Machine learning-driven prioritization of lipoprotein(a) testing: model development and validation
Authors' conclusion
A ML-guided approach to prioritise testing for elevated Lp(a) would require fewer tests to identify those above risk-enhancing thresholds or potentially eligible for emerging therapies, offering a scalable interim compromise between the low current testing rates and universal screening aspirations.
Study context
- Population
- UK Biobank participants ≥37 years old; external validation in NHANES IIIUK Biobank, split 20/60/20%
- Participants
- 438,579 participants
- Comparison
- Machine learning (ML)-targeted Lp testing vs universal screening
- Outcome
- Identification of individuals with elevated Lp(a) across thresholds (90, 125, 150, 175, 200, 430 nmol/L)
Methodological detailsethnicity-calibrated, external validation in NHANES III
- Ethnicity-calibrated
- External validation in NHANES III
Performance
As reported in the abstract
Cases identified per one million tests
Universal approach
- Cases >90 nmol/L
- 222,717
- Cases >430 nmol/L
- 1950
ML-targeted testing
- Cases >90 nmol/L
- 280,899
- Cases >430 nmol/L
- 6881
Cases >90 nmol/L
Universal approach222,717
ML-targeted testing280,899
Cases >430 nmol/L
Universal approach1950
ML-targeted testing6881
Yield increase
- 90 nmol/L
- +26% (95% CI 20–28%)
- 430 nmol/L
- +253% (95% CI 192–310%)
- Yield increase at 125 nmol/L
- 38% (95%CI:35-40%)
- Yield increase at 150 nmol/L
- 51% (95%CI:47-54%)
- Yield increase at 175 nmol/L
- 59% (95%CI:55-63%)
- Yield increase at 200 nmol/L
- 66% (95%CI:61-71%)
Further findings
External validation (NHANES III)
- Similar performance
Top 4 predictors
- Age, height (proxy for sex), total cholesterol and statin use
Original publication
Machine learning-driven prioritization of lipoprotein(a) testing: model development and validation. Eur J Prev Cardiol. Published 30 March 2026. doi:10.
Summarised from the published abstract; every number is checked against it