Bloodstream infection clusters for critically ill patients: analysis of two-center retrospective cohorts

Abstract

Abstract Background Bloodstream infections (BSI) are highly prevalent in hospitalized patients requiring intensive care. They are among the most serious infections and are highly associated with sepsis or septic shock, which can lead to prolonged hospital stays and high healthcare costs. This study aimed at establishing an easy-to-use nomogram for predicting the prognosis of patients with BSI. Methods In retrospective study, records of patients with BSI admitted to the intensive care unit (ICU) over the period from Jan 1st 2016 to Dec 31st 2021 were included. We used data from two different China hospitals as development cohort and validation cohort respectively. The demographic and clinical data of patients were collected. Based on all baseline data, k-means algorithm was applied to discover the groups of BSI phenotypes with different prognostic outcomes, which was confirmed by Kaplan-Meier analysis and compared using log-rank tests. Univariate Cox regression analyses were used to estimate the risk of clusters. Random forest was used to identified discriminative predictors in clusters, which were utilized to construct nomogram based on multivariable logistic regression in the discovery cohort. For easy clinical applications, we developed a bloodstream infections clustering (BSIC) score according to the nomogram. The results were validated in the validation cohort over a similar period. Results A total of 360 patients in the discovery cohort and 310 patients in the validation cohort were included in statistical analyses. Based on baseline variables, two distinct clusters with differing prognostic outcomes were identified in the discovery cohort. Population in cluster 1 was 211 with a ICU mortality of 17.1%, while population in cluster 2 was 149 with an ICU mortality of 41.6% (p < 0.001). The survival analysis also revealed a higher risk of death for cluster 2 when compared with cluster 1 (hazard ratio: 2.31 [95% CI, 1.53 to 3.51], p < 0.001), which was confirmed in validation cohort. Four independent predictors (vasoconstrictor use before BSI, mechanical ventilation (MV) before BSI, Deep vein catheterization (DVC) before BSI, and antibiotic use before BSI) were identified and used to develop a nomogram. The nomogram and BSIC score showed good discrimination with AUC of 0.96. Conclusion The developed score has potential applications in the identification of high-risk critically ill BSI patients.