Construction and assessment of a predictive framework for Enterobacterales bloodstream infection in real-world adult onco-hematology patients using Lasso-Cox regression

Bloodstream infections (BSIs) caused by <i>Enterobacterales</i>, particularly <i>Escherichia coli</i> (EC) and <i>Klebsiella pneumoniae</i> (KP), are the most prevalent infectious issues in individuals with hematologic malignancies (HM). Prognostic tools integrating pathogen-specific risks are urgently needed to optimize early clinical decision-making. This retrospective cohort study analyzed 1,299 HM patients with EC- or KP-BSI (2017-2023). Patients were randomly allocated to training (<i>n</i> = 909) and validation (<i>n</i> = 390) cohorts. Prognostic variables were selected through Lasso regression, followed by Cox proportional hazards modeling to construct a 30-day mortality prediction nomogram. Model performance was rigorously validated using receiver operating characteristic curves (AUCs), calibration curves, and decision-curve analysis. The cohort comprised 653 EC-BSI and 646 KP-BSI cases, predominantly with acute myeloid leukemia (54.3%). Pulmonary/skin-soft tissue infection origins, hematopoietic stem cell transplantation, concurrent septic shock/pneumonia, hypoproteinemia, prolonged neutropenia before BSI, and carbapenem-resistant strains emerged as independent predictors of mortality. Conversely, tumor consolidation stage status and ceftazidime-avibactam-containing regimens were predictive of survival benefit. The model demonstrated high predictive accuracy (AUC 0.881), excellent calibration, and superior clinical utility. We developed a simple model using routine clinical parameters that predicts 30-day mortality in HM patients with EC- or KP-BSI, guiding therapy without complex biomarkers.