Multivariable Confounding Adjustment in Distributed Data Networks without Sharing of Patient-Level Data

This article describes two approaches to conducting multivariable analyses for multi-site studies without sharing of patient-level data across sites: the case-centered logistic regression approach and the inverse variance-weighted meta-analysis approach.  The authors evaluate the above methods 1) by analyzing risks of angioedema associated with use of angiotensin-converting enzyme inhibitors, antiotensin II receptor blockers, and aliskiren, compared to use of beta-blockers in the Mini-Sentinel Distributed Database (MSDD), and 2) by performing simulations.