Machine Learning-driven Prioritisation of Lipoprotein(a) Testing: Model Development and Validation.

Area of Science:

• Cardiovascular Disease Research
• Biomarker Discovery
• Machine Learning in Healthcare

Background:

• Elevated lipoprotein(a) [Lp(a)] is a significant, yet under-tested, risk factor for cardiovascular disease (CVD) affecting billions globally.
• Current low testing rates (<1%) and challenges in universal screening necessitate innovative approaches for identifying at-risk individuals.
• Novel Lp(a)-lowering therapies are emerging, increasing the urgency for effective Lp(a) testing strategies.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for prioritizing patients for lipoprotein(a) [Lp(a)] testing.
• To assess the efficiency of ML-guided testing compared to universal screening for identifying elevated Lp(a) levels.
• To evaluate the ML model's performance across various clinical and trial-related Lp(a) thresholds.

Main Methods:

• Developed ethnicity-calibrated ML models using UK Biobank data (N=438,579) with participants aged ≥37.
• Split data into feature importance/selection (20%), derivation (60%), and validation (20%) sets.
• Evaluated model performance against guideline thresholds (90, 430 nmol/L) and trial entry criteria (150-200 nmol/L), with external validation in NHANES III.

Main Results:

• ML-targeted testing identified 26% more cases above 90 nmol/L and 253% more cases above 430 nmol/L compared to universal screening.
• Yield increases ranged from 38% to 66% at thresholds between 125-200 nmol/L.
• ML-targeted testing required 21%-72% fewer tests to identify one million cases, demonstrating improved efficiency.

Conclusions:

• A ML-guided approach significantly enhances the efficiency of identifying individuals with elevated Lp(a) compared to universal screening.
• This ML strategy offers a scalable solution to bridge the gap between current low testing rates and the need for broader screening.
• The model effectively prioritizes testing for individuals above risk-enhancing thresholds or eligible for emerging Lp(a)-lowering therapies.