Developing a Hybrid Risk Assessment Tool for Familial Hypercholesterolemia: A Machine Learning Study of Chinese...

Area of Science:

• Cardiovascular Medicine
• Genetics
• Artificial Intelligence in Healthcare

Background:

• Familial hypercholesterolemia (FH) is an autosomal-dominant disorder increasing premature cardiovascular disease (ASCVD) risk.
• Existing FH risk assessment tools like DLCN have variable accuracy across populations.
• There is a need for improved risk stratification tools for specific patient groups.

Purpose of the Study:

• To develop and evaluate a hybrid FH risk assessment tool (HFHRAT) integrating existing methods with machine learning (ML).
• To assess the performance of the HFHRAT in a cohort of Chinese patients with ASCVD.

Main Methods:

• 11 FH risk assessment tools were applied to 5,597 primary ASCVD patients.
• The top three performing tools were hybridized using a voting strategy.
• ML models were trained on hybrid results to create the HFHRAT, interpreted using PDP and ICE.

Main Results:

• The HFHRAT, a stacking ensemble method, demonstrated high performance (AUC_class: 94.85%, AUC_prob: 98.66%).
• Key risk indicators identified by HFHRAT include age <75 years and LDL-c >4 mmol/L.
• The hybridized model significantly improved risk discovery compared to individual tools.

Conclusions:

• The HFHRAT effectively reduces false-negative rates, potentially preventing atherosclerosis.
• This hybrid tool offers a more accurate risk assessment for specific populations, addressing limitations of current methods.