Perbandingan Metode Propensity Score Matching- Support Vector Machine dan Propensity Score Matching-Regresi Logistik Biner Pada Kasus HIV/AIDS

Abstract

In the health field, Randomized Controlled Trials (RCT) cannot be done because it relates to human life. The covariates variable as well as the presence of confounding variable in non-experimental research are usually not balanced between the treatment and control groups that cause the estimated treatment effect to be biased. One of the appropriate methods to overcome the biased treatment effect is the Propensity Score Matching (PSM). This study will compare the performance of the Propensity Score Matching-Support Vector Machine (PSM-SVM) method and the Propensity Score Matching-Binary Logistic Regression (PSM-RLB) method in cases of HIV/AIDS opportunistic infection. The confounding variable used was the opportunistic infection variable. The data used were data of HIV/AIDS patients treated at the Grati Public Health Center, Pasuruan Regency in 2016. The results showed that the variable of ARV therapy had a significant effect on opportunistic infections in HIV/AIDS patients. If we look at how much bias can be reduced, PSM-SVM is able to reduce bias more than the PSM-RLB method, which is 60.25%. However, the PSM-SVM method can produce a bias value (after matching) that is smaller than the PSM-RLB method, which is 0.044.