OBJECTIVES: Survival analysis plays an important role in assessing the effectiveness of medical product/treatments and the risk factors.  Particularly, the accelerated failure time (AFT) model provides the hazard/survival functions to the later health economic decision model to compute the cost-effectiveness. The model estimations of the usual survival models rely on the maximum likelihood estimation (MLE), which works under the assumption of independence of observations. However, this assumption is not always satisfied, especially in the chronic/relapsing disease. A patient with a chronic disease may experience recurrent disease progressions in the life course.  When the progression free survivals (PFS) are recorded longitudinally, the PFS of different patients can be considered independent. Nevertheless, owing to sharing some unobserved heterogeneity, PFS of the same patients tend to associate with each other. This within-patient association can affect the estimation accuracy, therefore may misinform the decision makers. Particularly designed for the multivariate survival analysis, the frailty model takes this issue into account. METHODS: Fist, in a simulation study, we compared the AFT model and the Frailty model, where 5000 hypothetical patients are assigned to two treatment arms, and each patient experiences 5 treatment lines. Second, we apply both the Weibull AFT model and the Weibull-Gamma frailty model to the real life data, where 254 patients of chronic lymphocytic leukemia(CLL) have been followed, and we conduct a hypothesis test on the significance of frailty term. RESULTS: The simulation study shows that the estimates of the AFT model deviate far from the true values when the unobserved heterogeneity is large. The real life study indicates that the AFT model should be replaced by the frailty model due to the significance of the frailty term. CONCLUSIONS: In modelling survivals for chronic disease, the frailty models provide more accurate effect estimation than the conventional survival model.