Senior Researcher PHARMO Institute for Drug Outcomes Research Utrecht, Netherlands
Background: Multi-database real-world studies may provide heterogeneous results leading to challenges in interpretation. This heterogeneity can also be a source of information on populations or healthcare systems
Objectives: To explore statistical heterogeneity and its possible sources in results reported in the EXACOS-CV multi-database study.
Methods: Four studies conducted in Germany, Canada, the Netherlands (NL) and Spain quantified the association between time periods following exacerbations of COPD (ECOPD) and the risk of non-fatal severe cardiovascular events or death. A common protocol was tailored to each country database. Heterogeneity between results was explored using the between-country variance estimated by the restricted maximum likelihood method. For results with noticeable statistical heterogeneity, two possible sources of heterogeneity were evaluated: clinical (participants, exposure and outcomes) and methodological diversity (design, risk of bias, definitions and data availability).
Results: Baseline characteristics across countries differed mainly in terms of sex (male, range 52% in Canada, 78% in Spain), prevalence of cardiac disease (e.g. hypertensive disease, range 37% in NL, 75% in Germany), and cardiac medication (range 47% in NL, 91% in Spain). In all countries, the risk of a CV event or death increased steeply in 1-7 days following an AECOPD and decreased during the 12 following months. However, between-country variance was observed for the 1-7 day period, the greatest being for heart failure (HF) decompensation (variance 2.3; 95%CI 0.7-32.3) with hazard ratios (HRs): Germany (HR 2.6; 95% CI 2.3-2.9), Spain (HR 5.1; 95% CI 4.4-6.0), NL (HR 27.4; 95% CI 17.3-43.4), and Canada (HR 72.3; 95% CI 64.4-81.2). Notably, risk of mortality in the 1-7 day period appeared higher in Germany (HR 18.1; 95%CI 17.3-19.0) than in Canada (HR 10.0; 95%CI 9.5- 10.6) suggesting that in Germany, the risk of mortality was higher than the risk of HF decompensation. Possible sources of heterogeneity include: differences in the distribution of possible effect modifiers (e.g., cardiac comorbidity), competing risk of death, hospitals coding practices (measurement bias), misdiagnosis (misclassification) and confounding.
Conclusions: Although the results were consistent across countries in terms of direction of associations, we have observed differences in absolute estimates and evaluated possible sources to explain these differences. Distributions of patients’ characteristics differed widely, partly driven by the database (bias), partly by local population, disease diagnosis and management practices (“true” heterogeneity), the latter providing insight on healthcare system differences.
