A type 1 diabetes shot that failed overall worked in one genetic subgroup

A vaccine-style injection designed to slow the autoimmune attack behind type 1 diabetes showed no overall benefit when first tested, and a 13-year follow-up has not changed that headline [s1]. What the new analysis adds is a split: in children carrying a particular immune-system gene variant, the treatment appeared to delay the disease by years, while in children without it, the same injection was associated with faster progression [s1]. The result is being framed as a case for precision-medicine trial design — but it rests on subgroups of 27 and 22 children, and the honest reading is that it generates a hypothesis rather than settling one.

Type 1 diabetes develops when the immune system destroys the insulin-producing beta cells of the pancreas. Long before symptoms appear, autoantibodies against beta-cell proteins can be detected in the blood, and children with two or more such autoantibodies are now considered to have early-stage disease that will, in most cases, progress to insulin dependence — so-called stage 3. GAD-alum is an alum-formulated version of one of those target proteins, glutamic acid decarboxylase (GAD65), given on the theory that exposing the immune system to the antigen in this form can retrain it towards tolerance rather than attack [s1].

What the trial did

DiAPREV-IT was an investigator-initiated, randomised, placebo-controlled trial — notable because it was funded through the Juvenile Diabetes Research Foundation rather than by a drug company, which removes the commercial stake that colours much of this field [s1]. It enrolled 50 children aged roughly 4 to 18 who had GAD autoantibodies plus at least one additional islet autoantibody, and randomly assigned them 1:1 to two subcutaneous injections of 20 µg GAD-alum or placebo, 30 days apart, with everyone masked to who got what [s1][s2]. The original five-year follow-up, reported in 2017, found no significant effect on the development of stage 3 type 1 diabetes [s1].

The new paper tracks the same children for a median of 12.9 years, pulling diagnosis dates from the Swedish national diabetes registry or, where needed, by telephone interview [s1]. One of the original 50 was excluded because baseline glucose values already met the stage 3 threshold, leaving 49. Of those, 35 went on to develop clinical type 1 diabetes [s1].

The discordant result

In the full group, GAD-alum did nothing: the hazard ratio for progression was 1.02 (95% CI 0.52–1.98; p=0.96) — essentially dead on "no difference" [s1]. The interest lies entirely in a pre-specified split by the HLA-DR3-DQ2 haplotype, a common high-risk immune genotype.

Among the 27 children who carried it, GAD-alum was linked to a delay in disease: a hazard ratio of 0.32 (95% CI 0.12–0.82; p=0.018), with the median time to diagnosis stretching to 9.0 years on treatment versus 3.4 years on placebo [s1]. Among the 22 who did not carry it, the direction flipped — a hazard ratio of 3.50 (95% CI 1.05–11.63; p=0.040), meaning faster progression on the drug [s1]. Those two findings are what the authors lean on to argue that genotype should guide who gets the injection.

Why to stay cautious

The confidence intervals tell the story the point estimates flatter. The protective subgroup rests on 27 children, and its separate measure of the effect — an incidence rate ratio of 0.42 — only just cleared significance (95% CI 0.18–0.997; p=0.049), while the gap in median onset time did not reach it on its own (p=0.096) [s1]. The apparent harm in the other subgroup sits on 22 children and an interval running from barely above 1 to nearly 12. Subgroup effects that point in opposite directions, from a trial that was negative overall and small to begin with, are exactly the pattern that later fails to replicate. A genuinely null drug can still throw up one "significant" subgroup and one harmful-looking one by chance when the numbers are this thin.

What makes the signal worth logging rather than dismissing is biological plausibility and independence: the HLA region governs how antigens are presented to the immune system, so a genotype-dependent response to an antigen-based therapy is at least coherent, and the trial had no sponsor pushing the interpretation [s1]. Antigen-specific immunotherapy for type 1 diabetes has a long record of promising subgroups that larger trials then flatten, which is the context any reader should hold here.

What it means

For now, GAD-alum remains an unapproved, unproven way to prevent type 1 diabetes. The follow-up does not show it works; it shows that if it works at all, the effect may depend on genotype — a hypothesis that only a prospective trial enrolling HLA-DR3-DQ2-positive children, and powered for that question, could confirm. Families of autoantibody-positive children should read this as a reason to watch the research, not as evidence of a treatment they are missing. The useful lesson is methodological: averaging a treatment effect across a genetically mixed group can hide benefit and harm that cancel out — but acting on the pieces before they are confirmed is how medicine repeatedly fools itself.

Sources

• [s1] Delayed time to stage 3 type 1 diabetes after GAD-alum treatment in HLA-DR3-DQ2-positive children: follow-up of the DiAPREV-IT trial. Diabetologia, 28 Sep 2026. https://doi.org/10.1007/s00125-026-06875-3
• [s2] Diabetes Prevention - Immune Tolerance (DiAPREV-IT), ClinicalTrials.gov NCT01122446. https://clinicaltrials.gov/study/NCT01122446