Learning Individual Survival Models from PanCancer Whole Transcriptome Data

Abstract

Purpose: Personalized medicine attempts to predict survival time for each patient, based on their individual tumour molecular profile. We investigate whether our survival learner in combination with a dimension reduction method can produce useful survival estimates for a variety of cancer patients. Experimental Design: This paper provides a method that learns a model for predicting the survival time for individual cancer patients from the PanCancer Atlas: given the (16335-dimensional) gene expression profiles from 10173 patients, each having one of 33 cancers, this method uses unsupervised NMF (non-negative matrix factorization) to re-express the gene expression data for each patient in terms of 100 learned NMF Factors. It then feeds these 100 Factors into the MTLR (Multi-Task Logistic Regression) learner to produce cancer-specific models for each of 20 cancers (with >50 uncensored instances); this produces “individual survival distributions” (ISDs), which provide survival probabilities at each future time for each individual patient – which provides a patient’s risk score and estimated survival time. Results: Our NMF-MTLR concordance indices outperformed the VAECox benchmark by 14.9% overall. We achieved optimal survival prediction using pan-cancer NMF in combination with cancer-specific MTLR models. We provide biological interpretation of the NMF model and clinical implications of ISDs for prognosis and therapeutic response prediction. Conclusions: NMF-MTLR provides many benefits over other models: superior model discrimination, superior calibration, meaningful survival time estimates, and accurate probabilistic estimates of survival over time for each individual patient. We advocate for the adoption of these cancer survival models in clinical and research settings.

Individual Survival Distributions Generated by Multi-Task Logistic Regression Yield a New Perspective on Molecular and Clinical Prognostic Factors in Gastric Adenocarcinoma

“…This study presents a survival modelling perspective called individual survival distributions with the MTLR model to provide novel insight into personalized survival predictions of gastric adenocarcinoma. Prior research has demonstrated that ISDs provide a valuable alternative to popular personalized survival modelling strategies, such as nomograms [12,15]. Using MTLR, we identified that the TME score and chemotherapy provide non-proportional but similar survival effects over time.…”

“…MTLR generates individual patient survival distributions for a specific number of times based on the number of uncensored patients in a dataset [26]. An empirical study demonstrated that MTLR provided equivalent or superior discrimination compared to other survival models and provided excellent model calibration in both low-and high-dimensional datasets [12,15,18].…”

“…For example, a chemotherapeutic treatment likely provides a proportionally greater survival effect around the time the treatment is administered, but longer-term survival may be more associated with the disease stage. Because MTLR is not bound by the proportional hazards assumption, the model can predict that patient P1 is more likely to survive 1 year than patient P2, but P2 is more likely to survive 5 years-the survival curves can cross [15]. We can also determine how the relative survival effects of a given variable change over time.…”