Li Ben, Eisenberg Naomi, Beaton Derek, Lee Douglas S, Al-Omran Leen, Wijeysundera Duminda N, Hussain Mohamad A, Rotstein Ori D, de Mestral Charles, Mamdani Muhammad, Roche-Nagle Graham, Al-Omran Mohammed
Department of Surgery, University of Toronto, Toronto, Ontario, Canada; Division of Vascular Surgery, St. Michael's Hospital, Unity Health Toronto, Toronto, Ontario, Canada; Institute of Medical Science, University of Toronto, Toronto, Ontario, Canada; Temerty Centre for Artificial Intelligence Research and Education in Medicine (T-CAIREM), University of Toronto, Toronto, Ontario, Canada.
Division of Vascular Surgery, Peter Munk Cardiac Centre, University Health Network, Toronto, Ontario, Canada.
J Vasc Surg. 2025 May 1. doi: 10.1016/j.jvs.2025.03.198.
Major lower extremity amputation for advanced vascular disease involves significant perioperative risks. Although outcome prediction tools could aid in clinical decision-making, they remain limited. To address this, we developed machine learning (ML) algorithms capable of predicting 1-year mortality following major lower extremity amputation.
The Vascular Quality Initiative (VQI) database was queried to identify patients who underwent major lower extremity amputation for non-traumatic and non-malignant causes between 2012 and 2024. A total of 75 features were collected from the index hospitalization, including 52 preoperative (demographic/clinical), five intraoperative (procedural), and 18 postoperative (in-hospital course/complications) variables. The primary outcome was 1-year all-cause mortality. The data was split into training (70%) and test (30%) sets. Six ML models were trained using preoperative features, employing 10-fold cross-validation, which included Extreme Gradient Boosting (XGBoost), random forest, Naïve Bayes classifier, support vector machine, artificial neural network, and logistic regression. The primary model evaluation metric was the area under the receiver operating characteristic curve (AUROC). The best-performing model was then further trained using intra- and postoperative features. Model robustness was evaluated through calibration plots and Brier scores. Model performance was assessed across various 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.
A total of 22,828 patients underwent major lower extremity amputation during the study period, with 5842 (25.6%) experiencing 1-year mortality. Patients who reached the primary endpoint were older with more comorbidities, had poorer functional status, and were more likely to undergo higher-level amputations. Despite having elevated cardiovascular risk, these patients were less likely to receive cardiovascular risk reduction medications. The best preoperative prediction model was XGBoost, which achieved an AUROC of 0.88 (95% confidence interval [CI], 0.87-0.89). In comparison, logistic regression showed an AUROC of 0.70 (95% CI, 0.68-0.72). The XGBoost model maintained excellent performance at the intra- and postoperative stages, with AUROCs of 0.88 (95% CI, 0.87-0.89) and 0.94 (95% CI, 0.93-0.95), respectively. Calibration plots indicated strong agreement between predicted/observed event probabilities, with Brier scores of 0.12 (preoperative), 0.11 (intraoperative), and 0.09 (postoperative). Among the top 10 predictors, 6 were preoperative features, including the level of and indication for amputation, comorbidities, and functional status. Model performance remained robust across all subgroups.
We developed ML models that can accurately predict 1-year mortality following major lower extremity amputation, outperforming logistic regression. These algorithms have potential for important utility in guiding patient selection, counseling, goals of care discussions, and clinical decision-making to support patient-centered care for a high-risk population.
因晚期血管疾病进行的下肢大截肢手术存在重大围手术期风险。尽管结局预测工具有助于临床决策，但它们仍然有限。为了解决这个问题，我们开发了能够预测下肢大截肢术后1年死亡率的机器学习（ML）算法。
查询血管质量倡议（VQI）数据库，以识别2012年至2024年间因非创伤性和非恶性原因接受下肢大截肢手术的患者。从首次住院期间收集了总共75个特征，包括52个术前（人口统计学/临床）、5个术中（手术过程）和18个术后（住院过程/并发症）变量。主要结局是1年全因死亡率。数据被分为训练集（70%）和测试集（30%）。使用术前特征训练了六个ML模型，采用10折交叉验证，其中包括极端梯度提升（XGBoost）、随机森林、朴素贝叶斯分类器、支持向量机、人工神经网络和逻辑回归。主要模型评估指标是受试者工作特征曲线下面积（AUROC）。然后使用术中和术后特征对表现最佳的模型进行进一步训练。通过校准图和布里尔评分评估模型的稳健性。根据年龄、性别、种族、民族、农村地区、地区贫困指数中位数、既往同侧小截肢、既往同侧开放/血管腔内血运重建、截肢水平、截肢指征和紧急程度，在各个亚组中评估模型性能。
在研究期间，共有22828例患者接受了下肢大截肢手术，其中5842例（25.6%）在1年内死亡。达到主要终点的患者年龄较大，合并症更多，功能状态较差，并且更有可能接受更高水平的截肢手术。尽管心血管风险升高，但这些患者接受心血管风险降低药物治疗的可能性较小。最佳术前预测模型是XGBoost，其AUROC为0.88（95%置信区间[CI]，0.87-0.89）。相比之下，逻辑回归的AUROC为0.70（95%CI，0.68-0.72）。XGBoost模型在术中和术后阶段均保持出色表现，AUROC分别为0.88（95%CI，0.87-0.89）和0.94（95%CI，0.93-0.95）。校准图表明预测/观察到的事件概率之间有很强的一致性，布里尔评分分别为0.12（术前）、0.11（术中）和0.09（术后）。在排名前十的预测因素中，6个是术前特征，包括截肢水平和指征、合并症和功能状态。模型性能在所有亚组中均保持稳健。
我们开发的ML模型能够准确预测下肢大截肢术后1年死亡率，优于逻辑回归。这些算法在指导患者选择、咨询、护理目标讨论和临床决策以支持高危人群的以患者为中心的护理方面具有重要实用价值。
