Development and validation of a predictive model for in-hospital mortality from perioperative bacteremia in gastrointestinal surgery.

Abstract

Purpose: Prognostic scores require fluctuating values, such as respiratory rate, which are unsuitable for retrospective auditing. Therefore, this study aimed to develop and validate a predictive model for in-hospital mortality associated with gastrointestinal surgery for retrospective auditing.

Methods: Data from patients with bacteremia related to gastrointestinal surgery performed at Shizuoka General Hospital between July 2006 and December 2021 were extracted from a prospectively maintained database. Patients suspected of having a positive blood culture with contaminating bacteria or missing laboratory data were excluded. The remaining patients were randomly assigned in a 2:1 ratio to the deviation and validation cohorts. A logistic regression model estimated the odds ratios (ORs) and created a predictive model for in-hospital mortality. The model was evaluated using receiver operating characteristic (ROC) curves and calibration plots.

Results: Of 20,637 gastrointestinal surgeries, 398 resulted in bacteremia. The median age of patients with bacteremia was 72 years, and 66.1% were male. The most common pathogens were Staphylococcus (13.9%), followed by Bacteroides (12.4%) and Escherichia (11.4%). Multivariable logistic regression showed that creatinine abnormality (P < 0.001, OR = 3.39), decreased prognostic nutritional index (P < 0.001, OR = 0.90/unit), and age ≥ 75 years (P = 0.026, OR = 2.89) were independent prognostic factors for in-hospital mortality. The area under the ROC curve of the predictive model was 0.711 in the validation cohort. The calibration plot revealed that the model slightly overestimated mortality in the validation cohort.

Conclusions: Using age, creatinine level, albumin level, and lymphocyte count, the model accurately predicted in-hospital mortality after bacteremia infection related to gastrointestinal surgery, demonstrating its suitability for retrospective audits.