Friday, February 24
PS2 Poster Session 2 and Refreshments Fri, Feb 24, 5:15 PM - 6:30 PM
Conference Center AB

Use of Longitudinal Models to Identify Subject-Specific Implausible Body Mass Index Measures: A Comparison with Screening for Population-Level Outliers (303474)

Janne Boone-Heinonen, The Ohio State University 
Andy Brickman, Health Choice Network 
Jennifer DeVoe, The Ohio State University 
Elizabeth Hooker, OCHIN, Inc. 
Kenneth Mayer, The Fenway Institute 
Jean O’Malley, The Ohio State University 
*Carrie Tillotson, OCHIN, Inc. 

Background: Invalid Body Mass Index (BMI) measures are typically identified as outliers based on population distributions, which conflate implausible and extreme measures. We demonstrate a longitudinal approach to identify BMI outliers in electronic health records. Study Population: Children 5-18 years with =3 BMI measures, from the ADVANCE PCORnet Clinical Data Research Network, a national network of Federally-Qualified Health Centers (n=94,627; n=797,281 BMI measures).

Methods: We used mixed effects regression to model BMI as a function of sex and age, with random intercepts and slopes for age, clustered by child. We identified longitudinal outliers as observations with studentized residual >|10|, compared to population outliers defined by the CDC (z-score>|5|).

Results: Mixed effects regression identified 447 longitudinal outliers; 207 were not population outliers. Conversely, mixed models classified 6404 (96%) of the population outliers as non-outliers. Among BMIs identifying severe obesity (n=81,803), 253 were identified as longitudinal outliers, compared to 5971 by CDC definitions.

Conclusions: Mixed effects regression is a useful method for differentiating true BMI extremes from measurement error in children and adolescents.