Prediction Model for Disease Progression and Relevant Prognostic Factors in Hodgkin Patients: An Application of Frailty Multistate Model

Abstract

Background: In clinical cancer studies, there has been a high tendency of searching for more specific and new prognostic factors in cancers in the last few years. This multistate study aimed to model the progression of Hodgkin's disease by accounting for individual effect (heterogeneity) using the joint and independent frailty models. Method: After the utilization of the illness-death model, joint-modeling accounted for the dependency between relapse and death by considering the individual characteristics as a frailty term. Therefore, the effect of influential prognostic factors was evaluated on disease progression by frailty and joint-frailty multistate models. Results: The individual predictions were determined using the frameworks of the both models. The model was applied to 389 Hodgkin lymphoma patients. Gender (male), age (over 55 years), and low level of hemoglobin (less than 10.5) were associated with an increased risk of death and relapse in patients. The likelihood cross-validation criterion was proposed to choose the joint frailty model as a better fitting model. Conclusion: Multistate models were appropriate tools to study the whole event history of the subjects, which provided a deep insight into the dynamics of the disease. The problem of events-subjects dependency in the survival data was clarified using the multistate model. Therefore, the heterogeneity and dependency between the states led to more accurate estimations of the effects of the prognostic factors, thereby improving the predictions.