This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial facts on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
STEP-IPF was a randomized, parallel, triple-masked phase 3 trial with 180 enrolled participants comparing sildenafil with placebo in pulmonary fibrosis and pulmonary hypertension. The registered primary endpoint was a binary measure of improvement in 6-minute walk distance at Week 12.
| Feature | STEP-IPF |
|---|---|
| Trial name | STEP-IPF |
| Brief title | Sildenafil Trial of Exercise Performance in Idiopathic Pulmonary Fibrosis |
| Phase | Phase 3 |
| Status | Completed |
| Therapeutic area | Pulmonology |
| Conditions | Pulmonary Fibrosis; Hypertension, Pulmonary |
| Enrollment | 180 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Triple |
| Primary purpose | Treatment |
| Primary endpoint type | Binary |
| Hypothesis type | Superiority |
| Lead sponsor | Duke University |
| ClinicalTrials.gov | NCT00517933 |
2. Clinical Question
The trial addressed whether sildenafil could improve a prespecified measure of exercise performance compared with placebo in the studied population. The primary question was defined around the proportion of participants achieving at least a 20% improvement in 6-minute walk distance from enrollment to Week 12.
Population
Participants enrolled in a phase 3 trial involving pulmonary fibrosis and pulmonary hypertension. The registry caveat states that findings are limited to patients with advanced IPF.
Intervention
Sildenafil citrate.
Comparator
Placebo.
Primary question
Whether the sildenafil and placebo groups differed in the binary primary endpoint of at least 20% improvement in 6-minute walk distance at Week 12.
3. Trial Design
Sildenafil
- Sildenafil citrate
- Serious adverse events: 13/89
Placebo
- Placebo
- Serious adverse events: 15/91
The ClinicalTrials.gov record identifies the study as randomized, parallel, and triple-masked. It does not provide a randomization ratio, dosing schedule, treatment duration beyond the endpoint time frames, or additional allocation details, so those features are not inferred here.
4. Endpoints
Primary Endpoint
| Endpoint | Registry definition | Time frame | Type |
|---|---|---|---|
| Change in 6-minute Walk Distance From Enrollment to Week 12 (≥ 20% Improvement) | This is a binary score (1 or 0) with 1 being better than 0. All models adjust for baseline values of age, height, sex, white race, and DLCO. Values from models are estimated changes from baseline to 12 weeks (with 95% confidence intervals). The difference estimate is based on sildenafil - placebo. | Measured at Week 12 | Binary |
The primary endpoint is therefore not the raw change in walking distance. It is a binary classification based on whether a participant achieved at least a 20% improvement. That distinction is statistically important: a binary endpoint discards some of the numerical information contained in the underlying continuous walking-distance measurement, but it directly answers the prespecified responder question.
Reported Secondary Endpoints
| Endpoint | Time frame | Outcome type | Reported method |
|---|---|---|---|
| Estimated Change From Baseline to 12 Weeks in 6-minute Walk Distance | Baseline, week 12 | Continuous | Linear regression |
| Change in Dyspnea | Enrollment to 12 weeks (phase I) | Continuous | Mixed-effects model |
| Change in Forced Vital Capacity (FVC) Adjusted Values | Baseline, Week 12 | Continuous | Mixed-effects model |
| Change in Diffusing Capacity of the Lung for Carbon Monoxide (DLCO) Adjusted Values | Baseline, Week 12 | Binary | Mixed-effects model |
| Change in Borg Dyspnea Index (BDI) After 6 Minute Walk Test (Adjusted Values) | Baseline, 12 week | Continuous | Linear regression |
| Change in St. George's Respiratory Questionnaire (Total Score) (Adjusted Values) | Baseline, 12 week | Continuous | Linear regression |
| Change in St. George's Respiratory Questionnaire (Symptoms Score) Adjusted Value | Baseline, 12 weeks | Continuous | Linear regression |
| Change in St. George's Respiratory Questionnaire (Activity Score) Adjusted Value | Baseline, 12 weeks | Continuous | Linear regression |
| Change in St. George's Respiratory Questionnaire (Impacts Score) Adjusted Value | Baseline, 12 weeks | Continuous | Logistic regression |
| Change in ICECAP-O Adjusted Value | Baseline, 12 weeks | Continuous | Linear regression |
| Change in EuroQOL Thermometer (Adjusted Value) | Baseline, 12 weeks | Continuous | Linear regression |
| Change in Short Form Health Survey (SF36) General Health - Adjusted Value | Baseline, 12 weeks | Continuous | Linear regression |
| Change in SF36 Aggregate Physical (Adjusted Value) | Baseline, 12 weeks | Continuous | Linear regression |
| Desaturation During 6-minute Walk Test (6MWT) | Week 12 | Time-to-event | Log-rank test |
5. Statistical Methodology
Primary analysis: chi-squared test
The registry reports a chi-squared test for the primary binary endpoint. The comparison is between sildenafil and placebo, with a superiority hypothesis. The posted P-value is 0.39.
The registry describes the endpoint as a binary score, with 1 representing better than 0. The ClinicalTrials.gov record does not provide the arm-specific responder counts, an effect estimate, or a confidence interval for this primary comparison.
The endpoint definition also states that models adjust for baseline age, height, sex, white race, and DLCO. The posted primary statistical-analysis record itself identifies the method as chi-squared, while the endpoint definition describes model-based adjustment. Those two pieces of registry information should not be silently treated as the same procedure.
Linear regression
Linear regression was used for several continuous outcomes, including estimated change in 6-minute walk distance, Borg Dyspnea Index, St. George's Respiratory Questionnaire scores, ICECAP-O, EuroQOL Thermometer, and SF36 measures. The reported effect measure was generally a mean difference in final values.
Mixed-effects models
Mixed-effects models were reported for change in dyspnea, adjusted FVC, and adjusted DLCO. A mixed-effects framework is useful for longitudinal data because measurements from the same participant are correlated rather than independent. The registry's analysis records provide estimates and confidence intervals for these outcomes but do not provide the full model specification.
Logistic regression
The registry reports logistic regression for the change in the St. George's Respiratory Questionnaire impacts score. Logistic regression is ordinarily used when the modeled outcome is categorical, with coefficients transformed into odds-based measures when appropriate. In this registry record, however, the posted effect measure is described as a mean difference in final values. The analysis should therefore be described according to the registry record rather than replacing its reported effect-measure label with an inferred one.
Log-rank test
The desaturation during 6-minute walk test endpoint was analyzed using a log-rank test in the ITT population. A log-rank comparison is appropriate for time-to-event data because it accounts for the ordering of event times and accommodates right-censored observations.
Intention-to-treat analysis
The registry explicitly identifies the FVC, DLCO, and desaturation analyses as using the ITT population. The ClinicalTrials.gov record does not provide an ITT definition beyond that designation, nor do they provide missing-data or censoring rules in sufficient detail to reconstruct the full analysis population.
6. Results: Primary Endpoint
Change in 6-minute Walk Distance From Enrollment to Week 12 (≥ 20% Improvement)
Chi-squared comparison
Hypothesis: superiority · Sildenafil vs placebo
The registry statistical-analysis record does not report an effect estimate or confidence interval for this primary binary comparison.
The reported P = 0.39 is the result of the registry-posted chi-squared comparison of the primary binary endpoint. It quantifies the statistical evidence against the null comparison specified by that test; it is not a measure of the magnitude of any treatment effect.
The ClinicalTrials.gov record does not provide the arm-specific number or percentage of participants achieving the ≥20% improvement, nor do they provide a risk difference, risk ratio, odds ratio, or confidence interval for the primary endpoint. Consequently, the magnitude and precision of the primary treatment difference cannot be reconstructed from the registry analysis record without adding information from another source.
A P-value of 0.39 should therefore not be translated into a statement such as "there was a 39% chance that the treatment did not work." A P-value does not provide that probability, and it does not describe the size of an effect.
The endpoint definition indicates that baseline age, height, sex, white race, and DLCO were included as adjustment variables in models. The posted primary method is nevertheless identified as a chi-squared test. The registry data do not provide enough detail to determine whether the posted chi-squared analysis and the model-adjusted endpoint description refer to separate analyses or to different layers of the registry's reporting.
7. Secondary Results
The registry posts 14 secondary statistical analyses in the ClinicalTrials.gov record. These analyses use several model families, so their estimates should be interpreted according to the scale and method associated with each endpoint rather than treated as interchangeable measures of one common outcome.
| Secondary endpoint | Method | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|
| Estimated Change From Baseline to 12 Weeks in 6-minute Walk Distance | Linear regression | Mean difference | 16.7 | -3.9 to 37.3 | 0.11 |
| Change in Dyspnea | Mixed-effects model | Slope / regression coefficient | -6.58 | -11.25 to -1.92 | 0.006 |
| Change in Forced Vital Capacity (FVC) Adjusted Values | Mixed-effects model | Mean difference | 0.01 | -0.05 to 0.07 | 0.70 |
| Change in Diffusing Capacity of the Lung for Carbon Monoxide (DLCO) Adjusted Values | Mixed-effects model | Mean difference | 1.6 | 0.1 to 3.0 | 0.04 |
| Change in Borg Dyspnea Index (BDI) After 6 Minute Walk Test (Adjusted Values) | Linear regression | Mean difference | -0.34 | -0.81 to 0.14 | 0.16 |
| Change in St. George's Respiratory Questionnaire (Total Score) (Adjusted Values) | Linear regression | Mean difference | -4.1 | -7.3 to -0.9 | 0.01 |
| Change in St. George's Respiratory Questionnaire (Symptoms Score) Adjusted Value | Linear regression | Mean difference | -5.7 | -10.6 to -0.9 | 0.02 |
| Change in St. George's Respiratory Questionnaire (Activity Score) Adjusted Value | Linear regression | Mean difference | -3.6 | -7.2 to -0.1 | 0.04 |
| Change in St. George's Respiratory Questionnaire (Impacts Score) Adjusted Value | Logistic regression | Mean difference | -3.7 | -7.8 to 0.4 | 0.07 |
| Change in ICECAP-O Adjusted Value | Linear regression | Mean difference | 0.02 | -0.01 to 0.06 | 0.18 |
| Change in EuroQOL Thermometer (Adjusted Value) | Linear regression | Mean difference | 2.3 | -2.8 to 7.3 | 0.37 |
| Change in Short Form Health Survey (SF36) General Health - Adjusted Value | Linear regression | Mean difference | 2.9 | 0.8 to 5.0 | 0.008 |
| Change in SF36 Aggregate Physical (Adjusted Value) | Linear regression | Mean difference | -0.2 | -2.1 to 1.7 | 0.86 |
| Desaturation During 6-minute Walk Test (6MWT) | Log-rank test | Not reported | Not reported | Not reported | 0.41 |
How to read the continuous estimates
For the linear-regression analyses, the registry identifies the effect measure as a mean difference in final values. The registry-reported endpoint definitions describe these as adjusted values where specified. A positive estimate means the reported final-value difference favors the sildenafil direction numerically; a negative estimate means the reported difference is negative on that endpoint's scale. Whether a negative value represents a clinically favorable change depends on the direction and meaning of the particular scale.
The confidence interval is especially important because it describes the precision of the estimated treatment difference. For example, the 6-minute walk distance estimate is 16.7 with a 95% confidence interval of -3.9 to 37.3. The interval includes zero, so the registry result does not provide strong statistical evidence of a nonzero mean difference under that analysis. The interval also communicates that the estimate is not precise enough to reduce the plausible range to a narrow positive effect.
Dyspnea
Change in Dyspnea
95% CI: -11.25 to -1.92 · P = 0.006
Mixed-effects model; effect measure reported as slope / regression coefficient.
The registry reports a negative estimated slope difference for change in dyspnea, with a confidence interval entirely below zero. The numerical direction is therefore clear on the reported scale, but the ClinicalTrials.gov record does not define the direction of clinical improvement for the dyspnea scale itself. The statistical result should consequently be reported as a negative between-group slope estimate rather than translated into an unverified clinical magnitude.
Quality-of-life and respiratory measures
Several St. George's Respiratory Questionnaire and SF36 measures have confidence intervals that exclude zero, including the St. George's total score, symptoms score, activity score, and SF36 general health measure. Other reported measures have intervals that include zero. These are secondary endpoints and were analyzed using multiple statistical procedures, so each result should be kept attached to its specific endpoint and scale.
| Endpoint | Estimate | 95% CI | P-value | Statistical reading |
|---|---|---|---|---|
| St. George's Respiratory Questionnaire — Total Score | -4.1 | -7.3 to -0.9 | 0.01 | Interval excludes zero |
| St. George's Respiratory Questionnaire — Symptoms Score | -5.7 | -10.6 to -0.9 | 0.02 | Interval excludes zero |
| St. George's Respiratory Questionnaire — Activity Score | -3.6 | -7.2 to -0.1 | 0.04 | Interval excludes zero |
| St. George's Respiratory Questionnaire — Impacts Score | -3.7 | -7.8 to 0.4 | 0.07 | Interval includes zero |
| SF36 General Health | 2.9 | 0.8 to 5.0 | 0.008 | Interval excludes zero |
| SF36 Aggregate Physical | -0.2 | -2.1 to 1.7 | 0.86 | Interval includes zero |
8. Desaturation and Time-to-Event Analysis
The registry classifies Desaturation During 6-minute Walk Test (6MWT) as a time-to-event endpoint measured at Week 12. The analysis population is the ITT population, and the reported method is a log-rank test.
Log-rank comparison
Endpoint: Desaturation During 6-minute Walk Test (6MWT)
ITT population · Sildenafil / Sildenafil vs Placebo / Sildenafil
The ClinicalTrials.gov record does not report a hazard ratio, event counts, median time, confidence interval, or Kaplan-Meier estimates for this endpoint. The appropriate interpretation is therefore limited to the reported log-rank P-value and the endpoint definition.
Unlike a simple comparison of proportions at one time point, a time-to-event analysis can use information about when events occur and can accommodate participants whose event time is not observed within the analysis period.
9. Safety
The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk.
| Safety measure | Sildenafil | Placebo |
|---|---|---|
| Serious adverse events | 13/89 | 15/91 |
The registry data do not provide a formal statistical comparison of serious adverse events, confidence intervals, or a P-value. Accordingly, the counts should not be converted into an inferred hypothesis test.
Sildenafil
13 affected participants among 89 at risk for the reported serious-adverse-event measure.
Placebo
15 affected participants among 91 at risk for the reported serious-adverse-event measure.
10. Statistical Methods Explained
Why was a chi-squared test used for the primary endpoint?
The primary endpoint is explicitly defined as binary: a participant receives a score of 1 or 0 according to whether the prespecified improvement criterion is achieved. A chi-squared test is a standard method for comparing categorical outcomes between treatment groups. The registry analysis reports this method and a P-value of 0.39.
What does the primary P-value of 0.39 tell us?
It describes the statistical evidence from the reported chi-squared comparison under its null hypothesis. It does not quantify treatment effect size, probability of treatment benefit, or clinical importance. Because the registry record does not provide the underlying responder counts or an effect estimate, the P-value cannot be converted into an absolute treatment difference from the ClinicalTrials.gov record.
Why analyze 6-minute walk distance both as a binary endpoint and as a continuous change?
The primary endpoint asks a responder question: did the participant reach at least the specified improvement threshold? The secondary analysis instead treats change in walking distance as a continuous measurement. The continuous analysis retains more numerical information, while the binary analysis directly represents the prespecified threshold. They therefore answer related but different statistical questions.
Why use a mixed-effects model for longitudinal outcomes?
Repeated measurements within the same participant are correlated. A mixed-effects model can account for this within-participant dependence while estimating treatment-related differences over the measurement period. In STEP-IPF, the registry reports mixed-effects analyses for dyspnea, FVC, and DLCO.
What does a confidence interval around a mean difference tell us?
A 95% confidence interval describes statistical uncertainty around the estimated mean difference under the specified analysis framework. An interval that includes zero is compatible with no between-group mean difference at that confidence level; an interval entirely above or below zero indicates that zero is outside the reported interval. The interval does not describe the range of individual patient responses.
Why use a log-rank test for desaturation?
The registry classifies desaturation during the 6-minute walk test as a time-to-event endpoint. A log-rank test compares the event-time experience between groups while accounting for the timing of events and censoring. It is therefore conceptually different from a chi-squared comparison of a binary endpoint observed at a single assessment.
11. P-values, Effect Size, and Statistical Precision
STEP-IPF illustrates why a P-value should not be treated as a substitute for an effect estimate. The secondary analyses provide examples where estimates, confidence intervals, and P-values can be read together.
| Measure | What it answers | What it does not answer |
|---|---|---|
| Estimate | What treatment difference was estimated on the specified scale? | How certain is the estimate by itself? |
| 95% confidence interval | How precise is the estimated treatment difference under the model? | What range contains every individual patient's response? |
| P-value | How compatible are the data with the null hypothesis used for the test? | How large or clinically important the treatment effect is |
The estimated mean difference was 16.7, with a 95% confidence interval of -3.9 to 37.3 and P = 0.11. The estimate is positive, but the confidence interval includes zero. The statistical evidence therefore does not establish a nonzero mean difference under this reported analysis. The interval also shows substantial uncertainty around the point estimate.
The reported slope difference was -6.58, with a 95% confidence interval of -11.25 to -1.92 and P = 0.006. The interval lies below zero. The numerical direction of the difference is clear, although the ClinicalTrials.gov record does not provide enough information about the scale's direction to convert the negative sign into an unqualified statement about clinical improvement.
12. Multiplicity and the Secondary Endpoints
The ClinicalTrials.gov record contains one registered primary endpoint and 14 posted secondary statistical analyses. Several secondary outcomes have P-values below conventional thresholds, while others do not. Because many endpoints are examined, the collection of secondary P-values should not automatically be interpreted as though each were an isolated confirmatory test.
The ClinicalTrials.gov record identifies the primary hypothesis as superiority but does not provide a multiplicity-adjustment strategy for the 14 posted secondary analyses. It would therefore be inappropriate to infer a formal familywise-error conclusion from the secondary P-values alone.
Primary endpoint
One registered binary primary endpoint, analyzed using a reported chi-squared test with P = 0.39.
Secondary endpoints
Fourteen posted statistical analyses spanning linear regression, mixed-effects models, logistic regression, and a log-rank test.
This distinction matters because statistical significance is a property of a specified hypothesis test and error-control framework, not simply a property of whether an individual number is below a familiar threshold.
13. Adjustment and Covariates
The primary endpoint definition states that all models adjust for baseline values of age, height, sex, white race, and DLCO. It also states that values from models are estimated changes from baseline to 12 weeks with 95% confidence intervals, and that the difference estimate is based on sildenafil minus placebo.
| Adjustment variable | Registry status |
|---|---|
| Age | Included as a baseline adjustment variable |
| Height | Included as a baseline adjustment variable |
| Sex | Included as a baseline adjustment variable |
| White race | Included as a baseline adjustment variable |
| DLCO | Included as a baseline adjustment variable |
Adjustment can improve precision when baseline covariates explain outcome variation. It does not turn a randomized comparison into an observational comparison; the treatment assignment remains randomized. At the same time, the exact model specification, coding of covariates, interaction terms, and missing-data handling are not reported in the ClinicalTrials.gov record.
14. Analysis Population and Missing Data
The registry explicitly identifies the FVC, DLCO, and desaturation analyses as using the ITT population. The ClinicalTrials.gov record does not provide a complete missing-data strategy, imputation method, withdrawal rules, or detailed censoring algorithm.
This is especially relevant to change-from-baseline analyses because the interpretation of a treatment difference depends not only on the fitted model but also on which observations contribute to the model and what assumptions are made about unavailable measurements.
15. Trial Timeline
Trial start
The STEP-IPF trial began in August 2007 according to the ClinicalTrials.gov record.
Primary endpoint assessment
The registered primary endpoint was measured at Week 12, with the endpoint defined around at least 20% improvement in 6-minute walk distance from enrollment.
Primary completion
The ClinicalTrials.gov record gives May 2009 as the primary completion date.
16. What the Primary Endpoint Does — and Does Not — Mean
The primary endpoint converts the underlying 6-minute walk-distance change into a binary outcome: 1 represents better than 0 under the registry's definition. The analysis therefore focuses on whether participants crossed the prespecified improvement threshold rather than estimating the average numerical change in walking distance.
P = 0.39 does not mean that sildenafil has a 39% probability of being ineffective, nor does it mean that the treatment effect is zero. It is the reported result of the specified chi-squared test. The underlying effect size and confidence interval are not reported in the ClinicalTrials.gov record.
The secondary 6-minute walk analysis reports an estimated mean difference of 16.7 with a 95% CI of -3.9 to 37.3 and P = 0.11. This provides a different view of the same general functional domain, but it is not interchangeable with the primary binary responder endpoint.
17. Important Limitations
- Advanced IPF population: the registry specifically states that findings are limited to patients with advanced IPF.
- Subgroup uncertainty: the registry states that it is unknown whether the treatment effect was driven by particular subgroup(s) of patients.
- Study duration and size: the registry states that the study was too short and enrolled too few participants to assess duration of effect.
- Primary effect size not posted in the ClinicalTrials.gov record: the primary chi-squared P-value is available, but the underlying responder counts and primary effect estimate/confidence interval are not reported.
- Secondary multiplicity: 14 secondary statistical analyses are reported, but the ClinicalTrials.gov record does not identify a multiplicity-adjustment strategy for these analyses.
- Incomplete model specification: the ClinicalTrials.gov record identifies methods and some adjustment variables but does not provide the complete model equations, covariance structure, coding decisions, or missing-data strategy.
- Endpoint-scale interpretation: several measures are reported as numerical differences, but the ClinicalTrials.gov record does not provide enough scale-specific information to establish the clinical meaning of every positive or negative direction.
- Serious adverse events: serious-adverse-event counts are provided by arm, but no formal comparative safety analysis is reported.
18. Why This Trial Matters Statistically
STEP-IPF is a useful teaching case because the same randomized comparison generated several types of statistical outcomes: a binary primary endpoint, continuous change outcomes, longitudinal mixed-effects analyses, a logistic-regression analysis, and a time-to-event analysis.
| Statistical concept | How it appears in STEP-IPF |
|---|---|
| Randomization | Participants were randomized to sildenafil or placebo. |
| Blinding | The registry identifies the study as triple-masked. |
| Binary endpoint | The primary endpoint classifies participants according to ≥20% improvement in 6-minute walk distance. |
| Chi-squared test | Reported for the primary binary endpoint. |
| Linear regression | Used for multiple continuous change outcomes. |
| Mixed-effects model | Used for dyspnea, FVC, and DLCO analyses. |
| Logistic regression | Reported for the St. George's Respiratory Questionnaire impacts score analysis. |
| Log-rank test | Used for desaturation during the 6-minute walk test. |
| ITT analysis | Explicitly identified for FVC, DLCO, and desaturation analyses. |
| Confidence intervals | Reported for the continuous secondary treatment-effect estimates. |
| Multiplicity | Multiple secondary endpoints require cautious interpretation of individual P-values. |
19. Related Tutorials
Learn more about the methods used in this trial:
20. Related Calculators
The corresponding statistical tools can reinforce the concepts used in the reported analyses:
21. Sources
- ClinicalTrials.gov: STEP-IPF, NCT00517933.
- PubMed: PubMed record: PMID 26111071.
- PubMed: PubMed record: PMID 25890798.
- PubMed: PubMed record: PMID 20484178.
Continue through the Clinical Biostats statistical pathway
Connect the trial's endpoints and methods to deeper tutorials and statistical calculators for clinical-trial analysis.
22. Record Summary
STEP-IPF provides a compact example of how one randomized clinical trial can require several statistical frameworks. Its registered primary endpoint is a binary responder outcome analyzed with a chi-squared test, while the posted secondary analyses include linear regression, mixed-effects models, logistic regression, and a log-rank test. The primary comparison reports P = 0.39, but the registry analysis record does not provide the corresponding primary effect estimate or confidence interval. Several secondary analyses do provide estimated differences and 95% confidence intervals, including analyses of 6-minute walk distance, dyspnea, FVC, DLCO, respiratory questionnaire scores, health-status measures, and desaturation.
The most important statistical lesson is to keep each result connected to its endpoint definition, analysis method, effect measure, confidence interval, and analysis population. A P-value without an effect estimate does not establish magnitude; a point estimate without a confidence interval does not communicate precision; and multiple secondary analyses require more cautious interpretation than a single prespecified primary hypothesis.