基于多维标志物构建老年骨折患者术后深静脉血栓形成的风险评估模型及性能验证

Development and validation of a predictive framework for post-surgical deep vein thrombosis risk in geriatric fracture cases utilizing multiple biomarker indicators

摘要:

目的 基于多维标志物构建老年骨折患者术后深静脉血栓形成（DVT）的风险评估模型，并验证其性能。方法 回顾性选取2022年1月—2025年12月宁夏医科大学附属固原市人民医院收治的320例老年骨折患者作为研究对象，以7∶3比例，采用留出法将患者随机分为建模组和验证组，分别有224、96例。收集患者临床资料，运用LASSO回归分析筛选关键变量并纳入多因素一般Logistic回归模型，获取独立危险因素用于构建列线图风险评估模型，采用受试者工作特征（ROC）曲线、校准曲线、决策曲线分析（DCA）、临床影响曲线（CIC）验证模型的预测性能和临床效能。结果 DVT组年龄、白细胞计数（WBC）、中性粒细胞计数、D-二聚体（D-D）、凝血酶原时间、纤维蛋白原（FIB）、白细胞介素-6、C反应蛋白（CRP）、全身免疫炎症指数均高于非DVT组，发病至入院时间、手术时间均长于非DVT组（P <0.05），白蛋白（ALB）、ALB与球蛋白比值、总蛋白水平均低于非DVT组（P <0.05）。多因素一般Logistic回归分析结果：年龄增加［O^R=1.082（95% CI：1.002，1.169）］、有吸烟史［O^R=4.512（95% CI：1.476，13.790）］、发病至入院时间延长［O^R=1.092（95% CI：1.034，1.153）］、WBC升高［O^R=1.580（95% CI：1.200，2.081）］、ALB降低［O^R=0.866（95% CI：0.761，0.984）］、D-D升高［O^R=1.517（95% CI：1.084，2.124）］、FIB升高［O^R=1.774（95% CI：1.150，2.735）］、CRP升高［O^R=1.109（95% CI：1.025，1.200）］均为老年骨折患者术后DVT的独立危险因素（P <0.05）。ROC曲线分析结果显示，建模组曲线下面积（AUC）为0.855（95% CI：0.789，0.920），敏感性、特异性分别为82.5%（95% CI：0.762，0.873）、75.5%（95% CI：0.691，0.810），约登指数为0.580；验证组AUC为0.901（95% CI：0.806，0.995），敏感性、特异性分别为88.9%（95% CI：0.800，0.948）、74.4%（95% CI：0.644，0.830），约登指数为0.633。Hosmer-Lemeshow检验结果显示，建模组和验证组模型拟合良好（P >0.10）。DCA曲线分析结果显示，在高风险阈值0.05～0.96和0.05～0.98采用列线图模型干预，存在净效益，当阈值概率为0.10～0.50时，模型净获益率显著优于全干预或不干预策略。验证组在阈值0.05～0.98均显示净获益，提示模型在不同风险偏好的临床决策场景中均具有实用价值。CIC曲线分析结果显示，在高风险阈值0.10～1.00采用列线图模型对患者进行术后DVT风险筛查，预测发生与实际发生DVT的患者例数高度吻合，绝对误差均<8例，相对误差<15%，表明模型预测结果与实际临床结局一致性良好。结论 该研究基于多维标志物构建的列线图风险评估模型，可准确预测老年骨折患者术后DVT的发生风险，可为患者术后DVT防治方案的合理制订提供循证依据。

Abstract:

Objective This study aimed to develop and validate a predictive nomogram model incorporating multiple biomarkers for assessing the likelihood of postoperative deep vein thrombosis (DVT) in elderly patients with fractures.Methods A retrospective analysis was conducted in 320 elderly patients with fractures treated at Guyuan People's Hospital Affiliated with Ningxia Medical University between January 2022 and December 2025. Using a hold-out method, patients were randomly allocated in a 7:3 ratio to a development cohort (n = 224) and a validation cohort (n = 96). Clinical data were collected, and LASSO regression was used to select key variables for inclusion in a multivariable logistic regression model. Independent risk factors were used to construct a nomogram. Receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and clinical impact curves (CIC) were used to evaluate predictive performance and clinical utility.Results Patients with DVT were older and had higher, white blood cell count (WBC), neutrophil count, D-dimer (D-D), prothrombin time, fibrinogen (FIB), interleukin-6, C-reactive protein (CRP), and systemic immune-inflammation index; longer onset-to-admission and operative times; and lower albumin (ALB), albumin-to-globulin ratio, and total protein than patients without DVT (P < 0.05). Multivariable logistic regression identified advanced age [O^R = 1.082 (95% CI: 1.002, 1.169) ], smoking history [O^R = 4.512 (95% CI: 1.476, 13.790) ], prolonged onset-to-admission time [O^R = 1.092 (95% CI: 1.034, 1.153) ], elevated WBC [O^R = 1.580 (95% CI: 1.200, 2.081) ], decreased ALB [O^R = 0.866 (95% CI: 0.761, 0.984) ], elevated D-D [O^R = 1.517 (95% CI: 1.084, 2.124) ], elevated FIB [O^R = 1.774 (95% CI: 1.150, 2.735) ], and elevated CRP [O^R = 1.109 (95% CI: 1.025, 1.200) ] as independent risk factors for postoperative DVT in elderly patients with fractures (P < 0.05). The AUCs were 0.855 (95% CI: 0.789, 0.920) in the development cohort and 0.901 (95% CI: 0.806, 0.995) in the validation cohort. Sensitivity, specificity, and the Youden index were 82.5% (95% CI: 0.762, 0.873), 75.5% (95% CI: 0.691, 0.810), and 0.580 in the development cohort and 88.9% (95% CI: 0.800, 0.948), 74.4% (95% CI: 0.644, 0.830), and 0.633 in the validation cohort, respectively. Hosmer-Lemeshow tests indicated good model fit in both cohorts (P > 0.10). DCA showed net benefit across threshold probabilities of 0.05-0.96 in the development cohort; at thresholds of 0.10-0.50, the model provided greater net benefit than either treat-all or treat-none strategies. The validation cohort showed net benefit across thresholds of 0.05-0.98. CIC analysis showed close agreement between predicted and observed DVT cases at thresholds of 0.10-1.00, with absolute errors of < 8 cases and relative errors of < 15%.Conclusion This nomogram based on multidimensional biomarkers accurately stratifies DVT risk following fracture surgery in elderly patients, offering evidence-based guidance for individualized thromboprophylaxis protocols.