JADE Mono-1: Complete Statistical Analysis of PF-04965842 in Moderate to Severe Atopic Dermatitis

1. Trial at a Glance

JADE Mono-1 was a randomized, parallel-group, quadruple-masked phase 3 trial testing whether two oral doses of PF-04965842 (100 mg and 200 mg) produced higher rates of skin clearance and extent-of-disease improvement than placebo after 12 weeks in adolescents and adults with moderate to severe atopic dermatitis.

2. Clinical Question

The trial asked whether once-daily oral PF-04965842, at either of two doses, increased the proportion of participants achieving clinically meaningful skin improvement at Week 12, measured by both a global physician rating (IGA) and a composite extent-and-severity index (EASI), compared with placebo.

3. Trial Design

4. Randomization, Stratification, and Analysis Populations

The registry identifies the randomization strata as baseline disease severity and age category. These strata were carried into the analyses: the binary endpoints were analysed with a Cochran-Mantel-Haenszel risk difference that weights stratum-specific differences, and the continuous pruritus-symptom score was analysed with a mixed model that included the strata as fixed factors.

The FAS is a modified intention-to-treat population: it keeps participants in their randomized arm but requires at least one dose. Because 156 + 154 + 77 equals the enrollment of 387, the at-risk counts suggest that essentially all enrolled participants contributed to safety, which limits the practical gap between the FAS and a strict all-randomized population.

5. Endpoints

Primary endpoints (registry wording, shortened)

Secondary endpoints with posted analyses

The ClinicalTrials.gov record contains many further secondary and other analyses beyond those summarized here.

6. Primary Endpoint Results: IGA Response at Week 12

Both doses were compared with placebo using a Cochran-Mantel-Haenszel approach. The registry describes the estimate and confidence interval as a weighted average of the stratum-specific differences in response percentage, using the normal approximation to binomial proportions. Arm-level response percentages are not reproduced here; the effect measure is the difference in percentage points.

7. Primary Endpoint Results: EASI-75 at Week 12

8. Secondary Results: Time Course of IGA and EASI-75 Response

The registry posts the same CMH risk-difference analysis at Weeks 2, 4 and 8. Read together with the Week 12 primary results, these describe how quickly the treatment differences emerged.

For 200 mg, every posted difference has a confidence interval excluding zero from Week 2 onward. For 100 mg, the Week 2 IGA and EASI-75 intervals and the Week 4 IGA interval include zero. This pattern is consistent with a slower or smaller early response at the lower dose; it should not be read as evidence that the 100 mg dose "did not work" at those visits, because a non-significant result is not a demonstration of no effect.

9. Secondary Results: Pruritus NRS Responders (≥4-Point Improvement)

For the pruritus NRS responder endpoint the registry describes a different analysis path from the skin endpoints: missing values were multiply imputed, each completed dataset was analysed with the CMH risk-difference method adjusting for randomization strata, separately for each week, and the results were combined with Rubin's rules to obtain the treatment difference, 95% CI and p-value. Results are posted for both the FAS and the PPAS.

Notably, the 200 mg pruritus difference is already 42.5 percentage points at Week 2 in the FAS, a larger early separation than either skin endpoint at the same visit. Itch is a patient-reported outcome, so this endpoint adds a dimension the clinician-rated primary endpoints cannot capture.

10. Secondary Results: PSAAD Total Score (MMRM)

The PSAAD total score is a continuous, patient-reported symptom measure, and the effect measure is a difference in least squares (LS) mean change from baseline. The registry describes a mixed model for repeated measures (MMRM) with fixed factors for treatment, week, treatment-by-week interaction, randomization strata (baseline disease severity and age category) and baseline value, using an unstructured covariance matrix. Negative differences favour PF-04965842 (greater symptom reduction).

These differences are in scale units, not percentage points, and the registry entries here do not state the PSAAD scale range or a minimal clinically important difference. Their magnitude therefore cannot be judged from the numbers alone; what can be said is that the 200 mg differences are consistently about twice the 100 mg differences at each visit, and the intervals for both doses exclude zero at every posted week.

11. Safety: Serious Adverse Events

Serious adverse events were infrequent in all three arms. With only 5, 5 and 3 affected participants, any between-arm comparison of serious adverse event rates is extremely imprecise; a 12-week trial of this size is not designed to detect differences in uncommon serious events, and the absence of an apparent imbalance should not be read as evidence of equivalent safety.

12. Statistical Methodology

Cochran-Mantel-Haenszel risk difference

The binary endpoints were analysed with a CMH approach adjusting for the randomization strata. Within each stratum k, the difference in response proportions is computed, and these are combined into a weighted average. The registry specifies that the confidence interval uses the normal approximation to binomial proportions.

Mixed model for repeated measures

The PSAAD analysis used an MMRM with treatment, week, treatment-by-week, strata and baseline score as fixed effects, and an unstructured covariance matrix for the within-participant correlation across visits. Including baseline value is an ANCOVA-style adjustment that improves precision; the treatment-by-week interaction allows a separate treatment effect at each visit.

Multiple imputation with Rubin's rules

For the pruritus NRS responder endpoint, missing responses were multiply imputed. Each completed dataset was analysed with the stratified CMH method, and estimates were pooled. Rubin's rules combine the average within-imputation variance with the between-imputation variance, so the final confidence interval reflects uncertainty due to missing data.

Analysis populations and hypotheses

All posted comparisons are labelled as superiority tests with two-sided 95% confidence intervals. Primary efficacy used the FAS; the PPAS served as a supportive sensitivity population for pruritus endpoints. The registry does not describe interim analyses, and the multiplicity strategy for the four primary comparisons and the secondary endpoints is not described in the record.

13. Statistical Methods Explained

Why use a Cochran-Mantel-Haenszel risk difference rather than a simple difference in proportions?

Randomization was stratified by baseline disease severity and age category. Response rates may differ substantially between strata (for example, between more and less severe disease), and the CMH approach estimates the treatment difference within each stratum before combining them. This respects the design, protects against chance imbalance across strata, and typically gives a more precise estimate than an unstratified comparison.

What does a risk difference of 51.0 percentage points mean?

It is an absolute difference in the proportion of responders. If the weighted EASI-75 response rate in the 200 mg arm exceeded placebo by 51.0 percentage points, then, on average, about 51 additional participants per 100 treated reached EASI-75 compared with placebo. Its reciprocal gives an approximate number needed to treat, but this should be computed with care because the interval for the difference also implies an interval for that number.

Why do the pruritus responder analyses use multiple imputation while the PSAAD analyses use MMRM?

A responder endpoint is a yes/no outcome at a single visit; a participant with a missing diary value has no observed response, so the analysis must either impute it or assign an outcome by rule. Multiple imputation generates plausible values several times, reflecting uncertainty. A continuous repeated outcome such as PSAAD can instead be modelled directly with MMRM, which uses all available visits and handles missing data under a missing-at-random assumption without imputing.

Why can the same effect be "significant" in the FAS but not in the PPAS?

At Week 4, the 100 mg pruritus difference was 15.0 in the FAS (P = 0.0251) and 12.5 in the PPAS (P = 0.1138). The PPAS excludes participants with major protocol violations, reducing sample size and widening the interval, and the point estimate itself moved slightly. Neither population is "the truth": the FAS better preserves randomization, while the PPAS asks what happens among participants who followed the protocol. Consistency of direction and magnitude across both is more informative than the 0.05 threshold in either.

Why does a smaller placebo arm matter?

The variance of a difference in proportions is the sum of the variances in each arm. With 77 participants at risk in the placebo arm versus about twice that in each active arm, the placebo rate contributes disproportionately to the uncertainty. Unequal allocation gives more participants access to active treatment and more information about the active drug's safety, at some cost in precision for each comparison.

Do four primary comparisons require a multiplicity adjustment?

With two doses and two co-primary endpoints, testing each at an unadjusted two-sided 0.05 level would inflate the chance of at least one false-positive claim. A trial of this design would ordinarily prespecify a procedure such as hierarchical testing or alpha splitting; the ClinicalTrials.gov record does not describe it. Because all four p-values are 0.0037 or smaller, the conclusions here are unlikely to be sensitive to reasonable adjustment, but formal conclusions should rest on the prespecified procedure.

14. Limitations and Interpretation Issues

• Short duration: the primary endpoints are at Week 12. The registry data summarized here do not address durability or long-term safety.
• Dichotomized endpoints: IGA response and EASI-75 discard information about the degree of improvement, and differences depend on the thresholds chosen.
• Correlated endpoints: IGA and EASI are both clinician-rated measures of the same disease process, so agreement between them is partly expected rather than independent confirmation.
• Shared control group: both dose comparisons depend on the same placebo arm, so a chance high or low placebo response would affect both estimates.
• Multiplicity: many secondary analyses across endpoints, weeks and populations are posted; isolated nominal p-values among them should not be treated as confirmatory.
• Missing data assumptions: both multiple imputation and MMRM typically rely on a missing-at-random assumption that cannot be verified from the data.
• Normal approximation: stratum-weighted differences with the normal approximation can be less accurate when some strata have few responders, as is plausible at early visits.
• Rare-event safety: with 13 participants with serious adverse events across all arms, safety comparisons are very imprecise.
• No dose-versus-dose test: the posted analyses compare each dose with placebo; the apparent dose gradient is descriptive.

15. Why This Trial Matters Statistically

JADE Mono-1 is a compact illustration of how modern dermatology trials analyse binary responder endpoints alongside continuous patient-reported outcomes, and how analysis choices are tied to the randomization design.

16. Related Tutorials

Learn more about the methods used in this trial:

17. Related Calculators

18. Sources

• ClinicalTrials.gov registry record: NCT03349060 — Study to Evaluate Efficacy and Safety of PF-04965842 in Subjects Aged 12 Years And Older With Moderate to Severe Atopic Dermatitis.
• PubMed: PMID 42162528.
• PubMed: PMID 40028832.
• PubMed: PMID 39104653.
• PubMed: PMID 38896380.
• PubMed: PMID 37988255.

19. Record Summary

In JADE Mono-1, both PF-04965842 doses produced higher Week 12 IGA and EASI-75 response rates than placebo, with stratum-weighted risk differences of 15.8 and 36.0 percentage points for IGA and 27.9 and 51.0 percentage points for EASI-75 (100 mg and 200 mg respectively), and confidence intervals excluding zero for all four primary comparisons. Secondary pruritus and symptom endpoints moved in the same direction, with consistently larger differences at 200 mg. The most useful interpretation combines absolute risk differences, their confidence intervals, the time course across visits, the sensitivity to analysis population, and the design features — stratification, a shared placebo arm and missing-data methods — that shape how these numbers should be read.