This technology is a software-based causal survival analysis and model-ranking workflow for estimating treatment effects to guide model or experiment selection from time-to-event cohort data.
Commonly used survival regression methods and model-selection approaches can produce biased or unstable conclusions from confounding, censoring, model misspecification, or limited validation. Reliable approaches are needed that estimate treatment benefit or harm from human cohort data when outcomes are time-to-event or follow-up is incomplete. Inaccurate treatment-effect estimates and poorly chosen mechanistic models can misdirect clinical interpretation, research priorities, and future experiment selection.
This technology is a software-based workflow for analyzing time-to-event cohort data, including datasets with right-censored outcomes. It applies causal survival analysis methods to account for measured confounding and incomplete follow-up, enabling treatment-effect estimation over time with reduced sensitivity to model misspecification. The workflow also ranks candidate mechanistic models based on out-of-sample prediction of a common measurable endpoint, rather than relying primarily on in-sample fit. It can further prioritize future experiments by balancing expected information gain against practical constraints, with validation performed on censored survival datasets using out-of-sample performance metrics. This provides a more objective and efficient basis for estimating treatment effects, selecting models, and planning follow-up studies.
IR CU27024
Licensing Contact: Joan Martinez
