Analysis of a Machine Learning-Based Risk Stratification Scheme for Chronic Limb-Threatening Ischemia

Area of Science:

• Vascular Surgery
• Data Science
• Biostatistics

Background:

• Effective risk stratification is crucial for comparative effectiveness research in chronic limb-threatening ischemia (CLTI).
• Existing risk stratification models for CLTI have demonstrated limited accuracy and efficacy.
• There is a need for improved, comprehensive, and reproducible risk prediction models for CLTI.

Purpose of the Study:

• To evaluate the utility of topic model cluster analysis for developing an accurate risk prediction model for CLTI.
• To identify distinct patient subgroups within CLTI based on clinical features.
• To assess the predictive performance of these subgroups for 1-year CLTI-free survival.

Main Methods:

• A multicenter, nested cohort study utilizing data from the PREVENT III clinical trial.
• Supervised topic model cluster analysis was applied to infrainguinal vein bypass patient data.
• The analysis identified patient clusters and their associated features to predict 1-year CLTI-free survival.

Main Results:

• The analysis identified three distinct CLTI stages (clusters) among 1238 patients.
• 1-year CLTI-free survival rates varied significantly across stages: 82.3% for stage 1, 61.1% for stage 2, and 53.4% for stage 3.
• Major limb amputation rates increased progressively with stage (4.2%, 10.8%, 18.4% respectively).

Conclusions:

• Topic model cluster analysis successfully identified three distinct stages of CLTI.
• CLTI-free survival can be accurately and reproducibly quantified as a patient-centric outcome.
• These findings support the use of this method for improved CLTI risk stratification.