The phrase "genetic responder" carries an implicit promise — that a variant in your genome tells a clinician which drug will work. The systematic literature does not support that promise in its strong form. What it supports is weaker and more interesting: a set of candidate polymorphisms, each with a pooled association of modest magnitude, sitting inside meta-analyses whose heterogeneity is often as informative as their point estimates.
Background and context
The question is old; the pooled evidence is recent. A systematic review and meta-analysis covering psoriasis, psoriatic arthritis, rheumatoid arthritis, and inflammatory bowel disease assembled 185 studies and 62,774 individuals to test associations between single nucleotide polymorphisms and response to biologic therapy. That is a substantial body of work, and it is almost entirely observational — cohorts followed for response, genotyped retrospectively, analyzed for association. For the combined diseases, the minor allele of MYD88 (rs7744) was associated with response. The framing matters: this is a candidate-gene literature, not a genome-wide one, and candidate-gene findings have a long history of failing to replicate at scale. The population is also narrow — patients with chronic inflammatory disease on biologics — and nothing here generalizes to healthy trainees or to training response itself.
Mechanism or physiology
The mechanistic logic is coherent enough. Biologic agents are proteins, and protein drugs are subject to immune recognition, receptor binding, and clearance — all processes with genetic components. Variants in Fc receptors plausibly alter how antibody-based drugs engage immune effector cells. In inflammatory bowel disease, a functional polymorphism in FCGR3A was significantly associated with anti-TNF treatment response among Crohn's patients on biologics, which fits a receptor-affinity story rather than a random hit. Variants in signaling adaptors such as MYD88 sit downstream of innate immune sensing and could shift the inflammatory set point a drug is trying to move. What the mechanism does not tell us is magnitude. A plausible pathway can produce a statistically detectable association of trivial clinical size, and the review literature rarely separates those two things cleanly.
Evidence summary
Pooled estimates in this field are typically reported as odds ratios with 95% confidence intervals, and the intervals are the story. The MYD88 finding is an association across a combined population, not a validated predictor in any single disease. The FCGR3A result is qualified by disease — Crohn's, on biologics — exactly the kind of population restriction that should travel with the claim. A separate systematic review and meta-analysis of genetic biomarkers predicting tocilizumab response in rheumatoid arthritis reaches for the same target with a narrower drug and disease, and the pattern repeats: candidate variants, modest pooled effects, calls for larger randomized work. Where the literature reports a strong responder rate — as in felbamate for drug-resistant genetic generalized epilepsy, with a pooled 65% responder rate (95% CI, 51–80) — that figure describes the drug's performance, not a genetic predictor's accuracy.
Practical application
For a clinician, the practical reading is that genotyping is not yet a treatment-selection tool in these diseases. It may become one; the candidate list is not empty. But a variant with a modest odds ratio and a confidence interval that brushes the null does not justify switching a patient's biologic, and it certainly does not justify withholding an effective drug. The more defensible use of this literature is in trial design — enriching cohorts by candidate genotype to test whether the signal holds prospectively. For everyone else, the application is epistemic: when a headline announces a "gene for drug response," ask whether the underlying design was a meta-analysis of observational cohorts or a prospective randomized trial, and ask what the odds ratio was. Those two questions deflate most of the coverage.
Caveats and limitations
The limitations are structural, not incidental. Candidate-gene studies are prone to publication bias toward positive findings, and meta-analysis cannot recover studies that were never submitted. Outcome definitions for "response" vary across the 185 studies — different drugs, different timepoints, different remission criteria — so pooling assumes a homogeneity the field does not have. Ancestry is another gap: most cohorts are of European descent, and allele frequencies differ across populations, which limits transportability. Sample sizes per variant, once split by disease and drug, are often small. The genetic associations reported for keratoconus, while a different disease entirely, illustrate the broader pattern — many reported associations, few that survive rigorous replication. The data are noisier than the slogan. Anyone making personal medical decisions about treatment should consult a physician or qualified clinician rather than a direct-to-consumer genotype report.
FAQ
If MYD88 rs7744 showed up in a meta-analysis, shouldn't my doctor test for it?
Not yet. That association was pooled across four diseases and 185 studies; it has not been validated as a predictor in any single disease, and a modest odds ratio with a confidence interval near the null is not a basis for changing treatment. Ask your clinician what the effect size was, not just whether the variant appeared.
Is a direct-to-consumer genotype report useful for picking a biologic?
No. These are candidate-gene findings from observational cohorts, and consumer reports typically test different variants than the ones in this literature. Treatment selection belongs with a clinician who can weigh the drug's evidence, your disease course, and your history — not a report that flags a polymorphism without context.
What would actually move this field forward?
Prospective trials that enrich cohorts by candidate genotype and test whether the signal holds — plus larger, more diverse cohorts and standardized outcome definitions. Until then, the honest summary is a candidate list with modest effects, not a treatment algorithm.




