(064) Real-world evidence generation methods used to estimate fracture risk with drugs to treat type 2 diabetes mellitus: a scoping review

Background: Population-based studies provide real-world evidence (RWE) about the long-term effectiveness and rare side effects of drugs. They complement randomized controlled trials for evidence-based regulatory and clinical decision making. Careful consideration must be given since methodological decisions can impact the validity of effect estimates.

Objectives: To summarize the RWE methods employed to estimate the effect of antidiabetic medications on fracture risk.

Methods: We searched MEDLINE (OVID), Embase (OVID), CINAHL (EBSCO) scientific journal databases for full-text studies of adults, published in English between January 2000 and January 2023. Studies were included if the primary exposure was at least one antidiabetic drug, the primary outcome was fracture, and both exposure and outcome were defined using a healthcare administrative data source. Data abstraction included study design, exposure and outcome measurement, analytical approaches, and effect estimate (null, decreased, or increased fracture risk). We were specifically interested to document if authors justified decisions for observation windows (induction, exposure risk window, and washout), and how authors handled time trends.

Results: We identified 50 eligible studies (40 cohort, 9 case-control, 1 self-control case series). Most cohort studies (60%) and all case-control studies defined medication exposure over time using an intention-to-treat approach. Only one cohort study considered time trends to account for drug availability and coverage. The expected duration between the start of exposure and the possible occurrence of the outcome was not clearly reported in 60% of studies, and few studies (12% of cohort and no case-control) provided justification for a biologically plausible risk window. More than half of the studies (54%) were not clear in defining how fractures were measured, and only 20% referenced a validation paper to verify the fracture coding in the data source leveraged. Inconsistent effect estimates were observed, where almost half of the studies (49%) reported an increased risk of fracture from the exposure medication, 13% reported a decrease in fracture and 39% reported no significant association.

Conclusions: The conflicting evidence regarding fracture risk associated with antidiabetic drug exposure in observational studies may relate in part to heterogeneity in study design and analytical approaches. Few studies considered biological plausibility in their design or considered drug escalation or switching. Proper methodological design and transparency in reporting are needed.