Biopsychosocial predictors of long-term major amputation following peripheral vascular intervention in chronic limb-threatening ischemia using a machine learning algorithm approach

<h2>Abstract</h2><h3>Background</h3> Patients with chronic limb-threatening ischemia (CLTI) remain at high risk of major amputation following revascularization. Although traditional clinical risk factors are well-established, the role of biopsychosocial factors in amputation risk remains unclear. This study aimed to identify the most important biopsychosocial clinical factors that predict the 3-year risk of major amputation following peripheral vascular intervention (PVI). <h3>Methods</h3> Patients who underwent PVI for CLTI between January 2017 and December 2018 were identified from the PVI module of the Vascular Quality Initiative, which is linked to Medicare outcomes data. We evaluated 63 pre-procedural variables, including peripheral arterial disease (PAD)/CLTI-related, demographic, functional, and behavioral characteristics. A random survival forest algorithm accounting for the competing risk of death (RSF-CR) was developed to rank pre-procedural variables based on their importance in predicting the 3-year major amputation risk. The relative importance [RI] of each predictor was calculated using the Breiman-Cutler importance approach. A sensitivity analysis was conducted using the minimal depth approach, which measures the importance of each variable in the decision-making process. We assessed the RSF-CR's accuracy and discriminative ability using the out-of-bag error rate and the Harrell C-index, respectively. <h3>Results</h3> A total of 7848 patients with CLTI were included (mean age, 76.5 ± 7.6 years; 57.2% male; 77.9% White). At 3-year follow-up, 40.9% died, and 13.8% underwent major amputation. A non-null RI was calculated for 52 of 63 preprocedural variables. Advanced chronic kidney disease (stage 5) was the most important predictor of 3-year major amputation risk (RI = 100%), followed by below-the-knee disease, prior major amputation, and diabetes (RI = 78%, 71%, and 57%, respectively). Beyond these medical history and PAD/CLTI-related factors, the key demographic, functional status, and behavioral factors were female sex, ambulation with wheelchair, and pain (RI = 28%, 28%, and 3%, respectively). Model demonstrated good performance (out-of-bag error rate=32.7%; C-index = 0.67). <h3>Conclusions</h3> Long-term limb outcomes are affected by a multitude of factors including traditional cardiovascular risk factors as well as functional and behavioral patient characteristics. A multidimensional risk assessment framework, incorporating both traditional clinical and biopsychosocial factors, is essential to shared decision-making and delivery of individualized care with multidisciplinary involvement.