Zhang Bo, Qin Yumei, Jiu Liandi, Qin Chunming, Wang Jiangbo, Zhao Haiqing
Digital Health China Technologies Co., Ltd., Beijing, China.
Nanxishan Hospital of Guangxi Zhuang Autonomous Region, The Second People's Hospital of Guangxi Zhuang Autonomous Region, Guilin, China.
Front Med (Lausanne). 2025 Jun 26;12:1574546. doi: 10.3389/fmed.2025.1574546. eCollection 2025.
To construct a venous thromboembolism (VTE) risk prediction model for orthopedic inpatients using machine learning modeling techniques, identify high-risk patients, and optimize clinical interventions.
This study involved a retrospective analysis of 286 orthopedic inpatients from Nanxishan Hospital of Guangxi Zhuang Autonomous Region (The Second People's Hospital of Guangxi Zhuang Autonomous Region) from January 1, 2022 to December 31, 2022. To ensure patient information security, all data were fully anonymized before access. The collected data included basic information such as gender, age, ethnicity, and body mass index (BMI), lifestyle factors and medical history (including smoking, alcohol use, diabetes, hypertension, and personal and family history of VTE), clinical test results (such as thrombin time, plasma D-dimer, total bilirubin, and urinary protein via dry chemistry), as well as genetic test results related to VTE risk. Feature analysis and data mining were conducted, and eight different machine learning algorithms were used to build the prediction model. The SHapley Additive exPlanation (SHAP) method was used to rank the feature importance and explain the final model.
Through a comprehensive evaluation and comparison of eight different machine learning models, the results clearly indicate that the XGBoost model outperforms the others across all performance metrics, achieving the highest accuracy of 0.828 and AUROC of 0.931, significantly surpassing the other models, particularly in prediction accuracy and discriminative ability. Compared to the traditional Caprini scoring model, XGBoost not only shows improvements in accuracy and specificity but also demonstrates a significant increase in Area Under the Curve (AUC), further validating its superior performance in VTE risk prediction.
This model can be effectively used for early risk prediction of VTE, helping to reduce the incidence of venous thromboembolism in orthopedic patients. Given its promising results, further validation and wider application of the model in clinical settings are warranted to enhance patient outcomes and improve preventive strategies.
运用机器学习建模技术构建骨科住院患者静脉血栓栓塞症（VTE）风险预测模型，识别高危患者，并优化临床干预措施。
本研究对广西壮族自治区南溪山医院（广西壮族自治区第二人民医院）2022年1月1日至2022年12月31日期间的286例骨科住院患者进行回顾性分析。为确保患者信息安全，所有数据在访问前均进行了完全匿名化处理。收集的数据包括性别、年龄、种族和体重指数（BMI）等基本信息、生活方式因素和病史（包括吸烟、饮酒、糖尿病、高血压以及VTE个人和家族史）、临床检验结果（如凝血酶时间、血浆D-二聚体、总胆红素和干化学法检测的尿蛋白）以及与VTE风险相关的基因检测结果。进行了特征分析和数据挖掘，并使用八种不同的机器学习算法构建预测模型。采用SHapley加性解释（SHAP）方法对特征重要性进行排名并解释最终模型。
通过对八种不同机器学习模型的综合评估和比较，结果清楚地表明，XGBoost模型在所有性能指标上均优于其他模型，准确率最高达到0.828，曲线下面积（AUROC）为0.931，显著超过其他模型，尤其是在预测准确性和判别能力方面。与传统的Caprini评分模型相比，XGBoost不仅在准确性和特异性方面有所提高，而且曲线下面积（AUC）也显著增加，进一步验证了其在VTE风险预测中的优越性能。
该模型可有效用于VTE的早期风险预测，有助于降低骨科患者静脉血栓栓塞症的发生率。鉴于其良好的结果，有必要在临床环境中对该模型进行进一步验证和更广泛的应用，以改善患者预后并优化预防策略。
