Outcomes and predictors of in-hospital mortality among patients admitted to the intensive care or step-down unit after a rapid response team activation: a retrospective cohort study

Introduction: It has been demonstrated that the implementation of rapid response teams (RRT) may improve clinical outcomes. Nevertheless, predictors of mortality among patients admitted to the intensive care unit (ICU) or to the step-down unit (SDU) after a RRT activation are not fully understood.

Objective: To describe clinical characteristics, resource use, main outcomes, and to address predictors of in-hospital mortality among patients admitted to the ICU/SDU after RRT activation.

Methods: Retrospective single-center cohort study conducted in a medical-surgical ICU/SDU located in a private quaternary care hospital. Adult patients admitted to the ICU or SDU between 2012 and 2020 were compared according to in-hospital mortality. A multivariate logistic regression analysis was performed to identify independent predictors of in-hospital mortality.

Results: Among the 3841 patients included in this analysis [3165 (82.4%) survivors and 676 (17.6%) non-survivors], 1972 (51.3%) were admitted to the ICU and 1869 (48.7%) were admitted to the SDU. Compared to survivors, non-survivors were older [76 (64-87) yrs. vs. 67 (50-81) yrs.; p<0.001], had a higher SAPS 3 score [64 (56-72) vs. 49 (40-57); p<0.001], and had a longer length of stay (LOS) before unit admission [8 (3-19) days vs. 2 (1-7) days; p<0.001). Non-survivors used more non-invasive ventilation (NIV) (42.2% vs. 20.9%; p<0.001), mechanical ventilation (MV) (36.7% vs. 9.3%; p<0.001), vasopressors (39.2% vs. 12.3%; p<0.001), renal replacement therapy (15.5% vs. 4.3%; p<0.001), and blood components transfusion (34.9% vs. 14.0%; p<0.001). Independent predictors of in-hospital mortality were the SAPS 3 score, the Charlson Comorbidity Index, LOS before unit admission, immunosuppression, respiratory rate <8 or >28 ipm criteria for RRT activation, RRT activation during the night shift, and the need for high-flow nasal cannula, NIV, MV, vasopressors, and blood components transfusion.

Conclusion: Multiple factors may affect outcomes of ICU/SDU-admitted patients after RRT activation. Therefore, efforts should be made to boost RRT effectiveness to improve patient safety.

Methods
Data Collection and Study Variables

All study data were retrieved from an institutional yellow code data bank, and from Epimed Monitor System® (Epimed Solutions, Rio de Janeiro, Brazil) (19), which are structured electronic case report forms where patients’ data are prospectively entered by trained hospital case managers. All data were extracted by an independent research assistant who did not participate in this study. All data were fully anonymized before being made available to the researchers. The data were accessed and extracted on 28/06/2021.

The collected variables included demographics, comorbidities, Simplified Acute Physiology Score (SAPS 3 score) (20), Sequential Organ Failure Assessment (SOFA) score (21) at ICU/SDU admission, Charlson Comorbidity Index (CCI) (22), Modified Frailty Index (MFI) (23), reason for ICU or SDU admission, the MEWS (10) at the moment of RRT activation, reason for RRT activation, patient’s location before unit admission, the time elapsed between patient deterioration and RRT activation, the time between RRT activation and the team’s arrival, destination after RRT activation (ICU or SDU), resource use and organ support [vasopressors, non-invasive ventilation (NIV), high flow nasal cannula (HFNC), mechanical ventilation (MV), renal replacement therapy (RRT) and blood components transfusion] during the first hour of ICU/SDU admission and during the ICU/SDU stay, unit and hospital length of stay (LOS), and hospital mortality.

Statistical analysis

Categorical variables were reported as absolute and relative frequencies, while continuous variables were presented as median with interquartile ranges (IQR). Normality was evaluated using the Kolmogorov-Smirnov test.

Comparisons were performed between survival and non-survival patients. Categorical variables were compared with the X2 test or Fisher’s exact test as appropriate. Continuous variables were compared using an independent t-test or the Mann–Whitney U test in cases of non-normal distribution.

Univariable logistic regression analysis was performed to identify which predictors were associated with in-hospital mortality. Multivariable logistic regression analyses with a backward elimination procedure, including all the predictors showing a p-value <0.20 in the univariable analysis, were undertaken to obtain an adjusted odds ratio (OR) along with 95% confidence interval (CI) and to identify which predictors were independently associated with in-hospital mortality. Therefore, the final model contained only variables significantly associated with in-hospital mortality after a multivariable backward logistic regression analysis.

To avoid collinearity, the characteristics included in the SAPS 3 score (namely, age) and the comorbidities included in the CCI [namely, diabetes mellitus, chronic heart failure (CHF), chronic kidney disease (CKD), solid tumor, liver cirrhosis, hematological malignancy, and severe chronic obstructive pulmonary disease (COPD)] were not individually added to the model. Collinearity was assessed using the variance inflation factor (VIF). A VIF >2.5 was arbitrarily defined as an indicator of collinearity.

We tested the linearity assumption for continuous variables included in logistic regression models by analyzing the interaction between each predictor and its own log (natural log transformation) (24). Whenever the linearity assumption was violated, continuous numerical variables were categorized (24). Final multivariable logistic regression model discrimination [area under a receiver operating characteristic curve (AUC)] and calibration (Hosmer-Lemeshow chi-square statistic) were reported (25).

Two-tailed tests were used, and statistical significance was set at p<0.05. All analyses were performed using the IBM Statistical Package for the Social Sciences (SPSS) Statistics for Macintosh, version 28 (IBM Corp., Armonk, NY, USA).