OBJECTIVE To establish a model for predicting the target achievement rate of anti-Ⅹa factor activity in the prevention of lower extremity deep vein thrombosis with low-molecular-weight heparin(LMWH) after orthopedic surgery and to identify its core influencing factors.

METHODS A retrospective analysis was conducted on 228 patients who received LMWH for lower extremity deep vein thrombosis prophylaxis after orthopedic surgery at the Sixth Affiliated Hospital of Xinjiang Medical University. Based on anti-Ⅹa factor activity, patients were divided into a target-achievement group(156 cases, ≥0.2 U·mL⁻¹) and a non-achievement group(72 cases, <0.2 U·mL⁻¹). They were randomly split into a training set(160 cases) and a validation set(68 cases) in a 7∶3 ratio. Patient demographic data, biochemical indices, bleeding events, and anti-Ⅹa activity results were collected. After comparing general data between the two groups and performing correlation heatmap analysis, machine learning methods including Binary Logistic Regression, LASSO Regression, and Random Forest were employed to evaluate and screen for influencing factors. Subsequently, a multivariable Logistic Regression was used to construct a nomogram prediction model. The models’ performance was evaluated using the area under the curve(AUC) of the receiver operating characteristic curve, calibration curves, and decision curve analysis.

RESULTS Based on Logistic Regression, four variables were identified: antithrombin Ⅲ(AT-Ⅲ) activity, daily dose, platelet count(PLT), and creatinine clearance(Ccr). LASSO Regression selected five variables: daily dose, AT-Ⅲ, PLT, Ccr, and prothrombin time(PT). Random Forest identified factors including daily dose, AT-Ⅲ, age, red blood cell count(RBC), and C-reactive protein(CRP). Using the core risk factors consistently retained across all three screening methods, model A was constructed via multivariable Logistic Regression. The regression equation was Y_A=0.052×daily dose+0.095×AT-Ⅲ. The nomogram’s total score was 109, and the final predicted probability of achieving target anti-Ⅹa activity was 0.794. The AUC in the training set was 0.756(sensitivity 98.2%, specificity 41.7%), and in the validation set, it was 0.722(sensitivity 81.8%, specificity 58.3%). Model B was built by combining the core risk factors with secondary risk factors retained in only two screening methods. Its regression equation was Y_B=0.04×daily dose+0.12×AT-Ⅲ−0.005×PLT−0.01×Ccr. The nomogram’s total score was 213, with a final predicted probability of 0.805. The training set AUC was 0.794(sensitivity 61.6%, specificity 83.3%), and the validation set AUC was 0.726(sensitivity 68.2%, specificity 79.2%). A full-variable model C incorporated all factors selected by the three methods. Its regression equation was Y_C=0.036×daily dose+0.115×AT-Ⅲ−0.004×PLT−0.01×Ccr−0.003×age−0.206×PT−0.251×RBC−0.004×CRP. The nomogram’s total score was 430, and the final predicted probability was 0.885. The training set AUC was 0.805(sensitivity 69.6%, specificity 79.2%), and the validation set AUC was 0.775(sensitivity 77.3%, specificity 66.7%). Calibration curves showed good alignment with the ideal curve, indicating satisfactory predictive accuracy for the models.

CONCLUSION This study identifies daily LMWH dose and AT-Ⅲ activity as the two core factors influencing the target achievement rate of anti-Ⅹa activity in orthopedic postoperative patients. The developed nomogram models demonstrate high predictive value and potential for aiding clinical decision-making. This series of models is expected to provide practical tools for formulating personalized anticoagulation regimens for orthopedic patients.