Establishment of a prediction model of postoperative infection complications in patients with gastric cancer and its impact on prognosis

Background: Postoperative infection delays postoperative adjuvant therapy and can lead to poor prognosis in gastric cancer patients. Therefore, accurately identifying patients at high risk of postoperative infection in patients with gastric cancer is critical. We therefore conducted a study to analyze the impact of postoperative infection complications on long-term prognosis.

Methods: From January 2014 to December 2017, we retrospectively collected the data of 571 patients with gastric cancer admitted to the Affiliated People's Hospital of Ningbo University. The patients were divided into an infection group (n=81) and control group (n=490) according to whether the patients experienced postoperative infection. The clinical characteristics of the 2 groups were compared, and the risk factors of postoperative infection complications in patients with gastric cancer were analyzed. Finally, the prediction model of postoperative infection complications was established.

Results: There were significant differences in age, diabetes, preoperative anemia, preoperative albumin, preoperative gastrointestinal obstruction, and surgical methods between the 2 groups (P<0.05). Compared with that in the control group, the mortality rate of patients in the infection group at 5 years after surgery was significantly increased (39.51% vs. 26.12%; P=0.013). Multivariate logistics regression analysis showed that age >65 years, preoperative anemia, albumin <30 g/L, and gastrointestinal obstruction were risk factors of postoperative infection in patients with gastric cancer (P<0.05). The data set was randomly divided into a training set and validation set; the sample size of the training set was 286 while the sample size of the validation set was 285. In terms of the predictive model's value in predicting postoperative infection in patients with gastric cancer, the area under the curve of the receiver operating characteristic (ROC) curve in the training set was 0.788 (95% confidence interval: 0.711-0.864), and the area under the curve of the ROC curve in the validation set was 0.779 (95% confidence interval: 0.703-0.855). In the validation set, the model was evaluated with the Hosmer-Lemeshow goodness-of-fit test, resulting in a chi-squared value of 5.589 and a P value of 0.693.

Conclusions: The present model can effectively identify patient as high risk of postoperative infection.