Target Population Statistical Inference With Data Integration Across Multiple Sources—An Approach to Mitigate Information Shortage in Rare Disease Clinical Trials
A major challenge for rare disease clinical trials is the limited amount of available information for making robust statistical inference. While external data present information integration opportunities to enhance statistical inference, conventional data combining methods, for example, meta-analysis, usually do not adequately address study population differences. Matching methods, on the other hand, directly account for population characteristics but often lead to inefficient use of data by underutilizing unmatched data points. Aiming at a better bias-variance tradeoff, we propose an intuitive integrated inference framework to borrow information from all relevant data sources and make inference on the response of interest over a target population precisely characterized by the joint distribution of baseline covariates. The method is easily implemented and can be complemented by modern statistical learning or machine learning tools. Statistical inference is facilitated by the bootstrap. We argue that the integrated inference framework not only provides an intuitive and coherent perspective for a variety of clinical trial inference problems but also has broad application areas in clinical trial settings and beyond, as a quantitative data integration tool for making robust inference in a target population precise manner for policy and decision makers.