To construct and validate an individualized nomogram to predict the probability of occurrence of portal vein thrombosis (PVT) after splenectomy in patients with hepatitis B cirrhosis.
We retrospectively collected the clinical data from 180 patients with hepatitis B cirrhosis undergoing splenectomy with postoperative anticoagulation therapy during the period from January, 2014 to January, 2020 in our hospital. The patients were randomized into modeling group (= 120) and validation group (=60), and the former group was further divided into PVT group (=49) and non-PVT group (=71) according to the occurrence of PVT occurred within 1 month after splenectomy. The independent risk factors of PVT after splenectomy were screened in the modeling group using univariate and multivariate binary logistic regression analyses and were used for construction of the nomogram prediction model. The area under the receiver-operating characteristic (AUROC) curve (C-index), GiViTI calibration belt and Hosmer-Lemeshow test, and the DCA curve were used to estimate the discrimination power, calibration and clinical efficiency of the prediction model in both the model construction group and validation group.
Univariate and multivariate logistic regression analyses showed that a history of hemorrhage, portal vein diameter, spleen vein diameter, spleen volume, varicose, postoperative platelet change, and postoperative D-dimer differed significantly between PVT group and non-PVT group ( < 0.05), and portal vein diameter, spleen vein diameter, and postoperative platelet change were independent risk factors of PVT after splenectomy ( < 0.05). The prediction model had a good discrimination power with AUROC (C-index) of 0.880 (95% : 0.818-0.942) in the modeling group and 0.873 (95% : 0.785-0.960) in the validation group. The 80% and 95% region of GiViTI calibration belt did not cover the 45-degree diagonal bisector line (=0.965 and 0.632, respectively), and the P-values of the Hosmer-Lemeshow test were 0.624 and 0.911, respectively, suggesting a high reliability of the predicted probability by the model. DCA curve analysis showed a threshold probability of 30.5%, with a net benefit of 30% in the modeling group and 34% in the validation group, indicating a good clinical efficiency of the model.
The model for predicting the risk of PVT after splenectomy in patients with hepatitis B cirrhosis can help in early identification of patients having high risks of PVT.