OBJECTIVES: Commercially-available real-world data (RWD) in the US are often sourced from employer-sponsored insurance plans or major provider networks, leading to underrepresentation of patients with lower socioeconomic status. InsightsDB, from the Institute for Health Metrics, collates electronic medical record (EMR) data from ~100 community hospitals (~85% governmental or voluntary non-profit) across 30 states with linkage to social determinants of health (SDOH) for over 5 million unique patients.

METHODS: We explored strengths and limitations of InsightsDB by replicating aspects of published studies conducted in widely-used databases. Where methodology could be adequately reproduced, we investigated whether differences in outcomes were associated with SDOH.

RESULTS: Use of recently-launched medications was less common in patients in InsightsDB compared to other databases. For example, a published study (Pantalone 2023; DOI: 10.2337/ds22-0064) found a ratio of ~1:34 patients taking a novel combination therapy for type 2 diabetes (T2D) vs a standard basal-bolus insulin regimen; the ratio in InsightsDB was 1:755. Although it was technically feasible to replicate most aspects of Pantalone 2023, limited data for certain variables (e.g. diagnoses or lab tests, which may have been performed outside the hospital setting) led to significantly higher attrition, e.g. only 2.6% of patients initiating basal-bolus insulin had a valid HbA1c reading vs 35.7% in Pantalone 2023. Conversely, strengths of InsightsDB included data on use of patient’s own medications (e.g. over-the-counter or prescribed outside the hospital setting), as well as non-formulary medications, that may not be visible in claims data.

CONCLUSIONS: Differences in treatment access and outcomes in the community hospital population vs patients in widely-used databases highlight the need for inclusion of underrepresented populations in real-world evidence (RWE) studies; data from community hospitals can fill some evidence gaps but studies should be carefully designed to maximize strengths and overcome limitations of hospital-level data.