📄 genetic and genomic medicine

European-derived coronary artery disease polygenic scores over-flag genetic risk in Vietnamese and Southeast Asian populations: a multi-score analysis in 1000 Genomes

This study demonstrates that European-derived coronary artery disease polygenic scores are inconsistently calibrated and systematically over-flag high genetic risk in Vietnamese and Southeast Asian populations when European thresholds are applied, highlighting the critical need for local validation and recalibration before clinical use.

Original authors: Hoang, Q. P., Le, T. X., Doan, D. D.

Published 2026-07-15

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Hoang, Q. P., Le, T. X., Doan, D. D.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you have a super-smart weather app that predicts rain. This app was trained exclusively on data from London. It knows exactly what "rainy" looks like in London: grey skies, puddles, and people with umbrellas. Now, imagine you take that same London-trained app and try to use it in Vietnam to predict the monsoon season.

The app might look at the sky, see the humidity, and scream, "RAIN! RAIN! RAIN!" with 100% confidence. But here's the twist: in Vietnam, that same sky might just mean a humid afternoon, not a flood. The app isn't necessarily "wrong" about the clouds, but it's using the wrong rulebook for the location.

This is exactly what a team of researchers from Vietnam discovered when they tested Coronary Artery Disease (CAD) polygenic scores. Think of these scores as "genetic weather apps" that predict your risk of heart disease. Almost all of these apps were built using data from people of European ancestry. The researchers wanted to see what happens when you run these European "apps" on Vietnamese and Southeast Asian DNA.

The Great Mismatch

The team grabbed four different "genetic weather apps" (called polygenic scores) from a public library of scores. They ran them on the DNA of 2,504 people from the 1000 Genomes Project, specifically focusing on 99 Vietnamese Kinh people and 93 Dai people (a group from China's border region).

Here is the big finding: The apps were shouting "High Risk!" way too often for these populations.

In the European world, these scores are calibrated so that the top 20% of people are flagged as "high genetic risk." It's like setting a line in the sand: if you are in the top 20%, you get a warning.

But when the researchers applied that same "top 20%" line to the Vietnamese and Dai groups, the results were wild:

For one specific score (PGS000349), 57.6% of the Vietnamese Kinh and 57.0% of the Dai were flagged as high risk. That's nearly three times the number the app was supposed to flag!
For another score (PGS000058), 35.4% of the Vietnamese and 43.0% of the Dai were flagged.
Even the "best" score (PGS004198), which was fairly accurate for East Asians, still flagged 22.2% of the Vietnamese and 21.5% of the Dai—slightly over the limit.

The researchers found that the genetic scores for Vietnamese and Dai people were shifted upward by about 0.47 standard deviations compared to Europeans. In plain English, the "genetic weather" for these groups looked "stormier" to the European app, even though the app didn't actually know if a storm was coming.

What This Means (and What It Doesn't)

The researchers are very clear about what they proved and what they didn't.

They proved: The scores are miscalibrated. If you use a European threshold on Vietnamese DNA, you will over-flag people. You will tell healthy people they are at high risk when the score just doesn't know how to read their specific genetic map. This could lead to unnecessary worry, extra doctor visits, or even unnecessary treatments.

They did NOT prove: They did not prove that Vietnamese people actually have more heart disease or that their genes are "worse." The paper explicitly states that this shift is likely due to differences in how DNA is arranged (allele frequencies and linkage-disequilibrium) between populations, not because the Vietnamese people are actually at higher risk.

They did NOT prove: They did not measure how well these scores actually predict heart attacks in these groups. Because the 1000 Genomes Project data doesn't include health records (like who actually had a heart attack), they couldn't calculate the "accuracy" (AUC) or the "odds ratio." They only measured the distribution (the shape of the curve). They suggest that to know if these scores are truly accurate, scientists need to test them on real Vietnamese patients with known heart disease history.

The Takeaway

The main message is a warning label for doctors and scientists: Do not just copy-paste European genetic risk scores into Southeast Asian healthcare.

The researchers tested four different scores, and three of them over-flagged the Vietnamese population. One score was okay for East Asians but went crazy for African populations (flagging 69.3% as high risk!). This shows that you can't guess how a score will behave in a new population just by looking at its performance in Europe.

The paper concludes that before these tools can be used fairly in Vietnam or Southeast Asia, they need local recalibration. You can't use a London rain gauge to measure the monsoon in Hanoi without adjusting the scale first. Until then, using these scores could widen health gaps rather than close them, systematically telling the wrong people they are in danger.

Technical Summary: European-Derived CAD Polygenic Scores in Southeast Asian Populations

Problem Statement
Coronary artery disease (CAD) polygenic risk scores (PRS) are increasingly proposed for clinical risk stratification. However, these scores are derived almost exclusively from European-ancestry discovery datasets. Due to differences in allele frequencies, linkage disequilibrium (LD) structures, and effect sizes across populations, PRS portability to non-European groups is known to be compromised. Southeast Asian populations, including the 97+ million people of Vietnam, are severely under-represented in genomic resources. The critical clinical concern is that applying European-derived thresholds to these populations could systematically misclassify individuals as "high risk," leading to health disparities through over- or under-identification. This study addresses the lack of characterization regarding how European-derived CAD scores perform specifically in Vietnamese and Southeast Asian cohorts.

Methodology
The authors conducted a secondary analysis of publicly available, de-identified data without collecting new human subjects.

Data Sources: Genotypes were obtained from the 1000 Genomes Project Phase 3 (GRCh37), focusing on 2,504 individuals. The target populations were the Vietnamese Kinh (KHV, n=99) and the Chinese Dai (CDX, n=93). European (EUR), East Asian (EAS), South Asian (SAS), Admixed American (AMR), and African (AFR) super-populations served as comparators.
Polygenic Scores: Four independent, European-derived CAD scores were retrieved from the PGS Catalog:
- PGS000058 (204 variants)
- PGS000349 (70 variants)
- PGS002809 (205 variants)
- PGS004198 (5,723 variants, LASSO-derived)
Computation: Individual scores were calculated using PLINK2 with allele-matched dosage weighting. Over 97% of variants for each score were successfully matched.
Statistical Analysis: Raw scores were standardized to a mean of zero and unit variance across the pooled sample. The study assessed:
1. Calibration: Cross-ancestry distribution differences (ANOVA, Welch t-tests).
2. Clinical Consequence: The proportion of each population flagged as "high genetic risk" when the European top-20% threshold (80th percentile) was applied. Under perfect calibration, this rate should be 20% for all groups.
Scope Limitation: As the 1000 Genomes data lacks cardiovascular phenotypes, the study did not estimate predictive accuracy (AUC) or effect sizes (odds ratios). It focused strictly on distribution shifts and threshold mis-stratification.

Key Results

Distribution Shifts: For the primary score (PGS000058), the standardized PRS differed significantly across super-populations ( $F(4, 2499) = 121.1, p < 0.001$ ). The Vietnamese Kinh mean was +0.47 SD above the European mean, and the Dai mean was +0.43 SD above. This indicates a systematic upward shift in genetic risk scores for Southeast Asians relative to Europeans.
Over-Flagging: Applying the European top-20% threshold resulted in significant over-flagging in Southeast Asian samples:
- Vietnamese Kinh (KHV): The fraction flagged as high risk ranged from 22.2% to 57.6% across the four scores.
- Dai (CDX): The fraction flagged ranged from 21.5% to 43.0%.
- Specific Score Performance:
  - PGS000349 (70 variants) was the most miscalibrated, flagging 57.6% of Kinh and 57.0% of Dai (nearly triple the intended rate).
  - PGS000058 flagged 35.4% of Kinh and 43.0% of Dai.
  - PGS002809 showed moderate over-flagging (25.3% for Kinh).
  - PGS004198 (the largest score) was approximately calibrated for East/Southeast Asians (~22%) but dramatically over-flagged Africans (69.3%).
Inconsistency: The magnitude and direction of miscalibration varied by score. No single score behaved consistently across all non-European populations; for instance, a score well-calibrated for East Asians could severely over-flag Africans.

Key Contributions

Empirical Evidence of Miscalibration: The study provides quantitative evidence that European-derived CAD PRS are inconsistently calibrated in Vietnamese and Dai populations, predominantly leading to the over-identification of high genetic risk.
Score-Specific Variability: It demonstrates that cross-ancestry performance cannot be assumed based on a score's European metrics; even large, LASSO-derived scores (PGS004198) behave differently across populations (calibrated for East Asians but biased for Africans).
Clinical Risk Quantification: By translating distribution shifts into "flagging rates," the paper quantifies the immediate clinical consequence of using unadjusted European thresholds: a substantial portion of Southeast Asian individuals would be incorrectly categorized as high risk.

Significance and Claims
The paper concludes that European-derived CAD polygenic scores cannot be assumed transferable to Vietnamese or Southeast Asian populations without local validation and recalibration.

Clinical Implication: Deploying these scores with European thresholds in Southeast Asian care settings would likely drive systematic over-referral and over-treatment, exacerbating health inequities.
Methodological Boundary: The authors explicitly state that while distribution shifts confirm miscalibration, they do not quantify predictive accuracy (discrimination). Determining the true loss of accuracy requires phenotyped cohorts (e.g., GenomeAsia 100K or national biobanks) to calculate odds ratios and AUCs.
Call to Action: The results reinforce the necessity for larger Southeast Asian genomic and phenotypic resources and the development of ancestry-specific scores before equitable clinical deployment is possible.

The study remains modest in its claims, strictly limiting its conclusions to calibration and threshold mis-stratification based on the available genotype data, while clearly delineating the need for future phenotyped studies to assess true predictive performance.

The Great Mismatch

What This Means (and What It Doesn't)

The Takeaway

Technical Summary: European-Derived CAD Polygenic Scores in Southeast Asian Populations

More like this