Using Parametric and Cox Models in Analysis of Factors Influencing the Diagnosis of Retinopathy in Type II Diabetes

Background and One of the major complications of diabetes is retinopathy which is a highly specific vascular complication of type I and II diabetes. This study aimed at investigating the factors influencing the diagnosis of retinopathy in patients with type II diabetes using Cox, parametric, exponential, weibull, Log normal, and Log-logistic models and comparing their functions. This prospective cohort study included 400 patients with type II diabetes without retinopathy who were registered at Yazd Diabetes Research Center in 2008. They were followed up for diagnosis of retinopathy until February 2013. Significant variables in the univariate model for investigating the effective factors on diagnosis of retinopathy were entered in Cox multivariate and parametric models. The criterion to compare the performance of models was Akaike information criterion. Data was then analyzed in R software and 0.05 was considered as the significance level. The mean and median periods to diagnose retinopathy were 46 and 58 months. Among the patients 3% were diagnosed with retinopathy in less than one year and in 16% the diagnosis was made in less than two years. Some variables including anemia, taking aspirin, duration of diabetes, and using insulin were amongst the main risk factors for diagnosis of retinopathy (P<0.05). According to Akaike information criterion we found that Cox model best fits for determining factors influencing the diagnosis of retinopathy.