(B-203) Aggregating and harmonizing registry databases for comparative analyses – lessons learnt

Background: Pulmonary arterial hypertension (PAH) is a rare disease with limited real-world evidence. Aggregating multiple fit-for-purpose data sources into one database has potential to mitigate this.

Objectives: This study aimed to evaluate the characteristics of PAH patients undergoing different treatment regimens and assess the feasibility of a comparative effectiveness analysis of PAH triple combination (target cohort) relative to double combination therapy (comparator cohort).

Methods: We aggregated data from four heterogeneous studies, involving adult patients diagnosed with PAH who initiated PAH-specific treatment between October 2013 and November 2021. An index date was determined for the target cohort as the 3rd drug add-on date, while for the comparator, it was the date of reaching an intermediate/high risk status. We employed a logistic regression for propensity-score (PS) modeling.

Results: Both cohorts (target n=521 and comparator n=1217) had mostly similar demographic and clinical characteristics. However, the target cohort had a substantially longer time from diagnosis to index (25.8 vs 9.6 months), i.e., when triple therapy started for the target cohort or should have been started for the comparator cohort, as per treatment guidelines. In addition, geographic distribution differed by cohort.
After matching by several strategies, residual imbalance persisted (absolute SMD>0.1), including time from PAH diagnosis to index, which is understood to be associated with study outcomes (e.g., time to disease worsening) making it a plausible confounder. Further, baseline conditions remained imbalanced after matching making confounding by initial health status a plausible threat to validity. Hence, the comparative effectiveness analysis was not carried forward due to high risk of bias as empirically demonstrated.

Conclusions: Although the clinically rich aggregated PAH database provided insights into the characteristics of PAH patients treated with different regimens, it appeared infeasible for PS matching to create adequately exchangeable exposure cohorts for valid comparative analyses. Post-hoc assessment suggested that our main limitations were difficulty with designing an appropriate index date for the comparator cohort, and differences between patient populations, especially in terms of time between diagnosis and treatment initiation.
This is an example where – despite in-depth study design comparisons, database evaluations, and initial fit-for-purpose assessments – comparative effectiveness analysis was deemed infeasible. The study highlights the importance of the PS matching diagnostics evaluation in comparative effectiveness research, in particular exemplifying an informed decision making to prevent biased results.