREAT: creatinine; eGFR: estimated glomerular filtration rate; GLU: glucose; HCT: hematocrit; HGB: hemoglobin; MCH: mean corpuscular hemoglobin; MCHC; mean corpuscular hemoglobin concentration; MCV: mean corpuscular volume; PLT: platelet; K: potassium; RBC: red blood cell; RDW: red blood cell distribution width; BILI: total bilirubin; PROT: total protein; white blood cell count; BASO%: basophil proportion among WBCs; EOS%: eosinophil proportion among WBCs; LYM%: lymphocyte proportion among WBCs; M0N0%: monocyte proportion among WBCs; NEUT%: neutrophil proportion among WBCs; BASO: basophil count; EOS: eosinophil count; LYM: lymphocyte count; MONO: monocyte count; NEUT: neutrophil count; BLR: basophil-to-lymphocyte ratio; ELR: eosinophil-to-lymphocyte ratio; MLR: monocyte-to-lymphocyte ratio; NLR: neutrophil-to-lymphocyte ratio; BMI: body mass index; ECOG-PS: Eastern Cooperative Oncology Group performance status. Therefore, multivariable analyses adjusting for PreChemo were performed when selecting the features associated with the two target variables (overall survival and clinical benefit).

[0201] Feature selection analyses were performed on the training set. A Cochran-Mantel-Haenszel test was used to find the association between features and clinical benefit (clinical benefit and no clinical benefit). Before applying the Cochran-Mantel-Haenszel test, each continuous feature w as dichotomized based on a cutoff by its median value from the training set. A Cox proportional-hazard regression was used to identify the variables associated with 70180755440.1084284.00347overall survival. The continuous values were directly subjected to the analysis in a Cox regression. Both tests were conducted, adjusting for PreChemo as a confounding factor. Features with significant false discovery rate (FDR) adjusted P-values (< 0.05) for each corresponding outcome were selected. The FDR method was applied separately to the P-values from the Cochran-Mantel-Haenszel test and Cox proportional -hazard regression test. As a result, 22 variables were identified as significantly associated with clinical benefit (Table 5), and 33 variables showed significant associations with overall survival as shown in FIGs. 10A- 10BModel construction

[0202] When constructing SCORPIO-CB first three different classifiers were trained, including one classic classifier (ridge logistic regression [RLR]) and two machine learning classifiers generally performed best among 179 classifiers (support vector machine [SVM], and random forest [RF]), using the 22 variables from the feature selection analysis as shown in FIG. 10B. Employed were three classifiers using the scikit-leam (v.1.2.2) package. The target outcome, clinical benefit, was coded as a dummy variable: 0 and 1 for no clinical benefit (SD<6m and PD) and clinical benefit (CR, PR, and SD>6m), respectively. For a hyperparameter tuning with a five-fold CV, the GridSearchCV function was used for the SVM and the RF, whereas the LogisticRegressionCV function w as applied for the RLR. The optimal hyperparameters for each algorithm were selected based on the highest average AUC value across folds (Table 6).

[0203] For SCORPIO, one classic survival model (ridge cox regression [RCOX]), and two survival models corresponding to SVM and RF (fast survival SVM [FSSVM], and random survival forest [RSF]), were trained by a hyperparameter tuning with a five-fold CV using the selected 33 variables as shown in FIG. 10A. The scikit-survival (v.0.20.0) package was used for the three survival models. The target outcome, overall survival, was coded as two fields with the survival time (in months) and the survival status (0 and 1 for censored and deceased, respectively). For hyperparameter tuning, a custom script performed grid search analysis with five-fold CV. The best hyperparameters were selected based on the highest average C-index value across folds. C-index was calculated by the concordance_index_censored function from the scikit-survival package (Table 6).

[0204] F or the RCox and the RLR, the feature values w ere normalized using Z-status method before running the algorithms. The original feature value (x_(i,j )) of a feature (feature i) in the jth sample was transformed (Z _(i J)) as follows:71180755440.1084284.00347

[0205] where p i and o_i denote the average value and the standard deviation value of a given feature (feature i) across the samples in the training set, respectively. For the test sets, the standard deviation and the average values from the training set were used.

[0206] To generate unweighted ensemble models, the risk statuses generated by the three algorithms as a previous study were averaged. For SCORPIO, first was applied a min-max normalization before averaging the values, because the three survival models resulted in different scales of risk status: RCox (from -1.14 to 2.05 in the training set), FSSVM (from -2.83 to -1.07 in the training set), and RSF (from 103.06 to 1628.80 in the training set). The scaled nsk status ( Krisk_score'3 J) in the jth sample was calculated from the original nsk status ( Erisk_scorel J) as follows:<where mintrainand maxtrainrepresent the minimum risk status and the maximum risk status across the samples in the training set, respectively. This approach transforms the raw risk status of each survival model into a value from 0 to 1 in the training set.

[0207] In the test sets, a risk_score’7- was calculated employing the minimum value (mintrain) and the maximum value (maxtrain) from the training set. In each survival model, this step sometimes results in greater than 1 of risk_score'j for patients predicted to show extremely poor response to ICI whose risk_scorej is greater than maxtrain. On the other hand, each survival model sometimes results in less than 0 of risk_score’j for patients predicted to show extremely good response to ICI whose risk_scorej is less than mintrain. For these patients, risk_score' was converted as 1 or 0 when they had a value greater than 1 or less than 0, respectively. After applying the aforementioned normalization step to each survival models, risk_score'j from three survival models was averaged to generate the unweighted risk status in SCORPIO.

[0208] In contrast to SCORPIO, an average of the predicted status from the three classifiers for SCORPIO-CB was directly calculated because of the classifiers used to generate the output with the same scale of the predicted probability (from 0 to 1 ).

[0209] All analyses related to the model construction were conducted using Python 3.8.8.72180755440.1084284.00347Patient stratification and outcome comparison

[0210] For the primary analysis of predicting clinical outcomes, patients in the test sets were stratified into three risk groups. Patients were stratified using the risk statuses according to the first quartile (0.24) and third quartile (0.47) of the risk statuses observed in the training set: high-risk group (risk status > 0.47), moderate-risk group (0.24 < risk status < 0.47), and low-risk group (risk status < 0.24). The same cutoff values were used in all cohorts regardless of data source and cancer type. To compare clinical benefit rates by the risk group, a Fisher’s exact test was performed. To compare overall survival by the risk group, Cox proportional hazards regression and log-rank test were used. A two-sided P < 0.05 was considered statistically significant. Kaplan-Meier plots, log-rank test P-values, and Cox proportional hazard ratios were generated by the survminer package (v.0.4.9). For the real-world cohorts, BLCA, hepatobiliary cancer, melanoma, NSCLC, RCC, and SCLC were analyzed separately since they were collected in all cohorts. The rest of the cancer types were grouped as ‘Others’ in each cohort and subsequently analy zed. In the MSK-II cohort, additionally performed were analyses on individual cancer types within the ‘Others’ group by taking advantage of the large sample size. All the statistical tests were performed with R programming language version 4.1.1 (https: / / www.r-project.org / ).Comparing the prognostic performance of the machine learning models with TMB, and PD-L1

[0211] In the hold-out test set, the best model between SCORPIO and SCORPIO-CB was selected for subsequent analyses on the MSK-II, clinical trials, and MSHS cohorts. For this, calculated were AUC values to measure the performance for clinical benefit classification and AUC(t) values to measure the performance for predicting overall survival. The receiver operating characteristic (ROC) curves and calculated AUC values using the precrec package (v.0.14.4). AUC(t) values were calculated using the timeROC package (v.0.4).

[0212] In the MSKCC cohorts, the predictive power of TMB was also evaluated along with the two machine learning models. In the clinical trial cohorts, the predictive power of PD-L1 staining was also evaluated when raw immunostaining values from the IC or TC were available.

[0213] All analyses regarding the AUC and AUC(t) were performed with continuous values.Comparing the performance of SCORPIO and other machine learning models

[0214] SCORPIO'S performance w as compared with previously developed machine learning models for predicting ICI efficacy in patients with NSCLC. The study by Vanguri et al. included 26 models across nine data categories: clinical, radiology, pathology, genomics,73180755440.1084284.00347dynamic deep atention-based multiple-instance learning model with masking (DyAM) unimodal, DyAM bimodal, DyAM multi-modal (automated), DyAM multi-modal (with PD-L1 tumor proportion status), and multi-modal average. Each category's best-performing model was subjected to the analysis. For a fair comparison with SCORPIO, data was utilized from 150 out of 237 patients from Vanguri et al., ensuring model statuses from all nine models were available. Model statuses were obtained from "Source Data Extended Data FIG. 9 of Vanguri et al.'s publication. Re-evaluated overall survival and clinical benefit were re-revaluated using RECIST vl .1 criteria, which is consistent with the study. Three patients with concurrent cancer diagnoses were included in analyses of index tumor response but not included in analyses for survival outcomes.

[0215] For visualization of ROC curves and calculation of AUC values, the precrec package was used. AUC(t) values were calculated using the timeROC package. All AUC and AUC(t) analyses were conducted with continuous values.Model InterpretabilityGlobal model explanation

[0216] In the training set, the SHAP method39 (v.0.44.1) was applied to examine the magnitude of the relative importance and the direction of the impact of each feature in SCORPIO as shown in FIG. 10A. Two different explainer functions were applied in this study: i) the Explainer function for the RCox and the RSF, and the KemelExplainer function for the FSSVM, To demonstrate the relative importance and the direction of impact of each variable in the ensemble models, the aggregated SHAP values were generated across three algorithms which form SCORPIO (RCox, FSSVM, and RSF) as shown in FIG.28A. Summary plots were produced for the distribution of feature impact on the predicted risk status from SCORPIO. These analyses were performed on the training set. Feature impact was measured by the SHAP approach. Each patient of the training set is illustrated as a data point per feature. The color of each point indicates the feature value of a given patient. Positive SHAP value represents that the feature raised the risk status in a given patient (positive impact), whereas negative SHAP value represents that the feature lowered the risk status in a given patient (negative impact). Features are ordered according to their relative feature importance by averaging the absolute SHAP values across patients in each feature. CL: chloride; ALB: albumin; ECOG-PS: Eastern Cooperative Oncology Group performance status; Smoking: smoking hi story ; AGAP: anion gap; RBC: red blood cell; EOS%: eosinophil proportion among white blood cells (WBCs); PROT: total protein; CREAT: creatinine; LYM% lymphocyte proportion among WBCs; LYM:74180755440.1084284.00347lymphocyte count; BMI: body mass index; NEUT: neutrophil count; Age: age at ICI; NEUT% neutrophil proportion among WBCs; Stage: tumor stage at ICE HCT: hematocrit; GLU: glucose; ALT: alanine aminotransferase; HGB: hemoglobin; MONO: monocyte count; AST: aspartate aminotransferase; eGFR: estimated glomerular filtration rate; BASO%: basophil proportion among WBCs; RDW: red blood cell distribution width; BLR: basophil-to-lymphocyte ratio; MLR: monocyte-to-lymphocyte ratio; PLT: platelet; NLR: neutrophil-to-lymphocyte ratio; ALK: alkaline phosphatase; MCHC: mean corpuscular hemoglobin concentration; BILL total bilirubin. Since the above three survival models have different scales of risk status and SHAP value, the SHAP values were normalized before generating the aggregative ones to avoid a biased result.

[0217] First, the mean of the |SHAP| value of all samples across the training set. which displays the average impact of a feature (feature i) on model output, ((|SHAP|) _(i, train)), was scaled by amin-max normalization as follows (Step 1):where mintrairiand maxtraindenotes the minimum \SHAP\trainvalue and the maximum \SHAP\trainvalue across 33 features in the training set, respectively. As a result, the average impact of a feature on model output (\SHAP\i'train) gets transformed into a decimal from 0 to 1 \SHAP\'iitrain).

[0218] Second, the min-max scaled values (\SHAP\'i train) were transformed into negative values when variables had negative direction of impact on the predicted risk status. The direction of impact was measured by the Spearman’s correlation coefficient between the original feature values and the SHAP values using all samples in the training set (Step 2). When the original feature values and the SHAP values in a given feature (feature i) had a negative correlation coefficient, this feature thus has a negative impact on the predicted risk status. The aforementioned two steps were individually applied to each survival algorithm as shown in FIG. 28B. Each feature importance bar indicates the magnitude of feature importance which was generated by averaging the absolute SHAP values across patients in each feature. The direction of impact was measured by the Spearman’s correlation coefficient between the feature values and the SHAP values using all samples in each feature Third, the average of75180755440.1084284.00347+ \SHAP\'i trairifrom the three survival algorithms was calculated as the aggregated SHAP values for SCORPIO as shown in FIG. 10A.Local model explanation

[0219] To generate the aggregated SHAP values for SCORPIO at a patient level as shown in, a similar approach was taken as the global level analysis. The min-max scaled SHAP (\SHAP\') values were first calculated for a feature (feature i) in the jthpatient as follows (Step 1):where minj and maxj denote the minimum \SHAP | value and the maximum |S7MP| value across the 33 features in the jthpatient, respectively.

[0220] The patient level SHAP values already provide the directions of effect as negative or positive values with their raw values. Hence, calculating Spearman’s correlation coefficient between the original feature values and the SHAP values which was performed in the global model explanation was not required in the local model explanation. Instead, transformed was the min-max normalized SHAP value of a feature (feature i), \SHAPinto the negative value when its original SHAPt j was negative (Step 2). These approaches were individually applied to the three different survival models. After the aforementioned two steps, the average of + \SHAP\'i j from the three survival models for the aggregated SHAP values at a patient level was calculated. As illustrated in FIG. 29, SHAP values for various local models trained and customized for various tissue ty pes were obtained to represent impact of various features on risk score. The impact was illustrated as a negative impact or positive impact on a gradient scale as shown in FIG. 29. The SHAP values were obtained for the global SCORPIO model, a bladder-specific model, a colorectal-specific model, an esophageal-specific model, a hepatobiliary-specific model, an H&N specific model, a melanoma-specific model, an NSCLC-specific model, an ovarian-specific model, a renal-specific model, and an SCLC-specific model as illustrated in FIG. 29. For each model, the SHAP values of each feature, including features comprising biomarkers measured from blood samples, were measured and evaluated for feature impact on risk score.Bulk RNA-seq analysis180755440.1084284.00347

[0221] RNA was isolated from formalin-fixed paraffin-embedded (FFPE) tumor samples of NSCLC. Bulk RNA-seq was performed using the Tempus xT RNA-seq protocol69, which involves exome capture with IDT xGen probes covering over 19,000 genes, requiring at least 50 ng of RNA for library construction. Sequencing was done to a minimum depth of 30 million reads on aNovaSeq 6000. Transcript abundances in transcripts per million (TPM) values were derived using Kallisto70 (v.0.44.0) pseudoalignments to Ensembl GRCh37 (Release 75). Gene-level TPM values were obtained by summing the transcript-level TPM of 20,061 genes with at least one annotated protein-coding transcript covered by the assay; then, the values were log2(TPM + l)-transformed. Batch correction was applied for samples sequenced with different probe designs by limma71 (v.3.54.2).

[0222] For the H&N cancer samples, bulk RNA-seq reads were aligned against the hgl9 reference genome by STAR (v.2.5.3a) 2-pass alignment. Raw read counts were computed using the R package GenomicAlignments (v.1.14.2) over aligned reads with UCSC KnownGene in hgl9 as the base gene model. The Union counting mode was used and only mapped paired reads after alignment quality filtering were considered. Finally, fragments per kilobase of transcript per million mapped reads (FPKM) values were computed by the R package DESeq2 (v.1.18.1).

[0223] The Danaher signature was employed to deconvolute 14 immune cell compositions from the two cohorts with bulk RNA-seq. For the NSCLC and H&N cancer cohorts, limma-corrected log2(TPM + 1) and log2(FPKM + 1) values were used, respectively. The expression values of marker genes for each cell type were averaged following the methodology described in the original paper.77180755440.1

Claims

084284.00347CLAIMSWhat is claimed is:

1. A method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, comprising:(a) obtaining a set of individual characteristic variables of said subject;(b) assaying a blood sample obtained or derived from said subj ect to obtain a set of laboratory measurements therefrom;(c) computer processing at least said set of individual characteristic variables and said set of laboratory measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (d) determining, based at least in part on said computer processing in (c), a predicted clinical outcome of said subject upon receiving said ICI.

2. The method of claim 1, wherein said cancer is selected from the group consisting of bladder cancer (BLCA), hepatobiliary cancer, melanoma, non-small cell lung cancer (NSCLC), renal cell carcinoma (RCC), small cell lung cancer, and a combination thereof.

3. The method of claim 1, wherein said predicted clinical outcome comprises a clinical benefit of said ICI or an overall survival of said subject.

4. The method of claim 3, wherein said predicted clinical outcome comprises said clinical benefit of said ICI.

5. The method of claim 4, wherein said clinical benefit of said ICI comprises a complete response to said ICI, a partial response to said ICI, a stable disease without progression for at least six months after initially receiving said ICI, a progression of said cancer, or a regression of said cancer.

6. The method of claim 4, further comprising determining a likelihood of said clinical benefit of said ICI.

7. The method of claim 6, wherein said likelihood comprises a probability of said clinical benefit of said ICI.

8. The method of claim 3, wherein said predicted clinical outcome comprises said overall survival of said subject.

9. The method of claim 8, wherein said overall survival of said subject comprises at least about 1 month, 2 months, 3 months, 4 months, 5 months, 6 months, 7 months, 8 months, 9 months, 10 months, 11 months, 12 months, 2 years, 3 years, 4 years, 5 years, or more than 5 years.78180755440.1084284.0034710. The method of claim 8, further comprising determining a risk status indicative of a likelihood of poor outcome.

11. The method of claim 1, wherein said trained machine learning algorithm comprises a member selected from the group consisting of a logistic regression, a Cox regression, a support vector machine, a random forest, and a combination thereof.

12. The method of claim 11, wherein said logistic regression comprises a ridge logistic regression.

13. The method of claim 11 , wherein said Cox regression comprises a ridge Cox regression.

14. The method of claim 11 , wherein said support vector machine comprises a fast survival support vector machine.

15. The method of claim 11, wherein said random forest comprises a random survival forest.

16. The method of claim 1, wherein (d) further comprises determining said predicted clinical outcome with an area under receiver operating characteristic curve (AUC) of at least about 0.

50. 0.55, 0.60, 0.65, 0.70, 0.75, 0.80, 0.85, 0.90, or 0.95.

17. The method of claim 1, wherein (d) further comprises determining said predicted clinical outcome with an accuracy of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.

18. The method of claim 1, wherein (d) further comprises determining said predicted clinical outcome with a sensitivity of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.

19. The method of claim 1, wherein (d) further comprises determining said predicted clinical outcome with a specificity of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%. or 95%.

20. The method of claim 1, wherein (d) further comprises determining said predicted clinical outcome with a positive predictive value of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.

21. The method of claim 1, wherein (d) further comprises determining said predicted clinical outcome with a negative predictive value of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.

22. The method of claim 1 , further comprising administering said ICI to said subj ect, based at least in part on said predicted clinical outcome determined in (d).79180755440.1084284.0034723. The method of claim 1, further comprising selecting said subject to not receive said ICI and administering an alternative therapy to said subject, based at least in part on said predicted clinical outcome determined in (d).

24. The method of claim 1, wherein said ICI comprises a first-line therapy, a second-line therapy, or a third-line therapy.

25. The method of claim 1, wherein said ICI is selected from the group consisting of an anti-cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) agent, an anti-programmed death 1 (PD-l) / programmed death ligand 1 (PD-L1) agent, and a combination thereof.

26. The method of claim 1 , wherein said set of individual characteristic variables comprises a member selected from the group consisting of: a demographic characteristic, a clinical characteristic, a risk group stratification of said subject, and a combination thereof.

27. The method of claim 26, wherein said demographic characteristic comprises age or sex of said subject.

28. The method of claim 26, wherein said clinical characteristic is selected from the group consisting of: body mass index (BMI), drug class (DrugClass), chemotherapy during immunotherapy (DunngChemo), systemic therapy history (PreChemo), Eastern Cooperative Oncology Group performance status (ECOG-PS), smoking history (Smoking), tumor stage (Stage), viral infection (Virus), and a combination thereof.

29. The method of claim 26, wherein said risk group stratification comprises a low-risk group, a moderate-risk group, or a high-risk group.

30. The method of claim 1 , wherein said blood sample comprises a whole blood sample, a serum sample, or a plasma sample.

31. The method of claim 1, wherein said set of laboratory measurements comprises a member selected from the group consisting of: a comprehensive metabolic panel (CMP) measurement, a complete blood count (CBC) measurement, a coagulation panel measurement, conjugated bilirubin (CB), direct bilirubin (DB), glucose-6-phosphate dehydrogenase (G6PD), ionized calcium (iCA), lactate dehydrogenase (LDH), lipase (LPS), and a combination thereof.

32. The method of claim 31, wherein said CMP measurement is selected from the group consisting of: albumin (ALB), alkaline phosphatase (ALK), alanine aminotransferase (ALT), anion gap (AGAP), aspartate aminotransferase (AST), blood urea nitrogen (BUN), calcium (CA), chloride (CL), carbon dioxide (CO2), creatine (CREAT), estimated glomerular fdtration rate (eGFR), glucose (GLU), potassium (K), bilirubin (BILI), total protein (PROT), magnesium (MG), phosphorus (P). and a combination thereof.80180755440.1084284.0034733. The method of claim 31, wherein said CBC measurement is selected from the group consisting of: white blood cell count (WBC), basophil count (BASO). eosinophil count (EOS), granulocytes count (GRAN), lymphocyte count (LYM), monocyte count (MONO), neutrophil count (NEUT), basophil proportion among WBC (BASO%), eosinophil proportion among WBC (EOS%), granulocytes proportion among WBC (GRAN%), lymphocyte proportion among WBC (LYM%), monocyte proportion among WBC (MONO%), neutrophil proportion among WBC (NEUT%), hematocrit (EICT), hemoglobin count (HGB), mean corpuscular hemoglobin concentration (MCHC), mean corpuscular hemoglobin (MCEI), mean corpuscular volume (MCV), platelet count (PLT), red blood cell count (RBC), red blood cell distribution width (RDW), basophil-to-lymphocyte rate (BLR), eosinophil-to-lymphocyte ratio (ELR), monocyte-to-lymphocyte ratio (MLR), and neutrophil-to-lymphocyte ratio (NLR). and a combination thereof.

34. The method of claim 31, wherein said coagulation panel measurement is selected from the group consisting of: activated partial thromboplastin time (APTT), international normalized ratio (INR), prothrombin time (PT), and a combination thereof.

35. A non-transitory computer-readable medium comprising machine-executable code that, upon execution by one or more computer processors, implements a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, said method comprising:(a) obtaining a set of individual characteristic variables of said subject;(b) assaying a blood sample obtained or derived from said subject to obtain a set of laboratory measurements therefrom;(c) computer processing at least said set of individual characteristic variables and said set of laboratory measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (d) determining, based at least in part on said computer processing in (c), a predicted clinical outcome of said subject upon receiving said ICI.

36. A computer system for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, comprising:(a) a database that is configured to store a set of individual characteristic variables of said subject; and(b) one or more computer processors operatively coupled to said database, wherein said one or more computer processors are individually or collectively programmed to:81180755440.1084284.00347(i) assay a blood sample obtained or derived from said subject to obtain a set of laboratory measurements therefrom;(ii) process at least said set of individual characteristic variables and said set of laboratory measurements (i) against a reference set of individual characteristic variables and laboratory measurements or (ii) with a trained machine learning algorithm; and (iii) determine, based at least in part on said computer processing in (ii), a predicted clinical outcome of said subject upon receiving said ICI.

37. A method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, comprising:(a) assaying a blood sample obtained or derived from said subj ect to obtain a set of laboratory measurements therefrom;(b) computer processing at least said set of laboratory measurements against (i) a reference set of laboratory measurements or (ii) with a trained machine learning algorithm, or both (i) and (ii); and(c) determining, based at least in part on said computer processing in (b), a predicted clinical outcome of said subject upon receiving said ICI.

38. The method of claim 37, wherein the method does not comprise obtaining a set of individual characteristic variables of said subject.

39. The method of claim 37, wherein said cancer is selected from the group consisting of bladder cancer (BLCA), hepatobiliary cancer, melanoma, non-small cell lung cancer (NSCLC), renal cell carcinoma (RCC), small cell lung cancer, and a combination thereof.

40. The method of claim 37, wherein said predicted clinical outcome comprises a clinical benefit of said ICI or an overall survival of said subject.

41. The method of claim 40. wherein said predicted clinical outcome comprises said clinical benefit of said ICI.

42. The method of claim 41, w herein said clinical benefit of said ICI comprises a complete response to said ICI, a partial response to said ICI, a stable disease without progression for at least six months after initially receiving said ICI, a progression of said cancer, or a regression of said cancer.

43. The method of claim 41, further comprising determining a likelihood of said clinical benefit of said ICI.

44. The method of claim 43, wherein said likelihood comprises a probability of said clinical benefit of said ICI.82180755440.1084284.0034745. The method of claim 37, wherein said predicted clinical outcome comprises said overall survival of said subject.

46. The method of claim 45, wherein said overall survival of said subject comprises at least about 1 month, 2 months, 3 months, 4 months, 5 months, 6 months, 7 months, 8 months, 9 months, 10 months, 11 months, 12 months, 2 years, 3 years, 4 years, 5 years, or more than 5 years.

47. The method of claim 41, further comprising determining a risk status indicative of a likelihood of poor outcome.

48. The method of claim 37, wherein said trained machine learning algorithm comprises a member selected from the group consisting of a logistic regression, a Cox regression, a support vector machine, a random forest, and a combination thereof.

49. The method of claim 48, wherein said logistic regression comprises a ridge logistic regression.

50. The method of claim 48, wherein said Cox regression comprises a ridge Cox regression.

51. The method of claim 48, wherein said support vector machine comprises a fast survival support vector machine.

52. The method of claim 48, wherein said random forest comprises a random survival forest.

53. The method of claim 37, wherein (d) further comprises determining said predicted clinical outcome with an area under receiver operating characteristic curve (AUC) of at least about 0.50, 0.55, 0.60, 0.65, 0.70, 0.75, 0.80, 0.85, 0.90, or 0.95.

54. The method of claim 37, wherein (d) further comprises determining said predicted clinical outcome with an accuracy of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%. or 95%.

55. The method of claim 37, wherein (d) further comprises determining said predicted clinical outcome with a sensitivity of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.

56. The method of claim 37, wherein (d) further comprises determining said predicted clinical outcome with a specificity of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.

57. The method of claim 37, wherein (d) further comprises determining said predicted clinical outcome with a positive predictive value of at least about 50%, 55%, 60%, 65%, 70%, 75%. 80%. 85%. 90%. or 95%.83180755440.1084284.0034758. The method of claim 37, wherein (d) further comprises determining said predicted clinical outcome with a negative predictive value of at least about 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%.

59. The method of claim 37, further comprising administering said ICI to said subject, based at least in part on said predicted clinical outcome determined in (d).

60. The method of claim 37. further comprising selecting said subject to not receive said ICI and administering an alternative therapy to said subject, based at least in part on said predicted clinical outcome determined in (d).

61. The method of claim 37, wherein said ICI comprises a first-line therapy, a second-line therapy, or a third-line therapy.

62. The method of claim 37. wherein said ICI is selected from the group consisting of an anti-cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) agent, an anti -programmed death 1 (PD-l) / programmed death ligand 1 (PD-L1) agent, and a combination thereof.

63. The method of claim 37, wherein said blood sample comprises a whole blood sample, a serum sample, or a plasma sample.

64. The method of claim 37, wherein said set of laboratory measurements comprises a member selected from the group consisting of: a comprehensive metabolic panel (CMP) measurement, a complete blood count (CBC) measurement, a coagulation panel measurement, conjugated bilirubin (CB), direct bilirubin (DB). glucose-6-phosphate dehydrogenase (G6PD), ionized calcium (iCA). lactate dehydrogenase (LDH), lipase (LPS), and a combination thereof.

65. The method of claim 64, wherein said CMP measurement is selected from the group consisting of: albumin (ALB), alkaline phosphatase (ALK), alanine aminotransferase (ALT), anion gap (AGAP), aspartate aminotransferase (AST), blood urea nitrogen (BUN), calcium (CA), chloride (CL), carbon dioxide (CO2), creatine (CREAT), estimated glomerular filtration rate (eGFR), glucose (GLU), potassium (K), bilirubin (BILI), total protein (PROT), magnesium (MG), phosphorus (P), and a combination thereof.

66. The method of claim 64, wherein said CBC measurement is selected from the group consisting of: white blood cell count (WBC), basophil count (BASO). eosinophil count (EOS), granulocytes count (GRAN), lymphocyte count (LYM). monocyte count (MONO), neutrophil count (NEUT), basophil proportion among WBC (BASO%), eosinophil proportion among WBC (EOS%), granulocytes proportion among WBC (GRAN%), lymphocyte proportion among WBC (LYM%), monocyte proportion among WBC (M0N0%), neutrophil proportion among WBC (NEUT%), hematocrit (HCT). hemoglobin count (HGB), mean corpuscular 84180755440.1084284.00347hemoglobin concentration (MCHC), mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV), platelet count (PLT), red blood cell count (RBC), red blood cell distribution width (RDW), basophil-to-lymphocyte rate (BLR), eosinophil-to-lymphocyte ratio (ELR), monocyte-to-lymphocyte ratio (MLR), and neutrophil-to-lymphocyte ratio (NLR), and a combination thereof.

67. The method of claim 64, wherein said coagulation panel measurement is selected from the group consisting of: activated partial thromboplastin time (APTT), international normalized ratio (INR), prothrombin time (PT), and a combination thereof.

68. A non-transitory computer-readable medium comprising machine-executable code that, upon execution by one or more computer processors, implements a method for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, said method comprising:(a) assaying a blood sample obtained or derived from said subject to obtain a set of laboratory7measurements therefrom;(b) computer processing at least said set of laboratory’ measurements (i) against a reference set of laboratory measurements or (ii) with a trained machine learning algorithm, or both (i) and (ii); and(c) determining, based at least in part on said computer processing in (c), a predicted clinical outcome of said subject upon receiving said ICI.

69. The non-transitory computer-readable medium of claim 68. wherein the method implemented by the machine-executable code upon execution by one or more processors does not comprise obtaining a set of individual characteristic variables of said subject.

70. A computer system for determining an efficacy of an immune checkpoint inhibitor (ICI) for a subject having cancer, comprising:one or more computer processors operatively coupled to said database, wherein said one or more computer processors are individually or collectively programmed to:(i) assay a blood sample obtained or derived from said subject to obtain a set of laboratory measurements therefrom;(ii) process at least said set of laboratory measurements (i) against a reference set of laboratory measurements or (ii) with a trained machine learning algorithm, or both (i) and (ii); and(iii) determine, based at least in part on said computer processing in (ii), a predicted clinical outcome of said subject upon receiving said ICI.85180755440.1084284.0034771. The computer system of claim 70. wherein the computer system does not comprise a database that is configured to store a set of individual characteristic variables of said subject.180755440.1

Images