This study aimed to investigate the predictive value of coagulation, thromboelastography, stress response, and immune function indicators for the occurrence of deep venous thrombosis (DVT) following radical resection of cervical cancer and ovarian cancer. We conducted a prospective, single-centre, case-control study that included 230 cervical cancer patients and 230 ovarian cancer patients. In the testing cohort, the final predictive model for cervical cancer patients was: Logit(P)=9.365-0.063(R-value)-0.112(K value) +0.386(α angle)+0.415(MA)+0.276(FIB)+0.423(D-D)+0.195(IL-6)+0.092(SOD). For ovarian cancer patients, the final model was: Logit(P)= -2.846-0.036(R-value)-0.157(K value) +0.426(α angle) +0.172(MA) +0.221(FIB)+0.375(CRP) -0.126(CD4+/CD8+). In the validation cohort, these models exhibited good predictive efficiency, with a false-positive rate of 12.5% and a false-negative rate of 2.9% for cervical cancer patients, and a false-positive rate of 14.3% and a false-negative rate of 0% for ovarian cancer patients. In conclusion, the risk prediction models developed in this study effectively improve the predictive accuracy of DVT following radical resection of cervical and ovarian cancer.
Keywords: Cervical cancer; machine learning; ovarian cancer; prospective study; venous thrombosis.