Improving RCT-Based Treatment Effect Estimation Under Covariate Mismatch via Calibrated Alignment

This paper addresses a practical problem in medical research: combining data from randomized controlled trials (RCTs, which are reliable but small) with larger observational studies (which have more data but less reliable) to better estimate how treatment effects vary across different patient types. The key challenge is that these two data sources measure different sets of patient characteristics, so the authors propose CALM, a method that translates both datasets into a common 'embedding space' and then uses the reliable RCT data to adjust predictions from the observational study, avoiding the need to guess missing information.