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
POPLAR was a randomized, double-masked, parallel-group phase 2 trial evaluating imsidolimab versus placebo in adults with palmoplantar pustulosis. The registry reports enrollment of 59 participants and two study arms.
| Feature | POPLAR |
|---|---|
| Phase | Phase 2 |
| Status | Completed |
| Population | Adults with palmoplantar pustulosis |
| Design | Randomized, double-masked, parallel |
| Allocation | Randomized |
| Enrollment | 59 |
| Interventions | Imsidolimab and placebo |
| Primary purpose | Treatment |
| ClinicalTrials.gov | NCT03633396 |
| Lead sponsor | Vanda Pharmaceuticals |
| Sponsor type | Industry |
2. Clinical Question
The central efficacy question was whether treatment with imsidolimab produced a different change from baseline in disease severity, measured by the Palmoplantar Pustulosis Psoriasis Area Severity Index (PPPASI), at Week 16 compared with placebo.
Population
Adults with palmoplantar pustulosis.
Intervention
Imsidolimab, a biological intervention.
Comparator
Placebo, classified in the registry as a drug intervention.
Primary question
How did the change from baseline in PPPASI at Week 16 compare between the imsidolimab and placebo groups?
The registry also identified the incidence of treatment-emergent adverse events (TEAEs) from first dose of any study drug through Week 24 as a primary endpoint. Thus, the primary endpoint framework included both an efficacy measure and a clinical safety measure.
3. Trial Design
Placebo
- Placebo drug intervention
- Randomized comparator group
- Serious adverse events: 1/29 affected / at risk
Imsidolimab
- Imsidolimab biological intervention
- Randomized treatment group
- Serious adverse events: 0/30 affected / at risk
The randomized, double-masked, parallel structure is important statistically because treatment assignment is established before outcome assessment and the comparison is between concurrent randomized groups. The double-masked design also reduces opportunities for knowledge of treatment assignment to influence participant or investigator behavior and assessment.
4. Randomization and Analysis Populations
The registry identifies the allocation as randomized and the design model as parallel. The posted primary and secondary efficacy analyses use an intent-to-treat analysis set with available data.
| Analysis population | Role in the registry analysis |
|---|---|
| Intent-to-treat analysis set with available data | Used for the posted PPPASI and response analyses. |
| Safety population | The serious-adverse-event arm counts are reported as affected participants over participants at risk. |
The phrase “intent-to-treat analysis set with available data” deserves careful interpretation. The intent-to-treat principle preserves the randomized comparison, while restricting a particular model contribution to observations with available information can affect how missing outcome measurements enter the analysis. For the binary secondary endpoints, the registry additionally specifies multiple imputation for missing data.
5. Endpoints
| Endpoint | Registry definition / assessment | Time frame |
|---|---|---|
| Change From Baseline in PPPASI | The PPPASI assesses erythema, pustules, and desquamation of both palms and both soles, each on a scale from 0 to 4, with affected area scored from 0 to 6. The total PPPASI score ranges from 0 to 72; higher scores indicate more severe disease and a negative change from baseline indicates improvement. | Baseline to Week 16 |
| Number of Participants With TEAEs | TEAEs are new events occurring during or after the first dose or events that worsen after the first dose. Serious AEs include death, life-threatening events, persistent disability/incapacity, congenital anomaly/birth defect, or important medical events requiring intervention to prevent such outcomes. | First dose of any study drug to Week 24 |
| PPPASI 50 | Percentage of participants achieving a 50% reduction or improvement from baseline in PPPASI. | Baseline to Week 16 |
| PPPIGA | Percentage of participants achieving a clear or almost clear Palmoplantar Pustulosis Investigator's Global Assessment score. | Week 16 |
Why PPPASI is a continuous outcome
PPPASI retains information about the magnitude of change rather than reducing each participant to a responder/non-responder classification. A participant improving by a small amount and a participant improving by a large amount can therefore contribute differently to the continuous outcome analysis.
By contrast, PPPASI 50 converts the continuous disease-severity measure into a binary endpoint: whether a participant achieved at least a 50% reduction. PPPIGA is also binary in the posted analysis because the endpoint asks whether a participant achieved a clear or almost clear assessment.
6. Primary Results: PPPASI
The primary efficacy analysis evaluated change from baseline in PPPASI at Week 16 using a general linear mixed model for repeated measures. The analysis population was the intent-to-treat analysis set with available data, and the comparison was placebo versus imsidolimab, with the reported least-squares mean difference defined as Imsidolimab − Placebo.
Week 16 PPPASI: Least-Squares Mean Difference
95% CI: −4.05 to 3.77 · P = 0.944
MMRM; least-squares mean difference = Imsidolimab − Placebo
| Primary efficacy measure | Estimate | 95% CI | P-value |
|---|---|---|---|
| Change from baseline in PPPASI at Week 16 | −0.1 | −4.05 to 3.77 | 0.944 |
The estimated least-squares mean difference was −0.1 PPPASI points, with the contrast defined as imsidolimab minus placebo. Because a negative change in PPPASI represents improvement, a negative treatment difference points numerically toward greater improvement with imsidolimab, but the estimated difference here is very close to zero.
The 95% confidence interval from −4.05 to 3.77 spans both negative and positive values. It therefore includes treatment differences favoring either group as well as a difference close to zero. The interval communicates substantial uncertainty around the estimated treatment contrast rather than establishing that the true difference is exactly zero.
The P-value of 0.944 is a measure of compatibility between the observed data and the statistical hypothesis being tested under the model. It is not a measure of the size of the treatment effect, and it does not mean that there is a 94.4% probability that the treatment is ineffective. Effect size and uncertainty are better represented here by the estimate and its confidence interval.
The result is model-based and applies to the specified MMRM analysis population and covariate structure. Interpretation also depends on the assumptions underlying the repeated-measures model and on how incomplete longitudinal observations contribute information to the analysis.
7. How the MMRM Was Used
The registry describes a general linear mixed model for repeated measures with fixed effects for treatment, history of plaque psoriasis (Yes/No), visit, and treatment by visit interaction, and Baseline PPPASI score as covariate. Least-squares mean difference = Imsidolimab - Placebo, together with baseline PPPASI score as a covariate.
The posted analysis reports the least-squares mean difference for imsidolimab minus placebo at the specified Week 16 comparison.
Why include baseline PPPASI?
Baseline PPPASI is a natural covariate because participants can begin the trial with different levels of disease severity. Covariate adjustment can improve precision by accounting for part of the outcome variation associated with baseline severity. The treatment comparison then represents a model-adjusted difference rather than a simple difference between unadjusted arithmetic means.
Why include visit?
Repeated measurements are collected over time, so disease severity can vary by visit independently of treatment. Including visit allows the model to represent the overall pattern of outcome measurements across visits rather than treating each observation as if it came from an unrelated cross-sectional study.
Why include treatment by visit?
The treatment effect need not be identical at every visit. The treatment-by-visit interaction permits the model to represent different trajectories over time. The Week 16 estimate can therefore be understood in the context of the longitudinal model rather than as an isolated comparison disconnected from earlier measurements.
Why include plaque-psoriasis history?
The registry specifically identifies history of plaque psoriasis as a model covariate. Including it adjusts the treatment comparison for this prespecified characteristic within the fitted model.
8. Secondary Endpoint: PPPASI 50
The registry reports a secondary analysis of the percentage of participants who achieved a 50% reduction or improvement from baseline in PPPASI by Week 16. The analysis used logistic regression with treatment as a fixed effect, history of plaque psoriasis and baseline PPPASI as covariates, and multiple imputation for missing data.
PPPASI 50: Odds Ratio
95% CI: 0.288 to 2.686
Logistic regression with covariate adjustment and multiple imputation
| Secondary endpoint | Effect measure | Estimate | 95% CI |
|---|---|---|---|
| Percentage achieving PPPASI 50 at Week 16 | Odds Ratio | 0.879 | 0.288 to 2.686 |
An odds ratio of 0.879 is the reported adjusted odds contrast for achieving PPPASI 50, based on the logistic-regression analysis. An odds ratio below 1 means the estimated odds of achieving the specified response were lower in the imsidolimab group relative to placebo under the direction of the reported comparison.
The confidence interval, 0.288 to 2.686, is wide and crosses 1. It therefore encompasses substantially lower odds, odds near parity, and higher odds for the imsidolimab group. The estimate should consequently be interpreted together with its considerable uncertainty.
An odds ratio is not a risk ratio or a percentage-point difference. For example, an odds ratio of 0.879 does not mean that the response probability was 12.1% lower. The conversion from odds to probability depends on the underlying response probability.
The analysis also used multiple imputation for missing data. That means the reported estimate reflects the specified missing-data approach rather than simply excluding every participant with an unavailable Week 16 binary outcome.
Why logistic regression was appropriate
PPPASI 50 is a binary response endpoint: each participant either achieved the 50% improvement criterion or did not. Logistic regression is designed for this type of outcome and permits adjustment for baseline PPPASI and history of plaque psoriasis while estimating a treatment odds ratio.
9. Secondary Endpoint: PPPIGA
The second posted secondary efficacy analysis evaluated the percentage of participants achieving a clear or almost clear PPPIGA score at Week 16. The registry again specifies logistic regression with treatment as a fixed effect, history of plaque psoriasis and baseline PPPIGA score as covariates, and multiple imputation for missing data.
PPPIGA at Week 16: Odds Ratio
95% CI: 0.5 to 13.9
Logistic regression with covariate adjustment and multiple imputation
| Secondary endpoint | Effect measure | Estimate | 95% CI |
|---|---|---|---|
| Clear or almost clear PPPIGA at Week 16 | Odds Ratio | 2.6 | 0.5 to 13.9 |
An odds ratio of 2.6 means that the estimated odds of achieving a clear or almost clear PPPIGA score were 2.6 times the corresponding odds under the reported logistic-regression comparison. This is an odds comparison, not a statement that 2.6 times as many participants responded.
The 95% confidence interval of 0.5 to 13.9 is very wide and includes 1. The data therefore permit a broad range of plausible treatment contrasts under the fitted model. The width of the interval is especially important when interpreting the point estimate of 2.6.
The large point estimate should not be read in isolation. A point estimate can appear substantial while remaining statistically imprecise when the number of observations or events is limited. The confidence interval communicates that uncertainty directly.
As with PPPASI 50, multiple imputation was used for missing data. The resulting odds ratio is conditional on the specified imputation and regression framework.
10. Comparing the Efficacy Endpoints
| Endpoint | Model | Effect measure | Estimate | 95% CI |
|---|---|---|---|---|
| Change from baseline in PPPASI at Week 16 | MMRM | LS Mean Difference | −0.1 | −4.05 to 3.77 |
| PPPASI 50 at Week 16 | Logistic regression | Odds Ratio | 0.879 | 0.288 to 2.686 |
| Clear or almost clear PPPIGA at Week 16 | Logistic regression | Odds Ratio | 2.6 | 0.5 to 13.9 |
These three analyses answer related but different questions. PPPASI change retains the continuous disease-severity scale and estimates a model-adjusted difference in mean change. PPPASI 50 asks whether participants reached a particular threshold of improvement. PPPIGA asks whether participants reached a categorical investigator assessment of clear or almost clear disease.
The estimates should therefore not be placed on a common numerical scale. An LS mean difference of −0.1, an odds ratio of 0.879, and an odds ratio of 2.6 represent different statistical quantities. Their confidence intervals likewise describe uncertainty for different parameters.
Continuous endpoint
PPPASI change uses the amount of disease-score change and is analyzed with an MMRM.
Threshold endpoint
PPPASI 50 converts improvement into a binary responder definition and is analyzed with logistic regression.
Global assessment
PPPIGA captures a categorical investigator assessment and is analyzed with logistic regression.
Different estimands
The numerical effect estimates answer different questions and should not be interpreted as interchangeable measures of treatment benefit.
11. Missing Data and Multiple Imputation
The registry explicitly states that multiple imputation was used for missing data in the logistic-regression analyses of PPPASI 50 and PPPIGA. This is an important design feature because a binary endpoint measured at a specific follow-up visit can be missing for some randomized participants.
Multiple imputation replaces missing values with plausible values generated from an imputation model, creating multiple completed datasets. The analysis is then performed across those datasets and combined to account for uncertainty introduced by the missing observations.
The goal is to avoid treating missing observations as if they contained no information while incorporating uncertainty about their unobserved values.
Multiple imputation does not automatically eliminate missing-data bias. Its validity depends on the assumptions and variables used by the imputation model. The registry identifies the use of multiple imputation but does not provide further details here about the imputation model, number of imputations, or missingness assumptions.
12. Statistical Methodology
Mixed-model repeated measures
MMRM is particularly suited to longitudinal clinical-trial outcomes because it uses repeated measurements from the same participants while modeling systematic effects of treatment and visit. The treatment-by-visit interaction allows the treatment contrast to vary across visits.
Least-squares means
Least-squares means, sometimes called estimated marginal means, are model-adjusted means. They represent the expected outcome under the fitted model after accounting for the specified covariates and factors. The reported treatment contrast was calculated as imsidolimab minus placebo.
For PPPASI, a negative value points toward greater improvement in the imsidolimab group because negative change from baseline indicates improvement.
Logistic regression
Logistic regression models the probability of a binary outcome through the log odds. In this trial, it was used for the responder-type endpoints PPPASI 50 and clear or almost clear PPPIGA.
The exponentiated treatment coefficient corresponds to the treatment odds ratio under the fitted model.
Odds ratios
An odds ratio compares odds, where odds are probability divided by one minus probability. An odds ratio of 1 represents equal odds between groups. Values above 1 indicate higher estimated odds in the numerator treatment group, while values below 1 indicate lower estimated odds.
Odds ratios become less intuitive when response probabilities are common rather than rare. They should therefore not be translated directly into relative risks or percentage-point changes without knowing the underlying response probabilities.
Confidence intervals
A two-sided 95% confidence interval describes the uncertainty around an estimated parameter under the statistical model and sampling framework. It is not the range in which 95% of individual participant responses must fall.
P-values
The posted PPPASI analysis has a P-value of 0.944. A P-value evaluates the data relative to the specified statistical hypothesis; it does not quantify the magnitude of the treatment effect, the probability that the null hypothesis is true, or the clinical importance of the result.
13. Statistical Methods Explained
Why was an MMRM used for PPPASI?
PPPASI is measured repeatedly over time, so the analysis benefits from accounting for the longitudinal structure of the data. MMRM allows the model to incorporate treatment, visit, their interaction, baseline disease severity, and plaque-psoriasis history while estimating the treatment contrast at the relevant visit.
What does an LS mean difference of −0.1 mean?
It is the model-adjusted difference in mean change from baseline at Week 16, defined as imsidolimab minus placebo. Because negative PPPASI change represents improvement, the negative direction favors greater improvement with imsidolimab numerically. However, the estimate is close to zero and its 95% CI extends from −4.05 to 3.77.
Why does the PPPASI confidence interval matter?
The interval of −4.05 to 3.77 shows that the estimated treatment difference is imprecise. It contains both negative and positive values, so the point estimate alone does not adequately describe the uncertainty surrounding the treatment comparison.
Why was logistic regression used for PPPASI 50?
PPPASI 50 is a binary responder endpoint. Logistic regression is designed for binary outcomes and permits adjustment for treatment, history of plaque psoriasis, and baseline PPPASI while producing an odds ratio.
What does an odds ratio of 2.6 mean for PPPIGA?
Under the reported model, the estimated odds of achieving a clear or almost clear PPPIGA score were 2.6 times the corresponding odds for the comparison group. It does not mean that 2.6 times as many participants achieved the outcome, and the 95% CI of 0.5 to 13.9 shows substantial uncertainty around the estimate.
Why does multiple imputation matter?
Some binary endpoint observations can be missing at the scheduled assessment. Multiple imputation allows plausible values to be incorporated across several completed datasets rather than relying only on participants with observed values. The validity of the resulting inference depends on the assumptions and specification of the imputation model.
Why should the three efficacy estimates not be combined into one number?
PPPASI change, PPPASI 50, and PPPIGA represent different estimands. The first measures a continuous change in disease severity, while the other two classify participants according to response criteria. Their estimates are therefore complementary rather than interchangeable.
14. Primary Safety Endpoint
The registry defines the number of participants with treatment-emergent adverse events as a primary endpoint measured from the first dose of any study drug through Week 24. TEAEs include new events occurring during or after the first dose and events that worsen after the first dose.
The registry also defines serious adverse events as medical occurrences resulting in death, being life-threatening, causing persistent disability or incapacity, constituting a congenital anomaly or birth defect, or representing an important medical event that may jeopardize the participant or require medical or surgical intervention to prevent such outcomes.
For the posted statistical information, an arm-level estimate or P-value for the overall TEAE primary endpoint is not included. The registry does report serious adverse events by arm.
15. Safety Results
| Safety measure | Placebo | Imsidolimab |
|---|---|---|
| Serious adverse events, affected / at risk | 1/29 | 0/30 |
Placebo
1 participant affected among 29 participants at risk for the reported serious-adverse-event measure.
Imsidolimab
0 participants affected among 30 participants at risk for the reported serious-adverse-event measure.
The serious-adverse-event counts are descriptive arm-level safety information. With only these counts, they should not be treated as a formal comparative safety test. A difference in observed counts can arise from random variation, and a formal comparison would depend on the prespecified safety-analysis framework and the full adverse-event dataset.
16. Understanding the Confidence Intervals
| Endpoint | Estimate | 95% CI | What the interval shows |
|---|---|---|---|
| PPPASI change | −0.1 | −4.05 to 3.77 | Includes both negative and positive treatment differences. |
| PPPASI 50 | 0.879 | 0.288 to 2.686 | Includes odds ratios below and above 1. |
| PPPIGA | 2.6 | 0.5 to 13.9 | Includes odds ratios below and above 1 and is wide relative to the point estimate. |
All three confidence intervals are two-sided 95% intervals. The intervals should be interpreted according to their respective effect measures. For PPPASI, zero is the natural null value because the parameter is a difference. For the odds ratios, one is the natural null value because equal odds correspond to an odds ratio of 1.
Always identify the effect measure before interpreting the confidence interval. Asking whether an interval crosses zero or one is meaningful only after establishing whether the parameter is a difference or a ratio.
17. Multiplicity and Endpoint Hierarchy
The registry identifies two primary endpoints: change from baseline in PPPASI at Week 16 and number of participants with TEAEs through Week 24. It also reports two secondary efficacy analyses: PPPASI 50 and clear or almost clear PPPIGA.
| Endpoint | Role | Statistical approach reported |
|---|---|---|
| Change from baseline in PPPASI | Primary | MMRM |
| Number of participants with TEAEs | Primary | Endpoint reported; no arm-level comparative statistical estimate is included in the posted statistical analysis information. |
| PPPASI 50 | Secondary | Logistic regression with multiple imputation |
| Clear or almost clear PPPIGA | Secondary | Logistic regression with multiple imputation |
The registry does not provide a multiplicity-adjustment strategy for these endpoints in the statistical information reported here. Consequently, the secondary endpoint estimates should be distinguished from the primary PPPASI analysis rather than treated as independent confirmatory tests without additional information about the prespecified error-control framework.
18. What the PPPASI Result Does — and Does Not — Mean
The reported LS mean difference of −0.1 represents the estimated Week 16 difference in change from baseline in PPPASI, with the contrast defined as imsidolimab minus placebo.
Because a negative PPPASI change indicates improvement, the direction of the estimate is numerically toward greater improvement with imsidolimab. However, the estimate is very close to zero, and the 95% CI of −4.05 to 3.77 includes both directions of effect.
The P-value of 0.944 does not tell us that the treatment effect is 94.4% of something, nor does it measure clinical importance. A P-value is a feature of the hypothesis-testing framework. The treatment estimate and confidence interval provide the direct description of the magnitude and precision of the estimated contrast.
PPPASI ranges from 0 to 72, with higher values representing more severe disease. A treatment difference of −0.1 therefore represents a difference of one tenth of a PPPASI point on the reported scale. Whether a particular magnitude of PPPASI change is clinically meaningful is a separate clinical question and is not determined by the P-value alone.
19. Interpreting the Odds Ratios Correctly
The two secondary response analyses use odds ratios rather than mean differences. This distinction is fundamental.
| Odds ratio | Interpretation |
|---|---|
| 1 | Equal odds of the binary outcome between groups. |
| >1 | Higher estimated odds in the numerator treatment group. |
| <1 | Lower estimated odds in the numerator treatment group. |
An odds ratio cannot be read directly as a percentage difference in participants. If the probability of response is low, odds and probabilities can be numerically similar, but the relationship becomes less intuitive as the probability increases.
The POPLAR results illustrate why confidence intervals should accompany odds ratios. The PPPASI 50 estimate of 0.879 has a 95% CI of 0.288 to 2.686, while the PPPIGA estimate of 2.6 has a 95% CI of 0.5 to 13.9. These intervals convey substantially more information about uncertainty than the point estimates alone.
20. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary PPPASI analysis estimated an LS mean difference of −0.1 with a two-sided 95% CI of −4.05 to 3.77 and P = 0.944. The secondary responder analyses produced odds ratios of 0.879 and 2.6, with wide confidence intervals that cross the null value of 1.
Clinical interpretation
The registry results describe treatment effects on disease-severity change and categorical response measures, but the statistical estimates alone do not establish the clinical importance of a particular magnitude of PPPASI change or odds ratio.
This distinction is particularly important for a small phase 2 randomized trial. Statistical precision is influenced by the amount of information available for estimating each parameter. A wide interval should be recognized as uncertainty about the magnitude and direction of the treatment contrast rather than converted into a categorical statement based solely on the point estimate.
21. Important Limitations and Interpretation Issues
- Limited enrollment: the registry reports enrollment of 59 participants. Estimates from a relatively small randomized population can be statistically imprecise, as reflected by the wide confidence intervals for the binary secondary endpoints.
- Wide confidence intervals: the PPPASI 50 interval extends from 0.288 to 2.686, while the PPPIGA interval extends from 0.5 to 13.9. These ranges show substantial uncertainty around the corresponding odds-ratio estimates.
- Different estimands: continuous PPPASI change and binary responder endpoints measure different aspects of outcome and should not be treated as interchangeable.
- Model dependence: the primary PPPASI estimate depends on the specified MMRM, including treatment, visit, treatment-by-visit interaction, plaque-psoriasis history, and baseline PPPASI.
- Missing data: multiple imputation was used for the logistic-regression analyses. Inference from imputation depends on the assumptions and specification of the imputation model.
- Available-data wording: the primary and secondary efficacy analyses are described using the intent-to-treat analysis set with available data, so the interpretation depends on which observations contributed to each analysis.
- Multiplicity: the registry information reported here does not specify a multiplicity-adjustment strategy for the collection of primary and secondary endpoints.
- Safety information: serious adverse-event counts are available by arm, but the posted statistical information does not provide a comparative hypothesis test for the overall TEAE primary endpoint.
- Generalizability: the trial population consists of adults with palmoplantar pustulosis enrolled in a phase 2 randomized study, and the extent to which its results apply to populations outside the study is not determined by the statistical estimates alone.
22. Why This Trial Matters Statistically
POPLAR is a useful teaching example because it places several common clinical-trial methods side by side within the same randomized study. The primary continuous endpoint uses an MMRM, while two secondary binary endpoints use covariate-adjusted logistic regression with multiple imputation.
| Concept | How it appears in POPLAR |
|---|---|
| Randomization | Participants were allocated randomly to two parallel study arms. |
| Double masking | The study used a double-masked design. |
| Repeated measures | PPPASI change was analyzed using an MMRM. |
| Covariate adjustment | Baseline PPPASI and history of plaque psoriasis were included in the primary MMRM. |
| Least-squares mean difference | The primary treatment effect was reported as the imsidolimab minus placebo LS mean difference. |
| Logistic regression | PPPASI 50 and PPPIGA were analyzed as binary outcomes using logistic regression. |
| Odds ratio | The binary secondary endpoints were summarized using odds ratios. |
| Multiple imputation | Missing data were addressed with multiple imputation for the two logistic-regression analyses. |
| Confidence intervals | All posted efficacy estimates include two-sided 95% confidence intervals. |
| Safety analysis | Serious adverse events are reported as affected participants over participants at risk by arm. |
23. Statistical Lessons From the POPLAR Results
Do not confuse a difference with a ratio
The PPPASI analysis estimates a difference, where zero is the null value. The responder analyses estimate odds ratios, where one is the null value.
Read the interval before the point estimate
The width and location of the confidence interval provide essential information about precision and plausible treatment effects.
Model adjustment changes the estimand
The LS mean difference is adjusted for specified covariates and repeated-measures structure rather than being a simple unadjusted difference.
Missing-data methods matter
Multiple imputation is part of the statistical definition of the posted responder analyses, not merely a data-cleaning step.
The trial also demonstrates why the same clinical question can produce different numerical summaries depending on endpoint definition. A continuous score preserves more information about magnitude of change, whereas a responder definition may be easier to interpret clinically but can discard distinctions among participants who fall on the same side of the response threshold.
24. Statistical Methods in Context
The POPLAR analysis can be understood as a sequence of increasingly specific statistical questions:
Did mean disease-severity change differ?
MMRM estimated the adjusted Week 16 difference in PPPASI change between imsidolimab and placebo.
Did the odds of achieving PPPASI 50 differ?
Logistic regression estimated an odds ratio while adjusting for history of plaque psoriasis and baseline PPPASI, with multiple imputation for missing data.
Did the odds of achieving clear or almost clear PPPIGA differ?
A second logistic-regression analysis used history of plaque psoriasis and baseline PPPIGA as covariates and multiple imputation for missing data.
What adverse events were observed?
The registry reports treatment-emergent adverse events as a primary safety endpoint and provides serious-adverse-event counts by randomized arm.
25. A Statistical Reading of the Primary Result
The primary PPPASI estimate is best understood by considering three quantities together: the direction of the estimate, its magnitude, and its uncertainty.
| Component | Reported value | Interpretation |
|---|---|---|
| Direction | −0.1 | Numerically favors greater improvement with imsidolimab because negative PPPASI change represents improvement. |
| Magnitude | −0.1 PPPASI points | The estimated adjusted difference is close to zero on the PPPASI scale. |
| Precision | 95% CI −4.05 to 3.77 | The interval spans both negative and positive treatment differences. |
| Hypothesis-testing result | P = 0.944 | The reported P-value is not a measure of effect size or clinical importance. |
This is a useful example of why a P-value should never be presented without the corresponding effect estimate and confidence interval. A P-value by itself cannot show whether an estimated difference is clinically large, clinically small, or highly uncertain.
26. Sources
- ClinicalTrials.gov: POPLAR (NCT03633396).
Continue through the Clinical Biostats knowledge graph
Explore the statistical methods behind randomized clinical-trial analyses, including repeated-measures models, logistic regression, confidence intervals, odds ratios, and missing-data methods.
27. Record Summary
POPLAR provides a compact example of how different statistical methods address different clinical-trial endpoints. The randomized, double-masked phase 2 design compared imsidolimab with placebo in adults with palmoplantar pustulosis. The primary efficacy endpoint, change from baseline in PPPASI at Week 16, was analyzed using an MMRM incorporating treatment, history of plaque psoriasis, visit, treatment-by-visit interaction, and baseline PPPASI. The reported LS mean difference was −0.1, with a two-sided 95% CI of −4.05 to 3.77 and P = 0.944.
The secondary responder endpoints illustrate a different approach. PPPASI 50 produced an odds ratio of 0.879 with a 95% CI of 0.288 to 2.686, while clear or almost clear PPPIGA produced an odds ratio of 2.6 with a 95% CI of 0.5 to 13.9. Both analyses used logistic regression with covariate adjustment and multiple imputation for missing data.
Statistically, the most important lesson is that these estimates cannot be interpreted in isolation. The LS mean difference, odds ratios, confidence intervals, P-value, model specification, analysis population, and missing-data approach all contribute to the interpretation. The safety information provides a separate perspective, with serious adverse events reported as 1/29 for placebo and 0/30 for imsidolimab.