SS18 - Predicting mortality risk following major lower extremity amputation using machine learning

Objectives: Major lower extremity amputation carries important perioperative risks. Outcome prediction tools may help guide clinical decision-making but remain limited. We developed machine learning (ML) algorithms that predict 1-year mortality following major lower extremity amputation.

Methods: The Vascular Quality Initiative (VQI) database was used to identify patients who underwent major lower extremity amputation for non-traumatic and non-malignant causes between 2012-2024. We identified 75 features from the index hospitalization (52 pre-operative [demographic/clinical], 5 intra-operative [procedural], and 18 post-operative [in-hospital course/complications]). The primary outcome was 1-year all-cause mortality. The data was divided into training (70%) and test (30%) sets. Six ML models were trained using pre-operative features with 10-fold cross-validation (Extreme Gradient Boosting [XGBoost], random forest, Naïve Bayes classifier, support vector machine, artificial neural network, and logistic regression). The primary model evaluation metric was area under the receiver operating characteristic curve (AUROC). The algorithm with the best performance was further trained using intra- and post-operative features. Model robustness was assessed using calibration plots and Brier scores. Performance was evaluated across subgroups based on age, sex, race, ethnicity, rurality, median Area Deprivation Index, prior ipsilateral minor amputation, prior ipsilateral open/endovascular revascularization, level of amputation, indication for amputation, and urgency.

Results: Overall, 22,828 patients underwent major lower extremity amputation during the study period and 5,842 (25.6%) suffered 1-year mortality. Patients who reached the primary endpoint were older with more comorbidities, had poorer functional status, and underwent higher level amputations. Despite having elevated cardiovascular risk, they were less likely to receive risk-reduction medications at discharge. The best pre-operative prediction model was XGBoost, achieving an AUROC (95% CI) of 0.88 (0.87-0.89). In comparison, logistic regression had an AUROC (95% CI) of 0.70 (0.68-0.72). The XGBoost model maintained excellent performance at the intra- and post-operative stages, with AUROC’s (95% CI’s) of 0.88 (0.87-0.89) and 0.94 (0.93-0.95), respectively (Fig 1). Calibration plots showed good agreement between predicted/observed event probabilities with Brier scores of 0.12 (pre-operative), 0.11 (intra-operative), and 0.09 (post-operative). The most important predictive features were level of and indication for amputation, pre- and post-operative ambulatory status, in-hospital complications, and discharge medications. Model performance remained robust across all subgroups.

Conclusions: We developed ML models that can accurately predict 1-year mortality following major lower extremity amputation for vascular disease, performing better than logistic regression. These algorithms have potential for important utility in guiding patient selection, counselling, goals of care discussions, and clinical decision-making to support personalized care for a high-risk population (Fig 2).