Key Takeaways
- Heritability estimates differed across biobanks: SNP-based heritability for atherosclerotic cardiovascular disease varied significantly between UK Biobank, Geisinger MyCode and the NIH All of Us Research Program.
- Underlying genetic effects remained highly similar: Genetic correlations across the three cohorts were consistently high, suggesting that the differences were unlikely to reflect fundamentally different genetic architecture.
- Phenotype capture matters: Differences in healthcare systems, EHR completeness and disease ascertainment may influence genetic estimates, reinforcing the importance of harmonized and well-characterized biobank data.
Large population biobanks are increasingly combined or compared to improve statistical power and validate genetic findings across different populations and healthcare systems. But a new study demonstrates why apparently comparable phenotypes may not always behave the same way across biobank datasets.
Researchers from Geisinger compared atherosclerotic cardiovascular disease (ASCVD) genetics across Geisinger’s MyCode Community Health Initiative, UK Biobank and the NIH All of Us Research Program, using genotype and electronic health record data from nearly one million participants.
The open-access study, published August 28 in the Journal of Human Genetics, examined SNP-based heritability and genetic correlation for ASCVD and two increasingly specific subphenotypes: coronary events and myocardial infarction.
The researchers found that heritability estimates differed significantly among the biobanks. UK Biobank generally produced higher estimates, while all three cohorts showed higher heritability for the narrowly defined myocardial infarction phenotype than for the broader ASCVD definition.
Yet when the researchers compared the genetic effects themselves, genetic correlations between the biobanks remained consistently high and were not significantly different from 1.0. This suggests that the common genetic contribution to cardiovascular disease was broadly similar across the cohorts despite the differing heritability estimates.
Differences in How Biobanks Capture Disease
The study points to phenotype ascertainment and clinical data capture as possible explanations for the discrepancy.
UK Biobank follows participants over time through linked healthcare records, including hospital inpatient records and other health-related datasets. MyCode draws longitudinal clinical information primarily from Geisinger’s integrated healthcare system, while All of Us receives EHR data from healthcare provider organizations and other participant-authorized sources.
These differences can affect whether clinical encounters and diagnoses are captured consistently. Missing events or differences in how cases are defined can result in phenotype misclassification, potentially lowering estimated heritability even when the underlying genetic effects are similar.
The challenge becomes particularly important as biobanks increasingly function not simply as collections of biological specimens, but as integrated genomic and clinical data resources. Biobanking.com has previously explored this broader transition in its expert article on Biobanking 4.0 and biospecimen-driven digital systems.
The researchers also tested part of the ascertainment explanation by restricting MyCode and All of Us to inpatient diagnoses to more closely resemble the UK Biobank approach used in the analysis. The overall pattern remained, and heritability estimates generally became lower under the inpatient-only definitions—illustrating the trade-off between phenotype specificity and case ascertainment.
Why It Matters for Biobanks
The findings have implications well beyond cardiovascular genetics.
As researchers increasingly conduct analyses across multiple large biobanks, harmonizing genomic variants alone may not be sufficient. Phenotype definitions, EHR coverage, healthcare-system structure and data completeness can all influence downstream results.
This makes the provenance and quality of associated clinical data an increasingly important component of biobank quality.
Biobanking.com recently covered the Phenotype-Genotype Reference Map (PGRM), a framework for assessing biobank data quality. The PGRM evaluates whether established genotype-phenotype associations can be reproduced in large DNA biobanks.
The new cross-biobank study highlights the same broader challenge from another direction: reliable genomic research depends not only on the quality of DNA and sequencing data, but also on how accurately and consistently phenotypes are represented.
For biobanks seeking to make their datasets interoperable and useful for large-scale collaborative studies, this reinforces the importance of standardized phenotype definitions, transparent metadata and clear documentation of how clinical information is obtained.
Interoperability is already becoming a central issue in distributed biobanking infrastructure. For example, Biobanking.com previously reported on the BBMRI-ERIC Federated Platform, which enables researchers to search distributed biobank samples and data and construct virtual cohorts without requiring all information to be centralized.
As these types of federated and cross-biobank analyses become more common, the ability to understand how phenotypes were generated in each contributing dataset becomes increasingly important.
The study also has an important limitation. The analysis was restricted to participants of European ancestry with available genotype and EHR data, meaning the findings should not automatically be assumed to apply across ancestrally diverse populations.
Nevertheless, the work demonstrates an increasingly important principle for biobank-enabled research: two biobanks can show highly similar underlying genetic effects for a disease while still producing different quantitative estimates because of how the phenotype itself is captured.
About the Author
Biobanking.com delivers valuable news and information on biobanking, biorepositories, cryopreservation, and biosample/biospecimen management, while connects buyers with top vendors globally.
