OBJECTIVES

: Early Warning Scores to predict patients’ clinical deterioration are recommended to improve outcomes. We aimed to validate a tool integrating real-time vital signs and laboratory results in predicting adverse outcomes in patients hospitalized in medical wards.

METHODS

: VITAL (surVeiIIlance, moniToring and ALert) is a score developed at Centro Hospitalar São João, a tertiary hospital with 1083 beds (Portugal). Each parameter (n=14) was classified as abnormal when below or above a threshold or if a significant change occurred. The algorithm was run in a retrospective dataset of patients admitted from February to June 2013 (402,044 measurements) to test the ability of the score to predict unanticipated admission to higher level of care (HLC) and death within 24 hours. The performance of VITAL (including a modified version with parameters weighing differentially) was assessed by the area under the receiver operating characteristic (AUROC) curve, and the number needed to evaluate (NNE) and rate of alerts for different cut-offs.

RESULTS

: Admission to HLC and death within 24 hours of score calculation occurred in 0.4% and 0.7% of patients, respectively. The AUROC (95%CI) values were 0.65 (0.62-0.68) for predicting admission to HLC and 0.69 (0.67-0.70) for death. For VITAL ≥3, 5 alerts per 100 patients per day would occur, with a positive predictive value of 1.7% (95%CI; 1.3%-2.0%) and 3.9% (3.4%-4.5%), respectively. VITAL ≥3 would elicit further evaluation of 69 and 33 patients to detect one admission to HLC and one death, respectively. Part of the apparently false positive alerts at 24 hours developed the adverse event in the following 3 to 7 days.

CONCLUSIONS

: VITAL performed well to identify patients at high short-term risk of admission to HLC and death, with a relatively low additional workload, and would also result in earlier detection of events, contributing to the improvement of patient care.