A covariate's effect can change over time, and differently for each patient.

In survival analysis, scientists study how different factors affect how long someone survives. Standard models assume these factors have the exact same effect over time and across different groups. But in reality, a factor's impact can change over time, differ between groups like patients or treatments, or even change over time in completely different ways for each group. The researchers built a new method using piecewise exponential additive mixed modelspiecewise exponential additive mixed modelsA statistical model that divides time into segments and assumes the risk of an event is constant within each segment, while also accounting for differences between individuals. to capture this complex behavior. Their system uses flexible curves called penalized splinespenalized splinesA flexible curve that is prevented from being too wiggly by a penalty, which helps avoid overfitting to noise. to map non-linear changes over time, along with random effects to handle differences between groups. A built-in penalty stops the model from overfitting when these complex patterns do not exist. It also helps solve the usual challenge of choosing time intervals. Tests on simulated data showed this new method outperforms competing options. The researchers also showed its real-world value using a brain tumor case study.

Why it matters

In medical studies, a treatment's effectiveness often changes over time and varies between patient groups. Existing models miss this combined variation, leading to inaccurate predictions. This work lets researchers capture these complex patterns, improving survival predictions for individual patients.

Questions answered