Background: Many studies rely on the International Classification of Diseases, 9th or 10th Revision, Clinical Modification codes to define obesity in electronic health records data. While prior studies found misclassification and low sensitivity of codes for pediatric obesity, it remains unclear whether this misclassification is random and what are the implications of combining different code types to define obesity. Methods: We assessed prevalence, sensitivity, and specificity of obesity codes among 7.4 million children aged 2-19 years over 2014-2021. Among those with obesity in 2021, we estimated the probability of receiving any code or a specific code type by patient characteristics. Results: Obesity code utilization increased in prevalence from 3.9% in 2014 to 9.8% in 2021; prevalence of obesity based on BMI increased from 17.4% to 20.5%. Code sensitivity increased from 19.8% to 40.8%. Among children with obesity in 2021, those with severe obesity (reference: no severe obesity) and chronic disease (reference: no chronic disease) were more likely to get a code, and the highest likelihood was associated with obesity diagnosis codes (vs. status codes). Conclusions: Despite increases, obesity code utilization remained low. Obesity code misclassification is not random and certain child characteristics (e.g., severe obesity or chronic disease) are associated with a higher probability of getting a code. There are also significant differences by code type; thus, caution should be taken before combining obesity codes as a proxy for obesity status, especially in longitudinal analyses. More universal documentation of obesity may improve the quality of care and the use of these data for evaluation and research purposes.
Keywords: ICD codes; obesity diagnosis; sensitivity; specificity.