Diagnostic definitions shape genome-wide association findings in atopic dermatitis

A new paper set to appear in the Journal of Investigative Dermatology went online October 14, 2026, under the title "Diagnostic definitions shape genome-wide association findings in atopic dermatitis." The full text sits behind a paywall at Elsevier's ScienceDirect, and the abstract didn't surface in open search as of this writing. What's checkable instead is the body of prior work the claim rests on, and that record already shows how much a single methodological choice can move a field's output.

Start with the count. A 2019 systematic review in the British Journal of Dermatology, led by Nakamura and colleagues, found 59 different definitions of atopic dermatitis across 45 studies. The team then looked at 26 of those studies that had been folded into a genome-wide association meta-analysis and found that switching the case definition changed both prevalence estimates and which risk factors turned up as significant, filaggrin gene associations among them.

The mechanism behind that shift has a name: misclassification. A validation study of 1,638 Japanese children compared self-reported, physician-diagnosed atopic dermatitis against a clinical exam. Self-report caught only 26.7% of true cases (sensitivity) while correctly ruling out 94.1% of non-cases (specificity). A cohort built on self-report misses roughly three out of four real cases, and a genome-wide association study run on that cohort loses statistical power exactly where the signal would otherwise be strongest.

UK Biobank illustrates the same problem at larger scale. The dataset holds two separate self-report fields, one asking about "rhinitis/hay fever or eczema ever" and another asking about "eczema or dermatitis." Critics writing in Medizinische Genetik have argued that genome-wide association studies built on the combined hay-fever-or-eczema field are effectively studying allergic disease in general, not eczema specifically, and that some true eczema loci drop out of the results as a consequence.

Scale has grown fast even as this definitional noise persists. The first atopic dermatitis genome-wide association study, in 2009, centered on filaggrin and returned a short list of loci. A 2024 meta-analysis covering 1,094,060 individuals identified 77 independent loci, with subgroup analyses by severity and age of onset built in specifically to address the heterogeneity problem. A 2025 paper in Nature Communications pooled 12 cohorts across five ancestries and reported 101 loci, 16 of them new.

Bigger numbers haven't settled the question of which hits are real. A UK Biobank prioritization analysis found that only about 25 atopic dermatitis loci are considered reproducible across independent studies, out of hundreds that have been reported somewhere in the literature. That gap between reported loci and reproducible ones is one of the scope limits this strand of research keeps running into: a result generated under one case definition doesn't automatically hold once the definition changes.

The gap matters past the level of locus-counting. Heritability estimates from these genome-wide association studies run 14% to 30%, while twin studies put heritability at 71% to 90%, a persistent shortfall researchers call missing heritability. Rare, low-frequency variants account for more than 20% of that gap in one estimate, and definitional noise across cohorts is a second plausible contributor, one that shrinks or grows depending on how tightly a study defines its cases.

That noise now feeds directly into clinical tools. A 2025 polygenic risk score study found that prediction accuracy for atopic dermatitis jumped when filaggrin loss-of-function variants were added to the model, moving the odds ratio from 2.16 to 3.23. A risk score is only as good as the genome-wide association data it was trained on, so a training set built from a loosely defined or self-reported case list hands that same imprecision forward into the score a clinic would eventually use.

Whether the October 14 paper adds a new dataset, a new comparison across definitions, or a review of the existing split between dermatologist-verified and self-reported cohorts isn't yet confirmed from the published record. Checking the abstract against PubMed once it clears or against the JID site directly would settle that detail. Until then, the surrounding literature already draws the shape of the problem: one review counting 59 definitions across 45 studies, and a validation study putting a hard number, 26.7%, on how often self-report catches a real case of atopic dermatitis.