(169) Identification of Obesity and Tobacco Use in Claims Compared to Electronic Health Record (EHR) Data

Sunday, August 28, 2022

8:00 AM – 1:45 PM CEST

Location: Hall C

Publication Number: 1214

Presenting Author(s)

Casie Horgan
– Harvard Pilgrim Health Care Institute, Boston, Massachusetts, United States

Co-Author(s)

Po-Yin Chang
– U.S. Food and Drug Administration, United States
Meg Her
– Harvard Pilgrim Health Care Institute, United States
Rebecca Hawrusik
– Harvard Pilgrim Health Care Institute/Harvard Medical School, Boston, MA, United States
Alexander Peters, MPH
– Associate Research Analyst, Harvard Pilgrim Healthcare Institute, MA, United States
Catherine Corey, MSPH
– U.S. Food and Drug Administration, United States
David Graham, MPH, MD
– Associate director, US Food and Drug Administration, Silver Spring, Maryland, United States
Judith C. Maro, PhD
– Harvard Pilgrim Health Care Institute, Boston, Massachusetts, United States
Noelle M. Cocoros, DSc
– Epidemiologist, Harvard Pilgrim Health Care Institute, Boston, Massachusetts, United States

Background: Incomplete capture of obesity and tobacco use in claims data can lead to misclassification of variables.

Objectives: To examine claims and EHR-based indicators of obesity and tobacco use among sodium-glucose co-transporter-2 inhibitor (SGLT-2i) users in the FDA Sentinel System.

Methods: Using data from 7 integrated delivery systems with both EHR and claims data we identified new SGLT-2i users aged ≥20 years (3/2013 - 6/2018; washout=365 d) and assessed obesity and tobacco use in the 365 days prior to and including index dispensing. EHR-based BMI was calculated using closest valid height and weight to index and categorized as underweight/normal (15<=BMI < 25), overweight (25<=BMI < 30), obese (30<=BMI < 40) and severely obese (40<=BMI<=90). Claims-based BMI and obesity was based on presence of a diagnosis or procedure code, using the code closest to index. For the obese and severely obese, we used a narrow claims definition (BMI-specific codes) and broad definition (BMI-specific and generic obesity codes). EHR-based tobacco use was defined by the populated record closest to index. Claims-based tobacco use was defined by diagnosis code or dispensing of a tobacco cessation product. We cross-stratified users’ claims and EHR data to compare designations across data sources.

Results: We identified 3,155 new SGLT-2i users, of which 1,936 (61%) had an EHR-based BMI, with 69% obese or severely obese; 43% had any claims code, with 98% of those obese or severely obese (broad definition). Among the 1,936 with an EHR BMI, 277 users (14%) had concordant claims data using the narrow obesity definition and 853 (44%) with the broad definition of obesity. Among all other users, 35% had neither EHR BMI nor code, 44% had an EHR BMI but no corresponding code, 18% had a code but no corresponding EHR BMI, and 3% had disagreement between the two sources. For the broad definition, concordance between EHR and claims for obesity and severe obesity was 54% and 86%.
Among the cohort of SGLT-2i users, 2,205 (70%) had an EHR record for tobacco use with 965 (44%) indicating current or former use. Among current/former users, 40% had a concordant claims-based code. Concordance was higher for current users (77%) than former users (32%). There were 1194 SGLT-2i users (38%) with an EHR record indicating ‘never smoker’; of these, 2% had a contradicting code for smoking. There were 143 users (5%) who had a claim for smoking but no EHR data.

Conclusions: Among those treated with an SGLT-2i, concordance between claims and EHR data was good for current tobacco use. For obesity, using a more sensitive definition in claims increased concordance with EHR data. Characterization of missingness and clinical measurement concordance for key confounders could inform quantitative bias analysis or imputation approaches for future work.