Online Program Home
My Program

Abstract Details

Activity Number: 358 - Contributed Poster Presentations: Section on Statistics in Epidemiology
Type: Contributed
Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistics in Epidemiology
Abstract #303077
Title: Optimal Subclassification via Propensity Scores Using Graphical Presentations
Author(s): Eiji Nakatani* and Sho Komukai and Takanobu Nomura
Companies: Shizuoka General Hospital and Osaka University Graduate School of Medicine and Kyoto University Graduate School of Medicine
Keywords: Lorenz curve; Observational studies; Propensity score; ROC curve; Stratification; Subclassification

The subclassification based on the propensity score is a substantial inference procedure for estimating average causal effects in non-randomized studies. One of the practical problems is optimally determining the number of subclasses and their cutpoints. Ideally, 1) variation, 2) predictivity to the treatment assignment and,3) association with the outcome of the propensity scores needs to be optimally little in each subclass. Further, 4) the optimal balance of assignments in each subclass is needed. From their perspectives; we proposed a new methodology on propensity scores subclassification using graphical presentations [indexes], such as receiver operating characteristic (ROC) curve [area under the curve (AUC)] and Lorenz curve [Gini index]. Researchers can visually check the properties, such as the positivity assumption, to be satisfied with a particular subclass, which leads to an accurate estimate of the causal effect. We conduct simulations to compare with quantile method, and an application to our proposed method, based on right heart catheterization dataset, is provided.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program