RESUMO
OBJECTIVE: To derive and validate clinical prediction models to identify children at low risk of clinically significant intoxications for whom intensive care admission is unnecessary. DESIGN: Retrospective review of data in the National Poison Data Systems from 2011 to 2014 and Georgia Poison Center cases from July to December 2016. SETTING: United States PICUs and poison centers participating in the American Association of Poison Control Centers from 2011 to 2016. PATIENTS: Children 18 years and younger admitted to a United States PICU following an acute intoxication. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: The primary study outcome was the occurrence of clinically significant intoxications defined a priori as organ system-based clinical effects that require intensive care monitoring and interventions. We analyzed 70,364 cases. Derivation (n = 42,240; 60%) and validation cohorts (n = 28,124; 40%) were randomly selected from the eligible population and had similar distributions of clinical effects and PICU interventions. PICU interventions were performed in 1,835 children (14.1%) younger than 6 years, in 374 children (15.4%) 6-12 years, and in 4,446 children (16.5%) 13 years and older. We developed highly predictive models with an area under the receiver operating characteristic curve of 0.834 (< 6 yr), 0.771 (6-12 yr), and 0.786 (≥13 yr), respectively. For predicted probabilities of less than or equal to 0.10 in the validation cohorts, the negative predictive values were 95.4% (< 6 yr), 94.9% (6-12 yr), and 95.1% (≥ 13 yr). An additional 700 patients from the Georgia Poison Center were used to validate the model and would have reduced PICU admission by 31.4% (n = 110). CONCLUSIONS: These validated models identified children at very low risk of clinically significant intoxications for whom pediatric intensive care admission can be avoided. Application of this model using Georgia Poison Center data could have resulted in a 30% reduction in PICU admissions following intoxication.