Development of a nomogram for predicting in-hospital mortality in Escherichia coli bloodstream infection: a retrospective cohort study

Abstract Background Escherichia coli is the leading cause of bloodstream infection globally, with in-hospital mortality ranging from 5% to 30%. Although a multicentre validated mortality score (PROBAC) has recently been reported, validated bedside risk stratification tools specific to E. coli bacteraemia remain scarce in Chinese populations. Methods This retrospective cohort study included 202 consecutive adult patients with E. coli bacteraemia at The University of Hong Kong-Shenzhen Hospital (2019–2023). Given the low event rate, Firth penalized (bias-reduced) logistic regression was used as the primary method to identify independent predictors of in-hospital mortality, with ridge (L2) and LASSO (L1) regression as sensitivity analyses. Internal validation was performed by bootstrap optimism correction (1,000 replicates) and 5-fold cross-validation; a temporal split (development 2019–2021, n = 118, 17 events; exploratory validation 2022–2023, n = 84, 2 events) was used only as an exploratory sensitivity analysis. Results In-hospital mortality was 9.4% (19/202). Three independent predictors were identified: respiratory failure (Firth odds ratio [OR] 100.13, 95% confidence interval [CI] 13.58–738.13), malignancy (OR 93.36, 95% CI 8.16–1068.64), and blood urea nitrogen (BUN; OR 1.25 per mmol/L, 95% CI 1.11–1.41). The model showed an apparent area under the receiver operating characteristic curve (AUC) of 0.974 and a bootstrap optimism-corrected AUC of 0.969, with an apparent calibration slope near 1.13 and positive net benefit on decision curve analysis at threshold probabilities of 5–50% in the development cohort. Conclusion This three-variable nomogram, fitted with penalized regression to mitigate overfitting, shows promise for bedside risk stratification in E. coli bacteraemia. The model is developmental and requires external validation in independent multicentre cohorts before any clinical application.