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
ESTEEM 2 was a randomized, quadruple-masked phase 3 trial comparing oral apremilast (APR 30 mg twice daily) with placebo in patients with moderate to severe plaque psoriasis. A 16-week placebo-controlled phase registry-reported the primary endpoint, and a later randomized treatment-withdrawal phase (Weeks 32 to 52) tested whether the response was maintained after stopping the drug.
| Feature | ESTEEM 2 |
|---|---|
| Phase | Phase 3 |
| Condition | Plaque psoriasis (moderate to severe) |
| Design | Randomized, parallel-group, quadruple-masked, placebo-controlled, with a randomized treatment-withdrawal phase |
| Interventions | Apremilast (drug); placebo (drug); topical or phototherapy (other) |
| Primary endpoint | PASI-75 response at Week 16 |
| Primary analysis | Risk difference; chi-squared test; FAS with LOCF imputation |
| Enrollment | 413 |
| Dates | Start 2010-11-22; primary completion 2012-03-15 |
| ClinicalTrials.gov | NCT01232283 |
| Sponsor | Amgen (industry) |
2. Clinical Question
ESTEEM 2 asked two linked questions: first, does oral apremilast produce a greater proportion of patients achieving at least a 75% reduction in psoriasis severity than placebo after 16 weeks; and second, among patients who responded, is the benefit lost more quickly when apremilast is withdrawn than when it is continued?
Population
Patients with moderate to severe plaque psoriasis enrolled in a placebo-controlled, masked trial.
Intervention
Oral apremilast 30 mg twice daily (APR 30 BID).
Comparator
Placebo during Weeks 0–16; placebo re-randomization during the Weeks 32–52 withdrawal phase.
Primary question
Is the percentage of participants achieving PASI-75 at Week 16 higher with apremilast than with placebo?
3. Trial Design
Apremilast 30 mg BID
- Oral apremilast twice daily from randomization
- Analysed in the FAS as randomized
- Eligible participants re-randomized at Week 32 to continue apremilast or switch to placebo
Placebo
- Oral placebo during the placebo-controlled phase
- Analysed in the FAS as randomized
- Topical or phototherapy listed as an additional intervention in the registry
4. Endpoints
| Endpoint | Registry definition (shortened) | Time frame |
|---|---|---|
| Primary: PASI-75 response | Percentage of participants who achieved at least a 75% reduction from baseline in PASI score at Week 16. PASI combines erythema, thickness and scaling (each 0–4) with area involvement (0–6) across head, trunk, upper limbs and lower limbs; total score ranges from 0 to 72, higher scores reflecting greater severity. PASI was set to missing if any severity score or degree of involvement was missing. | Baseline to Week 16 |
| Secondary: sPGA 0/1 | Static Physician Global Assessment of clear (0) or almost clear (1) with at least 2 points reduction from baseline | Baseline to Week 16 |
| Secondary: BSA | Percent change from baseline in affected body surface area | Baseline and Week 16 |
| Secondary: PASI percent change | Percent change from baseline in PASI score | Baseline and Week 16 |
| Secondary: PASI-50 | Percentage achieving a 50% improvement in PASI | Baseline and Week 16 |
| Secondary: Pruritus VAS | Change from baseline in pruritus visual analog scale score | Baseline and Week 16 |
| Secondary: DLQI | Change from baseline in Dermatology Life Quality Index total score | Baseline and Week 16 |
| Secondary: SF-36 MCS | Change from baseline in SF-36 v2.0 Mental Component Summary score | Baseline to Week 16 |
| Secondary: PASI-75 + sPGA 0/1 | Achieving both PASI-75 and sPGA clear/almost clear with ≥2-point reduction | Baseline to Week 16 |
| Secondary: Time to loss of effect | Loss of 50% of the PASI improvement obtained at Week 32 compared with baseline, during the randomized treatment-withdrawal phase | Weeks 32 to 52 |
5. Primary Endpoint Results: PASI-75 at Week 16
The primary analysis compared apremilast with placebo in the full analysis set (FAS), which included all participants randomized per protocol, analysed in the group to which they were randomized. Missing Week 16 values were handled by last observation carried forward (LOCF).
Risk difference in PASI-75 response (apremilast − placebo)
95% CI: 16.3 to 29.6 (two-sided) · P < 0.0001 (chi-squared)
Hypothesis: superiority · Population: FAS, LOCF
| Element | Posted value |
|---|---|
| Groups compared | Apremilast vs placebo |
| Effect measure | Risk difference (percentage of participants) |
| Estimate | 23 |
| 95% CI | 16.3 to 29.6 (two-sided) |
| P-value | <0.0001 |
| Method | Chi-squared test |
| Missing data | Last observation carried forward |
What the estimate means. A risk difference of 23.0 means that the estimated proportion of participants reaching PASI-75 at Week 16 was 23.0 percentage points higher with apremilast than with placebo. It is an absolute difference in proportions.
What it does not mean. It does not mean that 23.0% of apremilast-treated patients responded, nor that apremilast increased the chance of response by 23% in relative terms. The registry analysis reports the difference rather than the two arm-level rates, so the relative size of the effect cannot be read from this estimate alone.
Precision. The 95% CI of 16.3 to 29.6 excludes zero comfortably. Even the lower limit represents a difference of more than 16 percentage points, so the data are not compatible with a trivially small benefit under the analysis assumptions.
The p-value is not the effect size. P < 0.0001 says that a difference this large would be very unlikely if apremilast and placebo produced the same response rate. It does not indicate how large or how clinically important the difference is; the estimate and interval do that.
Cautions. LOCF assumes that a participant's last observed status would have persisted to Week 16. If participants who discontinued early would have changed afterwards, LOCF can bias the estimate in either direction and tends to understate uncertainty because imputed values are treated as observed. The chi-squared test is unstratified as posted; any stratification used in the protocol is not described in the registry analysis.
6. Secondary Endpoint Results
The registry posts formal analyses for eight secondary endpoints. Binary endpoints were analysed with a chi-squared test and a risk difference; continuous endpoints were analysed with ANCOVA including treatment group as a factor and the baseline value as a covariate. All used the FAS with LOCF, and all were framed as superiority comparisons.
Binary response endpoints (Week 16)
| Endpoint | Risk difference | 95% CI | P-value |
|---|---|---|---|
| PASI-75 (primary, for reference) | 23 | 16.3 to 29.6 | <0.0001 |
| PASI-50 | 35.8 | 26.9 to 44.7 | <0.0001 |
| sPGA clear/almost clear with ≥2-point reduction | 16.1 | 10.2 to 21.9 | <0.0001 |
| PASI-75 and sPGA 0/1 combined | 14.2 | 8.5 to 20.0 | <0.0001 |
The ordering is what would be expected from nested thresholds: a less demanding criterion (PASI-50) shows a larger absolute difference, and the joint requirement of PASI-75 plus sPGA 0/1 shows the smallest. Risk differences depend on the baseline (placebo) rate, so differences between these endpoints do not by themselves indicate stronger or weaker drug activity.
Continuous endpoints (ANCOVA, Week 16)
| Endpoint | LS mean difference | 95% CI | P-value | Direction |
|---|---|---|---|---|
| Percent change in PASI | -34.8 | -42.4 to -27.2 | <0.0001 | Negative favours apremilast |
| Percent change in affected BSA | -42.15 | -51.11 to -33.20 | <0.0001 | Negative favours apremilast |
| Change in pruritus VAS | -21.3 | -28.4 to -14.2 | <0.0001 | Negative favours apremilast |
| Change in DLQI total score | -4.0 | -5.3 to -2.8 | <0.0001 | Negative favours apremilast |
| Change in SF-36 MCS | 2.63 | 0.69 to 4.56 | 0.0078 | Positive favours apremilast |
For the percent-change endpoints, the LS mean difference is expressed in percentage points of change: for example, -34.8 for PASI means that the adjusted mean percent reduction in PASI was 34.8 percentage points greater with apremilast than with placebo. For DLQI and pruritus VAS, lower scores indicate fewer symptoms or less impairment, so negative differences favour apremilast; for the SF-36 MCS, higher scores indicate better mental health status. The MCS interval (0.69 to 4.56) is closer to zero than the others, and its P-value of 0.0078 is less extreme, which is consistent with a smaller and less precisely estimated effect.
Randomized treatment-withdrawal phase (Weeks 32–52)
Hazard ratio for loss of effect: re-randomized to placebo vs continued apremilast
95% CI: 3.408 to 17.399 · P < 0.0001 (log-rank)
Loss of effect: loss of 50% of the PASI improvement obtained at Week 32 compared with baseline
Participants who had received apremilast and were re-randomized to placebo at Week 32 had an estimated hazard of losing effect 7.700 times that of participants re-randomized to continue apremilast. The interval is wide (3.408 to 17.399), reflecting the limited number of participants and events in the withdrawal phase, but its lower bound is well above 1.
7. Safety: Serious Adverse Events
The registry reports serious adverse events by study period. Denominators differ because each period includes a different set of participants and a different length of exposure.
| Group and period | Participants with SAEs / at risk |
|---|---|
| Placebo, Weeks 0–16 (placebo-controlled phase) | 3 / 136 |
| Apremilast, Weeks 0–16 (placebo-controlled phase) | 5 / 272 |
| APR-APR-PBO, Weeks 32–52 (randomized withdrawal) | 2 / 62 |
| Apremilast, Weeks 0–260 (apremilast-exposure period) | 44 / 380 |
Only the Weeks 0–16 rows form a randomized comparison. The Weeks 0–260 apremilast-exposure count covers a much longer observation window and pools all participants who received apremilast, so it cannot be compared directly with the 16-week placebo count. No formal statistical comparison of SAEs is posted in the registry.
8. Statistical Methodology
Chi-squared test and risk difference for binary endpoints
For PASI-75, PASI-50, sPGA 0/1 and the combined endpoint, the proportion of responders in each arm was compared using a chi-squared test, and the effect was summarized as a risk difference with a two-sided 95% CI.
The chi-squared statistic tests whether response is independent of treatment assignment; the risk difference and its confidence interval quantify the size of the difference.
ANCOVA for continuous endpoints
Percent change in PASI and BSA, and change in pruritus VAS, DLQI and SF-36 MCS, were analysed with analysis of covariance. The model included treatment group as a factor and the baseline value as a covariate, and both the estimate and the p-value were adjusted for that covariate.
β1 is the LS (least-squares) mean difference: the estimated treatment effect at a common baseline value.
Log-rank test and hazard ratio for the withdrawal phase
Time to loss of effect during Weeks 32–52 was compared between the re-randomized groups with a log-rank test, and summarized as a hazard ratio with 95% CI. Participants who did not lose effect by Week 52, or who left the phase early, contribute follow-up time as censored observations.
Analysis population and missing data
Efficacy analyses used the FAS: all participants randomized per protocol, analysed according to randomized group. This follows the intention-to-treat principle. For the continuous endpoints, the analysis included participants with a baseline value and at least one post-baseline value. Missing Week 16 values were imputed by LOCF across all Week 16 analyses.
Multiplicity
The registry lists one primary endpoint and a set of secondary endpoints but does not describe a hierarchical testing procedure or other multiplicity adjustment in the posted analyses. With nine secondary comparisons, the secondary p-values are most safely read as supportive evidence rather than independent confirmatory claims, although most are far below conventional thresholds.
9. Statistical Methods Explained
These outcomes are continuous and correlated with their baseline values. Adjusting for baseline in ANCOVA removes chance baseline imbalance from the comparison and reduces residual variance, which narrows the confidence interval compared with a simple comparison of change scores.
After adjusting for baseline DLQI, the average reduction in DLQI total score at Week 16 was estimated to be 4.0 points larger with apremilast than with placebo (95% CI -5.3 to -2.8). It is an average difference between groups, not the change experienced by any individual patient.
LOCF fills a missing Week 16 value with the last observed value. It is simple but assumes no change after dropout, which is often implausible in a chronic, fluctuating disease. It also treats imputed values as if they were observed, so standard errors can be too small. Multiple imputation or likelihood-based methods relax these assumptions.
It indicates that among initial apremilast recipients, stopping treatment was associated with a much higher rate of losing the Week 32 PASI improvement than continuing. Because both groups were created by re-randomization, the comparison is causal within this selected population. The wide interval (3.408 to 17.399) means the true ratio could be several-fold smaller or larger than 7.700.
They are two summaries of the same underlying score. PASI-75 dichotomizes each patient's change at a 75% threshold, which is clinically intuitive but discards information. PASI percent change keeps the full continuous information and is analysed on the mean scale. Consistent direction across both supports the robustness of the primary finding.
A generic mental-health summary score is less directly tied to skin disease than PASI or DLQI, so a smaller effect relative to its variability is expected. Its P-value of 0.0078 is still below 0.05, but its interval (0.69 to 4.56) approaches zero, and without a stated multiplicity procedure it should be treated with more caution.
10. Limitations and Interpretation Issues
- LOCF imputation: a single-imputation method that assumes stable outcomes after dropout and can understate uncertainty.
- Multiplicity: the registry does not describe a multiplicity-control strategy for the secondary endpoints.
- Arm-level rates not in the posted analyses: the analyses report differences, so relative measures and the number needed to treat are not directly available from them.
- Withdrawal-phase selection: the hazard ratio applies only to participants re-randomized at Week 32; it does not describe the full randomized population.
- Small withdrawal sample: the wide CI for the withdrawal HR reflects limited information, and proportional hazards cannot be assessed from the registry.
- Safety periods differ: SAE counts cover different durations and populations; only the Weeks 0–16 counts form a randomized comparison.
- Short placebo-controlled period: the randomized comparison with placebo ends at Week 16.
11. Why This Trial Matters Statistically
ESTEEM 2 brings together three common analysis frameworks in one trial: binary response analysis, covariate-adjusted continuous outcomes, and a time-to-event analysis inside a randomized withdrawal design.
| Concept | How it appears in ESTEEM 2 |
|---|---|
| Risk difference | PASI-75, PASI-50, sPGA 0/1 and combined response |
| Chi-squared test | Hypothesis test for all binary Week 16 endpoints |
| ANCOVA | Baseline-adjusted LS mean differences for five continuous endpoints |
| Intention-to-treat analysis | FAS analysed according to randomized group |
| Missing data | LOCF imputation for Week 16 endpoints |
| Randomized withdrawal | Re-randomization at Week 32 to continued apremilast or placebo |
| Log-rank test and hazard ratio | Time to loss of effect: HR 7.700 |
| Confidence intervals | Precision of every posted comparison |
12. Related Tutorials
Learn more about the methods used in this trial:
13. Related Calculators
14. Sources
- ClinicalTrials.gov: NCT01232283 — Study to Evaluate Safety and Effectiveness of Oral Apremilast (CC-10004) in Patients With Moderate to Severe Plaque Psoriasis.
- PubMed: PMID 37316690.
Explore more trial statistics
Connect this trial's endpoints and methods to tutorials and calculators for response rates, ANCOVA and time-to-event analysis.
15. Record Summary
In ESTEEM 2, apremilast increased the proportion of participants achieving PASI-75 at Week 16 by an estimated 23.0 percentage points relative to placebo (95% CI 16.3 to 29.6; P < 0.0001). Binary and ANCOVA-based secondary endpoints pointed in the same direction, and in the randomized withdrawal phase, switching to placebo was associated with a hazard ratio of 7.700 for loss of effect. Interpretation should weigh the reliance on LOCF, the absence of a described multiplicity procedure for secondary endpoints, and the selected population of the withdrawal phase.