Journal of orthopaedic surgery and research · 2026
Machine learning versus clinical scoring systems for predicting poor wound healing after posterior surgery for thoracolumbar tuberculosis.
СтатьяСравнительное 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.
Xiong Tuotuo, Huo SiYan, Zhang Ye, Qin Wanyuan, Fang Mingyang, Zhu Yong
Аннотация
OBJECTIVE: Postoperative poor wound healing (PWH) is a significant complication following posterior surgery for thoracolumbar tuberculosis, leading to prolonged recovery and increased healthcare costs. Traditional prognostic markers often lack precision. This study aimed to develop and validate machine learning (ML)-based models to predict PWH and compare their performance with conventional logistic regression and established clinical scoring systems.
METHOD: A retrospective analysis of 188 patients undergoing posterior debridement and internal fixation for thoracolumbar tuberculosis was conducted. Patients were randomly partitioned into a training set (70%, n = 132) and a test set (30%, n = 56). Four predictive models were constructed: Model I (Multivariate Logistic Regression), Model II (LASSO Regression), Model III (Random Forest), and Model IV (Artificial Neural Network). The performance of these models was compared against the Prognostic Nutritional Index (PNI) and the Naples Prognostic Score (NPS) using the Area Under the Curve (AUC) and Decision Curve Analysis (DCA).
RESULTS: The incidence of PWH was analyzed alongside preoperative and intraoperative variables. LASSO regression identified six significant predictors: albumin, lymphocytes (%), C-reactive protein (CRP), white blood cell (WBC) count, hemoglobin, and the number of instrumented levels. Based on these features, a nomogram (Model II) was developed. In the test set, the LASSO-based model achieved the highest predictive accuracy with an AUC of 0.879, significantly outperforming the traditional logistic regression model (AUC: 0.733), the Random Forest model (AUC: 0.860), and the Artificial Neural Network (AUC: 0.840). Furthermore, the LASSO model demonstrated superior discrimination compared to clinical scoring systems (PNI: 0.760; NPS: 0.746). DCA confirmed that the LASSO-based nomogram provided the highest clinical net benefit.
CONCLUSION: The LASSO-based nomogram demonstrated superior predictive accuracy and clinical net benefit for PWH compared to traditional regression methods and established prognostic scores. For limited clinical cohorts, penalized regression techniques effectively mitigate overfitting risks and provide robust interpretability compared to highly parameterized algorithms. This validated nomogram serves as a practical and reliable tool for early risk stratification, thereby optimizing personalized perioperative management for patients with thoracolumbar tuberculosis.
METHOD: A retrospective analysis of 188 patients undergoing posterior debridement and internal fixation for thoracolumbar tuberculosis was conducted. Patients were randomly partitioned into a training set (70%, n = 132) and a test set (30%, n = 56). Four predictive models were constructed: Model I (Multivariate Logistic Regression), Model II (LASSO Regression), Model III (Random Forest), and Model IV (Artificial Neural Network). The performance of these models was compared against the Prognostic Nutritional Index (PNI) and the Naples Prognostic Score (NPS) using the Area Under the Curve (AUC) and Decision Curve Analysis (DCA).
RESULTS: The incidence of PWH was analyzed alongside preoperative and intraoperative variables. LASSO regression identified six significant predictors: albumin, lymphocytes (%), C-reactive protein (CRP), white blood cell (WBC) count, hemoglobin, and the number of instrumented levels. Based on these features, a nomogram (Model II) was developed. In the test set, the LASSO-based model achieved the highest predictive accuracy with an AUC of 0.879, significantly outperforming the traditional logistic regression model (AUC: 0.733), the Random Forest model (AUC: 0.860), and the Artificial Neural Network (AUC: 0.840). Furthermore, the LASSO model demonstrated superior discrimination compared to clinical scoring systems (PNI: 0.760; NPS: 0.746). DCA confirmed that the LASSO-based nomogram provided the highest clinical net benefit.
CONCLUSION: The LASSO-based nomogram demonstrated superior predictive accuracy and clinical net benefit for PWH compared to traditional regression methods and established prognostic scores. For limited clinical cohorts, penalized regression techniques effectively mitigate overfitting risks and provide robust interpretability compared to highly parameterized algorithms. This validated nomogram serves as a practical and reliable tool for early risk stratification, thereby optimizing personalized perioperative management for patients with thoracolumbar tuberculosis.