Developing and evaluating a machine learning based algorithm to predict the need of pediatric intensive care unit transfer for newly hospitalized children

摘要: a b s t r c Background: Early warning scores (EWS) are designed to identify early clinical deterioration by combining physiologic and/or laboratory measures generate quantified score. Current EWS leverage only small fraction of Electronic Health Record (EHR) content. The planned widespread implementation EHRs brings the promise abundant data resources for prediction purposes. three specific aims our research are: (1) develop an EHR-based automated algorithm predict need Pediatric Intensive Care Unit (PICU) transfer in first 24 h admission; (2) evaluate performance new on held-out test set; and (3) compare effectiveness algorithm's with those two published Warning Scores (PEWS). Methods: cases were comprised 526 encounters 24-h transfer. In addition cases, we randomly selected 6772 control from 62516 inpatient admissions that never transferred PICU. We used 29 variables logistic regression compared against PEWS set. Results: achieved 0.849 (95% CI 0.753-0.945) sensitivity, 0.859 0.850-0.868) specificity 0.912 0.905-0.919) area under curve (AUC) Our AUC was significantly higher, 11.8 22.6% set, than PEWS. Conclusion: novel higher specificity, reported literature. © 2014 Authors. Published Elsevier Ireland Ltd. This is open access article CC BY-NC-SA license (http://creativecommons.org/licenses/by-nc-sa/3.0/).