Revisiting methods for modeling longitudinal and survival data: Framingham Heart Study

Abstract Background Statistical methods for modeling longitudinal and time-to-event data has received much attention in medical research is becoming increasingly useful. In clinical studies, such as cancer AIDS, biomarkers are used to monitor disease progression predict survival. These measures often missing at failure times may be prone measurement errors. More importantly, time-dependent survival models that include the raw measurements lead biased results. previous studies these two types of frequently analyzed separately where a mixed effects model applied event outcome. Methods this paper we compare joint maximum likelihood methods, two-step approach time dependent covariate method link with emphasis on using We apply Bayesian semi-parametric maximizes measures. also implement Two-Step approach, which estimates random separately, classic Time Dependent Covariate Model. use simulation assess bias, accuracy, coverage probabilities parameter connects times. Results Simulation results demonstrate performed best estimating when variability measure low but somewhat downwards high. yield higher high measure. The resulted consistent underestimation parameter. illustrate from Framingham Heart Study lipid Myocardial Infarction were collected over period 26 years. Conclusions Traditional data, method, observed tend provide downwardly estimates. better estimates, although comparison depend underlying residual variance.