Establishment and validation of a predictive nomogram for preoperative deep vein thrombosis in elderly patients with isolated femoral neck fracture

Abstract

Current assessment of preoperative deep vein thrombosis (DVT) in elderly patients with femoral neck fracture is limited. We aimed to identify independent risk factors for preoperative DVT in patients with femoral neck fracture and to develop and validate a predictive nomogram. Patients diagnosed with femoral neck fracture from October 2014 to April 2019 were retrospectively analyzed. Baseline data, treatment information and laboratory test results were collected, and the occurrence of DVT in the preoperative period was regarded as the study outcome event. Multivariable logistic regression identified independent risk factors associated with a higher incidence of preoperative DVT. The predictive nomogram was constructed based on the analysis results. The stability of the model was further assessed in this study using patients from May 2019 to September 2022 as an external validation set. A total of 921 patients were enrolled in the study, of which 639 were used in the training cohort and the other 282 for the validation cohort. Multivariate analysis revealed age (OR 1.100, 95% CI 1.042-1.162), BMI ≥ 28 (OR 3.969, 95% CI 1.792-8.793), smoking (OR 2.998, 95% CI 1.255-7.165), LDL-C > 3.4 mmol/L (OR 2.628, 95% CI 1.316-5.250), and d-dimer > 0.475 mg/L (OR 3.157, 95% CI 1.565-6.368) were the independent risk factors of preoperative DVT. The concordance index (C-index) of the nomogram were 0.832 in the training set, and the corrected values after internal validation were 0.759. The receiver-operating characteristic (ROC) curve, the calibration curve, the Hosmer-Lemeshow test and the decision curve analysis (DCA) performed well in both the training and validation cohorts. In this study, we developed a personalised predictive nomogram containing five high-risk factors, which can help surgeons stratify the risk of preoperative DVT in elderly patients with femoral necks and guide high-risk patients to ultrasound scans or prophylactic anticoagulation as soon as possible.

Keywords: Femoral neck fracture; Prediction model; Preoperative deep vein thrombosis; Risk factor.

Fig. 1

Patients selection flow-chart.

Fig. 2

Nomogram for predicting preoperative DVT in patients with femoral neck fracture. Five factors were calculated into the preoperative DVT prediction nomogram, with each predictor assigned a given score on the top points axis, and the predicted probability of DVT corresponding to the total points was shown on the bottom probability axis.

Fig. 3

Receiver-operating characteristic (ROC) curves for the nomogram in the training (a) and validation sets (b). The predictive accuracy of the nomogram was positively correlated with the area under the curve (AUC). The AUC of the nomogram was 0.759 and 0.774 in the training and validation sets, respectively, indicating that the model had good discriminative ability. The specificity and sensitivity of the training set are 0.811 and 0.660, and the specificity and sensitivity of the validation set are 0.699 and 0.769.

Fig. 4

Calibration curves of nomogram in the training set and validation set. X-axis represents the predicted probability of the model and y-axis represents the actual probability. The closer the red and green curves fit the ideal dashed line, the better the predictive consistency of the nomogram.

Fig. 5

Decision curve analysis (DCA) of nomogram in the training set (a) and validation set (b). DCA illustrated that the net benefit of the training model is higher in the threshold probability interval of 6–85%, and the net benefit of the validation model is higher in the threshold probability interval of 3–78%.