Annals of vascular surgery · 2026
A Dynamic Machine Learning Approach to Complement Nurse-Led Clinics in Identifying High-Risk Patients with Intermittent Claudication.
СтатьяВалидационное исследованиеНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Ravindhran Bharadhwaj, Hatfield-Chetter Georgina, Morris-Jarvis Josephine, Stedman Natalie, Carradice Daniel, Smith George, Thakker Dhaval, Bicknell Colin, Chetter Ian, Pymer Sean
Аннотация
BACKGROUND: To develop and validate dynamic machine learning (ML) models to predict 6-month adverse outcomes in intermittent claudication (IC), enabling risk-ranked prioritization for surveillance and timely revascularization.
METHODS: This study included patients from a prospectively maintained nurse-led IC clinic registry (2020-2024). Predictors included demographics, comorbidities, medications, smoking exposure, frailty, ankle-brachial pressure index [ABPI], and treadmill performance. We first trained feedforward neural networks (multilayer perceptron and radial basis function) and then developed ensemble models (Least Absolute Shrinkage and Selection Operator logistic regression, gradient boosting, random forest, and XGBoost; stacking meta-learner logistic regression with 3-fold cross-validation). Data were split 70/30 with stratification. Model performance was assessed using discrimination (area under the receiver operator characteristic curve) and F1 score, and calibration (Brier score).
RESULTS: In the cohort (n = 629), IC deterioration requiring revascularization occurred in 132 patients (21.0%), chronic limb-threatening ischemia (CLTI) progression in 27 (4.3%), and mortality in 34 (5.4%) at 6 months. Ensemble models achieved area under the curve values of 0.712 (IC revascularization), 0.754 (CLTI), and 0.828 (mortality), and Brier scores of 0.154, 0.039, and 0.047, respectively. F1 scores were >0.70 for all model iterations. Earlier neural network models demonstrated low cross-entropy but less favorable probability accuracy (Brier ∼0.22) and F1 performance, motivating transition to ensemble approaches. Key predictors included ABPI, maximum walking distance and initial claudication distance, age, renal disease, chronic obstructive pulmonary disease, smoking exposure, and frailty.
CONCLUSION: Ensemble ML models provide calibrated 6-month risk estimates for revascularization, CLTI, and mortality after nurse-led IC clinic assessment, which may support patient prioritization for surveillance and early escalation of care. External validation and prospective impact evaluation are warranted and represent our next steps.
METHODS: This study included patients from a prospectively maintained nurse-led IC clinic registry (2020-2024). Predictors included demographics, comorbidities, medications, smoking exposure, frailty, ankle-brachial pressure index [ABPI], and treadmill performance. We first trained feedforward neural networks (multilayer perceptron and radial basis function) and then developed ensemble models (Least Absolute Shrinkage and Selection Operator logistic regression, gradient boosting, random forest, and XGBoost; stacking meta-learner logistic regression with 3-fold cross-validation). Data were split 70/30 with stratification. Model performance was assessed using discrimination (area under the receiver operator characteristic curve) and F1 score, and calibration (Brier score).
RESULTS: In the cohort (n = 629), IC deterioration requiring revascularization occurred in 132 patients (21.0%), chronic limb-threatening ischemia (CLTI) progression in 27 (4.3%), and mortality in 34 (5.4%) at 6 months. Ensemble models achieved area under the curve values of 0.712 (IC revascularization), 0.754 (CLTI), and 0.828 (mortality), and Brier scores of 0.154, 0.039, and 0.047, respectively. F1 scores were >0.70 for all model iterations. Earlier neural network models demonstrated low cross-entropy but less favorable probability accuracy (Brier ∼0.22) and F1 performance, motivating transition to ensemble approaches. Key predictors included ABPI, maximum walking distance and initial claudication distance, age, renal disease, chronic obstructive pulmonary disease, smoking exposure, and frailty.
CONCLUSION: Ensemble ML models provide calibrated 6-month risk estimates for revascularization, CLTI, and mortality after nurse-led IC clinic assessment, which may support patient prioritization for surveillance and early escalation of care. External validation and prospective impact evaluation are warranted and represent our next steps.
Предметные рубрики MeSH
HumansIntermittent ClaudicationFemalePredictive Learning ModelsRisk AssessmentRisk FactorsAgedMalePredictive Value of TestsRegistriesTime FactorsDecision Support TechniquesPeripheral Arterial DiseaseMiddle AgedRandom ForestReproducibility of ResultsClassification AlgorithmsBoosting Machine Learning AlgorithmsProspective StudiesClinical Decision-MakingPrediction AlgorithmsNurse's RoleMachine Learning