GSK1070806 in atopic dermatitis: when a 44 percent placebo response ends an IL-18 program

What was tried

GlaxoSmithKline ran a Phase 2b randomized, double-blind, placebo-controlled, parallel-group, dose-finding study of GSK1070806 in adults with moderate-to-severe atopic dermatitis (NCT05999799). GSK1070806 is a humanized monoclonal antibody that binds interleukin-18 (IL-18) and neutralizes its signaling. The study randomized 161 participants across placebo and four subcutaneous dose levels, with a 16-week treatment period and a primary endpoint of percent change from baseline in the Eczema Area and Severity Index (EASI) at Week 16. Eligibility required an EASI of at least 16, an Investigator's Global Assessment of at least 3, and disease involving at least 10 percent of body surface area. The study began on 16 November 2023 and reached its primary completion on 23 July 2025, at which point GSK terminated it. The recorded reason was a sponsor decision to stop early on a pre-specified efficacy futility criterion, with no safety concerns identified.

The biological hypothesis

IL-18 is an IL-1 family cytokine that promotes interferon-gamma production, activates NK and T cells, and amplifies inflammatory circuits depending on the surrounding cytokine context. In atopic dermatitis, IL-18 is elevated in lesional skin and serum, and it has been proposed as a driver of barrier disruption and itch (Front Immunol, 2023, 10.3389/fimmu.2023.955369; JACI, 2025, 10.1016/j.jaci.2025.07.025). The therapeutic logic was that neutralizing IL-18 would dampen this upstream amplifier and reduce eczema severity in a way that complements the established IL-4 and IL-13 axis. The genetic and clinical support for IL-18 as a causal node in atopic dermatitis was thin going in. The Open Targets overall association score for IL18 against atopic eczema is 0.0987, carried almost entirely by literature co-occurrence rather than human genetic evidence, and Mendelian randomization work in atopic dermatitis has not nominated IL-18 as a causal driver (J Invest Dermatol, 2024, 10.1016/j.jid.2023.10.016).

What actually happened

The primary endpoint did not separate from placebo. Mean percent change from baseline in EASI at Week 16 was -44.53 percent for placebo, compared with -42.86, -69.95, -41.49, and -58.08 percent for dose levels 1 through 4. Two active arms came in below placebo, and the response did not rise with dose. Secondary EASI-75 responder rates told the same story: 12.8 percent for placebo (5 of 39), against 20.0, 47.6, 16.7, and 26.3 percent across the four dose arms. Only dose level 2 produced a responder rate that clearly exceeded placebo, and it did so without support from the adjacent doses. The IGA 0 or 1 endpoint with a 2-point drop reached 7.7 percent on placebo and stayed in single digits to low double digits on active treatment. Safety was unremarkable, with serious adverse events in one participant each in dose levels 3 and 4, none on placebo, and no new signals. The decision was therefore an efficacy decision, taken at a planned futility look rather than after a fully enrolled comparison.

Failure mechanism, best guess

Two facts dominate. The first is the placebo response of -44.5 percent on EASI, which is high for atopic dermatitis and compresses the room any active drug has to demonstrate separation. The second is the absence of a coherent dose-response. When a biologic engages its target across a wide dose range and only one non-terminal dose beats control, the most parsimonious reading is that IL-18 neutralization does not move the disease much in unselected moderate-to-severe atopic dermatitis, and the dose level 2 result reflects variation in small arms rather than a real exposure-response relationship. This fits the weak causal evidence for IL-18 in this indication. The cytokine is elevated in lesional skin, but elevation is not the same as causation, and the IL-4 and IL-13 axis appears to carry most of the type 2 signal that current biologics exploit.

How to prevent this next time

The first lever is a Bayesian predictive-probability framework with a pre-specified stopping rule. The predictive probability of trial success integrates over the posterior for the treatment effect given interim data:

PP(success)=∫θ​P(final success∣θ,interim data)⋅π(θ∣interim data)dθ

A low predictive probability at the interim look is exactly what should trigger an early stop, and that is what happened here, which is the system working as designed.

The second lever is an honest power calculation against a realistic placebo response. To detect an EASI-75 difference of 13 percent versus 28 percent at two-sided alpha 0.05 with 80 percent power, a two-proportion calculation requires roughly 110 participants per arm. With about 40 participants in each active arm, power for that contrast was approximately 40 percent, so the dose-finding stage was structurally underpowered to distinguish a true moderate effect from a high placebo response. A historical base-rate adjustment matters here, because atopic dermatitis placebo responses frequently land between 10 and 20 percent on EASI-75 and can run higher with background topical therapy, so the analysis plan should have anchored on a responder endpoint with a pre-registered placebo assumption.

The single highest leverage change would have been powering the dose-finding stage around an EASI-75 responder contrast against a realistically high placebo response, instead of a continuous EASI percent-change endpoint that the observed placebo effect could absorb.

What this means for similar programs

IL-18 remains an active target in inflammation, and a separate anti-IL-18 antibody (aletekitug) has reported clinical and molecular effects in atopic dermatitis (Allergy, 2026, 10.1111/all.70172). The lesson is not that IL-18 is undruggable. It is that biology with literature-only association and no human genetic anchor should carry a higher evidentiary bar before a dose-ranging trial, and that small per-arm sizes plus a continuous endpoint are a fragile combination in indications with large placebo responses. Programs targeting cytokines that are elevated but not genetically validated should plan for biomarker-defined enrichment from the start.

Open questions

Did any biomarker-defined subgroup, such as participants with high baseline IL-18 or a type 1 skewed signature, show a stronger effect that a pooled analysis masked? Would a higher or more frequent dose have shifted the exposure-response, or was target engagement already saturated at the tested levels? And does the dose level 2 EASI-75 signal of 47.6 percent reflect anything real, or is it small-sample variation that the adjacent doses fail to corroborate?

Sources

1. • ClinicalTrials.gov, NCT05999799, study record and results posting (enrollment, arms, EASI primary and secondary outcomes, adverse events, termination reason). https://clinicaltrials.gov/study/NCT05999799 - Open Targets Platform, IL18 (ENSG00000150782) association with atopic eczema (EFO_0000274), overall score 0.0987. https://platform.opentargets.org/target/ENSG00000150782 - ChEMBL, GSK1070806 (CHEMBL2109606), maximum phase 2. https://www.ebi.ac.uk/chembl/ - IL-18 in atopic dermatitis, a multifaceted driver of skin inflammation. JACI, 2025. https://- Interleukin-18 and IL-18BP in inflammatory dermatological diseases. Front Immunol, 2023. https://- Interleukin-18 cytokine in immunity, inflammation, and autoimmunity. Front Immunol, 2022. https://- A narrative review of IL-18 and IL-37 in the pathogenesis of atopic dermatitis. Int J Mol Sci, 2024. https://- Mendelian randomization studies in atopic dermatitis, a systematic review. J Invest Dermatol, 2024. https://- Clinical and molecular effect of the anti-IL-18 antibody aletekitug in adults with atopic dermatitis. Allergy, 2026. https://- Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics, 2019. https://.