| 王颖,李可清.青少年特发性脊柱侧弯患者术后慢性疼痛预测及护理决策系统研究.骨科,2026,17(2): 160-166. |
| 青少年特发性脊柱侧弯患者术后慢性疼痛预测及护理决策系统研究 |
| Chronic post-surgical pain prediction and nursing decision system for adolescent idiopathic scoliosis patients |
| 投稿时间：2025-09-09 |
| DOI：10.3969/j.issn.1674-8573.2026.02.012 |
| CN KeyWords: 青少年特发性脊柱侧弯 慢性术后疼痛 自适应机器学习 风险预测模型 护理因素 |
| EN KeyWords: Adolescent idiopathic scoliosis Chronic post-surgical pain Adaptive machine learning Risk prediction model Nursing factors |
| Fund Project: |
|
| View Counts: 19 |
| PDF Download Counts: 0 |
| CN Abstract: |
| 目的 构建并验证一个整合多维临床和生物心理社会因素的改进型吕佩尔狐优化算法（IRFO）自适应机器学习（AutoML）框架，精准预测青少年特发性脊柱侧弯（AIS）患者脊柱融合术后慢性疼痛（CPSP）风险。方法 纳入668例AIS手术患者，采用8∶2比例随机划分训练集（534例）与测试集（134例）。开发IRFO-AutoML框架，通过双阶段优化流程。特征筛选应用AutoML识别初始特征，再以LASSO回归验证稳健性；沙普利加法解释（SHAP）模型解析可解释性。评价指标包括分类性能（ROC-AUC、PR-AUC）、校准性能（Brier分数）及临床应用（决策曲线分析，DCA）。结果 训练集与测试集结局事件（CPSP）比例一致（33.90% vs. 34.33%，χ2=0.009，P=0.925）。模型识别7项核心预测因子：患者疼痛灾难化量表（PCS）评分、C反应蛋白（CRP）峰值、术前疼痛数字评分量表（NRS）评分、监护人焦虑评分、主弯Cobb角、下床活动时间、术中失血量。IRFO-AutoML在测试集性能显著优于对比模型（ROC-AUC：0.923 9；PR-AUC：0.843 4；F1：0.745 5）。SHAP分析揭示高PCS与高CRP组合、重度侧弯与大量失血使风险增加。DCA证明模型在1%~99%的风险阈值内具有较好的临床净获益。结论 本研究开发的IRFO-AutoML框架有效整合多维因素，解决高维数据处理瓶颈并优化预测精度，显著提升AIS患者术后CPSP预测精度，其闭环决策支持系统为临床早干预提供新工具。 |
| EN Abstract: |
| Objective To construct and validate an improved Rüppell's fox optimizer (IRFO)-based adaptive machine learning (AutoML) framework integrating multi-dimensional clinical and biopsychosocial factors, for the precise prediction of chronic post-surgical pain (CPSP) risk in adolescent idiopathic scoliosis (AIS) patients after spinal fusion surgery. Methods A total of 668 AIS surgical patients were included, and they were randomly divided into training set (534 cases) and testing set (134 cases) at a ratio of 8∶2. The IRFO-AutoML framework was developed through a dual-stage optimization process. For feature selection, AutoML was applied to identify initial features, and LASSO regression was used to verify robustness. The SHAP model was employed to analyze interpretability. Evaluation metrics included classification performance (ROC-AUC, PR-AUC), calibration performance (Brier score), and clinical utility (decision curve analysis, DCA). Results Outcome event (CPSP) rates were consistent between training and testing sets (33.90% vs. 34.33%, χ2=0.009, P=0.925). The model identified seven core predictors: patient pain catastrophizing scale (PCS) score, CRP peak value, preoperative NRS score, guardian anxiety score, main curve Cobb angle, time to ambulation, and intraoperative blood loss. IRFO-AutoML significantly outperformed comparison models in testing set performance (ROC-AUC: 0.923 9; PR-AUC: 0.843 4; F1: 0.745 5). SHAP analysis revealed clinical interactions: the combination of high PCS and high CRP significantly increased risk; severe scoliosis and massive blood loss also elevated risk. DCA demonstrated net benefit within 1% to 99% risk thresholds. Conclusion The IRFO-AutoML framework effectively integrates multi-dimensional factors, addresses high-dimensional data processing bottlenecks, optimizes predictive accuracy, and significantly improves CPSP prediction for AIS patients. Its closed-loop decision support system provides a novel tool for early clinical intervention. |
|
查看全文
Download PDF Reader |
| CLose |
|
|
|
