Nonfasting, Telehealth-Ready LDL-C Testing With Machine Learning to Improve Cardiovascular Access and Equity.

Area of Science:

• Cardiovascular disease prevention and management.
• Application of artificial intelligence in healthcare diagnostics.
• Health equity and access to care.

Background:

• Current low-density lipoprotein cholesterol (LDL-C) testing mandates fasting and clinic visits, creating significant access barriers.
• Approximately 40% of lipid panels are performed outside fasting windows, yielding unreliable results.
• Existing workflows are outdated, leading to millions of unnecessary repeat visits annually, disproportionately affecting vulnerable populations.

Purpose of the Study:

• To demonstrate machine learning (ML) can transform lipid testing into an accurate, equitable, and telehealth-ready service.
• To eliminate simultaneous barriers of fasting requirements, in-person visits, and algorithmic bias in LDL-C testing.
• To provide a solution for the access crisis in cardiovascular risk assessment.

Main Methods:

• Cross-sectional analysis of the All of Us Research Program dataset (n=3,477; tests n=696).
• Evaluation of ML model performance stratified by fasting status, including non-fasting real-world data (40.1%).
• Assessment of telehealth feasibility (labs-only configuration), racial equity, and economic impact.

Main Results:

• The ML system demonstrated superior accuracy in non-fasting conditions compared to traditional equations (e.g., Friedewald), with a 17.2% improvement.
• A labs-only configuration was non-inferior, enabling retail pharmacy and home-testing workflows.
• The ML system achieved racial equity without race input and demonstrated significant economic savings by reducing repeat visits.

Conclusions:

• The ML approach addresses critical quality gaps in cardiovascular prevention, including delayed treatment and access barriers.
• It enables accurate, non-fasting, telehealth-compatible, and race-free LDL-C estimation, transforming lipid testing into an access enabler.
• Feasible implementation with existing infrastructure offers a scalable solution for value-based care, particularly for underserved populations.