|
Abstract:
|
Evidence derived from existing health care data can augment knowledge gained from randomized trials when trials are lacking, and help inform clinical decision making. Observational research is often criticized with many claiming it cannot be used to estimate causal treatment effects. The main reason cited is confounding, but other concerns include p-hacking and publication bias. We seek to address these concerns by formulating a new approach to observational research. Informaticists, statisticians and clinicians from the Observational Health Data Sciences and Informatics collaborative developed the Large-Scale Evidence Generation and Evaluation across a Network of Databases (LEGEND) initiative that aims to generate evidence on the effects of medical interventions using observational data while addressing the above concerns by following a recently proposed paradigm. We define ten guiding principles of LEGEND that enshrine this new paradigm. We generate evidence at large scale in a network of databases using propensity scores, negative and positive control questions, empirical calibration, full transparency, and other best practices.
|
