Novel risk prediction models for deep vein thrombosis after thoracotomy and thoracoscopic lung cancer resections, involving coagulation and immune function

Abstract

The main focus of this study was to compare the predictive value of coagulation, fibrinolysis, thromboelastography, stress response, and immune function in predicting the incidence of deep venous thrombosis (DVT) in lung cancer (LC) patients undergoing thoracoscopic LC resection vs thoracotomy LC resection. To do that, a prospective, single-center, case-control study involving 460 LC patients was conducted. The risk indicators affecting patients with DVT after LC resection in the testing cohort were determined using logistic regression and receiver operator characteristic (ROC) analyses. One validation cohort was used to assess the risk prediction models. DVT incidence was higher in the thoracoscopic group (18.7%) than in the thoracotomy group (11.2%) in the testing cohort (χ ² = 4.116, P = 0.042). The final model to predict the incidence of DVT after thoracoscopic LC excision (1 day after surgery) was as follows: Logit(P) = 9.378 - 0.061(R-value) - 0.109(K value) + 0.374(α angle) + 0.403(MA) + 0.298(FIB) + 0.406(D-D) + 0.190(MDA) - 0.097(CD4⁺/CD8⁺). For thoracotomy LC resection, the final model (3 days after operation) was: Logit(P) = -2.463 - 0.026(R-value) - 0.143(K value) + 0.402(α angle) + 0.198(D-D) + 0.237(MDA) + 0.409(SOD). In the validation cohort, this risk prediction model continued to demonstrate good predictive performance. As a result, the predictive accuracy of postoperative DVT in patients who underwent thoracoscopic LC resection and thoracotomy LC resection was improved by risk prediction models.

Keywords: lung cancer; lung cancer resection; machine learning; prospective study; venous thrombosis.