New approaches of causal inference techniques in retrospective clinical studies applied in explanatory survival models.

C. M. Rodríguez Leal, R. Susi García, T. Pérez Pérez, M. Á. Luque Fernández

Randomised trials are considered the gold standard for studying causality in medicine. They are not always feasible, and observational studies sometimes are the only alternative. They are prone to confounding bias, so several methods can be used to adjust for confounding and to approximate the intervention effect as causal.
A common approach for causal analysis in time-to-event settings for observational data is an extension of the Cox regression model. The resulting hazard ratio (HR) can be interpreted as a causal association. However, it does not estimate the absolute risk modification over the study period. A more suitable approach is to use a generalization of standardization via the G-formula, giving information about the effect of interventions over the follow-up period.
Classical and causal inference methods are applied and contrasted using three real-world clinical studies using multiple regression to estimate adjusted HR and G-methods to produce standardized survival functions

Keywords: Causal inference, survival analysis, G-methods, standardization, Cox regression model

Biostatistics II
June 13, 2025  9:00 AM
MR 3

M. Durban Reguera, G. Camarda

J. M. Vera Verdezoto, P. R. Cano Ortiz, E. Gutiérrez Álvarez

R. Alonso Sanz, A. M. Franco Pereira, M. C. Pardo Llorente