Propensity Score Matching Using Support Vector Machine in Case of Type 2 Diabetes Mellitus (DM)

摘要

Randomization in the treatment and control group was not appropriate for non-experimental studies because it will produced bias estimation of treatment effects. In addition to randomization, the presence of confounding variables will also produce bias estimation of treatment effect. This bias estimation of treatment effect can be handled using Propensity Score (PS) method. One of the methods that have been developed from the propensity score is Propensity Score Matching (PSM). In this study, the propensity score is estimated using Support Vector Machine (SVM). Confounding variables that used in this study is exercise activities. The purpose of this study is to apply the PSM using SVM method and calculate the accuracy and Percent Bias Reduction (PBR) on type 2 Diabetes mellitus (DM) disease complications case. The data used in this study is type 2 Diabetes Mellitus (DM) patients data treated at Pasuruan regional public hospital on March 2017. The results of PSM-SVM analysis shows that there are 40 of 96 patients with type 2 DM who have enough exercise activities paired with patients who have less exercise activities. Average Treatment of Treated (ATT) estimation result shows that exercise activity variables (Z) has significant effect on disease complication variables (Y). The accuracy of the PSM-SVM method is 70.00% and 16.65% of the bias can be reduced.