Background — Obesity has a significant impact on population health and healthcare. Administrative databases might be a useful tool to study obesity at a population level. We aimed to examine the validity of hospital codes for obesity in Ontario, Canada.
Methods — Using linked healthcare databases (ICES), we conducted a validation study in adults ≥18 years who had a height and weight recorded during a hospitalization in Southwestern Ontario. We considered a body mass index ≥30 kg/m2 as our gold standard definition for obesity. We then examined the validity of two International Classification of Diseases 10th Revision coding algorithms for obesity (algorithm 1 ICD 10 E66.X and algorithm 2 ICD 10 E65.X-68.X). As additional analyses, we examined the validity of algorithms in different obesity classes (i.e. obese class 1, 2, 3), and in patients with diagnosed diabetes and hypertension.
Results — There were 34 588 patients included in our study (mean age 62 years, 47% female). Algorithm 1 performed best, with a sensitivity, specificity, positive predictive value and negative predictive value of 8.8%, 99.8%, 95.4% and 65.1% respectively. The sensitivity of this algorithm was highest in patients with obesity class 3 (27.4%) and in those with diagnosed diabetes.
Conclusions — Hospital codes for obesity have a high PPV and specificity. These codes can be used to build and study cohorts of patients with obesity in administrative database studies. However, given their limited sensitivity, administrative codes provide inaccurate incidence and prevalence estimates.