PREDICTION OF DIABETIC RETINOPATHY AMONG DIABETIC NEUROPATHY IN T2DM PATIENTS USING DATA MINING ALGORITHM

Abstract

Diabetic retinopathy (DR) and diabetic neuropathy (DN) are the most common complications among diabetes mellitus (DM) patients. Despite the widespread awareness, the implications of these serious diabetes complications remain poorly understood. Hence, this study aims to determine the association between DR and DN, predict DR and identify the significant risk factors associated with DR among DN patients based on the best predictive model obtained. Three models are employed in this study; Logistic Regression (LR) (Forward, Backward, Enter and Optimize), Decision Tree (Information Gain, Gini Index and Gain Ratio) and Artificial Neural Network with a splitting of 70-30. This study involved 361 T2DM patients who had undergone DM screening at the Ophthalmology Clinic, UiTM Medical Specialist Centre. Results of this study show that the prevalence of DR in individuals with DN was 1.75 times more than in individuals without DN. The LR (Optimize Evolutionary) is the best model for LR with accuracy=68.42% and AUC =0.423, compared to the other models; LR Forward (Accuracy=68.42%, AUC = 0.731), LR Backward ((Accuracy=57.89%, AUC=0.487) and LR Enter (Accuracy=57.89%, AUC =0.487). The DT Information Gain is the best model for the Decision Tree model (Accuracy=92.31%, AUC=0.667) compared to the DT Gini Index (Accuracy=92.31%, AUC=0.333) and DT Gain Ratio (Accuracy=84.62%, AUC=0.50). The ANN model gives an accuracy of 68.42% and ROC=0.50. Thus, the DT Information Gain is the best model to predict the presence of DR in T2DM patients with significance factors; duration of DM, Age, diastolic BP and BMI. The significance of this study can be applied globally to promote better health understanding in predicting the presence of DR among T2DM with DN patients and future prevention.

Keywords

• data mining
• diabetes complications
• diabetic neuropathy
• diabetic retinopathy
• risk factors