Enhancing Survival Analysis of Contraceptive Use with Time-Varying Covariates: Evidence from Longitudinal Data

Statistical methodologies for medical and health research have changed significantly, bringing out the dynamics pertaining to disease progression and treatment outcomes. Methods for analyzing survival data help understand the changes in subjects over time, including assessing the time to event. However, the standard survival models assume only time-invariant relationships, ignoring the clinical variable’s time-varying nature, which is observed mostly in chronic diseases. This study addresses this limitation by creating a time-varying covariate model for longitudinal data. The model incorporates a shared random-effects structure for longitudinal and survival components, facilitating correlation between the two processes. We develop a Cox proportional hazard model that incorporates a time-varying covariate, validate its applicability to data, and compare the predictive accuracy of the proposed model to the standard model. Existing data from Performance Monitoring for Action (PMA) was used to compare the performance of the standard Cox model with the time-varying covariate model. Key findings indicate that age, education, intention of using contraception in the future, and method switching significantly influence the risk of discontinuation in contraceptive use. The time-varying model shows the best prediction values based on AIC, BIC, and the concordance index, demonstrating the advantage of employing such models in reproductive health studies. Informed by the policy implications, there should be some strategies targeting younger, educated women, as well as those who will use contraceptives in the future. Other than improving the methodologies in the field of survival analysis, this study introduces a more accurate and adaptable framework for clinical forecasting, ultimately leading to improved treatment outcomes.