Development and validation of a multidimensional predictive model for 28-day mortality in ICU patients with bloodstream infections: a cohort study.

BACKGROUND

Bloodstream infections (BSI) are a leading cause of sepsis and death in intensive care unit (ICU). Traditional severity scores, including the Sequential Organ Failure Assessment (SOFA), Acute Physiology Score III (APSIII), and Simplified Acute Physiology Score II (SAPS II), exhibit limitations in effectively predicting mortality among BSI patients, primarily due to their reliance on a narrow range of clinical variables. This study aimed to develop and validate a comprehensive nomogram model for 28-day all-cause mortality prediction in BSI patients.

METHODS

A retrospective cohort study was conducted using data from 3,615 patients with positive blood cultures from the MIMIC-IV database, divided into training (n=2,532) and validation (n=1,083) cohorts. Through a two-step variable selection process combining LASSO regression and Boruta algorithm, we identified 12 predictive variables from 58 initial clinical parameters. The model's performance was evaluated using AUROC, net reclassification improvement (NRI), integrated discrimination improvement (IDI), and decision curve analysis (DCA).

RESULTS

The nomogram demonstrated superior discrimination (AUROC: 0.760 vs. 0.671, P<0.001 for SOFA; 0.760 vs. 0.705, P<0.001 for APSIII; 0.760 vs. 0.707, P<0.001 for SAPS II) in the training cohort, with consistent performance in the validation cohort (AUROC: 0.742). Key predictors identified in our model included the need for mechanical ventilation, the presence of malignancy, platelet count, and scores on the Glasgow Coma Scale (GCS). The model showed significant improvements in NRI and IDI, with consistent net benefit across a wide range of threshold probabilities in DCA.

CONCLUSIONS

This study developed and validated a predictive model for 28-day mortality in BSI patients that demonstrated superior performance compared to traditional severity scores. By integrating clinical, laboratory, and treatment-related variables, the model provides a more comprehensive approach to risk stratification. These findings highlight its potential for improving early identification of high-risk patients and guiding clinical decision-making, though further prospective validation is needed to confirm its generalizability.