Methods This study retrospectively collected data on patients with urosepsis from the MIMIC-IV database and Bijie Hospital of Zhejiang Provincial People’s Hospital. Multiple statistical methods were employed to explore the association between the RAR and short-term adverse outcomes, including multivariable Cox regression, restricted cubic spline (RCS) regression, and Kaplan–Meier (KM) survival analyses. Subsequently, three machine learning algorithms were utilized to screen for important features, followed by the construction of a multivariable Cox regression model for risk prediction. The performance of the risk prediction model was evaluated using receiver operating characteristic (ROC) curve analysis, with comparative validation performed via DeLong’s test. Results This study ultimately included 3,374 patients with urosepsis. The 28-day ICU mortality and in-hospital mortality rates were 15.20 and 13.75%, respectively. In the fully adjusted multivariate models, RAR, whether treated as a continuous or categorical variable, remained significantly associated with both 28-day ICU mortality and in-hospital mortality. For each unit increase in continuous RAR, the hazard ratios (HRs) were 1.10 (95% confidence interval CI: 1.05–1.16) and 1.09 (95% CI: 1.04–1.15), respectively. Compared with the low-RAR group, the high-RAR group showed HRs of 1.55 (95% CI: 1.19–2.01) and 1.39 (95% CI: 1.06–1.82) for the two outcomes. RCS analysis indicated a positive dose–response relationship between RAR and short-term adverse prognosis. DeLong’s test and ROC curve analysis demonstrated that RAR can appropriately enhance the predictive ability of routine critical illness scores for adverse outcomes. Moreover, a risk-prediction model incorporating RAR Slightly better than traditional severity scores (such as SOFA and SAPS II) in identifying high-risk patients. All findings were validated in an external cohort. Conclusion This study suggests that the RAR could serve as a predictor of short-term mortality risk in patients with urosepsis, with potential for translation into a clinical stratification tool to aid early identification of high-risk patients and guide intervention. However, its clinical utility needs to be further validated in larger prospective studies.
Yu et al. (Tue,) studied this question.