老年髋部骨折术后1年内再骨折危险因素预测模型的构建

Construction of prediction model of risk factors of re-fracture within 1 year after hip fracture surgery in the elderly

摘要

目的 分析老年髋部骨折术后1年内再骨折的危险因素，据此构建预测模型。方法 2018年7月至2024年1月鄂钢医院骨科收治老年髋部骨折患者1 413例，采用留出法，按照4∶1比例将患者随机分为建模队列（1 130例）和内部验证队列（283例）。另在其他医院收集同时期收治408例老年髋部骨折患者作为外部验证队列。采用中国知网和PubMed检索国内外相关文献，收集与老年髋部骨折术后1年内再骨折相关的变量，结合本研究纳入样本的临床资料，最终确定纳入分析的相关变量。采用LASSO-Logistic回归分析筛选关键变量进行多因素分析获得髋部骨折术后1年内再骨折的独立危险因素。据此分别构建Logistic回归模型、决策树模型、神经网络模型、支持向量机模型、K近邻模型及极限梯度提升模型。采用受试者工作特征曲线（ROC）验证模型的预测性能，选择最优预测模型与FRAX模型进行比较。结果 本研究髋部骨折术后1年内再骨折发生率为8.7%（123/1 413）。根据LASSO回归的λ_min共筛选出8个关键变量。多因素分析结果显示，年龄、周围血管病、骨质疏松、慢性阻塞性肺疾病、阿尔茨海默病、视觉障碍、肺部感染、尿路感染为再骨折的独立危险因素（P<0.05）。基于多因素分析获得的8个独立危险因素，构建了6种预测模型。综合考虑预测性能与临床适用性，以Logistic回归模型性能和临床适用性最佳。与FRAX模型比较，Logistic回归模型的预测性能明显高于FRAX模型。结论 利用老年髋部骨折术后1年内再骨折相关独立危险因素，结合机器学习算法构建预测模型，能较准确地预测老年髋部骨折患者术后1年再骨折的发生风险，可为患者再骨折的早期预防提供参考依据。

Abstract

Objective To analyze the risk factors for re-fracture within 1 year after hip fracture in the elderly, and construct a prediction model accordingly. Methods Retrospectively, 1 413 cases of elderly hip fracture patients admitted to Department of Orthopedics, Egang Hospital, from July 2018 to January 2024, were collected. The patients were randomly divided into modelling cohort (1 130 patients) and internal validation cohort (283 patients) according to the ratio of 4∶1 by the leave-out method; 408 elderly hip fracture patients admitted to other hospitals during the same period of time were collected as the external validation cohort. China National Knowledge Infrastructure and PubMed were used to search domestic and international relevent literature, from which variables related to re-fracture within 1 year after hip fracture in the elderly were collected, combined with the clinical data of the samples included in this study. Variables were included, and key variables were screened byLASSO-Logistic regression analysis, and multifactorial analyses were carried out to obtain the independent risk factors for re-fracture within 1 year after hip fracture, according to which the Logistic. Accordingly, Logistic regression model, decision tree model, neural network model, support vector machine model, K-nearest neighbour model and extreme gradient boosting model were constructed. The predictive performance of the models was verified by receiver operating characteristic curve (ROC), and the optimal predictive model was selected for comparison with the FRAX model. Results The incidence of re-fracture within 1 year after hip fracture was 8.70% (123/1 413). A total of 8 key variables were screened according to the λmin ofLASSO regression. The results of multifactorial analysis showed that age, peripheral vascular disease, osteoporosis, chronic obstructive pulmonary disease, Alzheimer's disease, visual impairment, pulmonary infection, and urinary tract infection were the independent risk factors for re-fracture (P<0.05). Six prediction models were constructed based on the eight independent risk factors obtained from the multifactorial analysis. Considering the prediction performance and clinical applicability, theLogistic model was the best in terms of performance and clinical applicability. Compared with the FRAX model, the predictive performance of the Logistic model was significantly higher. Conclusion The independent risk factors related to re-fracture within 1 year after surgery of elderly hip fracture combined with the machine learning algorithm, a prediction model was constructed, which can more accurately predict the risk of re-fracture of elderly hip fracture patients in the first year after surgery, and can provide a reference basis for early prevention of re-fracture of patients.

关键词

髋骨折 / 老年人 / 再发 / 机器学习算法

Key words

Hip fractures / Aged / Recurrence / Machine learning algorithm

参考文献

[1] Veronese N, Maggi S. Epidemiology and social costs of hip fracture[J]. Injury, 2018, 49(8): 1458-1460. DOI: 10.1016/j.injury.2018.04.015.

[2] Hjelholt TJ, Johnsen SP, Brynningsen PK, et al. Impact of stroke history on the risk of recurrent hip fracture or major osteoporotic fractures among patients with incident hip fracture: a nationwide cohort study[J]. J Bone Miner Res, 2023, 38(2): 278-287. DOI: 10.1002/jbmr.4760.

[3] Wang L, Yang M, Ge Y, et al. Risk prediction of second hip fracture by bone and muscle density of the hip varies with time after first hip fracture: a prospective cohort study[J]. Bone Rep, 2023, 20: 101732. DOI: 10.1016/j.bonr.2023.101732.

[4] Nymark T, Lauritsen JM, Ovesen O, et al. Short time-frame from first to second hip fracture in the funen county hip hracture study[J]. Osteoporos Int, 2006, 17(9): 1353-1357. DOI: 10.1007/s00198-006-0125-y.

[5] 刘四海, 王飞, 崔志刚, 等. 老年患者髋部骨折术后再骨折的相关危险因素分析[J]. 中国卫生统计, 2023, 40(3): 412-414. DOI: 10.11783/j.issn.1002-3674.2023.03.022.

[6] Liu S, Zhu Y, Chen W, et al. Risk factors for the second contralateral hip fracture in elderly patients: a systematic review and meta-analysis[J]. Clin Rehabil, 2015, 29(3): 285-294. DOI: 10.1177/0269215514542358.

[7] Wendt K, Heim D, Josten C, et al. Recommendations on hip fractures[J]. Eur J Trauma Emerg Surg, 2016, 42(4): 425-431. DOI: 10.1007/s00068-016-0684-3.

[8] 中华人民共和国国家卫生健康委员会医政司. 老年髋部骨折诊疗与管理指南（2022年版）[J]. 骨科临床与研究杂志, 2023, 8(2): 77-83. DOI:DOI: 10.3969/j.issn.1672-7185.2023.01.002.

[9] Riley RD, Ensor J, Snell KIE, et al. Calculating the sample size required for developing a clinical prediction model[J]. BMJ, 2020, 368: m441. DOI: 10.1136/bmj.m441

[10] 吴晓天, 潘福根, 付文芹, 等. 老年髋部骨折术后对侧髋部再骨折危险因素的荟萃分析[J]. 中国矫形外科杂志, 2019, 27(2): 149-154. DOI: 10.3977/j.issn.1005-8478.2019.02.12.

[11] Riley RD, Archer L, Snell KIE, et al. Evaluation of clinical prediction models (part 2): how to undertake an external validation study[J]. BMJ, 2024, 384: e074820. DOI: 10.1136/bmj-2023-074820.

[12] Shrestha B, Dunn L. The declaration of helsinki on medical research involving human subjects: a review of seventh revision[J]. J Nepal Health Res Counc, 2020, 17(4): 548-552. DOI: 10.33314/jnhrc.v17i4.1042.

[13] Jager-Wittenaar H, Sealy M, Naumann E, et al. Nutritional risk screening: a need to guide Alice in Nutritionland[J]. Curr Opin Clin Nutr Metab Care, 2024, 27(5): 381-386. DOI: 10.1097/MCO.0000000000001051.

[14] Charlson ME, Carrozzino D, Guidi J, et al. Charlson comorbidity index: a critical review of clinimetric properties[J]. Psychother Psychosom, 2022, 91(1): 8-35. DOI: 10.1159/000521288.

[15] 张立峰, 冯卫, 贾燕飞, 等. 老年髋部骨折术后对侧髋部再骨折危险因素的回顾性研究[J]. 中国医师杂志, 2017, 19(3): 447-449. DOI: 10.3760/cma.j.issn.1008-1372.2017.03.037.

[16] 郑繁荣, 王其飞, 贾俊秀, 等. 老年髋部骨折后对侧再骨折相关风险[J]. 中华骨质疏松和骨矿盐疾病杂志, 2017, 10(1): 43-47. DOI: 10.3969/j.issn.1674-2591.2017.01.008.

[17] Yu S, Li C, Zhong Y, et al. Risk factors for contralateral hip re-fractures in patients aged over 80 years with intertrochanteric femoral fractures[J]. Front Surg, 2022, 9: 924585. DOI: 10.3389/fsurg.2022.924585.

[18] 高宽惠, 李志超, 谭国庆, 等. 骨碎补抑制骨髓腔内脂肪异常积累引起的脂毒性介导的成骨细胞焦亡及其机制[J]. 中国骨质疏松杂志, 2024, 30(8): 1133-1139. DOI: 10.3969/j.issn.1006-7108.2024.08.007.

[19] 叶子, 王芬. 类风湿关节炎合并骨质疏松症患者外周血T淋巴细胞亚群、炎症因子水平与骨代谢指标的相关性分析[J]. 浙江医学, 2019, 41(9): 927-929. DOI: 10.12056/j.issn.1006-2785.2019.41.9.2018-1656.

[20] 汪洋, 竺义亮, 席金涛, 等. 血清炎症因子与老年慢性阻塞性肺疾病病人并发骨质疏松症的相关性分析[J]. 临床外科杂志, 2023, 31(5): 466-469. DOI: 10.3969/j.issn.1005-6483.2023.05.017.