(128) The State of Use and Utility of Negative Controls in Pharmacoepidemiologic Studies

Publication Number: 459

Background: Pharmacoepidemiologic studies often involve use of real-world data, which are challenged by various sources of bias, including confounding, selection bias, and information bias. Use of negative controls to address bias, while not a new concept, has recently gained more attraction in drug safety and effectiveness studies. Nevertheless, to what extent these methods have been applied in pharmacoepidemiology studies is not well known.

Objectives: To evaluate the state of use and utility of negative controls in pharmacoepidemiologic studies.

Methods: We searched PubMed, EMBASE, CINAHL, Cochrane Library, Scopus, and Dissertations and Theses Global for the published literature through September 2020 on drug safety and effectiveness studies that deployed negative controls to address potential bias. Two reviewers independently assessed study eligibility and abstracted data. We summarized the studies based on the utility of negative controls and identified the approaches for testing the underlying assumptions of negative controls.

Results: We identified 177 eligible articles. Most studies were published after 2013 (167, 94%). Cohort study accounted for 64% (114), followed by case-control studies (18, 10%) and self-control case series (10, 6%). Administrative data (111, 63%) and electronic health records (60, 34%) were the most common data sources used in these studies. Ninety (51%) studies used negative control outcome only, 55 (31%) used negative control exposure only, and 28 (16%) used both. Negative controls were applied mainly to address unmeasured confounding (88, 50%), with fewer addressing selection bias (8, 5%), information bias (6, 3%), or competition bias (1, 1%). We identified four domains of the utility of negative controls: 1) detection of bias (173, 98%), 2) correction of bias with regard to point estimate or confidence interval calibration (15, 8%), 3) p-value calibration (12, 7%), and 4) performance comparison of different analytical methods used in drug safety and effectiveness studies (23, 13%). We summarized the systematic steps to examine the underlying assumptions for negative controls.

Conclusions: The use of negative controls in pharmacoepidemiologic studies are mainly to detect bias due to unmeasured confounding. Lacking in many studies was empirical tests of the underlying assumptions. Negative controls are a powerful method but to ensure rigorous and widespread application, future work is needed to empirically test underlying assumptions or explore innovative approaches with less restrictive assumptions.