DOI: 10.1161/circ.148.suppl_1.18246 ISSN: 0009-7322

Abstract 18246: Determining the Most Important Factors for 1-year Health Status Using a Machine-Learning Algorithm in Patients With Peripheral Artery Disease Undergoing Revascularization for Claudication

Gaelle Romain, Jacob Cleman, Kim G Smolderen, Carlos Mena-Hurtado
  • Physiology (medical)
  • Cardiology and Cardiovascular Medicine

Introduction: Over time, patients with peripheral artery disease (PAD) are at high risk of poor cardiovascular outcomes, but most immediately, their health status is impacted due to lifestyle-limiting symptoms. To improve functioning, peripheral vascular intervention (PVI) may be offered. We aimed to understand which factors predict subsequent health status outcomes following PVI.

Methods: We used data from the international ILLUMENATE examining the safety of peripheral angioplasty with vs. without Stellarex drug-coated balloon for femoropopliteal artery disease in randomized trial between 2013 and 2019. Treatment arms were combined for this analysis. EQ-5D 3L Index and Visual Analog Scale (VAS) were assessed at baseline and 1 year. Random forest regression, a base-trees machine-learning algorithm, was built in the training sample (80% of the cohort) and evaluated in the validation sample (20% of the cohort). The importance of 58 included pre-procedural variables was ranked. The most important variables were defined based on visual inspection of the relative importance plots to evaluate the information gained value of each included variable.

Results: In the 473 included patients, the mean age was 68.2 ± 9.2 years, 32.1% were female and 4.2% were Black. For the 1-year 3L Index, the most important pre-procedural variables were liver disease and the 3L Index (54.1%). For the 1-year VAS, the most important variables were angioplasty/drug eluting stent following by chronic lung disease (94.9%), mechanical thrombectomy (81.0%), anticoagulant (80.1%) and congestive heart failure (76.5%). (Figure)

Conclusions: As part of an effort to develop an individualized health status profile with machine learning, a predictable ranking of pre-procedural variables can inform 1-year health status outcomes in patients with PAD undergoing PVI for claudication. These approaches can inform clinical decision-making and pre-habilitation efforts to maximize treatment success.

More from our Archive