Abstract

International studies show that including Early Warning Score (EWS) Systems in clinical care processes allows caregivers to anticipate patient decompensations and make timely decisions. Currently, it is possible to find different perspectives in the construction of EWS; however, the centile-based approach has not been applied to pediatric patients. This study aims to develop a pediatric EWS using the statistical properties of historical vital sign records. A scoring method with a supervised statistical approach was developed and evaluated. 178.970 vital sign records were obtained from 4,104 patients aged 0 to 18 years, in conjunction with information from Electronic Health Records from a pediatric hospital. Based on a 7-band scoring method and using different vital signs, a model was generated that allows us to predict an unplanned transfer of patients to ICU, 8 hours before it happened. The prediction obtained in the best case showed results in areas under the ROC Curve (AUC) of 0.71. In addition, the calculated cut-off thresholds are sensitive in the lower risk categories (0.877 at low-risk levels) and possessing greater specificity in the higher-risk categories (0.89 in the High-Risk category). The results of this study allow correcting the normal ranges in the vital signs in the Expert-based model, which is currently in operation. Our supervised statistical approach provides an auditable and reproducible method.