Al-Juhani Abdulkreem, Desoky Rodan, Abdullah Abdullah, Younes Elsayed, Khoja Sultan, Aljohani Sereen S, Desoky Abdalrahman
General Surgery, King Abdulaziz University Faculty of Medicine, Jeddah, SAU.
Medicine, College of Medicine, Alfaisal University, Riyadh, SAU.
Cureus. 2025 Jul 23;17(7):e88568. doi: 10.7759/cureus.88568. eCollection 2025 Jul.
Chronic limb-threatening ischemia (CLTI) and peripheral arterial disease (PAD) sometimes lead to non-healing lesions and amputations, despite revascularization efforts. Current clinical instruments for prognostication exhibit insufficient personalized precision. This systematic research sought to assess the predictive efficacy of machine learning models in forecasting wound healing and limb preservation after lower limb revascularization. A comprehensive literature search was conducted in PubMed, Web of Science, Embase, Scopus, and IEEE Xplore from January 2018 to March 2025. Studies were considered if they utilized machine learning techniques to forecast outcomes following surgical or endovascular lower limb revascularization. The inclusion criteria adhered to the Population, Index model, Comparator, and Outcome (PICO) framework. The Prediction model Risk Of Bias ASsessment Tool (PROBAST) was utilized to evaluate the risk of bias. Only studies that presented quantifiable performance measurements (e.g., area under the receiver operating characteristic curve [AUROC], calibration, sensitivity) were included. Data extraction and risk evaluation were performed separately by two reviewers. Out of 450 records reviewed, five studies satisfied the inclusion criteria. The majority of studies utilized extensive registry data, with sample sizes varying from 392 to 235,677 patients. Machine learning techniques, such as XGBoost, neural networks, and Bayesian algorithms, surpassed standard logistic regression in prognostic accuracy (AUROC 0.78-0.95). Three studies exhibited a little risk of bias in all domains. Nevertheless, two investigations indicated a high or ambiguous risk owing to restricted sample size or absence of external validation. The variability in outcome definitions and model inputs prevented meta-analysis. External validation was infrequent, and practical applicability remains unsubstantiated. Machine learning models exhibit significant predictive capability in forecasting wound healing and limb salvage results following revascularization, frequently surpassing conventional clinical instruments. Nonetheless, extensive validation and prospective assessment are necessary prior to clinical application.
尽管进行了血运重建努力，但慢性肢体威胁性缺血（CLTI）和外周动脉疾病（PAD）有时仍会导致伤口不愈合和截肢。目前用于预后评估的临床工具个性化精准度不足。这项系统性研究旨在评估机器学习模型在预测下肢血运重建后伤口愈合和肢体保留方面的预测效果。2018年1月至2025年3月在PubMed、科学网、Embase、Scopus和IEEE Xplore中进行了全面的文献检索。如果研究利用机器学习技术预测手术或血管腔内下肢血运重建后的结果，则予以考虑。纳入标准遵循人群、指标模型、对照和结局（PICO）框架。使用预测模型偏倚风险评估工具（PROBAST）来评估偏倚风险。仅纳入呈现可量化性能测量值（如受试者操作特征曲线下面积[AUROC]、校准、敏感性）的研究。两名评审员分别进行数据提取和风险评估。在审查的450条记录中，有五项研究符合纳入标准。大多数研究使用了广泛的登记数据，样本量从392例至235,677例患者不等。诸如XGBoost、神经网络和贝叶斯算法等机器学习技术在预后准确性（AUROC为0.78 - 0.95）方面超过了标准逻辑回归。三项研究在所有领域均显示出较小的偏倚风险。然而，两项调查由于样本量有限或缺乏外部验证而表明存在高风险或偏倚不明确。结局定义和模型输入的变异性妨碍了荟萃分析。外部验证很少见，实际适用性仍未得到证实。机器学习模型在预测血运重建后伤口愈合和肢体挽救结果方面具有显著的预测能力，常常超过传统临床工具。尽管如此，在临床应用之前需要进行广泛的验证和前瞻性评估。
