This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
DELTA 2 was a randomized, double-blind, parallel-group phase 3 trial testing whether a topical Janus kinase inhibitor, delgocitinib cream 20 mg/g, increased the proportion of adults with moderate to severe chronic hand eczema who achieved Investigator's Global Assessment treatment success at Week 16, compared with the cream vehicle alone.
| Feature | DELTA 2 |
|---|---|
| Phase | Phase 3 |
| Condition | Moderate to severe chronic hand eczema in adults |
| Design | Randomized, parallel-group, double-masked, vehicle-controlled |
| Arms | 2: delgocitinib cream 20 mg/g; cream vehicle |
| Enrollment | 473 participants |
| Primary endpoint | Number of participants with IGA-CHE treatment success (IGA-CHE TS) at Week 16 |
| Primary analysis | Cochran-Mantel-Haenszel test; Mantel-Haenszel risk difference stratified by region and baseline IGA-CHE score |
| Start / primary completion | 2021-05-25 / 2022-12-27 |
| Lead sponsor | LEO Pharma (industry) |
| ClinicalTrials.gov | NCT04872101 |
2. Clinical Question
The central question was whether applying delgocitinib cream 20 mg/g, rather than an identical-appearing vehicle cream, would increase the proportion of patients whose hand eczema was rated clear or almost clear with at least a 2-step improvement from baseline after 16 weeks of treatment.
Population
Adults with moderate to severe chronic hand eczema.
Intervention
Delgocitinib cream 20 mg/g, a topical Janus kinase inhibitor.
Comparator
Cream vehicle, the same base formulation without the active drug.
Primary question
Does delgocitinib cream increase the proportion of participants achieving IGA-CHE treatment success at Week 16 relative to vehicle?
A vehicle comparator is the natural control for a topical product. Emollient bases can themselves improve dry, fissured skin, so the vehicle arm isolates the contribution of the active ingredient rather than the contribution of applying a cream at all. This has a direct statistical consequence: the vehicle response rate is not a "no treatment" rate, and the risk difference estimates the effect attributable to delgocitinib on top of the vehicle.
3. Trial Design
Delgocitinib cream 20 mg/g
- Topical Janus kinase inhibitor cream
- Double-masked application
- Assessed through Week 16
Cream vehicle
- Same cream base without active drug
- Double-masked application
- Assessed through Week 16
The arm sizes above are the numbers of participants at risk in the serious adverse event summary on the ClinicalTrials.gov record. The registry does not list the planned allocation ratio, but the unequal arm sizes indicate that more participants were assigned to active treatment than to vehicle, a common choice in vehicle-controlled dermatology trials to gain additional safety exposure to the active drug.
4. Endpoints
Primary endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Number of participants with IGA-CHE TS at Week 16 | The IGA-CHE rates global chronic hand eczema severity on a 5-point scale from 0 (clear) to 4 (severe). IGA-CHE treatment success is an IGA-CHE score of 0 (clear) or 1 (almost clear) with a ≥2-step improvement from baseline. | 16 weeks |
The primary endpoint is a binary responder endpoint built from an ordinal scale. The dual requirement (an absolute state of clear or almost clear, plus a relative improvement of at least 2 steps) prevents a participant who starts at "moderate" and moves to "mild" from counting as a success, and prevents a participant who was only mildly affected from counting after a trivial change.
Key secondary endpoints with posted analyses
| Endpoint family | Measure | Time points |
|---|---|---|
| IGA-CHE TS | Responder (binary) | Weeks 4 and 8 |
| HESD itch score (weekly average) reduction ≥4 points | Responder (binary) | Weeks 2, 4, 8 and 16 |
| HESD pain score (weekly average) reduction ≥4 points | Responder (binary) | Weeks 4, 8 and 16 |
| HESD score (weekly average) reduction ≥4 points | Responder (binary) | Weeks 4, 8 and 16 |
| HECSI-75 and HECSI-90 | Responder (binary) | HECSI-75 at Weeks 8 and 16; HECSI-90 at Week 16 |
| DLQI reduction ≥4 points | Responder (binary) | Week 16 |
| Percentage change in HECSI score | Continuous | Baseline to Week 16 |
| Change in DLQI, HESD, HESD itch, HESD pain, HEIS and HEIS PDAL scores | Continuous | Baseline to Week 16 |
5. Estimand, Stratification and Missing Data
Every posted analysis on the ClinicalTrials.gov record is described as the primary analysis of the primary estimand "composite". A composite estimand folds intercurrent events into the outcome definition itself rather than treating them as missing data. In DELTA 2 this was done in two ways, depending on the endpoint type.
6. Primary Endpoint Result
The primary endpoint was analyzed with the Cochran-Mantel-Haenszel (CMH) test, and the effect was summarized as a Mantel-Haenszel risk difference stratified by region and baseline IGA-CHE score. Data were not available for 1 participant in the delgocitinib group.
IGA-CHE treatment success at Week 16 — risk difference
95% CI: 15.8 to 28.5 (two-sided) · P < 0.001
Delgocitinib cream 20 mg/g vs cream vehicle · Superiority hypothesis · Cochran-Mantel-Haenszel
| Item | Value |
|---|---|
| Comparison | Delgocitinib cream 20 mg/g vs cream vehicle |
| Effect measure | Risk difference (Mantel-Haenszel, stratified) |
| Estimate | 22.2 |
| 95% confidence interval | 15.8 to 28.5 |
| P-value | <0.001 |
| Strata | Region; baseline IGA-CHE score |
| Intercurrent events | Missing data, rescue treatment and permanent IMP discontinuation counted as non-response |
What the estimate means. A risk difference of 22.2 means that, averaged across the region and baseline-severity strata, the proportion of participants achieving IGA-CHE treatment success at Week 16 was estimated to be 22.2 percentage points higher with delgocitinib cream than with vehicle. This is an absolute difference in proportions, which is the most directly interpretable effect scale for a responder endpoint.
What it does not mean. It is not a 22.2% relative improvement, and it does not say that 22.2% of delgocitinib-treated participants responded. The individual arm response proportions are needed to express the effect in relative terms, and a risk difference says nothing about which individual patients benefited.
Precision. The two-sided 95% confidence interval of 15.8 to 28.5 excludes zero by a wide margin. Even the lower bound corresponds to a meaningful absolute difference in treatment success, so the conclusion does not depend on the point estimate being exactly right.
The p-value. P < 0.001 indicates that a difference of this size would be very unlikely if the true stratified difference were zero. It does not measure how large the effect is; the size of the effect is conveyed by the risk difference and its interval.
Cautions. The estimate belongs to the composite estimand: rescue treatment and discontinuation are counted as failures. A different estimand (for example, a hypothetical strategy that imputes what would have happened without rescue) could produce a different number. The non-responder rule was applied identically in both arms, so the comparison remains fair, but the absolute response levels in each arm depend on how often rescue and discontinuation occurred.
7. Secondary Endpoint Results
The ClinicalTrials.gov record posts formal analyses for a large set of secondary endpoints. All were tested as superiority comparisons of delgocitinib cream 20 mg/g against vehicle, using the same composite estimand.
IGA-CHE treatment success over time
| Time point | Risk difference | 95% CI | P-value |
|---|---|---|---|
| Week 4 | 6.5 | 0.8 to 12.3 | 0.043 |
| Week 8 | 22.9 | 16.0 to 29.8 | <0.001 |
| Week 16 (primary) | 22.2 | 15.8 to 28.5 | <0.001 |
The Week 4 difference is small and its interval lies close to zero, while the Week 8 and Week 16 differences are of similar size. This pattern suggests that the separation between arms for global clearance develops between Week 4 and Week 8 and is then maintained. Overlapping intervals at Weeks 8 and 16 mean that the data do not distinguish between a plateau and a modest change after Week 8.
Symptom responder endpoints (Hand Eczema Symptom Diary)
| Endpoint (≥4-point reduction, weekly average) | Week | Risk difference | 95% CI | P-value |
|---|---|---|---|---|
| HESD itch score | 2 | 6.6 | 1.1 to 12.0 | 0.031 |
| HESD itch score | 4 | 18.3 | 11.0 to 25.6 | <0.001 |
| HESD itch score | 8 | 29 | 21.3 to 36.7 | <0.001 |
| HESD itch score | 16 | 27.4 | 19.0 to 35.8 | <0.001 |
| HESD pain score | 4 | 20.5 | 13.1 to 27.8 | <0.001 |
| HESD pain score | 8 | 29.6 | 21.7 to 37.4 | <0.001 |
| HESD pain score | 16 | 26 | 17.0 to 35.1 | <0.001 |
| HESD score | 4 | 16.9 | 10.2 to 23.7 | <0.001 |
| HESD score | 8 | 25 | 17.5 to 32.5 | <0.001 |
| HESD score | 16 | 23.7 | 15.1 to 32.2 | <0.001 |
The number of participants with unavailable data differed across these analyses: 5 (delgocitinib) and 3 (vehicle) for itch, 6 and 6 for the overall HESD score, and 20 and 18 for pain. Responder analyses of patient-reported scores require a sufficiently high baseline score for a 4-point reduction to be possible, which likely explains why these analysis populations are smaller than the primary analysis population.
Sign-based and quality-of-life responder endpoints
| Endpoint | Week | Risk difference | 95% CI | P-value |
|---|---|---|---|---|
| HECSI-75 | 8 | 31 | 22.7 to 39.3 | <0.001 |
| HECSI-75 | 16 | 31.3 | 23.1 to 39.5 | <0.001 |
| HECSI-90 | 16 | 22.2 | 15.4 to 29.0 | <0.001 |
| DLQI reduction ≥4 points | 16 | 26.4 | 17.0 to 35.9 | <0.001 |
Continuous endpoints (ANCOVA)
| Endpoint (baseline to Week 16) | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Percentage change in HECSI score | Mean difference | -45.5 | -56.4 to -34.6 | <0.001 |
| Change in DLQI score | Mean difference | -3.9 | -5.0 to -2.8 | <0.001 |
| Change in HESD score (weekly average) | Mean difference | -1.9 | -2.4 to -1.4 | <0.001 |
| Change in HESD itch score (weekly average) | Median difference (as posted) | -2.0 | -2.5 to -1.4 | <0.001 |
| Change in HESD pain score (weekly average) | Mean difference | -2.0 | -2.6 to -1.5 | <0.001 |
| Change in HEIS score | Median difference (as posted) | -0.81 | -0.99 to -0.62 | <0.001 |
| Change in HEIS PDAL score | Mean difference | -0.82 | -1.01 to -0.62 | <0.001 |
Negative values favor delgocitinib because lower scores indicate less severe disease, symptoms or impact. For example, the HECSI estimate of -45.5 means that the adjusted mean percentage change from baseline was 45.5 percentage points more negative (greater improvement) with delgocitinib than with vehicle.
The secondary results are directionally consistent with the primary endpoint across clinician-rated signs (IGA-CHE, HECSI), patient-reported symptoms (HESD itch, pain, overall) and quality of life (DLQI, HEIS). Consistency across differently measured outcomes strengthens the overall evidence more than any single additional p-value does.
The ClinicalTrials.gov record does not describe the multiplicity strategy used across these many endpoints. With more than twenty posted comparisons, the nominal p-values should be read in light of whatever hierarchical testing procedure the protocol specified; they should not each be treated as an independent confirmatory claim.
For the continuous outcomes, WOCF imputation assigns the worst observed value after rescue, discontinuation or missed visits. This penalizes treatment failure, but it is a single-imputation method that does not propagate uncertainty about the imputed values, so the confidence intervals may be somewhat narrower than a multiple-imputation approach would give.
8. Safety: Serious Adverse Events
| Arm | Participants with serious adverse events / at risk |
|---|---|
| Delgocitinib cream 20 mg/g | 5 / 313 |
| Cream vehicle | 3 / 159 |
Serious adverse events were infrequent in both arms. With so few events, the trial has very little power to detect differences in serious adverse event rates, and no formal comparison of safety is posted on the registry. Safety summaries of this kind are descriptive and are generally based on participants who received at least one application of study treatment.
9. Statistical Methodology
Cochran-Mantel-Haenszel test
The CMH test compares binary outcomes between two groups while conditioning on strata. Within each combination of region and baseline IGA-CHE score, a 2×2 table of treatment by response is formed; the test combines the information across tables, giving each stratum a weight related to its size. This controls for imbalances in stratification factors and typically improves precision when the factors predict response.
where p̂1k and p̂0k are the observed response proportions for delgocitinib and vehicle in stratum k, and the weights favor strata with more balanced and larger samples.
ANCOVA for continuous outcomes
Analysis of covariance models the Week 16 change (or percentage change) as a function of treatment, adjusted for region, baseline IGA-CHE score and the baseline value of the same scale. Adjusting for the baseline value removes a large source of between-patient variability and corrects for chance baseline imbalance, which narrows the confidence interval for the treatment difference compared with an unadjusted comparison of change scores.
The coefficient β1 is the adjusted mean difference between arms, reported on the registry as the "mean difference (net)".
Handling of intercurrent events and missing data
Non-responder imputation for binary endpoints and WOCF for continuous endpoints both implement the composite strategy. Each is transparent and conservative with respect to treatment failure, but each relies on a single fixed value per participant. The registry does not report sensitivity analyses under alternative estimands or multiple imputation.
Hypothesis and test sidedness
All posted analyses are superiority comparisons with two-sided 95% confidence intervals. A two-sided interval that excludes zero corresponds to rejection of the null hypothesis of no difference at the two-sided 5% level.
10. Statistical Methods Explained
Why use a risk difference rather than an odds ratio for the primary endpoint?
A risk difference is on the same scale that clinicians and patients use to think about response: "how many more people out of 100 reach clear or almost clear skin?" Odds ratios are mathematically convenient but exaggerate the apparent effect when response is common and are frequently misread as relative risks. The Mantel-Haenszel risk difference keeps the stratified structure of the analysis while reporting an absolute effect.
Why stratify by region and baseline IGA-CHE score?
Baseline severity strongly predicts whether a participant can reach a score of 0 or 1 with a 2-step improvement: a participant starting at 4 (severe) must move further than one starting at 3 (moderate). Region can capture differences in climate, occupational exposure and clinical practice. Stratifying both randomization and analysis on these factors keeps the comparison balanced and removes their influence from the treatment effect estimate.
What does a negative mean difference of -45.5 in HECSI percentage change mean?
It is the adjusted difference between arms in the average percentage change from baseline. If vehicle-treated participants improved by a given percentage on average, delgocitinib-treated participants improved by an estimated additional 45.5 percentage points. It is a difference of averages, not the percentage of participants who improved.
Why were participants who used rescue therapy counted as non-responders?
Rescue medication can itself clear eczema. If rescued participants were analyzed on their observed scores, the vehicle arm could look better than vehicle alone really is, diluting the estimated effect of delgocitinib. Counting rescue as failure keeps the estimand focused on success achieved with the randomized treatment.
Why does the Week 4 IGA-CHE result have P = 0.043 while later time points have P < 0.001?
At Week 4 the estimated difference (6.5 percentage points) is much smaller relative to its standard error, so the evidence against no difference is weaker. The p-value reflects both the size of the difference and its precision; it is not itself a measure of effect size, and a smaller p-value at Week 16 does not mean the drug works "more" at Week 16 in any direct sense.
Is worst observation carried forward a conservative choice?
WOCF penalizes treatment failure and so tends to disadvantage the arm with more rescue or discontinuation. It is conservative in that direction, but it is not a principled model of missing data; it understates variability because the imputed values are treated as if they were observed.
11. Limitations
- Multiplicity: the registry does not describe the testing hierarchy across the many secondary endpoints and time points, so secondary p-values should not be interpreted as independent confirmatory findings.
- Single-imputation methods: non-responder imputation and WOCF are simple and transparent but can understate uncertainty.
- Estimand specificity: results apply to the composite estimand; the effect under alternative handling of rescue or discontinuation is not reported on the registry.
- Arm-level proportions: the registry analyses emphasize between-arm differences; relative measures and numbers needed to treat require the arm-level response proportions.
- Duration: the posted comparisons cover 16 weeks; longer-term durability of the effect is not addressed by these analyses.
- Safety precision: serious adverse events were rare (5/313 vs 3/159), so the trial cannot characterize uncommon serious risks precisely.
- Subgroups: the registry does not report subgroup analyses, so consistency of the effect across baseline characteristics cannot be assessed from these data.
12. Why This Trial Matters Statistically
DELTA 2 is a clear teaching example of how a modern dermatology trial analyzes responder and continuous endpoints under an explicitly named estimand.
| Concept | How it appears in DELTA 2 |
|---|---|
| Composite estimand | Rescue and IMP discontinuation defined as treatment failure |
| Cochran-Mantel-Haenszel test | Stratified comparison of binary responder endpoints |
| Risk difference | Absolute effect scale for the primary and responder endpoints |
| ANCOVA | Baseline-adjusted analysis of continuous score changes |
| Covariate adjustment | Region, baseline IGA-CHE and baseline outcome value |
| Missing data | Non-responder imputation and WOCF |
| Confidence intervals | Two-sided 95% intervals for every posted comparison |
| Repeated time points | Effects tracked at Weeks 2, 4, 8 and 16 |
13. Related Tutorials
Learn more about the methods used in this trial:
14. Related Calculators
15. Sources
- ClinicalTrials.gov: NCT04872101 — Efficacy and Safety of Delgocitinib Cream in Adults With Moderate to Severe Chronic Hand Eczema (DELTA 2).
- PubMed: PMID 39033766.
- PubMed: PMID 42143645.
- PubMed: PMID 41775339.
- PubMed: PMID 41240207.
- PubMed: PMID 40081663.
Explore the methods behind the results
Each trial's endpoints and analyses connect to a deeper statistical tutorial and to calculators for working through similar problems.
16. Record Summary
DELTA 2 randomized 473 adults with moderate to severe chronic hand eczema to delgocitinib cream 20 mg/g or cream vehicle. Under a composite estimand that treated rescue, discontinuation and missing data as failure, the stratified Mantel-Haenszel risk difference for IGA-CHE treatment success at Week 16 was 22.2 percentage points (95% CI 15.8 to 28.5; P < 0.001). Secondary responder and ANCOVA analyses of signs, symptoms and quality of life pointed in the same direction, and serious adverse events were infrequent in both arms. The most informative reading combines the absolute risk difference, its confidence interval, the estimand that defines what was measured, and awareness of the multiplicity across the secondary endpoints.