1592-P: Prediction Method for Infection Risk in Perioperative Patients with Diabetes

Abstract

Background/Purpose: Perioperative blood glucose management is important for complication prevention and early hospital discharge. The relation between blood glucose fluctuations and infection is few reported. This study aimed to find predictive risk factors to identify high-risk patients with diabetes requiring more extensive management. Methods: A retrospective study was performed on the perioperative blood glucose management of 105 patients. The presence of infection within one month after surgery was the primary outcome, and the predictive factors were identified. For statistical analysis, analysis of variance or Wilcoxon rank sum test was used. Based on univariate analysis, principle component analysis and logistic analysis were performed to prepare a prediction model of infection onset within one month after surgery using predictive risk factors. The area under the receiver operating characteristic curve (AUC) was evaluated. Results: Patients with infection within one month after surgery had higher mean blood glucose levels (185.1±28.7 vs. 168.3±33.6 mg/dL, P=0.013), larger preoperative blood glucose fluctuations (54.9±24.1 vs. 37.7±23.1 mg/dL, P=0.006), lower albumin level (3.6±0.6 vs. 3.9±0.6 g/dL, P=0.046), longer operation time (432.5±179.6 vs. 282.5±178.3 minute, P<0.001), more bleeding (972.3±920.1 vs. 436.4±795.8 mg/dL, P=0.003), and longer postoperative hospitalization periods (51.9±56.1 vs. 22.4±24.3 days, P<0.001) than patients without infection. They also had greater frequency of adverse events (hazard ratio (HR)=3.19, P=0.013) and mortality within 1 year (HR=4.03, P=0.026). Preoperative blood glucose fluctuation and operation time were considered as more reliable factors of the prediction model. The model had high prediction accuracy with AUC 0.801. Discussion: Preoperative blood glucose fluctuations and long operation time were risk factors for infection in perioperative patients with diabetes. This prediction model can detect these high-risk patients. Disclosure M. Koshizaka: Research Support; Self; Astellas Pharma Inc., Pfizer Health Research Foundation, Taisho Pharmaceutical Co., Ltd. R. Ishibashi: None. Y. Maeda: None. T. Ishikawa: None. Y. Maezawa: None. M. Takemoto: None. K. Yokote: None. Funding Pfizer Health Research Foundation