Talk by Russ Greiner, University of Alberta

Hosts: Siamak Ravanbakhsh and Mark Schmidt

Title: An Effective Way to Estimate an Individual's Survival Distribution

Abstract: The 'survival prediction' task requires learning a model that can es timate the time until an event will happen for an instance; this differs from standard regression problems as the training survival dataset may inc lude many 'censored instances', which specify only a lower bound on that instance's true survival time. This framework is surprisingly common, as it includes many real-world situations, such as estimating the time untila customer defaults on a loan, until a game player advances to the next level, until a mechanic device breaks, and customer churn. This presenta tion focuses the most common situation: estimating the time until a patien t dies. An accurate model of a patient’s individual survival distribu tion can help determine the appropriate treatment and care of terminal pat ients. The common practice of estimating such survival distributions uses only population averages for (say) the site and stage of cancer; however , this is not very precise, as it ignores many important individual diffe rences among patients. This paper describes a novel technique, PSSP (pati ent-specific survival prediction), for estimating a patient’s individual survival curve, based on the characteristics of that specific patient, u sing a model that was learned from earlier patients. We describe how PSSP works, and explain how PSSP differs from the more standard tools for surv ival analysis (Kaplan-Meier, Cox Proportional Hazard, etc). We also showthat PSSP is 'calibrated', which means that its probabilistic estimates are meaningful. Finally, we demonstrate, over many real-world datasets (various cancers, and liver transplantation), that PSSP provides survivalestimates that are helpful for patients, clinicians and researchers.

Link to Bio: https://docs.google.com/document/d/1U72jjpomVwbsC3ydFaqQfrizVnSgB9pBRdmMUuhi4BQ/pub