A cautionary note on the naive use of general-population biobanks to study pulmonary arterial hypertension, with a focus on Mendelian randomization
File(s)
Published version
Author(s)
Type
Journal Article
Abstract
Pulmonary hypertension (PH) is defined by a mean pulmonary artery pressure >20 mmHg [1]. Patients with PH are assigned to one of five internationally recognised groups. Pulmonary arterial hypertension (PAH), or group 1 PH, is a heterogeneous collection of conditions characterised by increased precapillary pulmonary vascular resistance. Groups 2 to 5 PH comprise PH caused, in turn, by left heart disease, lung diseases (e.g. COPD), chronic thromboembolism, and miscellaneous causes such as haematological diseases. PAH is increasingly being investigated using data from biobanks with a general population sampling frame (figure 1a), as opposed to disease-specific cohorts. General population biobanks tend to define PAH with a single medical record code. Many studies use these data to perform Mendelian randomisation (MR; figure 1b). MR is a study design that uses genetic variants specifically associated with an exposure of interest to test causal claims [2, 3]. We demonstrate two issues with existing general population biobank PAH data: low power and non-random misclassification. These result in a failure to replicate findings from gold standard PAH datasets in general population biobanks, and spurious findings from general population biobanks that fail to replicate in gold standard datasets. Because PAH is rare, with a prevalence below 50 cases per million [1], population-based biobanks have few cases. Power in case-control studies is not substantially improved by increasing the case-to-control ratio beyond 1:4 [4]. Consequently, biobanks are typically less well powered than PAH-specific cohorts, despite having tens or hundreds of times more participants. Low power has two implications: true-positive associations are more likely to be missed, and detected associations have lower odds of being true. Indeed, a reported association between variants proxying IL6R signalling and PAH risk observed in an early release of FinnGen failed to replicate in much larger cohorts [5]. Meta-analyses can address low power. Researchers therefore meta-analysed three general-population biobanks (UK Biobank (UKB), FinnGen release 12, and Million Veteran Program (MVP)) with 3302 apparent PAH cases and 1 205 457 controls (https://mvp-ukbb.finngen.fi/pheno/I9_HYPTENSPUL) [6-8]. The prevalence in this meta-analysis is much higher than expected based on general population surveys. This might imply the presence of misclassification. Misclassification (when people categorised as cases do not have the condition, or people categorised as controls do) reduces power [9]. The RHODES et al. [10] genome-wide association study (GWAS) contained 2085 PAH cases with gold standard expert centre diagnoses, and 9659 controls. The FinnGen-UKB-MVP meta-analysis is theoretically better powered because it has more "cases" and controls. Failure to detect an association observed in RHODES et al. [10] in the FinnGen-UKB-MVP meta-analysis would suggest misclassification of cases in the general population biobanks. RHODES et al. [10] detected associations with three independent variants at genome-wide significance (p<5×10 −8): rs2856830 in the HLA-DPA1/DPB1 gene cluster, and rs13266183 and rs10103692 near the
Date Issued
2025-10-16
Date Acceptance
2025-08-29
Citation
European Respiratory Journal, 2025, 66 (4)
ISSN
0903-1936
Publisher
European Respiratory Society
Start Page
2500436
Journal / Book Title
European Respiratory Journal
Volume
66
Issue
4
Copyright Statement
Copyright ©The authors 2025. This version is distributed under the terms of the Creative Commons Attribution Licence 4.0.
License URL
Identifier
https://www.ncbi.nlm.nih.gov/pubmed/40967765
Publication Status
Published
Coverage Spatial
England
Article Number
2500436
Date Publish Online
2025-09-18