Premium
Use of Real‐World Data to Emulate a Clinical Trial and Support Regulatory Decision Making: Assessing the Impact of Temporality, Comparator Choice, and Method of Adjustment
Author(s) -
Abrahami Devin,
Pradhan Richeek,
Yin Hui,
Honig Peter,
Baumfeld Andre Elodie,
Azoulay Laurent
Publication year - 2021
Publication title -
clinical pharmacology and therapeutics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.941
H-Index - 188
eISSN - 1532-6535
pISSN - 0009-9236
DOI - 10.1002/cpt.2012
Subject(s) - medicine , liraglutide , confidence interval , hazard ratio , propensity score matching , clinical trial , type 2 diabetes , diabetes mellitus , endocrinology
External controls have been primarily used in the setting of single‐arm trials of rare diseases; their use in common diseases has not been readily investigated, nor is there guidance on how to best select comparators. Thus, the objective of this study was to emulate a large cardiovascular outcome trial of type 2 diabetes to compare associations of effectiveness with different comparator groups to those reported in the trial. Using the Liraglutide Effect and Action in Diabetes: Evaluation of Cardiovascular Outcome Results (LEADER) trial, we investigated six comparator groups using three calendar time periods (Early: 1999–2003; Later: 2004–2008, and Contemporaneous: 2009–2013) and two comparators (sulfonylureas and other second‐to‐third‐line antidiabetic drugs). Hazard ratios (HRs) of the three‐point composite cardiovascular outcome were estimated using four variations of the propensity score (adjustment, stratification, fine stratification, and matching) and compared with the LEADER trial (HR, 0.87; 95% confidence interval, 0.78–0.97). When comparing users of liraglutide with users of sulfonylureas, the HRs ranged from 0.57 to 1.03, with estimates in the early period most closely reflecting the LEADER trial (HR, 0.57–0.88). In contrast, the HRs ranged from 0.73 to 0.97 when comparing liraglutide users with users of any second‐to‐third‐line antidiabetic drugs, although the later period generated estimates closest to the LEADER trial (HR, 0.77–0.84). Different methods of adjustment led to generally consistent HRs, aside from the fine stratification in the early period. This study highlights the complex interplay between comparator, temporality, and method of adjustment when selecting comparators using real‐word data. These design choices must be considered in the design of trial emulation studies.