More than 60% of clinically obese patients cannot be identified using diagnosis codes alone. For manufacturers, payers, and providers, that means prior authorizations fail, commercial opportunity is underestimated, and real-world evidence built on ICD-10 codes alone systematically miss eligible patients.
Anti-obesity medications are the most consequential pharmacological development in obesity treatment in a generation. Semaglutide and tirzepatide have fundamentally redefined clinical expectations for weight loss, and payer, provider, and manufacturer attention to the obesity population has never been higher. However, the data infrastructure the healthcare system relies on to identify, treat, and measure that population is not keeping pace, limiting both patient access to these medications and commercial opportunities for pharma manufacturers.
To assess the disconnect between clinical reality vs. coded reality, we turned to NorstellaLinQ’s integrated claims and EHR database, spanning 650 U.S. health systems and 37.8 million patients with recorded BMI. We aimed to measure the gap between who is objectively obese by biometric measurement and who carries an ICD-10-CM obesity diagnosis code. Here’s what Norstella data reveals about how clinical overcoding and undercoding of obesity diagnoses are distorting physician behavior, patient access, and commercial opportunities for anti-obesity medications.
ICD-10 diagnosis codes capture fewer than 4 in 10 patients who are clinically obese by measured BMI.
The ICD-10 coding gap
ICD-10 diagnosis codes capture fewer than 4 in 10 patients who are clinically obese by measured BMI. This is not a documentation nuance. It’s a systematic limitation that cascades into denied prior authorizations, understated disease burden, and commercial opportunity models built on a foundation that excludes the majority of the population they’re meant to measure.
The scale of the problem is consistent and directional:
- Undercoding is the norm, not the exception. Only 39.5% of objectively obese patients, measured by BMI, the clinical standard reference, carry a corresponding ICD-10 obesity code. Among patients who do carry a Z68 BMI code, 82% still lack a formal obesity diagnosis code, meaning the code most likely to signal obesity to a payer is present without the actional diagnosis attached.
- The blind spot is worst where treatment opportunity is highest. Undercoding peaks at 74.6% among Class I obese males aged 35-49, a commercially active, treatment-naïve segment that is nearly invisible to code-based identification algorithms. Lower BMI classes are undercoded at rates of 67-72%, meaning the patients closest to the intervention threshold are the least likely to be found.
- A missing code does not mean a healthy patient. The negative predictive value of an absent obesity code is just 67.8%, meaning nearly 1 in 3 patients with no obesity diagnosis is nevertheless clinically obese based on their actual BMI. Coverage algorithms and opportunity models that use diagnosis codes as a proxy for eligibility are systematically excluding patients who qualify.
- 1 in 4 obese patients is invisible to the systems designed to reach them. No ICD-10 code, no prior authorization pathway, no commercial signal, despite meeting clinical criteria for treatment.
A clean ICD record is not a clean bill of health. The trend line is improving, but not fast enough to matter. Coding sensitivity has risen modestly over five consecutive years, from 30.1% in 2021 to 37.6% in 2025. That’s eight percentage points of progress across four years. At that rate, reaching even 60% sensitivity—still a failing grade by any clinical standard—would take more than a decade. Meanwhile, the absolute number of uncoded obese patients is growing. The gap is narrowing in percentage terms while the population it excludes continues to expand.
Nearly 1 in 3 patients with no obesity diagnosis is nevertheless clinically obese based on their actual BMI.
Why undercoding happens
Obesity is documented in the vitals because a medical assistant records height and weight. It appears far less often as a billable diagnosis code because physicians managing a 15-minute encounter prioritize the reason for the visit. Hypertension gets coded. Diabetes gets coded. Obesity, even when it directly underlies both, is routinely left out.
There’s also an element of stigma reflected in the data. Some clinicians have historically avoided coding obesity explicitly when no actionable treatment pathway existed. Effective pharmacotherapy changes that calculus, and the data suggests it already is: Coding sensitivity improved eight percentage points between 2021 and 2025, a period that coincides with the rise of semaglutide and tirzepatide as mainstream clinical tools. Whether prior authorization requirements, which create a direct documentation incentive, accelerate that trend further remains an open question.
| Obesity class | BMI range | Undercoding rate |
| Class I | 30 to 35 | 67 to 72% |
| Class II | 35 to 40 | 55 to 62% |
| Class III | 40+ | 39 to 47% |
| Most undercoded: Class I males, aged 35 to 49 | 74.6% | |
Undercoding decreases with disease severity, meaning the patients closest to the intervention threshold, where earlier treatment has the most to offer, are the least likely to appear in any code-based analysis. Women are consistently more likely to be coded than men across all BMI classes. The structural blind spot is concentrated precisely where the opportunity for prevention is highest and the administrative record is least reliable.
The overcoding complication
Undercoding gets the attention, but overcoding creates an equally important distortion from the opposite direction, and it is largely invisible in the literature.
Our data shows that 14% of patients with a current BMI below 30 carry an active obesity diagnosis code. This is not simply a coding error. Many of these patients were legitimately obese when the code was applied and have since lost weight through anti-obesity medication, bariatric surgery, or sustained behavioral change. The diagnosis persists in the record while the clinical condition has resolved.
The downstream consequences differ by stakeholder. For payers, overcoding means some coverage and utilization management decisions rest on a historical diagnosis that no longer reflects the patient’s current status, potentially authorizing therapy for patients who no longer meet clinical criteria while simultaneously denying it to uncoded patients who do. For researchers constructing real-world evidence cohorts from claims, the coded obesity population is not equivalent to the currently obese population, and analyses built on that assumption will produce systematically biased estimates of prevalence, treatment rates, and outcomes.
Together, undercoding and overcoding mean ICD-10-based identification overestimates obesity prevalence in some analytic contexts and dramatically underestimates it in others. There is no direction in which diagnosis codes alone produce a reliable count.
What this means for patient access and commercial opportunity
The documentation gap becomes an access gap at exactly the moment it matters most: when a payer evaluates coverage for anti-obesity medication. Many commercial prior authorization processes rely on diagnosis coding and supporting clinical documentation to establish eligibility. A patient with a confirmed BMI of 36 but no documented obesity diagnosis may face additional administrative barriers, not because they fail the clinical criteria, but because the documentation used to support coverage is incomplete. The patient qualifies. The paperwork may not.
The commercial implication is direct. Pharmaceutical teams building opportunity models on diagnosis-code counts are working with a structurally compressed view of the eligible market. The uncoded population is not uniformly distributed. It skews toward Class I patients, male patients, and geographies and specialties where obesity documentation rates are structurally lower. Field-force allocations and market-access strategies built on coded-patient counts will systematically underweight these pockets, misallocating resources away from the segments where the access gap is most acute and the untreated burden is highest.
The true addressable population for anti-obesity medications is substantially larger than ICD-based datasets suggest. The question is not whether that population exists, because the biometric data makes clear that it does. The question is whether manufacturers, payers, and health systems have the data infrastructure to find it.
From better data to better decisions
Correcting the obesity coding gap requires measuring patients using clinical evidence rather than relying solely on administrative diagnosis codes. That means integrating biometric measurements, longitudinal clinical data, pharmacy claims, and outcomes into a single patient view capable of identifying patients based on their actual clinical status—not simply the diagnoses that were billed.
Using NorstellaLinQ’s integrated claims and EHR platform, we evaluated more than 3.17 million real-world anti-obesity medication initiators by linking longitudinal height and weight measurements with treatment exposure and outcomes. This enables analyses that diagnosis codes alone cannot support, including quantifying real-world BMI change, identifying patients who achieve clinically meaningful weight-loss thresholds, and understanding how effectiveness varies across patient populations and care settings.
| Real-world outcome at ~6 months | Tirzepatide (Zepbound) | Semaglutide (Wegovy) |
| Mean BMI reduction | 2.3 kg/m² (6.3%) | 1.6 kg/m² (4.2%) |
| Achieved ≥5% BMI reduction | 52.8% | 43.6% |
| Achieved ≥10% BMI reduction | 33.5% | 23.4% |
These findings illustrate a broader point: integrating biometric measurements with longitudinal clinical and claims data does more than improve patient identification. It enables robust measurement of treatment effectiveness, persistence, access, and outcomes in the patients actually receiving therapy. As obesity care continues to evolve, data infrastructure must evolve alongside it, reflecting patients’ clinical reality rather than administrative coding alone.
For manufacturers, payers, and health systems alike, understanding the true addressable population, and how those patients move through diagnosis, treatment, and outcomes, will require integrating multiple complementary data sources. The organizations that can accurately identify these patients will be better positioned to improve access, generate more representative evidence, and make more informed clinical and commercial decisions.
Real-world outcomes run below clinical trial benchmarks, as expected in a heterogeneously treated population, but both medications produce clinically meaningful weight loss at scale. To put the difference in concrete terms, for a patient starting at BMI 36, tirzepatide’s advantage translates to roughly an additional five to six pounds of weight loss at six months compared to semaglutide. At the population level, the gap in patients achieving ≥10% BMI reduction, 33.5% vs. 23.4%, represents hundreds of thousands of patients crossing a clinically significant threshold.
The obesity market is no longer constrained by therapeutic innovation. It is constrained by data infrastructure. Organizations that continue relying on diagnosis codes alone will systematically underestimate disease burden, misallocate commercial resources, and overlook clinically eligible patients. The future of obesity real-world evidence begins with measuring patients as they are—not merely as they are coded.
NorstellaLinQ can deliver timely, clinically rich insights at scale and help teams move faster, publish sooner, and act with greater confidence. Contact us to learn more.
Limitations
These analyses are observational and descriptive; all figures are unadjusted and no causal claims are made. Several limitations bear directly on interpretation.
On coding analyses: BMI-derived obesity status depends on the accuracy and timing of recorded height and weight measurements. A single reference BMI can misclassify patients whose weight changed between measurement and code assessment. Coding completeness varies across EHR vendors, care settings, and specialties, and estimates may differ in datasets with different institutional composition. Active obesity codes follow claims and problem-list conventions that are not standardized across systems.
On effectiveness analyses: The GLP-1 initiator cohort is heterogeneously treated without randomization or covariate adjustment. Figures are not adjusted for dose, baseline BMI, adherence, comedications, or follow-up completeness. The tirzepatide-semaglutide comparison should not be interpreted as a head-to-head efficacy finding; confounding by indication and prescribing patterns is likely. These figures describe what happened in a real-world population, not what would happen under controlled conditions.
These limitations define the boundaries of what this data can answer. They do not diminish the core finding—that diagnosis codes alone produce a systematically incomplete and directionally biased picture of the obesity population, and that biometric-linked data is necessary to correct it.
Methods
Data are drawn from the NorstellaLinQ integrated claims and EHR database covering 650+ U.S. healthcare organizations. The identification cohort comprised 37,751,422 patients with at least one recorded height and weight between January 2017 and March 2026; obesity was defined as a measured BMI of 30 kg/m² or higher and treated as the reference standard. ICD-10-CM obesity codes (E66.x) and Z68 body mass index codes were evaluated against that standard to compute sensitivity, negative predictive value, and undercoding rates, overall and stratified by obesity class, sex, and age. Overcoding was measured as the proportion of patients with BMI below 30 carrying an active obesity code, and coding-sensitivity trends were assessed by calendar year from 2021 to 2025. Real-world effectiveness was assessed in a cohort of 3,171,331 GLP-1 initiators with linked pharmacy claims, comparing tirzepatide and semaglutide on mean BMI reduction and the proportion achieving 5% and 10% BMI reduction at approximately six months after initiation. All figures are descriptive.
Data source: NorstellaLinQ linked claims and EHR database, 650+ U.S. healthcare organizations. Analyses based on 37,751,422 patients with recorded BMI and 3,171,331 GLP-1 initiators, January 2017 to March 2026.
