Survival analysis in clustered data

Aim: Survival analysis in clustered data occurs when survival times are sampled in clusters. Thus, the survival times within the same cluster are correlated. The aim of this study is to compare advanced statistical analysis methods used in the analysis of clustered survival data with classical survival analysis. Material and Method: The simulation studies were designed to include two explanatory variables. Parameter estimates were obtained with the data produced in different sample sizes and different cluster sizes. In the study, model performances were compared for three different situations. The frailty, stratified, marginal Cox, marginal Weibull and classical Cox model were applied to the data obtained by simulating and their performances were mixed according to the akaie information criterion (AIC). Results: The results of the simulation scenario made in the study were found to be similar to each other in all three cases. In cases where the number of observations is small, the model with the worst performance was the marginal weibull. However, as the number of observations increases, the classical Cox model has the worst performance. In addition, the best performing model in all cases was obtained as the frailty model. Conclusion: In the survival analysis of clustered data, ignoring intra-cluster dependency may not always be biased enough to result in false statistical results. However, this bias is likely to increase as the number of observations increases. Therefore, it is recommended to use analysis methods that consider intra cluster correlation in the analysis of clustered survival data.