基于可解释机器学习的后路腰椎椎体间融合术后慢性疼痛风险预测模型构建

doi:10.6040/j.issn.1671-7554.0.2025.1184

摘要/Abstract

摘要： 目的基于机器学习构建高精度、可解释性好的后路腰椎椎体间融合术后慢性疼痛(chronic post-surgical pain, CPSP)风险预测模型,为临床早期识别高危人群、实现精准预防提供可靠工具。方法回顾性分析2019年1月至2023年12月在我院行后路腰椎椎体间融合术患者759例,其中男375例,女384例,33~80(55.28±9.94)岁。按7∶3比例分层随机分为训练集(n=531)和测试集(n=228)。收集所有患者术前、术中、术后共40项特征数据。行数据预处理和LASSO回归特征筛选后,构建7种机器学习模型,以受试者工作特征曲线下面积(area under the curve, AUC)、F1分数等作为核心指标筛选模型,使用沙普利加法解释(Shapley additive explanations, SHAP)工具对其进行可解释性分析。结果最终筛选出CPSP核心预测特征10项,其中疼痛灾难化量表(pain catastrophizing scale, PCS)、术前手术部位疼痛、并发症为三大核心驱动因素(累计贡献48.21%)。朴素贝叶斯(AUC=0.914)、逻辑回归(AUC=0.913)模型表现出较好的模型效果,朴素贝叶斯特异度达0.958,逻辑回归综合性能更均衡(F1分数=0.685)。SHAP分析明确了各特征对于预测结果的影响作用方向及强度,揭示PCS评分超过30分时,CPSP风险显著升高的阈值效应。结论利用解释性强的可解释机器学习算法所构建的后路腰椎椎体间融合术后CPSP风险预测模型,具有良好的判别效能和临床可解释性,且核心预测因子有助于指导临床个性化防控策略的开展,推动脊柱外科从“经验医学”向“数据驱动精准预防”转变。

关键词: 后路腰椎椎体间融合术, 术后慢性疼痛, 机器学习, 风险预测模型, 可解释人工智能, 算法比较

Abstract: Objective To construct a high-precision and well-interpretable risk prediction model for chronic post-surgical pain(CPSP)after posterior lumbar interbody fusion(PLIF), so as to provide a reliable tool for the early identification of high-risk populations and precise prevention in clinical practice. Methods A retrospective study was conducted on 759 patients who underwent PLIF at our hospital from January 2019 to December 2023, including 375 males and 384 females. The patients were 33-80(55.28±9.94)years old. All cases were stratified and randomly divided into a training set(n=531)and a testing set(n=228)at a ratio of 7∶3. A total of 40 characteristic variables were collected from all patients before, during, and after surgery. After data pre-processing and LASSO regression feature screening, seven machine learning models were constructed, with area under the curve(AUC), F1 score, and other indicators as the core indicator to select the optimal model, and the Shapley additive explanations(SHAP)tool was used for interpretability analysis. Results Ten core predictive features of CPSP were ultimately selected, among which pain catastrophizing scale(PCS)score, preoperative surgical site pain, and complications were the three core driving factors(cumulative contribution: 48.21%). The Naive Bayes(AUC=0.914)and logistic regression(AUC=0.913)models exhibited excellent performance, and logistic regression had a more balanced overall performance(F1 score=0.685). The Naive Bayes specificity was as high as 0.958. SHAP analysis clarified the direction and magnitude of the influence of each feature on the prediction results, revealing the threshold effect of significantly increasing CPSP risk when the PCS score exceeds 30 points. Conclusion The CPSP risk prediction model based on machine learning has good discriminative power and clinical interpretability. The core predictive factors provide clear targets for personalized prevention and control strategies in clinical practice, which helps to promote the transformation of spinal surgery from “empirical medicine” to “data-driven precision prevention”.

Key words: Posterior lumbar interbody fusion, Chronic post-surgical pain, Machine learning, Risk prediction model, Explainable artificial intelligence, Algorithm comparison

中图分类号:

R687.3

王建民,李晓峰,由志涛,董圣杰,赵宇驰,李占菊,邹德鑫,张剑锋,孙涛,杜伟. 基于可解释机器学习的后路腰椎椎体间融合术后慢性疼痛风险预测模型构建[J]. 山东大学学报 (医学版), 2026, 64(2): 78-88.

WANG Jianmin, LI Xiaofeng, YOU Zhitao, DONG Shengjie, ZHAO Yuchi, LI Zhanju, ZOU Dexin, ZHANG Jianfeng, SUN Tao, DU Wei. Construction of a chronic post-surgical pain prediction model for posterior lumbar interbody fusion surgery based on interpretable machine learning[J]. Journal of Shandong University (Health Sciences), 2026, 64(2): 78-88.

参考文献

[1] Wan ZY, Shan H, Liu TF, et al. Emerging issues questioning the current treatment strategies for lumbar disc herniation[J]. Front Surg, 2022, 9: 814531.doi:10.3389/fsurg.2022.814531
[2] Mills ES, Shelby T, Bouz GJ, et al. A decreasing national trend in lumbar disc arthroplasty[J]. Glob Spine J, 2023, 13(8): 2271-2277.
[3] Dantas F, Dantas FLR, Botelho RV. Effect of interbody fusion compared with posterolateral fusion on lumbar degenerative spondylolisthesis: a systematic review and meta-analysis[J]. Spine J, 2022, 22(5): 756-768.
[4] Wu QC, Ford NC, He SQ, et al. Characterizing a new rat model of chronic pain after spine surgery[J]. Bone Res, 2025, 13(1): 34. doi:10.1038/s41413-025-00408-1
[5] Carreon LY, Glassman SD, Howard J. Fusion and nonsurgical treatment for symptomatic lumbar degenerative disease: a systematic review of Oswestry Disability Index and MOS Short Form-36 outcomes[J]. Spine J, 2008, 8(5): 747-755.
[6] Wooden S. Chronic postsurgical pain[J]. Crit Care Nurs Clin N Am, 2024, 36(4): 585-595.
[7] Rosenberger DC, Segelcke D, Pogatzki-Zahn EM. Mechanisms inherent in acute-to-chronic pain after surgery-risk, diagnostic, predictive, and prognostic factors[J]. Curr Opin Support Palliat Care, 2023, 17(4): 324-337.
[8] Tassou A, Richebe P, Rivat C. Mechanisms of chronic postsurgical pain[J]. Reg Anesth Pain Med, 2025, 50(2): 77-85.
[9] Pak DJ, Yong RJ, Kaye AD, et al. Chronification of pain: mechanisms, current understanding, and clinical implications[J]. Curr Pain Headache Rep, 2018, 22(2): 9. doi:10.1007/s11916-018-0666-8
[10] 陈群生, 夏炳江. 脊柱手术患者术后慢性疼痛的危险因素及对策分析[J]. 浙江创伤外科, 2023, 28(11): 2058-2061. CHEN Qunsheng, XIA Bingjiang. Risk factors and countermeasures of chronic pain after spinal surgery[J]. Zhejiang Journal of Traumatic Surgery, 2023, 28(11): 2058-2061.
[11] Schober P, Vetter TR. Logistic regression in medical research[J]. Anesth Analg, 2021, 132(2): 365-366.
[12] Bayman EO, Dexter F. Multicollinearity in logistic regression models[J]. Anesth Analg, 2021, 133(2): 362-365.
[13] Wani SUD, Khan NA, Thakur G, et al. Utilization of artificial intelligence in disease prevention: diagnosis, treatment, and implications for the healthcare workforce[J]. Healthcare(Basel), 2022, 10(4): 608. doi:10.3390/healthcare10040608
[14] Karim MR, Islam T, Shajalal M, et al. Explainable AI for bioinformatics: methods, tools and applications[J]. Brief Bioinform, 2023, 24(5): bbad236. doi:10.1093/bib/bbad236
[15] Rathbone J, Rackham M, Nielsen D, et al. A systematic review of anterior lumbar interbody fusion(ALIF)versus posterior lumbar interbody fusion(PLIF), transforaminal lumbar interbody fusion(TLIF), posterolateral lumbar fusion(PLF)[J]. Eur Spine J, 2023, 32(6): 1911-1926.
[16] Shen N, Zhang YS, Ding ZD, et al. Predicting intrao-perative blood loss risk in severe lumbar disc herniation patients undergoing PLIF: a multicenter cohort study using ensemble learning[J]. Int J Surg, 2025, 111(9): 5904-5913.
[17] Mehren C, Ostendorff N, Schmeiser G, et al. Do TLIF and PLIF techniques differ in perioperative complications?-comparison of complications rates of two high volume centers[J]. Glob Spine J, 2025, 15(1): 84-93.
[18] Amaral R, Pokorny G, Marcelino F, et al. Lateral versus posterior approaches to treat degenerative lumbar pathologies-systematic review and meta-analysis of recent literature[J]. Eur Spine J, 2023, 32(5): 1655-1677.
[19] Schug SA, Lavandhomme P, Barke A, et al. The IASP classification of chronic pain for ICD-11: chronic postsurgical or posttraumatic pain[J]. Pain, 2019, 160(1): 45-52.
[20] Martinez V, Lehman T, Lavand'homme P, et al. Chronic postsurgical pain: a european survey[J]. Eur J Anaesthesiol, 2024, 41(5): 351-362.
[21] 许颖, 何龙, 王洁, 等. 脊柱侧弯矫形术患者术后慢性疼痛的危险因素[J]. 临床麻醉学杂志, 2022, 38(9): 944-948. XU Ying, HE Long, WANG Jie, et al. Risk factors for chronic post-surgical pain after scoliosis orthopaedics[J]. J Clin Anesthesiol, 2022, 38(9): 944-948.
[22] 宋子扬, 罗浩威, 贺毅, 等. 基于前瞻性队列建立与验证骨科术后慢性疼痛的预测模型[J]. 中国疼痛医学杂志, 2025, 31(9): 686-695. SONG Ziyang, LUO Haowei, HE Yi, et al. Development and validation of a prediction model for chronic post-surgical pain in orthopedic surgery based on a prospective cohort study[J]. Chin J Pain Med, 2025, 31(9): 686-695.
[23] Zhang J, Liu TX, Wang WX, et al. Effects of ultrasound-guided erector spinae plane block on postoperative acute pain and chronic post-surgical pain in patients underwent video-assisted thoracoscopic lobectomy: a prospective randomized, controlled trial[J]. BMC Anesthesiol, 2023, 23(1): 161. doi:10.1186/s12871-023-02100-5
[24] 中国康复医学会脊柱脊髓专业委员会腰椎研究学组, 赵杰, 海涌, 等. 后路腰椎椎体间融合技术规范的专家共识[J]. 中国脊柱脊髓杂志, 2024, 34(12): 1326-1334. Lumbar Spine Research Group, Spine and Spinal Cord Professional Committee, Chinese Rehabilitation Medical Association, ZHAO Jie, HAI Yong, et al. Expert consensus on technical specifications for posterior lumbar interbody fusion[J]. Chinese Journal of Spine and Spinal Cord, 2024, 34(12): 1326-1334.
[25] Chen D, Yang H, Yang L, et al. Preoperative psychological symptoms and chronic postsurgical pain: analysis of the prospective China Surgery and Anaesthesia Cohort study[J]. Br J Anaesth, 2024, 132(2): 359-371.
[26] 王丽云, 高天勤, 刘雨佳, 等. 基于机器学习产后压力性尿失禁风险预测模型的构建及验证[J]. 山东大学学报(医学版), 2025, 63(6): 55-66. WANG Liyun, GAO Tianqin, LIU Yujia, et al. Development and validation of a postpartum stress urinary incontinence risk prediction model based on machine learning[J]. Journal of Shandong University(Health Sciences), 2025, 63(6): 55-66.
[27] 贾若曦, 吕丽, 刘涵云, 等. 血细胞计数相关标志物对慢加急性乙型肝炎肝衰竭患者28天预后的诊断价值[J]. 山东大学学报(医学版), 2025, 63(6): 89-99. JIA Ruoxi, LYU Li, LIU Hanyun, et al. Diagnostic value of blood cell count-related biomarkers in predicting 28-day mortality in acute-on-chronic hepatitis b liver failure[J]. Journal of Shandong University(Health Sciences), 2025, 63(6): 89-99.
[28] Setiawan D, Wiranto Y, Girard JM, et al. Individualized machine-learning-based clinical assessment recommendation system[J]. PLoS Digit Health, 2025, 4(9): e0001022. doi:10.1371/journal.pdig.0001022
[29] Azodi CB, Tang JL, Shiu SH. Opening the black box: interpretable machine learning for geneticists[J]. Trends Genet, 2020, 36(6): 442-455.

多维度评价

Viewed

Full text

Abstract

Cited

Shared

Discussed