Despite initial recovery from critical illness requiring intensive care unit (ICU) admission, many patients remain at risk of subsequent deterioration and unplanned readmission to the ICU. Readmitted patients have higher mortality rates and significantly greater lengths of stay. Scoring systems designed to measure the severity of illness for patients admitted to the ICU have been used to predict the risk of readmission to the ICU, but are often too complex to be practical. Recently, the Stability and Workload Index For Transfer (SWIFT) has been reported to predict unplanned ICU readmission. The aim of this study is to validate the SWIFT score as a useful tool to predict readmissions to the ICU.
Consecutive adults discharged from the ICU of two hospitals between January 2009 and June 2011 were studied retrospectively. Patients at Hospital A, a large tertiary care hospital, were limited to medical critical care discharges. At Hospital B, a 250 bed community hospital, a mixture of medical and surgical ICU discharges were examined. The dependent variables were: source of the ICU admission, ICU length of stay, latest PaO2 to FIO2 ratio, PaCO2, and Glasgow Coma Scale prior to discharge. Each of the five SWIFT predictor variables was assigned points as in the original study. The total score was used to predict the 7day risk of ICU readmission or unexpected death. Logistic regression was used to construct receiver operating characteristic (ROC) curves. Area under the ROC curve (AUROC) was used to judge predictive power of the SWIFT tool.
Complete information was available for 5290 cases from Hospital A and 1089 cases from Hospital B. At Hospital A, 352 (6.7%) were readmitted or died unexpectedly within seven days. At Hospital B, 61 (5.6%) suffered the primary event. For Hospital A, logistic regression was highly significant (p<0.0001) but calibrated poorly (HL chi sq. p=0.0025). For Hospital B, the model was also significant (p=0.007) and calibrated well (HL chi sq. p=0.92). The AUROC for Hospital A was 0.66 and for Hospital B was only 0.60. Using the cutoff score of 15 as in the original study, the sensitivity, specificity, PPV and NPV were 40%, 80%, 12%, 95% respectively at Hospital A and 23%, 89%, 11% and 95% respectively at Hospital B.
Given the relatively low sensitivity and AUROC, SWIFT is a weak tool for identifying ICU patients at high risk of readmission. Despite the promising results shown in the original study, the SWIFT score does not appear generalizable to other hospitals. Scoring systems that predict the risk of ICU readmission that are simple enough to use at the bedside and show good ability to identify patients at risk remain the goal. Developing practical scoring systems to predict ICU readmission should be pursued by other studies in the future.
To cite this abstract:Buran G, Abuawwad R. Failure of the Swift Score to Predict Readmission to the ICU. Abstract published at Hospital Medicine 2012, April 1-4, San Diego, Calif. Abstract 97674. Journal of Hospital Medicine. 2012; 7 (suppl 2). https://www.shmabstracts.com/abstract/failure-of-the-swift-score-to-predict-readmission-to-the-icu/. Accessed March 28, 2020.