Method for determining risk of death in cancer patients, method for evaluating anti-cancer treatment, and method for selecting cancer patients for treatment

Abstract

A method for determining mortality risk is disclosed. In one configuration, the method includes receiving patient data representing information about a cancer patient. A mathematical model of mortality risk is used to determine the mortality risk of the cancer patient based on the received patient data. The mathematical model models the relationship determined from training data between the mortality risk and values ​​of at least 16 model parameters.

Description

[Technical Field]

[0001] The present invention relates to a method for determining the risk of death in a cancer patient. The method finds application in situations where quantifying such risk is beneficial, such as in the evaluation of anti-cancer therapies and / or in patient management regimens, treatment plans or patient selection for clinical trials. [Background technology]

[0002] Factors that affect life expectancy are important in public health. In oncology, predicting patient survival helps guide optimal patient management. Understanding which variables are prognostic can provide insight into disease biology and improve the design, implementation, and data analysis of clinical trials and real-world data.

[0003] Current research on prognostic factors in oncology is largely based on relatively small sample sizes and studies one risk factor at a time. Examples of such studies include: - Banks, E. et al. Erectile dysfunction severity as a risk marker for cardiovascular disease hospitalisation and all-cause mortality: a prospective cohort study.PLoS Med.10,e1001372(2013). - Hu,FBet al.Adiposity as compared with physical activity in predicting mortality among women.N.Engl.J.Med.351,2694-2703(2004). - McGee,DL,Liao,Y.,Cao,G.&Cooper,RSSelf-reported health status and mortality in a multiethnic US cohort.Am.J.Epidemiol.149,41-46(1999). -Thun,MJet al.Alcohol consumption and mortality among middle-aged and elderly USadults.N.Engl.J.Med.337,1705-1714(1997). - Tota-Maharaj,R.et al.Coronary artery calcium for the prediction of mortality in young adults <45 years old and elderly adults >75 years old.Eur.Heart J.33,2955-2962(2012).

[0004] Existing commonly used prognostic scores are constructed from a relatively small number of risk factors, including: - Royal Marsden Hospital Score (RMHS) (Nieder, C. & Dalhaug, AA. A new prognostic score derived from phase I study participants with advanced solid tumors is also valid in patients with brain metastasis. Anticancer Res. 977-9 (2010)). - International Prognostic Index (IPI) (International Non-Hodgkin's Lymphoma Prognostic Factors Project. A predictive model for aggressive non-Hodgkin's lymphoma. N. Engl. J. Med. 329, 987-994 (1993)). - IMDC risk model (Ko, JJ et al. The International Metastatic Renal Cell Carcinoma Database Consortium model as a prognostic tool in patients with metastatic renal cell carcinoma previously treated with first-line targeted therapy: a population-based study. Lancet Oncol. 16, 293-300 (2015)). - Glasgow Prognostic Score (Kinoshita, A. et al. The Glasgow Prognostic Score, an inflammation-based prognostic score, predicts survival in patients with hepatocellular carcinoma. BMC Cancer 52 (2013) doi:10.1186 / 1471-2407-13-52). The latest UK Biobank initiative (Sudlow, C. et al. UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age. PLoS Med. 12, e1001779 (2015)) constitutes an important addition to data availability. This approach was used by Ganna and Ingelsson (Ganna, A. & Ingelsson, E. 5-year mortality predictors in 498 103 UK Biobank participants: a prospective population-based study. The Lancet 386, 533-540 (2015)) to examine life expectancy in a population-based sample of approximately 500,000 participants and construct a mortality risk score that surpasses the Charlson Comorbidity Index (Charlson, M.E., Pompei, P., Ales, K.L. & C. Ronald, M.A. A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. J Chronic Dis. 373-383 (1987) doi:10.1016 / 0021-9681(87)90171-8). Summary of the Invention

[0005] The object of the present invention is to improve the quantitative determination of mortality risk.

[0006] According to one aspect, there is provided a computer-implemented method for determining a risk of mortality for a cancer patient, the method including receiving patient data representing information about the cancer patient; determining a risk of mortality for the cancer patient based on the received patient data using a mathematical model of mortality risk; and outputting the determined risk of mortality, wherein the mathematical model is configured to calculate a risk of mortality and at least the following model parameters: (i) age; (ii) gender; (iii) hemoglobin or hematocrit levels in the blood; (iv) serum or plasma urea nitrogen levels (v) alkaline phosphatase enzyme activity level in serum or plasma; (vi) protein levels in serum or plasma; (vii) serum or plasma albumin levels; (viii) serum or plasma chloride or sodium levels; (ix) blood eosinophil-to-leukocyte ratio; (x) serum or plasma lactate dehydrogenase enzyme activity level; (xi) heart rate, (xii) systolic blood pressure, (xiii) Eastern Cooperative Oncology Group (ECOG) performance status; (xiv) blood neutrophil-to-lymphocyte ratio; (xv) the ratio of the serum or plasma aspartate aminotransferase enzyme activity level to the serum or plasma alanine aminotransferase enzyme activity level, and (xvi) TNM classification of tumor stage The model then models the relationship determined from the training data between the values ​​of

[0007] Thus, a method for determining and outputting a mortality risk is provided. The output may be provided in the form of a numerical score. The numerical score may be output as data or displayed on a display. The determination is performed using a model based on a specific selection of parameters that provides an optimized balance between high sensitivity and specificity on the one hand, and manageable computation and data collection requirements on the other hand. As the data contained herein demonstrate, a core selection of 16 model parameters was found to provide significantly higher sensitivity and specificity than observed to be possible with alternative prior art prognostic scores. While performance improvements can be achieved by including additional model parameters, the possible improvements are relatively small compared to the advantages already achieved compared to alternative prior art prognostic scores. Thus, the selection of 16 core model parameters is demonstrated to provide a good balance between sensitivity and specificity on the one hand, and manageable computation and data collection requirements on the other hand.

[0008] This model was developed by the inventors by performing prognostic modeling of overall survival (OS) on cohort data (122,694 patients) from the Flatiron Health database. The inventors validated their results in two independent clinical studies. They investigated demographic, clinical, hematological, and blood chemistry parameters (focusing on routinely collected data), cancer diagnosis, and real-world mortality as endpoints, and assessed survival time from first treatment.

[0009] The model parameters represent selected factors that reflect both tumor biology and patient characteristics, and the resulting score has been shown to have prognostic value that significantly exceeds contemporary risk scores. The score can be used to improve risk stratification, patient matching, and interpretation of clinical trial results in early and late-stage oncology drug development.

[0010] We found that the determined score correlated not only with OS but also with dropout in certain early-stage studies. Therefore, using the determined score to exclude very high-risk patients may help protect patients from unnecessary exposure to the procedural burden and potential adverse events of the study.

[0011] The inventors also found a surprising and significant increase in the determined score relative to death. During dose escalation in first-in-human studies, it is a reasonable hypothesis that the determined score increases under ineffective drug doses or treatments, but remains stable or even improves under effective treatments. Thus, group-level analysis of the time course of the determined score by treatment group or drug dose can provide valuable insight into potential drug efficacy. The increase in the determined score, in combination with mortality events, can be used as a surrogate endpoint in clinical trials. In one embodiment, the mathematical model correlates the risk of death with model parameters (i) through (xvi) and at least the following parameters: (xvii) smoking history, (xviii) number of metastatic sites; (xix) blood platelet levels; (xx) serum or plasma calcium levels; (xxi) glucose levels in the blood; (xxii) lymphocyte to leukocyte ratio in blood; (xxiii) serum or plasma bilirubin levels; (xxiv) levels of monocytes in the blood; (xxv) the level of oxygen saturation in arterial blood; and (xxvi) Body mass index value The relationship between the training data is modeled.

[0012] In one embodiment, the mathematical model models the relationship determined from training data between the risk of death and the model parameters (i)-(xxvi) and at least the following parameter: (xxvii) the value of the alanine aminotransferase enzyme activity level in serum or plasma.

[0013] This particular choice of 27 model parameters was found by the inventors to provide a particularly beneficial balance of performance, and we observed significantly improved separation of Kaplan-Meier survival curves compared to RMHS.

[0014] In one embodiment, the mathematical model comprises the mortality risk and model parameters (i)-(xxvii) and at least the following parameters: (xxviii) blood eosinophil levels, and (xxix) Diastolic blood pressure value The relationship between the training data is modeled.

[0015] We found that adding these two additional model parameters improved sensitivity and / or specificity without unduly increasing the computational and / or data collection burden.

[0016] The inventors have demonstrated that this approach is robust to missing data at a level of up to about 10 missing values ​​of a parameter. Missing values ​​can be replaced, for example, by the mean value of the parameter of interest obtained from the model's training data. In one embodiment, the patient data includes values ​​for at least 16 of the model parameters, where at least model parameters (i)-(xxvi) are used. When only model parameters (i)-(xvi) are used, the approach is typically robust to data inconsistencies at a level of up to about 3 or 4 missing values ​​of a parameter.

[0017] In one embodiment, the model comprises a weighted sum of deviations in the patient data from the mean values ​​of the model parameters in the training data, and the model is formed by determining the weights from the training data. Forming the model in this manner is computationally efficient and provides high performance. The trained model can be represented in a compact form and applied with modest computational resources.

[0018] Embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which: [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 1 is a flowchart illustrating a method framework for determining mortality risk according to an embodiment of the present disclosure. [Figure 2] Figure 1. Plot visualizing the model performance of mortality risk for increasing regularization strength. [Figure 3] Figure showing the significance of model parameters in terms of HR estimates and corresponding confidence intervals, where HRs for protective parameters are less than 1, indicating that higher levels of the parameter are beneficial, and HRs for harmful parameters are greater than 1, implying higher risk at higher variance values ​​(abbreviations are used - see below). [Figure 4] Kaplan-Meier curves of overall survival (OS) for high / low RMHS scores. [Figure 5] 10A and 10B show Kaplan-Meier curves of overall survival (OS) for the highest risk 10% model score group versus the remaining 90% model score group for scores determined using the methods of the present disclosure. [Figure 6] 10 is a Kaplan-Meier curve of overall survival (OS) for scores determined using the methods of the present disclosure ranging from low (decile 1) risk to high (decile 10) risk. [Figure 7] The ROC curves for 3-month survival in the phase I study and RMHS are shown, with an AUC of 72.3. [Figure 8]The ROC curves for 3-month survival from the phase I study and this model are shown, with an AUC of 84.1. [Figure 9] The ROC curves for 3-month survival in the phase III study and RMHS are shown, with an AUC of 64.2. [Figure 10] The ROC curves for 3-month survival from the phase III study and this model are shown, with an AUC of 81.7. [Figure 11] Diagram showing the results of longitudinal monitoring of model scores by response group in the OAK Phase III clinical trial and landmark analysis. The x-axis corresponds to different time points. The date of occurrence of the outcome event is the right-most point. On the left, the time course (100 days) before the event is shown. Only data from patients who were observable for at least 100 days contribute to the figure. The y-axis corresponds to the determined score (group mean). Each curve represents one of the three outcome groups (death, progression, response as a combination of complete and partial response). [Figure 12] Kaplan-Meier for landmark analysis under landmarks on day 145 (= 75% of patients). [Figure 13] Graph corresponding to that of Figure 3 except using a reduced model containing 16 selected model parameters (called ROPRO16). [Figure 14] 1 is a bar graph comparing performance between models of the present disclosure with different numbers of model parameters and prior art alternative prognostic scores. Two comparisons are shown: C-index (comparing general performance for predicting OS) and AUC (comparing performance for predicting 12-week survival). [Figure 15] Bar graph comparing the performance of the disclosed model (27 model parameters, called ROPRO) with alternative prognostic scores from the prior art using a database covering 20 clinical trials from phase 1 to phase 3 across different cancer indications. [Figure 16]Bar graph comparing the performance between two different models of the present disclosure (with 27 and 26 model parameters) and alternative prognostic scores of the prior art using a database covering 20 clinical trials from phase 1 to phase 3 across different cancer indications. [Figure 17] Graph showing Kaplan-Meier survival analysis of the OAK study. This study demonstrated that atezolizumab treatment resulted in a statistically significant and clinically relevant improvement in OS compared with docetaxel in 2L / 3L NSCLC. This survival analysis confirms this finding. [Figure 18] Graph showing delta ROPRO analysis of the two groups from the study referenced in connection with FIG. 17 above. ROPRO scores were significantly increased in patients in the docetaxel group compared to the atezolizumab group, indicating that delta ROPRO is indicative of treatment efficacy. This trend is also seen earlier than the survival analysis. DETAILED DESCRIPTION OF THE INVENTION

[0020] The following abbreviations are used herein: Abbreviation ALP alkaline phosphatase ALT alanine aminotransferase AST aspartate aminotransferase BMI Body Mass Index CI confidence interval CLL chronic myeloid leukemia CRC colorectal cancer CSF1 colony-stimulating factor-1 DLBCL Diffuse large B-cell carcinoma ECOG Eastern Cooperative Oncology Group (Performance Status) EHR Electronic Health Record HCC hepatocellular carcinoma HER2 human epidermal growth factor receptor 2 HR hazard ratio IPI International Prognostic Index KM Kaplan-Meier LDH lactate dehydrogenase NLR Neutrophil-to-lymphocyte ratio NSCLC non-small cell lung cancer OS overall survival PDL1 Programmed death ligand 1 PFS Progression-free survival RMHS Royal Marsden Hospital Prognostic Score RCC renal cell carcinoma RoPro Roche Prognostic Score rSq Generalized r-squared (r 2 ) SCLC: Small cell lung cancer TNM TNM System (Tumor, Node, Metastasis)

[0021] The following describes a method for determining the mortality risk of cancer patients. The determined risk can be expressed as a score (i.e., a quantitative measure such as a real number). This score may be referred to herein as Roche Prognostic Score (RoPro). The model used to obtain the score may be referred to herein as RoPro model.

[0022] The disclosed methods may be computer-implemented. Each step of the disclosed methods may be performed by a computer in the most general sense of the term, which refers to any device capable of performing the data processing steps of a method, including dedicated digital circuitry. A computer may include various combinations of computer hardware, including, for example, a CPU, RAM, SSD, motherboard, network connections, firmware, software, and / or other elements known in the art that enable computer hardware to perform the necessary computing operations. The necessary computing operations may be defined by one or more computer programs. The one or more computer programs may be provided in the form of a medium or data carrier, optionally a non-transitory medium, that stores computer-readable instructions. When the computer-readable instructions are read by a computer, the computer performs the necessary method steps. The computer may consist of a self-contained unit, such as a general-purpose desktop computer, laptop, tablet, mobile phone, smart device (e.g., smart TV), etc. Alternatively, the computer may consist of a distributed computing system having multiple different computers connected to each other via a network, such as the Internet or an intranet.

[0023] 1 is a flowchart illustrating a method for determining a cancer patient's risk of mortality according to the present disclosure. In step S1, the method includes receiving patient data representing information about the cancer patient. The patient data can be received by a computer configured to perform the method using any of a variety of known methods for providing data to a computer. In step S2, a mathematical model of mortality risk is used to determine the cancer patient's risk of mortality based on the received patient data. In step S3, the determined mortality risk is output, for example, as data and / or directly on a local display.

[0024] In some embodiments, the mathematical model models the relationship determined from training data between the risk of mortality and values ​​of a selected group of model parameters. Thus, the model is a trained model. The training data can include historical patient data from a database. Any of a variety of known methods for training a mathematical model using training data can be used.

[0025] We used training data obtained from the Flatiron Health database (Flatiron: https: / / flatiron.com). We conducted a retrospective cohort analysis using electronic health records (EHRs) from the Flatiron Health database. The Flatiron Health database contains demographically and geographically diverse longitudinal data from over 280 oncology clinics in the United States. Institutional review board approval of the study protocol was obtained prior to study implementation and included a waiver of informed consent. Data were extracted from the February 2020 data release, including patient demographic and clinical data (e.g., cancer type, disease stage, comorbidities, medication regimens, and routine blood biomarkers). Patient records with missing first-line treatment information and patient records / variables with high missing data rates were excluded. Missing values ​​in the final analysis dataset were imputed.

[0026] In some embodiments, the model is formed by performing a multivariate Cox regression analysis on training data from multiple subjects, preferably at least 1000 subjects.

[0027] We used a Cox proportional hazards model (Cox DR. Regression Models and Life-Tables. Journal of the Royal Statistical Society. Series B (Methodological) Vol. 34). Risk of death was examined as overall survival (OS). Survival time was calculated from the start of the first treatment (defined as TO) to the event of "death" (death from all causes) as coded in the Flatiron Health real-world mortality tables (Stock, C., Mons, U. & Brenner, H. Projection of cancer incidence rates and case numbers until 2030: A probabilistic approach applied to German cancer registry data (1999-2013). Cancer Epidemiol 57, 110-119 (2018)). For cohorts including only patients with advanced / metastatic disease, the first systemic treatment for advanced / metastatic disease was classified as the first database event. Monitored follow-up time was calculated as the number of days elapsed from T0 to the date of the patient's last recorded clinical contact. The patient's last available prior T0 measurement was used for modeling. A time limit of 30 days before T0 was applied. See Connolly, JG, Schneeweiss, S., Glynn, RJ & Gagne, JJ. Quantifying bias reduction with fixed-duration versus all-available covariate assessment periods. Pharmacoepidemiol. Drug Saf. 28, 665-670 (2019).

[0028] We found from early experiments with a smaller data version from 2018 that there was little evidence that the routine variables available in Flatiron Health had different importance across cohorts. Therefore, we constructed a total index score (representing the determined risk of death).

[0029] For model comparison, consider the following for censored data and refer to: - Generalized r2 (Kalbfleisch, JD & Prentice, RL. The statistical analysis of failure time data, Second Edition. (Wiley Series in Probability and Statistics, 2002)). - Concordance Index (C-index) (Steck, H., Krishnapuram, B., Dehing-Oberije, C., Lambin, P. & Raykar, VCOn Ranking in Survival Analysis:Bounds on the Concordance Index.in Advances in Neural Information Processing Systems 20 (eds. Platt, JC, Koller, D., Singer, Y. & Roweis, ST) 1209-1216 (Curran Associates, Inc., 2008)). - Area under the ROC curve (ROC-AUC) (Blanche, P, Dartigues, JF & Jacquim-Gadda, H. Estimating and comparing time-dependent areas under receiver operating characteristic curves for censored event times with competing risks).

[0030] All analyses were performed using the statistical analysis package R (R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http: / / www.R-project.org / ).

[0031] Model selection was performed using a family-wise error rate-controlled (FWER-controlled) elimination procedure for Cox regression models, starting with 46 initial variables. Variables with the least impact on model performance were iteratively removed until all remaining parameters produced measurable improvements in the concordance index and generalization r2, which were significant at α = 0.05 / 46 = 0.0011 (Bonferroni correction). By construction, this procedure controls the family-wise error rate at α = 0.05, i.e., all parameters are significant after adjusting for multiple testing. In parallel, we derived a regularized Lasso model with 10-fold cross-validation (Simon N, Friedman J, Hastie T, Tibshirani R. Regularization Paths for Cox's Proportional Hazards Model via Coordinate Descent. J. Stat. Softw. Artic. 2011;39(5):1-13). Model performance is visualized in Figure 2 with respect to increasing regularization strength (larger λ). The number of variables (indicator and non-indicator) selected is shown at the top of the plot. λ is chosen to extract the most regularized model with a C-index within one standard error (right vertical dashed line in Figure 2) of the best-performing model (left vertical dashed line).

[0032] For continuous variables, the Cox model yields a hazard ratio (HR) per unit of the investigated variable (model parameter), e.g., the HR for age per 1-year age difference. For better comparability between variables, the "4SD-HR" is provided as a descriptive measure, specifically the HR for patients with high variable values ​​(equal to the mean + 2 standard deviations (SD)) versus those with low values ​​(mean - 2SD).

[0033] Overall, 122,694 patients across 17 cancer-type-specific cohorts were eligible for analysis. Table 1 summarizes patient characteristics. [Table 1] TIFF0007814326000002.tif237170

[0034] After quality control, data were available for 46 general measures, an overall missingness rate of 21% before imputation, and 99 additional cohort-specific measures (cytogenetics, biomarkers). Median survival was 19.0 months (95% confidence interval [CI] 18.8-19.2).

[0035] ROPRO The initial unadjusted analysis identified 44 variables significantly associated with OS. The elimination regression model selected 27, and the cross-validated Lasso model selected 28 independently contributing variables, 26 of which were concordant (Table 1). In both cases, plots of the concordance index by the number of included variables showed no evidence of early saturation in the model with fewer variables (Figure 2). The prognostic risk scores implied by the two modeling approaches correlated with r = 0.993, demonstrating the compatibility and stability of the modeling. All further result descriptions referring to the "ROPRO" or "RoPro" model or score refer to a model with 27 independently contributing variables in the multivariate modeling (Table 1, Figure 3). This model had an overall shrinkage factor of 0.998, with an apparent r and a Nagelkerkes / Cragg & Uhlers pseudo-r. 2 The difference is 1.0 x 10 -6The absolute margin of error for the 3-month survival time is small (0.0017). Therefore, all modeling conditions recently required for the following are met: Riley, RD et al. Minimum sample size for developing a multivariable prediction model: PART II - binary and time-to-event outcomes. Stat. Med. 38, 1276-1296 (2019).

[0036] The 27 variables found to be predictive can be used as model parameters in a model of mortality risk. In the specific example described here, the resulting determined risk is output as a prognostic score called the RoPro score. The score is based on a weighted sum of the patient's differences from the respective reference means for the variables (see Table 1 for details). An example RoPro score can be determined from: 0.00948(age-67.15) + 0.14162(gender-0.502) + 0.19521(smoking-0.576) + 0.02833(number of metastatic sites-0.897) - 0.04178(Hgb-12.087) + 0.19097(urea nitrogen-2.777) - 6e-04(platelets-267.423) + 1. 0619 (Calcium -2.231) + 0.06098 (Glucose -4.733) - 0.11658 (Lymphocyte-Leukocyte Ratio -2.953) + 0.19019 (ALP -4.582) - 0.00896 (Protein -68.888) - 0.05113 (ALT -3.008) - 0.03988 (Albumin - 37.851) + 0.08189 (bilirubin -(-0.773)) - 0.02462 (chloride -101.434) + 0.10671 (monocytes -(-0.548)) - 0.0543 (eosinophil-leukocyte ratio -0.307) + 1.22733 (LDH -1.694) + 0.42372 (heart rate -4.402) - 0.3 9878(SBP-(4.846))-0.02083(Oxygen-96.487)+0.20066(ECOG-0.84)+0.0905(NLR-1.156)-0.17076(BMI-3.301)+0.13122(AST-ALT-ratio-0.092)+0.081(Tumor stage-3.098).

[0037] The 27 model parameters are: (i) age; (ii) gender; (iii) smoking history, (iv) number of metastatic sites; (v) blood hemoglobin or hematocrit levels; (vi) serum or plasma urea nitrogen levels (vii) blood platelet levels; (viii) serum or plasma calcium levels; (ix) blood glucose levels; (x) lymphocyte-to-leukocyte ratio in the blood; (xi) alkaline phosphatase enzyme activity level in serum or plasma; (xii) protein levels in serum or plasma; (xiii) serum or plasma albumin levels; (xiv) serum or plasma bilirubin levels; (xv) serum or plasma chloride or sodium levels; (xvi) levels of monocytes in the blood; (xvii) blood eosinophil to leukocyte ratio; (xviii) serum or plasma lactate dehydrogenase enzyme activity level; (xix) heart rate, (xx) systolic blood pressure, (xxi) the level of oxygen saturation in arterial blood; (xxii) Eastern Cooperative Oncology Group (ECOG) performance status; (xxiii) the ratio of neutrophils to lymphocytes in the blood; (xxiv) body mass index; (xxv) the ratio of the level of aspartate aminotransferase enzyme activity in serum or plasma to the level of alanine aminotransferase enzyme activity in serum or plasma; (xxvi) TNM classification of tumor stage, and (xxvii) serum or plasma alanine aminotransferase enzyme activity level As stated above.

[0038] In the above example, the model of mortality risk models the relationship determined from training data between mortality risk and the values ​​of all model parameters (i) through (xxvii), which has been found to provide particularly high performance, however, acceptable performance may also be obtained using a smaller number of available model parameters.

[0039] In some embodiments, the model models the relationship determined from training data between risk of mortality and at least 16, optionally at least 18, optionally at least 20, optionally at least 22, optionally at least 24, optionally all values ​​of model parameters (i)-(xxvi). In some embodiments, the model models the relationship determined from training data between risk of mortality and at least 16, optionally at least 18, optionally at least 20, optionally at least 22, optionally at least 24 values ​​of model parameters (i)-(xxvii).

[0040] In some embodiments, the following additional model parameters are used: (xxviii) level of eosinophils in the blood, and (xxix) diastolic blood pressure.

[0041] The use of these additional model parameters increases the complexity and data requirements of the model, but can provide further performance improvements. In such embodiments, the model can model the relationship determined from the training data between mortality risk and values ​​of at least 16, optionally at least 18, optionally at least 20, optionally at least 22, optionally at least 24, and optionally all of the model parameters (i)-(xxix).

[0042] For a particular patient, some values ​​of the model parameters used by the model may not be available. The inventors have found that the model can still provide a reliable score in these situations up to a discrepancy of about 10 values. Therefore, the patient data should preferably include values ​​for at least 16, optionally at least 18, optionally at least 20, optionally at least 22, and optionally at least 24 of the model parameters used by the model. Alternatively, the patient data includes values ​​for all of the model parameters (i)-(xxvi). Alternatively, the patient data includes values ​​for all of the model parameters (i)-(xxvii). Alternatively, the patient data includes values ​​for all of the model parameters (i)-(xxix).

[0043] ROPRO16 As noted above, in some embodiments, the mathematical model models the relationship between mortality risk and values ​​of at least 16 model parameters. In one class of embodiments, the 16 model parameters are: (i) age; (ii) gender; (iii) hemoglobin or hematocrit levels in the blood; (iv) serum or plasma urea nitrogen levels (v) alkaline phosphatase enzyme activity level in serum or plasma; (vi) protein levels in serum or plasma; (vii) serum or plasma albumin levels; (viii) serum or plasma chloride or sodium levels; (ix) blood eosinophil-to-leukocyte ratio; (x) serum or plasma lactate dehydrogenase enzyme activity level; (xi) heart rate, (xii) systolic blood pressure, (xiii) Eastern Cooperative Oncology Group (ECOG) performance status; (xiv) blood neutrophil-to-lymphocyte ratio; (xv) the ratio of the serum or plasma aspartate aminotransferase enzyme activity level to the serum or plasma alanine aminotransferase enzyme activity level, and (xvi) TNM classification of tumor stage Includes.

[0044] When only these 16 model parameters are used, the model can be referred to as a 16-variable reduced model or "ROPRO16" model.

[0045] In some embodiments, the model comprises the following parameters: (xvii) smoking history, (xviii) number of metastatic sites; (xix) blood platelet levels; (xx) serum or plasma calcium levels; (xxi) glucose levels in the blood; (xxii) lymphocyte to leukocyte ratio in blood; (xxiii) serum or plasma bilirubin levels; (xxiv) levels of monocytes in the blood; (xxv) the level of oxygen saturation in arterial blood; and (xxvi) Body mass index is expanded to 26 parameters by including

[0046] In some embodiments, the model is expanded to 27 parameters by including the following parameters: (xxvii) alanine aminotransferase enzyme activity level in serum or plasma.

[0047] In some embodiments, the model comprises the following parameters: (xxviii) blood eosinophil levels, and (xxix) Expanded to 29 parameters by including diastolic blood pressure.

[0048] Use cases for determined mortality risk As described above, in some embodiments, determining the risk of death includes calculating a numerical score representing the risk of death. In some embodiments, determining the risk of death further includes comparing the calculated score to one or more predetermined thresholds or to the calculated scores of other cancer patients. For example, the risk of death can be determined to be "high" if the calculated score is above a predetermined threshold or above the average score calculated for a selected group of cancer patients. The risk of death can be determined to be "low" if the calculated score is below a predetermined threshold or below the average score calculated for a selected group of cancer patients.

[0049] The score thus provides a practical quantitative tool for representing the patient's physical condition. Scores can be obtained in the same way for different patients or for the same patient at different times. The score can therefore be used as an objective and quantitative basis for comparing different patients with each other and / or the condition of the same patient at different times. The score therefore provides a quantitative tool that can be used in various ways to support objective decision-making regarding the provision of medical care to patients and / or to facilitate the extraction of objective information from studies involving patients.

[0050] In some embodiments, the model comprises a weighted sum of deviations in the patient data from means of the model parameters in the training data, and the model is formed by determining weights from the training data.

[0051] In one embodiment, the model assigns each model parameter a respective weighting w i and assign the respective mean m of the values ​​of each model parameter across the training data. i The risk of death is then determined using the following formula: TIFF0007814326000003.tif14170(in the formula, w i is the weighting of the i-th model parameter, and m iis the mean of the ith model parameter, and m ij is the value of the ith model parameter for the jth cancer patient for whom the score is being calculated) The method may include calculating a numerical score according to:

[0052] In one embodiment, the weighting w i Each of the scores is as follows: TIFF0007814326000004.tif14170(where HR(x i ) is the estimated HR for variable i, and m i is the variable mean in the training data (e.g., the Flatiron Health database), and m ij is the value of patient j in variable i ∈ I) It includes the natural logarithm of the hazard ratio (HR), given as:

[0053] The determined mortality risk (eg, as a score) can be used in a variety of situations, as described above.

[0054] In one class of embodiments, a method for evaluating an anti-cancer treatment using the determined score can be provided, the method comprising: determining a patient's risk of mortality at a plurality of different time points while the patient is undergoing anti-cancer treatment, the method comprising performing a method for determining risk of mortality according to any of the embodiments of the present disclosure at each of the plurality of time points; and analyzing the resulting determined risk to determine the effectiveness of the anti-cancer treatment.

[0055] Alternatively or additionally, a method for selecting cancer patients for treatment with an anti-cancer therapy using the determined score can be provided, the method comprising determining a risk of death for a candidate patient using a method for determining a risk of death according to any of the embodiments of the present disclosure. The risk determination is then used to determine whether to select each candidate patient.

[0056] Further examples of applications of the determined risk (eg, score) are described below.

[0057] Performance evaluation of RoPro The performance of an exemplary embodiment of a method for determining mortality risk is discussed in this section. The exemplary embodiment uses 27 model parameters (i)-(xxvii). In this case, the determined risk is expressed as a score called the RoPro score, or simply "RoPro."

[0058] Individual patient RoPro scores (derived by inputting measurements for each of the variables into a formula) ranged from -3.49 to 4.12, with a 99% range of [-2.21; 2.09]. Higher scores indicate higher risk.

[0059] The model in this example was found to significantly outperform RMHS with respect to all model performance metrics (r2 = 0.32, C-index = 0.747, AUC at 3 months = 0.82 vs. r2 = 0.03, C-index = 0.54, AUC at 3 months = 0.58).

[0060] Figures 4-6 show a comparison of RMHS and RoPro. Figure 4 demonstrates a clear separation of survival curves according to high and low RMHS (HR 2.31; 95% CI 2.26-2.36). Figure 5 shows the highest 10% of patients with RoPro scores versus the remaining 90%, demonstrating improved separation of survival curves (HR 4.85; 95% CI 4.75-4.94). Further subdividing the sample into 10 subgroups of equal size but increasing RoPro (deciles) demonstrated clear separation of the respective Kaplan-Meier survival curves (Figure 6). Median survival was clearly separated along deciles, with a median survival of 2,975 days in the lowest deciles compared to 114 days in the highest deciles. Patients at highest risk (RoPro2 score >1.18) had a HR of 27.6 (95% CI 26.37-28.88), P < 2.23 × 10 compared with those at lowest risk (score < -1.29). -308 had the following characteristics:

[0061] The RoPro score demonstrated robust performance in all 17 cohorts based on highly consistent variable (model parameter) estimate size and direction across cohorts. Cohort-specific re-estimated variable weights ln(HR(x i ) showed a strong improvement in model performance for chronic lymphocytic leukemia (CLL) (r² = 0.13, C-index = 0.704, 3-month AUC = 0.82 for CLL-general RoPro; r² = 0.17, C-index = 0.74, 3-month AUC = 0.84 for CLL-specific RoPro).

[0062] Model validation of RoPro in clinical trial populations For validation purposes, the (RoPro) model was retrospectively applied to patients from two independent clinical studies, without reestimation of variable weights, using Cox regression to fit the RoPro score as a single variable. The first analysis was based on a phase I first-in-human study, BP29428 (NCT02323191), which investigated the combination of emactuzumab, a monoclonal antibody targeting the colony-stimulating factor-1 (CSF1) receptor (Gomez-Roca, CA et al. Phase I study of emactuzumab single agent or in combination with paclitaxel in patients with advanced / metastatic solid tumors reveals depletion of immunosuppressive M2-like macrophages. Ann. Oncol. 30, 1381-1392 (2019)), with atezolizumab, a monoclonal antibody targeting PDL1, in patients (n=216) with locally advanced or metastatic solid tumors unamenable to standard treatment. The second analysis used data from the phase III OAK trial, which evaluated the efficacy and safety of atezolizumab monotherapy versus single-agent docetaxel in participants (n=1187) with locally advanced or metastatic non-small-cell lung cancer (NSCLC) after failure of platinum-containing chemotherapy (Rittmeyer, A. et al. Atezolizumab versus docetaxel in patients with previously treated non-small-cell lung cancer (OAK): a phase 3, open-label, multicentre randomized controlled trial. Lancet 389, 255-265 (2017)).

[0063] First, we investigated the correlation between RoPro and OS in the phase I study BP29428 (n = 216, P = 6.08 × 10 -15The results were reproducible in the Flatiron Health data (r2 = 0.23, C-index = 0.81). Patients with an ROC > 1.18 (cutoff equal to the 90th percentile in Flatiron Health, n = 8) had a particularly poor OS prognosis (HR 11.29; 95% CI 5.52-23.12). The 3-month survival ROC-AUC values ​​were comparable to those in Flatiron Health (Figures 7 and 8).

[0064] Second, the correlation between RoPro and OS survival was also replicated in the phase III OAK trial (n = 1,187, P = 8.38 × 10 -60 , r2=0.20, C-index=0.69). Patients with an RoPro>0.82 (the cutoff equal to the 90th percentile in Flatiron Health for dedicated advanced NSCLC RoPro, n=34) had a worse OS prognosis (HR 3.65; 95% CI 2.59-5.12). The area under the curve value also exceeded that of RMHS (Figures 9 and 10).

[0065] In a phase I study, we observed a correlation between high RoPro (>1.18, see definition above, n=9) and study dropout. All high RoPro patients withdrew from the study very early due to either death (n=5) or progressive disease (n=4). Only one patient was able to receive more than two treatment cycles, but all patients had an ECOG performance status of 1 at baseline (study entry criteria). Typically, study protocols require a life expectancy of >12 weeks for enrollment, as drug efficacy is potentially masked in patients with a particularly poor prognosis. However, RoPro analysis demonstrates that, in current practice, it is not always possible to identify all high-risk patients in advance. However, if RoPro is available, the respective information can be retrospectively included to exclude patients from the analysis or to perform stratified analyses.

[0066] In a phase III study, we evaluated the potential impact of using a prognostic score as a covariate parameter in comparing OS between treatment groups. In a plain analysis, a HR of 0.794 (95% CI 0.690-0.913, P = 0.0012) was observed for atezolizumab versus docetaxel, in accordance with published results (Connolly, JG, Schneeweiss, S., Glynn, RJ, & Gagne, JJ. Quantifying bias reduction with fixed-duration versus all-available covariate assessment periods. Pharmacoepidemiol. Drug Saf. 28, 665-670 (2019)). Although not required in randomized trials, we adjusted for baseline RoPro in efficacy analyses. We were able to confirm increased efficacy with atezolizumab, obtaining a HR of 0.760 (95% CI 0.666-0.875, P = 0.00013) in the RoPro-adjusted analysis.

[0067] Finally, we performed an initial evaluation of whether changes in RoPro over time could indicate a subsequent event. An ad hoc visualization is shown in Figure 11. In patients who died (top curve), scores significantly worsened during the last 100 days before the event. A less pronounced increase toward progression was observed (middle curve), whereas partial responders (n = 191) and complete responders (n = 11) showed no increase (bottom curve). To formalize this observation, we performed a landmark analysis as described in Houwelingen, HCVD Dynamic Prediction by Landmarking in Event History Analysis. Scand. J. Stat. 34, 70-85 (2007). The landmark was set at 145 days, when 75% of study participants were alive. These patients were classified into two groups defined by either an increase or a decrease in RoPro from baseline to the landmark. Figure 12 shows the post-landmark survival curves. Patients with an increase in RoPro to the landmark (yellow) had significantly shorter survival times after landmark than patients with an early decrease in RoPro (blue), with a HR of 1.90 (95% CI 1.57-2.31, P = 8.1 × 10 -11 )

[0068] Consideration In this analysis of 122,694 patients from 17 cancer cohorts, 27 independent OS risk factors were identified and used to define a prognostic model, RoPro, which showed a previously unseen level of correlation with OS. Validation analyses using data from two independent clinical studies confirmed these results. To our knowledge, RoPro is based on the largest dataset used for prognostic modeling to date and overcomes typical limitations. Furthermore, within a classical statistical framework, we demonstrated that multiple prognostic factors independently contribute to risk modeling after adjusting for correlations between risk factors. Results were highly consistent when using different modeling approaches (backward elimination and regularized regression).

[0069] RoPro can include 27 clinical parameters combined into a single score, can be easily applied to new datasets, and has demonstrated substantial improvement in performance over alternative scores from models such as RMHS, IPI, or IMDC.

[0070] By construction, all RoPro variables are typically available in routine clinical practice in the United States. Most of the variables are also available in clinical practice in other countries and in clinical trials conducted by the pharmaceutical industry. A patient-specific error of 5 to 10 variables is allowed to calculate a patient's RoPro.

[0071] Finally, RoPro can be applied across cancer indications beyond the 17 used to build the model, as demonstrated by its performance in BP29428, where 40% of patients had cancer indications not available in Flatiron Health.

[0072] Our analysis of study BP29428 (the combination of emactuzumab and atezolizumab, NCT02323191) showed that RoPro not only correlated with OS but also with dropout in certain early-stage studies. Therefore, using RoPro to exclude very high-risk patients may help protect them from the procedural burden of a trial and unnecessary exposure to potential adverse events, but this must be balanced against the benefit of allowing such patients access to potentially effective novel drugs.

[0073] Landmark analysis in the OAK study demonstrated that early increases / decreases in RoPro indicate later survival events. While biologically plausible, we note that this observation is not trivial. A possible alternative scenario could be one of early and late prognostic factors, where the early factor does not deteriorate after a certain point, and the late mechanism indicates the occurrence of an event very close to the fatal event.

[0074] The increase in RoPro relative to death is a feature that generates potential applications. First, during dose escalation in first-in-human studies, it is a reasonable hypothesis that RoPro increases under ineffective drug doses or treatments but remains stable or even improves under effective treatments. Thus, group-level analysis of RoPro time course by treatment group or drug dose can provide valuable insight into potential drug efficacy. Increased RoPro, in combination with death events, can be used as a surrogate endpoint in clinical trials.

[0075] Continuous monitoring of RoPro over time may inform treatment decisions. Our analysis of the OAK study data showed that high-risk patients with baseline scores who worsened over time were less likely to benefit from treatment, and the majority of such patients died within a very short time. Therefore, patients with increasing RoPro may be switched to a different treatment. Observing a stable or slightly improving score may indicate benefit from treatment and, along with many other considerations, can be used to inform decisions regarding continued treatment.

[0076] The fact that RoPro successfully addresses a "novel" cancer indication, as exemplified by its application in a Phase I trial that recruited many patients outside the United States, increases confidence that RoPro is robust across geographic regions and ethnicities.

[0077] Although uncertainties and imprecision are expected to be encountered in retrospective real-world data analyses, our findings suggest that the increased power afforded by sample size overcomes potential biases.

[0078] Further performance evaluation including evaluation of RoPro16 13-18 show results from additional exemplary embodiments of the models of the present disclosure, including a model using 27 parameters (ROPRO), a model using 16 parameters (ROPRO16), and variations of these two models that omit Eastern Cooperative Oncology Group (ECOG) performance status parameters (ROPRO minus ECOG and ROPRO16 minus ECOG).

[0079] Results are compared to prior art alternative prognostic scores, including RMHS, IPI, IMDC and ECOG (all mentioned above).

[0080] Figure 13 is a graph corresponding to that of Figure 3, except that the ROPRO16 model is used rather than the full ROPRO model with 27 parameters. The parameter weightings shown in Figure 13, derived from HR values ​​as described above, are believed to be broadly similar to the corresponding parameter weightings in Figure 3.

[0081] Figure 14 is a bar graph comparing the performance of ROPRO (B1), ROPRO minus ECOG (B2), ROPRO16 (B3), and ROPRO16 minus ECOG (B4) with the prior art prognostic scores RMHS (B5), IPI (B6), IMDC (B7), and ECOG (B8). Two comparisons are shown: C-index (comparing general performance for OS prediction) and AUC (comparing performance for predicting 12-week survival). As can be seen, all of the disclosed models, even when reduced to 16 or 15 parameters, outperform all of the prior art prognostic scores considered for both C-index and AUC. The relatively small difference between using 15 and 16 parameters demonstrates the relative robustness of the approach to data inconsistencies, even when the models are in their reduced (16 parameter) form.

[0082] Figure 15 is a bar graph comparing the performance of ROPRO (labeled C1 in the left-most subchart) with the prior art alternative prognostic scores RMHS (labeled C2 in the left-most subchart), IPI (labeled C3 in the left-most subchart), IMDC (labeled C4 in the left-most subchart), Glasgow (referenced above and labeled C5 in the left-most subchart), and SLD (representing the sum of the longest diameters of the tumor and labeled C6 in the left-most subchart). The bars in the five other subcharts follow the same sequence of C1–C6, but these labels have been omitted for clarity. The database used in this demonstration encompassed 20 clinical trials from phase 1 to phase 3 across different cancer indications. The bar graphs show that ROPRO outperformed other scores in terms of both C-index and AUC in phase 1, phase 2, and phase 3.

[0083] FIG. 16 is a bar graph comparing the performance of ROPRO (labeled D1) and ROPRO without ECOG (which is the same as ROPRO minus ECOG mentioned above and labeled D6) with prior art alternative prognostic scores RMHS (labeled D2), IPI (labeled D3), IMDC (labeled D4), and ECOG (labeled D5). The database used in this demonstration covered 20 clinical trials from phase 1 to phase 3 across different cancer indications. The bar graph again shows that both models of the present disclosure outperform all of the alternative prognostic scores considered.

[0084] Figure 17 is a graph showing the Kaplan-Meier survival analysis of the OAK study. This study demonstrated that atezolizumab treatment resulted in a statistically significant and clinically relevant improvement in OS compared with docetaxel in 2L / 3L NSCLC. This survival analysis confirms this finding.

[0085] Figure 18 is a graph showing delta ROPRO analysis of the two groups from the study referenced in connection with Figure 17 above. ROPRO scores were significantly increased in patients in the docetaxel group compared to the atezolizumab group, indicating that delta ROPRO is indicative of treatment efficacy. This trend is also seen earlier than the survival analysis.

[0086] The structure of this disclosure is defined in the following numbered sections:

[0087] 1. A computer-implemented method for determining a risk of mortality in a cancer patient, comprising: receiving patient data representing information about a cancer patient; determining the risk of death of the cancer patient based on the received patient data using a mathematical model of the risk of death; Outputting the determined mortality risk Including, The mathematical model correlates the risk of death with at least 16 of the following model parameters: (i) age; (ii) gender; (iii) smoking history, (iv) number of metastatic sites; (v) blood hemoglobin or hematocrit levels; (vi) serum or plasma urea nitrogen levels (vii) blood platelet levels; (viii) serum or plasma calcium levels; (ix) blood glucose levels; (x) lymphocyte-to-leukocyte ratio in the blood; (xi) alkaline phosphatase enzyme activity level in serum or plasma; (xii) protein levels in serum or plasma; (xiii) serum or plasma albumin levels; (xiv) serum or plasma bilirubin levels; (xv) serum or plasma chloride or sodium levels; (xvi) levels of monocytes in the blood; (xvii) blood eosinophil to leukocyte ratio; (xviii) serum or plasma lactate dehydrogenase enzyme activity level; (xix) heart rate, (xx) systolic blood pressure, (xxi) the level of oxygen saturation in arterial blood; (xxii) Eastern Cooperative Oncology Group (ECOG) performance status; (xxiii) the ratio of neutrophils to lymphocytes in the blood; (xxiv) body mass index; (xxv) the ratio of the aspartate aminotransferase enzyme activity level in the serum or plasma to the alanine aminotransferase enzyme activity level in the serum or plasma; and (xxvi) Value of TNM classification of tumor stage and modeling the relationship determined from training data between the

[0088] 2. The method of clause 1, wherein the patient data includes values ​​for at least 16 of each of the model parameters used by the model.

[0089] 3. The method of clause 1 or 2, wherein the patient data includes values ​​for all of the model parameters (i) through (xxvi).

[0090] 4. The model parameters are (xxvii) The method of any one of clauses 1 to 3, further comprising measuring serum or plasma alanine aminotransferase enzyme activity levels.

[0091] 5. The method of clause 4, wherein the patient data includes values ​​for all of the model parameters (i) through (xxvii).

[0092] 6. The model parameters are (xxviii) blood eosinophil levels, and (xxix) Diastolic blood pressure 6. The method of clause 4 or 5, further comprising:

[0093] 7. The method of clause 6, wherein the patient data includes values ​​for all of the model parameters (i) through (xxix).

[0094] 8. The method of any one of clauses 1 to 7, wherein the mathematical model models the relationship determined from training data between the risk of mortality and values ​​of all of the model parameters (i) to (xxvi).

[0095] 9. The method of any one of clauses 4 to 7, wherein the mathematical model models the relationship determined from training data between the risk of mortality and all values ​​of model parameters (i) to (xxvii).

[0096] 10. The method of clause 6 or 7, wherein the mathematical model models the relationship determined from training data between the risk of death and all values ​​of model parameters (i) through (xxix).

[0097] 11. The method of any one of clauses 1 to 10, wherein determining the risk of mortality comprises calculating a numerical score representative of the risk of mortality.

[0098] 12. The method of clause 11, wherein determining the risk of death further comprises comparing the calculated score to one or more predetermined thresholds, or comparing the calculated score to the calculated scores of other cancer patients.

[0099] 13. The method of any one of clauses 1 to 12, wherein the model comprises a weighted sum of deviations in the patient data from means of the model parameters in the training data, and the model is formed by determining weights from the training data.

[0100] 14. The method of any one of clauses 1 to 13, wherein the model is formed by performing a multivariate Cox regression analysis on training data of a plurality of subjects, preferably at least 1000 subjects.

[0101] 15. The model assigns each model parameter its own weighting w i and assign the respective mean m of the values ​​of each model parameter across the training data. i is formed by determining The risk of death is determined using the following formula: TIFF0007814326000005.tif14170(in the formula, w i is the weighting of the i-th model parameter, and m i is the mean of the ith model parameter, and m ij is the value of the ith model parameter for the jth cancer patient for whom the score is being calculated) 15. The method of any one of clauses 1 to 14, comprising calculating a numerical score according to:

[0102] 16. A method for evaluating an anti-cancer treatment, comprising determining a patient's risk of death at a plurality of different time points while the patient is receiving the anti-cancer treatment by carrying out the method of any one of clauses 1 to 15 at each of the plurality of time points, and analyzing the resulting determined risks to determine the effectiveness of the anti-cancer treatment.

[0103] 17. A method for selecting cancer patients for treatment with an anti-cancer therapy, the method comprising determining a candidate patient's risk of death using the method of any one of clauses 1 to 15, and using the determined risk to determine whether to select each candidate patient.

[0104] 18. The method of any one of clauses 1 to 17, wherein the method is computer-implemented.

[0105] 19. A computer program comprising instructions, which when the program is executed by a computer, cause the computer to perform any of the methods set out in clause 18.

Claims

1. 1. A computer-implemented method for determining a risk of mortality in a cancer patient, comprising: receiving patient data representing information about a cancer patient; determining a mortality risk for the cancer patient based on the received patient data using a mathematical model of mortality risk; outputting the determined mortality risk; Including, The mathematical model combines the mortality risk with all of the following model parameters: (i) age, (ii) gender; (iii) hemoglobin or hematocrit levels in the blood; (iv) serum or plasma urea nitrogen levels; (v) alkaline phosphatase enzyme activity levels in serum or plasma; (vi) protein levels in serum or plasma; (vii) serum or plasma albumin levels; (viii) serum or plasma chloride or sodium levels; (ix) blood eosinophil to leukocyte ratio; (x) serum or plasma lactate dehydrogenase enzyme activity level; (xi) heart rate; (xii) systolic blood pressure; (xiii) Eastern Cooperative Oncology Group (ECOG) performance status; (xiv) blood neutrophil to lymphocyte ratio; (xv) the ratio of the serum or plasma aspartate aminotransferase enzyme activity level to the serum or plasma alanine aminotransferase enzyme activity level; and (xvi) TNM classification of tumor stage and modeling the relationship determined from training data between values ​​of

2. The mathematical model combines the mortality risk with the model parameters (i) to (xvi) and at least the following parameters: (xvii) smoking history, (xviii) number of metastatic sites; (xix) platelet levels in the blood; (xx) serum or plasma calcium levels; (xxi) glucose levels in the blood; (xxii) lymphocyte to leukocyte ratio in the blood; (xxiii) serum or plasma bilirubin levels; (xxiv) levels of monocytes in the blood; (xxv) the level of oxygen saturation in arterial blood; and (xxvi) body mass index The method of claim 1 , wherein the method models a relationship determined from training data between values ​​of

3. The mathematical model combines the mortality risk with the model parameters (i) to (xxvi) and at least the following parameters: (xxvii) serum or plasma alanine aminotransferase enzyme activity level 3. The method of claim 2, wherein the method models a relationship determined from training data between values ​​of

4. The mathematical model combines the mortality risk with the model parameters (i) to (xxvii) and at least the following parameters: (xxviii) blood eosinophil levels, and (xxix) diastolic blood pressure 4. The method of claim 3, wherein the method models a relationship determined from training data between values ​​of

5. 5. The method of claim 1, wherein determining the risk of mortality comprises calculating a numerical score representative of the risk of mortality.

6. 6. The method of claim 5, wherein the determination of the risk of death further comprises comparing the calculated score to one or more predetermined thresholds or to the calculated scores of other cancer patients.

7. 7. The method of claim 1, wherein the model comprises a weighted sum of deviations in the patient data from mean values ​​of the model parameters in the training data, and the model is formed by determining weights from the training data.

8. 8. The method of any one of claims 1 to 7, wherein the model is formed by performing a multivariate Cox regression analysis on the training data of a plurality of subjects, preferably at least 1000 subjects.

9. The model assigns each of the model parameters a respective weighting w ｉ and assigning the respective mean m of the values ​​of each model parameter across the training data ｉ is formed by determining The determination of the risk of mortality is based on the following formula: (In the formula, w ｉ is the weighting of the i-th model parameter, and m ｉ is the mean of the i-th model parameter, and m ｉｊ is the value of the i-th model parameter for the j-th cancer patient for which the score is calculated.

9. The method of claim 1, comprising calculating a numerical score according to:

10. 10. A method for evaluating an anti-cancer treatment, comprising determining a patient's risk of death at a plurality of different time points while the patient is receiving the anti-cancer treatment by carrying out the method of any one of claims 1 to 9 at each of the plurality of time points, and analyzing the resulting determined risk to determine the effectiveness of the anti-cancer treatment.

11. 10. A method of selecting cancer patients for treatment with an anti-cancer therapy, the method comprising determining a risk of mortality of candidate patients using the method of any one of claims 1 to 9, and using the determined risk to decide whether to select each candidate patient.

12. A computer program comprising instructions, which when executed by a computer, cause the computer to carry out the method of any one of claims 1 to 11.

Images