Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 54 - Methods and Modeling for Medical Device and Clinical Studies
Type: Contributed
Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
Sponsor: Section on Medical Devices and Diagnostics
Abstract #309641
Title: Dynamic ICU Predictor Model of Pediatric in Hospital Mortality
Author(s): James Bost* and Anita Patel and Eduardo Antonio Trujillo Rivera and Qing Zeng and Murray M Pollack
Companies: Children's National Hospital and Children's National Hospital and George Washington University and George Washington University and Children's National Hospital
Keywords: predictive modeling; EMR; Big Data

Children’s National Hospital in partnership with Cerner Corp. acquired the HealthFacts® data registry which contains approx. 10 years’ worth of hospital admission data for over 500 Cerner EMR supported hospitals. The goal of this effort was to determine what numeric intervals of laboratory data best predict mortality, what combination of labs best predict mortality and how can this be used dynamically within the EMR system to provide hourly “probability of in hospital death” scores to support clinical decision making. 33 labs were assessed and include in the analysis. For each lab we determined optimal prediction intervals at the patient level. We started with 10 intervals based on standard percentile cut points and collapsed those in order, using a logistic regression model, based on the extent of overlap around their adjacent odds ratio 95% confidence intervals. Once the final intervals for each lab were determined we added labs intervals one at a time (based on their univariate predictive ability) until we had a model such that the addition of a new lab did not improve the models predictive ability based on sensitivity and positive predictive value assessments using a prob

Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program