(ASPOT07) Exploring sources of statistical heterogeneity in a multi-database observational study: results from a case study using EXACOS-CV

Background: The EXACOS-CV multi-country programme is a set of observational cohort studies that used a common protocol to characterize the association between time periods following exacerbations of COPD and severe cardiovascular (CV) events. All studies consistently showed that the risk of severe CV events is substantially increased for up to 1 year following an exacerbation. However, statistical heterogeneity across studies was observed in the strength of association.

Objectives: To identify and explore sources of heterogeneity among EXACOS-CV studies.

Methods: EXACOS-CV studies conducted in Germany, Canada, the Netherlands, and Spain deviated least from the common protocol and were included. Focus was on adjusted hazard ratio estimates for post-exacerbation risk of decompensated heart failure (HF) and all-cause death, which represent the most and least heterogeneous sets of results. The conceptual framework differentiated methodological (database-related) diversity from true variation. For the former, a checklist of possible concerns was created and completed by the research team. Ways to explore and rule-out methodological diversity were proposed. In turn, hypotheses about true heterogeneity sources were generated.

Results: Adjusted hazard ratios (aHR) for the risk of decompensated HF were most heterogeneous during the first 7 days following an exacerbation; aHRs ranged from 2.6 (95%CI, 2.3 - 2.9) in Germany to 72.3 (64.4 - 81.2) in Canada. Differences across studies in ability to adjust for all key confounders was a concern for heterogeneity. Applying a common adjusted model per study reduced between-study variance among aHRs by 33% from 2.9 to 1.9 but heterogeneity remained substantial. Levels of comorbidity differed across studies; in particular, baseline HF was most prevalent in Germany at 21% and Spain at 18% (vs. 7% and 9.4% in the Netherlands and Canada, respectively). Concerns about overdiagnosis of baseline HF in Germany were not supported by the literature. In Germany and Spain, the risk of all-cause death during 1-7 days post-exacerbation was higher than the risk of HF decompensation, showing differences across studies in competing risk of death.

Conclusions: Heterogeneity in risk of decompensated HF mainly concerns the first week following an exacerbation. One hypothesis is that this heterogeneity is driven by misdiagnosis – HF and COPD exacerbations share similar symptoms. This highlights differences in patient management, or physicians’ ability to detect HF decompensation in COPD patients. Although prone to statistical heterogeneity, multi-database studies can provide directional insights on the study research question whilst considering healthcare system and population differences.