Uzunlamasına Veriler İle Sağkalım Verilerinin Birlikte Modellenmesinde Model Performansının Bilgi İçeriği Yaklaşımı İle Değerlendirilmesi

Abstract

Joint modeling is a frequently used method in clinical studies for analyzing the longitudinal and survival data simultaneously. In this method, the joint model is fitted by considering the relationship between two processes, and it enables all the measurements repeatedly taken to contribute to the joint model. Another advantage of joint modeling is estimating the subject-specific dynamic risk predictions. The results can be interpreted by considering dynamic risk predictions of the individual characteristics of each patient rather than overall estimations. The subject-specific dynamic predictions contribute to clinical decisions (changing treatment etc.) taken by experts for patients. Hence, the risk estimations obtained from the joint model should be reliable and accurate. In the joint modeling, the accuracy of dynamic predictions is evaluated using the measures such as the area under the ROC curve (AUC) and Brier score (BS). In this study, we propose a new approach based on the weighted mutual information (WMI) for evaluating the performance of joint modeling. This approach measures the amount of information gained by the risk predictions about the true status of patients. The performance of the proposed approach is evaluated using both the real data set and the simulation study. The results obtained showed that the WMI can be used together with other approaches for evaluating the performance of joint models. According to the results of the simulation study, it was seen that the predictive performance of the joint model increased as the sample size increased; however, increasing the number of repeated measurements taken did not contribute to the joint model after some points. Furthermore, with the help of WMI, we were able to find the time point(s) in the real data set which provided information the most and found that the late time points provided much information about the predictions made at time points far from the last longitudinal measurement. In conclusion, we showed that the WMI can be used together with AUC and BS for evaluating the prediction accuracy of the joint model, and using several criteria instead of a single criterion might be much informative while evaluating the predictive of joint the model.