The study is a retrospective chart review, a design in which researchers go back through patient records to check whether a particular data-mining strategy matches clinical reality. In this case, the team examined how well different approaches to using the International Classification of Diseases, 10th Revision, better known as ICD-10, performed at distinguishing true rosacea cases from lookalike conditions. Billing codes are assigned for administrative reasons, not research purposes, and clinicians may use a code for a suspected diagnosis, a rule-out, or even a billing workaround. That gap between what a code says and what a patient actually has is precisely the kind of hidden error that can quietly poison large-scale studies.

The stakes are higher than they might first appear. A recent worldwide epidemiological study cited by the authors estimated that acne and rosacea together affect enormous numbers of people across the globe, making rosacea one of the most prevalent reasons patients seek dermatological care. The condition typically produces facial flushing, persistent redness, visible blood vessels, and in some cases acne-like bumps or thickened skin around the nose. Beyond the physical symptoms, rosacea is associated with significant psychological burden, ocular complications, and, when left untreated, progressive changes that are difficult to reverse. Understanding its true prevalence, treatment patterns, and long-term outcomes depends entirely on being able to count the right patients.

Electronic health records have transformed dermatology research by offering access to millions of patients whose care would otherwise be invisible to scientists. But as Mostaghimi and Noe argued in a widely discussed 2021 paper on the challenges of big data in dermatology, the sheer scale of these datasets can create a false sense of confidence. Garbage in, garbage out remains the governing law of computational research. If a cohort defined by billing codes contains even a modest proportion of misclassified patients, effect sizes shrink, spurious associations emerge, and conclusions about drug safety or disease trends can flip entirely. Validation studies like this one are the quality-control step that keeps the pipeline honest.

The Brigham and Women’s team has now built something of a validation program across inflammatory skin diseases. In 2021, Barbieri and colleagues published a similar analysis for acne, determining which ICD-10 code strategies most accurately identified dermatology patients with that condition. In 2025, Ershadi, Biba, Bensellam, and colleagues, many of them on the current paper, extended the approach to perioral dermatitis, a facial eruption that shares features with rosacea and is easily confused with it. The new rosacea study completes a natural trilogy, giving researchers validated code definitions for three of the most common papulopustular and erythematous facial dermatoses seen in clinic.

Why does rosacea present a particular classification challenge? Part of the answer lies in the structure of the ICD-10 system itself. Rosacea occupies a cluster of codes within the dermatitis section of the classification, with separate codes for different subtypes and presentations, and clinicians vary widely in how precisely they code. Some records carry a specific rosacea code, others a broader facial dermatitis code, and still others a code for an unrelated condition that the clinician used to describe a rosacea-like presentation. In addition, conditions such as seborrheic dermatitis, lupus erythematosus, steroid-induced facial erythema, and perioral dermatitis can mimic rosacea clinically, meaning that a code-based cohort will inevitably pull in some patients who do not have the disease at all unless the coding strategy is chosen carefully.

The methodological logic of a chart review validation is straightforward even if the work is painstaking. The investigators define a set of candidate code strategies, for example a single specific code versus a broader set of codes, or codes restricted to dermatology encounters versus codes from any clinical setting. They then identify patients whose records match each strategy and manually review the clinical documentation, including physician notes and diagnoses, to determine whether the patient truly has rosacea. Each strategy can then be scored on measures such as positive predictive value, the proportion of code-identified patients who genuinely have the disease. A strategy that casts a wide net may capture nearly everyone but drown the cohort in false positives, while a narrow strategy may be highly accurate but miss many patients. The optimal choice depends on the research question, and the value of this study is that it tells researchers exactly where those trade-offs fall.

The research was reviewed and approved by the Mass General Brigham Institutional Review Board, and the authors report their funding as none, with disclosed consulting relationships for several senior authors unrelated to the present work. The study appears as a research letter, a compact format that journals use for focused validation analyses whose conclusions are precise but narrow. That format suits the findings well: the deliverable is not a sweeping new theory of disease but a set of practical, evidence-based recommendations that other investigators can adopt immediately when constructing rosacea cohorts from claims databases or electronic health record systems.

The implications ripple outward across dermatology research. Observational studies of rosacea treatments, investigations of reported associations between rosacea and cardiovascular, gastrointestinal, or neurological disease, and pharmacoepidemiology of systemic therapies all depend on cohort definitions built from billing codes. A misclassified cohort could exaggerate or erase a genuine association, and inconsistent definitions across studies make meta-analysis nearly impossible. By establishing a validated standard, the Boston team gives the field a common yardstick, much as their earlier acne and perioral dermatitis validations did for those conditions. Journals and reviewers can now reasonably ask authors of rosacea database studies which validated code strategy they used and why.

There is also a broader lesson here about the infrastructure of modern medical science. As health systems digitize their records and machine learning models are trained on clinical data, the humble diagnostic code has become one of the most consequential pieces of metadata in medicine. Studies like this one, unglamorous as they may seem, are the calibration experiments that determine whether the resulting science is trustworthy. For the millions of people worldwide who live with the flushing, burning, and stigma of rosacea, better-validated data means faster progress toward understanding who develops the disease, why it flares, and which treatments genuinely work. The Brigham and Women’s group has shown, once again, that in the era of big data, the most important findings sometimes begin with a careful look at how we count.

Subject of Research: Validation of ICD-10 code-based classification strategies for identifying rosacea patients in retrospective medical record data

Article Title: Validating optimal ICD-10 classification approaches to identify patients with rosacea: a retrospective chart review

Article References: Acevedo-Fontanez, L. A., Bensellam, N., Ershadi, S., Biba, U., Sanchez, K., Cheng, D., Gregoire, S., Vazquez-Machado, M. C., Mostaghimi, A., & Barbieri, J. S. (2026). Validating optimal ICD-10 classification approaches to identify patients with rosacea: a retrospective chart review. Archives of Dermatological Research, 318(1), Article 452. https://doi.org/10.1007/s00403-026-04817-y

Image Credits: AI Generated

DOI: 10.1007/s00403-026-04817-y

Keywords: rosacea, ICD-10, medical coding, electronic health records, dermatology, retrospective chart review, validation study, diagnostic classification, dermatoepidemiology, Brigham and Women's Hospital, big data, skin disease