Dermatology
Atopic dermatitis and prurigo nodularis provided a distinct use case for testing broader applicability.
Home / Accomplishment / AD & PN
Case 02 · Dermatology
The framework carried into a second therapeutic area
The programme applied its common-data-model approach to atopic dermatitis and prurigo nodularis, adding prevalence estimation and allowing site-specific disease definitions within a shared framework.
- Diseases
- Atopic dermatitis and prurigo nodularis
- New module
- Prevalence estimate
Study cohorts
Cohorts in each data environment
Cohort size, study period, and the share of female patients reported by each participating data environment.
Why dermatology
An extension beyond cardiovascular research
The case tested whether the framework could support disease-burden, standard-of-care, and clinical-outcome analyses in an additional therapeutic area.
Prevalence estimation
The objectives included disease burden, prevalence, standard of care, and clinical outcomes.
Site-specific algorithms
The shared framework retained documented local code lists and disease-defining algorithms where needed.
Prevalence by age
Age-specific prevalence (2020)
Standard of care
Compare index medication profiles across databases
The central finding
How much does the definition move the answer?
The programme applied three definitions of increasing specificity and compared the resulting prevalence estimates across participating environments.
Explore how prevalence changes with the operational definition
Adult population · 2020
Prevalence by case definition
Definition 1 (D1)Any diagnosis
Definition 2 (D2)Two diagnoses within 180 days
Definition 3 (D3)Two diagnoses within 180 days plus systemic AD/PN medication
Database-specific rule: for CDARS, D3 is any diagnosis plus systemic AD medication.
Loading the evidence figure…
View accessible data table
What the comparison showed
Prevalence falls substantially as the definition tightens in every database that reported all three definitions. A single diagnosis code admits a much larger population than a definition requiring repeated diagnoses plus systemic treatment.
The comparison shows why operational definitions must be reported explicitly: estimates can differ because teams apply different diagnosis-frequency or treatment requirements.
The second finding
Cross-environment variation differed by outcome
The programme reported three clinical outcomes as incidence rates per person-year with confidence intervals. They behave very differently from one another, and that pattern is itself the result.
Complications
Point estimates ranged from 0.04 to 0.55 events per person-year across the five environments shown.
Treatment change
Point estimates ranged from 0.03 to 5.54 events per person-year across the five environments shown.
Interpretation boundary
The observed variation should not be attributed to a single cause from these results alone. Differences may reflect population characteristics, coding practices, data coverage, clinical practice, or local implementation. The reported rates describe the cross-environment variation but do not establish which mechanisms produced it.
Compare three clinical outcomes across five environments
Confidence-interval dot plot
Reported incidence rates per person-year
Loading the evidence figure…
View accessible data table
How flexibility was preserved
Standardization did not require erasing local coding
-
Shared
The data structure, the modules, and the quality standard
The programme used a common-data-model structure, shared analysis modules, and common quality expectations to support comparable outputs.
-
Local
The source code lists behind each concept
A coding library carries site-specific definitions. Ischemic stroke, for example, is I63 in NHIRD, CGRD, and DeSC; I63 and I64 in NHIS and HIRA; and 433 and 434 in CDARS. The framework does not pretend these coding systems are identical.
-
Negotiated
The disease algorithm, where local practice requires it
CDARS uses a different D3 rule. That is recorded and applied rather than forced into conformity, because forcing it would produce a comparable-looking number that means something different.
The conclusion reported by the programme
The CDM approach was feasibly applied to an additional therapeutic area; disease burden, standard of care, and clinical outcomes were consistently characterized; and flexible disease-defining algorithms were accommodated within the framework.