A time-varying covariate model for survival Analysis using longitudinal data

Abstract

Statistical methodologies for medical and health science research have changed significantly, bringing out the dynamics of disease progression and treatment outcomes. Methodologies for analyzing survival data help understand the changes in subjects over time, including the assessment of time to event. However, the standard survival models assume only time-invariant relationships, ignoring the changing character of reproductive indicators. This research introduces a modified approach to studying survival analysis, addressing the limitation by developing a time-varying covariate survival analysis model within a longitudinal data modeling framework. A Cox proportional hazards model that incorporates time-varying covariates was developed and validated. A comparison of the predictive accuracy of the modified model with that of the standard model was made. The model was validated using secondary data from Performance Monitoring and Account ability (PMA) data, Kenya. Several key covariates in the traditional model exhibited time-dependent effects, undermining the assumption of constant hazard. The modified model addresses this limitation by incorporating a shared random-effects structure for longitudinal data, which accommodates time-varying effects and yields enhanced and more accurate hazard estimation over time. The modified model proved better than the traditional Cox model at achieving predictive accuracy outcome measures. The modified model increased fitting statistics by reaching new AIC levels of 237,513.6 and BIC levels of 237,573.9 over traditional AIC (238,047.5) and BIC (238,085.2) statistics. Additionally, it displayed a 1.18 percent superior predictive capability through a C-index of 0.858 versus the traditional model’s 0.848 index. The study brings new techniques to survival analysis research while delivering practical recommendations for reproductive health service development through time-varying variable integration. Other than an improvement of methodology in the field of survival analysis, this model produces a more accurate and adaptable framework for clinical forecasting, ultimately leading to improved treatment outcomes. In line with this study, future research might seek to incorporate time series models into the survival analysis approach and dynamic prediction models that merge repeated measures with survival results to obtain more personalized predictions.