Comparing Cox regression and parametric models in the analysis of factors influencing event time of neuropathy in patients with type 2 diabetes

Abstract

Background: Cox proportional hazard model is the most common method for analyzing the effects of several variables on survival time. However,  under certain circumstances, parametric models give more precise estimates to analyze survival data than Cox. The purpose of this study was to investigate the comparative performance of Cox and parametric models in a survival analysis of factors affecting the event time of neuropathy in patients with type 2 diabetes. Materials and Methods: This study included 371 patients with type 2 diabetes without neuropathy who were registered at Fereydunshahr diabetes clinic. Subjects were followed up for the development of neuropathy between 2006 to March 2016. To investigate the factors influencing the event time of neuropathy, significant variables in univariate model (P < 0.20) were entered into the multivariate Cox and parametric models (P < 0.05). In addition, Akaike information criterion (AIC) and area under ROC curves were used to evaluate the relative
goodness of fitted model and the efficiency of each procedure,  espectively. Statistical computing was performed using R software
version 3.2.3 (UNIX platforms, Windows and MacOS). Results: Using Kaplan–Meier, survival time of neuropathy was computed 76.6 ± 5 months after initial diagnosis of diabetes. After multivariate analysis of Cox and parametric models, ethnicity, high?density lipoprotein and family history of diabetes were identified as predictors of event time of neuropathy (P < 0.05). Conclusion: According to AIC, “log?normal” model with the lowest Akaike’s was the best?fitted model among Cox and parametric models.
According to the results of comparison of survival receiver operating  haracteristics curves, log?normal model was considered as the most efficient and fitted model.