Comparison of different prediction models based on machine learning algorithms for the risk of poor postoperative wound healing in patients with spinal tuberculosis
10.3969/j.issn.1671-8348.2025.11.012
- VernacularTitle:基于机器学习算法的不同脊柱结核患者术后伤口愈合不良风险预测模型的比较
- Author:
Jinglian WEN
1
;
Wei TANG
;
Chengli WU
;
Run LI
;
Qing YE
;
Guoxuan PENG
Author Information
1. 贵州医科大学附属医院手术室,贵阳 550001
- Keywords:
spinal tuberculosis;
poor postoperative wound healing;
risk factors;
machine learning algo-rithm;
prediction model;
prediction scoring table
- From:
Chongqing Medicine
2025;54(11):2552-2558
- CountryChina
- Language:Chinese
-
Abstract:
Objective To analyze risk factors for poor postoperative wound healing in spinal tuberculo-sis patients and construct prediction models along with a risk scoring table using machine learning algorithms,providing references for early prevention and management.Methods Clinical data from 420 spinal tuberculo-sis patients treated at four tertiary hospitals in Guizhou Province between January 2017 and February 2024 were retrospectively analyzed.Risk factors were identified through univariate and multivariate analyses.Logis-tic regression,random forest,and support vector machine prediction models were constructed.Model perform-ance was evaluated using receiver operating characteristic(ROC)curves,precision,recall,accuracy,and F1-score.A risk prediction scoring table for poor postoperative wound healing was subsequently developed.Re-sults Among the 420 patients with spinal tuberculosis,132 experienced poor postoperative wound healing,with an incidence rate of 31.43%.Logistic regression analysis showed that BMI≤18.5 kg/m2,preoperative albumin≤30 g/L,combined surgical procedure,intraoperative blood loss>1 000 mL,direct bilirubin>8 μmol/L within 3 days after surgery,neutrophil count≤75×109/L within 3 days after surgery,and postopera-tive drainage volume>500 mL were risk factors for poor postoperative wound healing(P<0.05).In the comparison among the three risk prediction models constructed based on machine learning algorithms,the ran-dom forest model demonstrated the best predictive performance.A risk prediction scoring table was construc-ted based on the partial regression coefficients from the multivariate analysis,with a total score range from 0 to 11 points.A score>6 points indicated an increased risk of poor postoperative wound healing.The area un-der the ROC curve(AUC)for this risk prediction scoring table was 0.846(95%CI:0.769 to 0.923),indica-ting good predictive performance.Conclusion The random forest model based on machine learning algorithms and the risk prediction scoring table have certain predictive value for assessing the risk of poor postoperative wound healing in patients with spinal tuberculosis;they can assist healthcare professionals in early identifica-tion of high-risk patients.
