This technology is a computational framework that uses observational clinical data to estimate how treatment effects on patient survival change over time, supporting comparative effectiveness research and personalized treatment decisions.
Observational clinical datasets are a valuable source of evidence on real-world treatment effectiveness, but treatment is not randomly assigned and follow-up is often incomplete due to right censoring. Conventional survival analyses typically summarize a treatment’s effect with a single metric, such as a hazard ratio, or assume its relative benefit stays constant over time. These simplifications can obscure how treatment benefit changes over time and mask differences across patient subgroups, limiting their value for precision medicine, treatment sequencing, and comparative effectiveness research. There is a need for robust analytical methods that account for these limitations and accurately characterize how treatment effects vary over time across patient populations.
This technology is a computational framework that uses observational clinical data to estimate how a treatment’s effect on survival changes continuously over time. It models treatment assignment using propensity scores, then combines causal survival forests with doubly robust estimation and censoring adjustment to generate treatment-effect estimates across a range of follow-up times. Parametric and spline-based smoothing methods convert these estimates into a continuous treatment-effect trajectory, from which the framework derives interpretable features such as time of peak effect, maximum effect magnitude and treatment-effect half-life. It can also characterize how treatment effects vary across patient subgroups and applies multiple statistical checks to assess the robustness of its causal estimates.
This technology has been validated using clinical data from 2,651 head and neck squamous cell carcinoma patients drawn from the publicly available RADCURE dataset.
IR CU27023
Licensing Contact: Joan Martinez