DISCUSSION

The present real-world study observed a contemporary cohort of almost 1 million patients with obesity based on BMI to assess coding of the condition in medical claims. The study evaluated the demographic and clinical characteristics of these patients and calculated the prevalence of obesity coding among patients with and without these characteristics. Furthermore, it ascertained the validity of the ICD-10-CM codes for obesity. Finally, it investigated predictors of coding among patients with a BMI ≥30 kg/m2. The research undertaken enhances our understanding of obesity identification in real-world healthcare settings using medical claims data.

Compared with those without obesity diagnosis codes in their medical claims within 6 months before or after documentation of a BMI indicative of obesity in the EHR, patients with diagnosis codes displayed higher proportions of comorbid conditions, medication usage, and healthcare interactions across all measurable variables. The observed trend of patients with codes having on average a higher comorbidity burden could be attributed to various factors. One potential explanation is that these patients may be under the care of healthcare providers and practices who are more conscientious in their medical billing coding practices. In such instances, these patients might appear relatively “sicker” when compared with those under the care of less accurate/comprehensive coders. An alternative explanation is that patients grappling with more severe health conditions generally experience more frequent touchpoints with the healthcare system.10,11 These increased encounters with healthcare providers enhance opportunities for patients’ conditions to be adequately identified and coded. Thus, the prevalence of coding might be reflective of the patients’ frequent interaction with the healthcare system. Importantly, these differences should be interpreted as predictors of obesity documentation and coding behavior, rather than predictors of obesity itself, since all patients in this comparison had BMI-defined obesity.

The results of this study underscore that if an obesity code is observed in a patient’s medical claims, we can confidently presume that the patient has a BMI of ≥30 kg/m2, as suggested by the high PPV of 95.6%. This finding corroborates earlier estimates by Ammann et al and Suissa et al, who reported high PPVs of 92.4% and 97.3%, respectively, reinforcing the accuracy of recording obesity codes in medical claims for patients diagnosed with obesity.12,13 However, the absence of an obesity code in the medical claims does not conclusively guarantee that a patient has a BMI <30 kg/m2. This indicates that obesity remains undercoded, with only a 39.4% (sensitivity) chance that a patient with a BMI ≥30 kg/m2 will have a code for obesity within 60 days prior to or after the observed BMI measure. Compared with previous studies investigating ICD-10-CM codes for obesity, our findings concur with the commonly observed trend of undercoding in the medical claims (sensitivity <40%).12–14 These observations suggest that reliance on codes alone is not sufficient to accurately identify the full obesity population, and that a substantial number of individuals thought not to have obesity would be miscategorized. This misclassification greatly limits the ability to perform research that assesses the prevalence and incremental burden attributed to obesity in a claims database. In addition, based on current medical claims coding practices, a significant segment of the population with obesity would likely be overlooked if coding was the sole method used to identify patients with obesity from a claims database. This highlights the importance of complementary methods of identifying cases of obesity.

These findings have important implications for studies that rely on administrative claims data to identify patients with obesity. Although the high PPV suggests that patients with an obesity diagnosis code are very likely to have BMI-defined obesity, the low sensitivity indicates that claims-based coding alone may substantially underidentify the full obesity population. As a result, studies using obesity diagnosis codes as the sole method of cohort identification may underestimate obesity prevalence and may select a subset of patients with greater clinical complexity, higher healthcare utilization, or more frequent obesity-related treatment. This has implications for health economics and outcomes research, including estimates of disease burden, healthcare resource utilization, treatment patterns, costs, treatment eligibility, and unmet need. Whenever possible, claims-based studies of obesity should consider supplementing diagnosis codes with additional data sources, such as EHR-derived BMI, procedure codes, medication claims, or validated algorithms, to improve case identification.

Undercoding of obesity appears consistent across patients regardless of most clinical characteristics. The prevalence of obesity coding did not show noticeable variation between patients with and without associated clinical characteristics, and no specific clinical characteristic notably increased the odds of coding. This suggests that coding discrepancies exist across all clinical subgroups, with no group more or less likely to have accurate obesity documentation. The issues with medical claims coding, thus, appear to be relatively evenly spread across all groups. However, higher coding prevalence was seen among patients receiving obesity-related services, including bariatric surgery, diet counseling and surveillance, chronic weight management medication, and among patients with higher BMI values. These findings may reflect, in part, greater clinical recognition or documentation of obesity when it is directly relevant to treatment planning, medical necessity, or coverage requirements.15–17 These findings are also consistent with broader literature suggesting that reimbursement policies and payment models may influence diagnostic coding accuracy across a variety of medical conditions.18–21 However, because the present study did not directly evaluate provider billing behavior, reimbursement policies, or payer-specific documentation requirements, the observed differences in obesity coding should be interpreted as associations rather than evidence of a causal effect of reimbursement incentives. Future studies could more directly evaluate whether coverage requirements, risk-adjustment programs, or clinical documentation workflows influence obesity coding practices.

Study findings should be interpreted in light of several limitations. Results may not generalize beyond commercially insured and Medicare Advantage populations or to individuals without similar insurance coverage. BMI is an imperfect proxy for adiposity and may misclassify obesity status because it does not directly measure body composition, body fat percentage, fat distribution, or individual health risk.2,9 This limitation may be particularly relevant across racial and ethnic groups, as prior research has shown that BMI-adiposity relationships can vary by race/ethnicity; therefore, a single BMI cutoff may not reflect the same level of adiposity or cardiometabolic risk across all populations.22 Because EHR-derived BMI was available only for a subset of HIRD patients, selection bias is possible if patients with BMI data differ systematically from those without it. Claims data may also contain coding omissions or errors and may incompletely capture sample medications, medications filled but not taken, health conditions not recorded in billing data, and unmeasured confounders. Overall, these limitations may affect generalizability and could lead to under- or overestimation of coding validity and predictors of obesity documentation.