Selection of propensity score models and inference of causal efects in clinical studies

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K20227
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 60030:Statistical science-related
• Research Institution: Tokyo University of Science
• Principal Investigator: Ando Shuji 東京理科大学, 工学部情報工学科, 助教 (40803226)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 統計的因果推論 / 変数選択法 / モデル選択 / モデル選択

Outline of Final Research Achievements

In order to estimate causal effects without bias in clinical research, we evaluated and developed the following three methods for data analysis. (1) Investigation of the importance of the types of variables to be included in the propensity score model, (2) Development of a method for estimating a propensity score, and (3) Development of unbiased estimators of causal effects. The results of the research on issues (1) and (2) were presented at several conferences. For issue (3), the research results have been summarized in a paper and submitted to a journal (under review).

Free Research Field

統計科学

Academic Significance and Societal Importance of the Research Achievements

ランダムに処置の割り当てを行わない臨床研究では，因果効果を推定する際，交絡を調整する必要がある．交絡の調整法として，傾向スコア法がよく用いられている．交絡因子を特定できれば偏りなく因果効果を推定できるが，近年は，観測できる共変量の数が多くなり，交絡因子の特定がさらに難しくなっている．本研究では，観測されたデータから，交絡因子を特定することができる傾向スコアに対応した変数選択法と推定法を開発した．本研究の成果により，観測できる共変量の数が多い状況でも，偏りなく因果効果を推定できることになる．