Abstract Klebsiella pneumoniae is a major cause of severe infections among critically ill patients and is frequently associated with multidrug resistance (MDR). However, the relative prognostic importance of antimicrobial resistance, illness severity, bacteraemia, and Systemic Inflammation Score (SIS) remains incompletely understood. We evaluated the independent associations of organ dysfunction severity, bacteraemia, SIS, and MDR with in-hospital mortality among critically ill patients with microbiologically confirmed K. pneumoniae infection. This prospective observational study included 144 adult ICU patients with microbiologically confirmed monomicrobial K. pneumoniae infection. Clinical, laboratory, and microbiological variables were collected at the index infectious episode. Multivariable logistic regression was performed to identify variables independently associated with in-hospital mortality. Internal validation of the regression model was performed using 1,000 bootstrap resamples with bias-corrected and accelerated (BCa) confidence intervals. Exploratory machine learning analyses were performed using six supervised algorithms, and model interpretability was evaluated using Shapley Additive Explanations (SHAP). Overall in-hospital mortality was 52.8%. MDR and carbapenem-resistant isolates were identified in 67.4% and 54.9% of patients, respectively. In the primary multivariable model, SOFA score (adjusted odds ratio aOR 1.158, 95% confidence interval CI 1.071–1.252; p < 0.001), bacteraemia (aOR 3.609, 95% CI 1.504–8.657; p = 0.004), and SIS (aOR 3.260, 95% CI 1.241–8.565; p = 0.016) were independently associated with in-hospital mortality. Internal bootstrap validation using 1,000 resamples supported the robustness of the independent associations identified in the primary multivariable model (BCa 95% CIs: SOFA 1.06–1.31, bacteraemia 1.26–15.83, SIS 1.04–21.20). MDR was not independently associated with mortality after adjustment (aOR 1.352, 95% CI 0.396–4.615; p = 0.631). Among the exploratory machine learning models, Gaussian Naïve Bayes demonstrated the highest predictive performance (AUROC 0.78; Brier score 0.19). Global SHAP analysis ranked SOFA score, empirically active antimicrobial therapy, bacteraemia, and SIS as the most influential predictors within the prediction model. In this prospective observational cohort of critically ill patients with microbiologically confirmed K. pneumoniae infection, higher SOFA score, bacteraemia, and higher SIS were independently associated with in-hospital mortality, whereas MDR was not independently associated with mortality after multivariable adjustment, although this finding should be interpreted cautiously given the modest sample size and the wide confidence interval. The bootstrap analysis demonstrated the internal stability of the primary regression model. The exploratory machine learning analyses yielded predictor rankings broadly consistent with the regression findings but should be interpreted as complementary predictive analyses rather than evidence of causal or biological importance. External validation in independent cohorts is warranted before clinical implementation.
Nagendra et al. (2026) studied this question.