This page separates reported trial results from statistical interpretation. Numerical results and trial facts on this page are restricted to the ClinicalTrials.gov record. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
INSTAGE was a randomized, double-blind, parallel-group phase 3 trial evaluating nintedanib co-administered with sildenafil in patients with idiopathic pulmonary fibrosis and advanced lung function impairment. The registry reports 274 participants and two treatment groups.
| Feature | INSTAGE |
|---|---|
| Trial name | INSTAGE |
| Phase | Phase 3 |
| Condition | Idiopathic Pulmonary Fibrosis |
| Population | Idiopathic pulmonary fibrosis patients with advanced lung function impairment |
| Design | Randomized, double-blind, parallel-group |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 274 |
| Interventions | Nintedanib, placebo, and sildenafil |
| Primary endpoint | Change From Baseline in St George's Respiratory Questionnaire (SGRQ) Total Score at Week 12 |
| Results posted | Yes |
| ClinicalTrials.gov | NCT02802345 |
| Lead sponsor | Boehringer Ingelheim |
2. Clinical Question
The trial's primary statistical question was whether the treatment groups differed in change from baseline in SGRQ Total Score at Week 12, comparing nintedanib plus placebo with nintedanib plus sildenafil.
Population
Patients with idiopathic pulmonary fibrosis and advanced lung function impairment.
Intervention
Nintedanib co-administered with sildenafil.
Comparator
Nintedanib co-administered with placebo.
Primary question
Does the addition of sildenafil change the SGRQ Total Score from baseline to Week 12 relative to nintedanib plus placebo?
3. Trial Design
Nintedanib + Placebo
- Nintedanib
- Placebo
Nintedanib + Sildenafil
- Nintedanib
- Sildenafil
The registry identifies the allocation as randomized, the design model as parallel, and the masking as double. These design features are important because the treatment comparison is based on randomized assignment while blinding is intended to reduce the influence of treatment knowledge on trial conduct and outcome assessment.
4. Endpoints
| Endpoint | Time frame | Type | Analysis reported |
|---|---|---|---|
| Change From Baseline in St George's Respiratory Questionnaire (SGRQ) Total Score at Week 12 | Baseline and week 12 | Continuous | MMRM |
| Change From Baseline in Dyspnoea Using the University of California San Diego Shortness of Breath Questionnaire (UCSD SOBQ) at Week 12 | Baseline and week 12 | Continuous | MMRM |
| Change From Baseline in SGRQ Total Score at Week 24 | Baseline and week 24 | Continuous | MMRM |
| Change From Baseline in Dyspnoea Using UCSD SOBQ at Week 24 | Baseline and week 24 | Continuous | MMRM |
| Percentage of Patients With On-treatment Serious Adverse Events (SAE) From Baseline to Week 24 | Baseline and week 24 | Binary | Cochran-Mantel-Haenszel |
Primary endpoint definition
The registered primary endpoint was Change From Baseline in St George's Respiratory Questionnaire (SGRQ) Total Score at Week 12, with a time frame of baseline and week 12. The SGRQ is a 50-item questionnaire developed to measure health status and quality of life. Its scores cover Symptoms, Activity, Impacts (Psycho-social), and a total score. Scores range from 0 to 100, with higher scores indicating more limitations. The registry definition states that a minimum change of 4 units was established as clinically relevant after patient and clinician testing.
This endpoint is continuous: rather than classifying each participant simply as having or not having an event, the analysis compares treatment groups on a quantitative change-from-baseline outcome.
5. Statistical Methodology
Primary analysis: Mixed Model for Repeated Measures
The primary endpoint was analyzed using a Mixed Model for Repeated Measures (MMRM). The model included fixed effects for treatment and visit, presence of any echocardiographic signs, baseline SGRQ total score as a covariate, treatment-by-visit interaction, and baseline SGRQ total score-by-visit interaction. An unstructured covariance matrix was used.
The important statistical feature is that the analysis is longitudinal. Rather than treating the Week 12 measurement as an isolated observation, the MMRM framework models repeated outcome information across visits and estimates the treatment comparison while accounting for the covariance among repeated measurements.
The actual registry-described model additionally included the presence of any echocardiographic signs and used an unstructured covariance matrix.
Covariate adjustment
The primary MMRM adjusted for baseline SGRQ total score and the presence of any echocardiographic signs. Baseline adjustment is especially relevant for change-from-baseline outcomes because it can account for differences in starting level while estimating the treatment-group contrast at the analysis visits.
No imputation planned
The registry states that no imputation was planned. This is distinct from saying that missing observations were ignored without an analytical framework. In an MMRM, the model can use available repeated measurements under its underlying missing-data assumptions. The absence of a planned imputation procedure should therefore be reported explicitly rather than silently interpreted as a complete-case analysis.
Week 24 MMRM analyses
The Week 24 SGRQ and UCSD SOBQ analyses used the same general MMRM structure, including treatment, visit, the presence of any echocardiographic signs, baseline score as a covariate, treatment-by-visit and baseline-score-by-visit interactions, and an unstructured covariance matrix. The Week 24 analyses also report use of the Roger-Kenward approximation to estimate denominator degrees of freedom.
Categorical safety analysis
The serious-adverse-event endpoint was analyzed with a Cochran-Mantel-Haenszel approach. The comparison was adjusted for the categorical covariate consisting of the presence of any echocardiographic signs indicative of right heart dysfunction, coded yes/no. The registry reports adjusted Mantel-Haenszel-type risk ratios and risk differences with 95% confidence intervals.
6. Results
Primary endpoint: SGRQ Total Score at Week 12
Adjusted mean difference
95% CI: -3.33 to 2.30 · P = 0.7191
Nintedanib + placebo vs nintedanib + sildenafil
| Primary endpoint | Estimate | 95% CI | P-value |
|---|---|---|---|
| Change From Baseline in SGRQ Total Score at Week 12 | Adjusted mean difference: -0.52 | -3.33 to 2.30 | 0.7191 |
The adjusted mean difference of -0.52 compares the modeled change in SGRQ Total Score between nintedanib plus placebo and nintedanib plus sildenafil. With the comparison defined in that order, the negative estimate indicates a slightly lower modeled change in the nintedanib-plus-placebo group relative to the nintedanib-plus-sildenafil group.
The estimate does not mean that one treatment produced a 0.52-unit improvement for every patient. It is a treatment-group difference in an adjusted mean outcome, estimated from the longitudinal model.
The 95% confidence interval extends from -3.33 to 2.30. Its width indicates that the estimate is not highly precise enough to distinguish the observed small difference from a range of values on either side of zero under the stated model and sampling framework.
The P-value of 0.7191 addresses the compatibility of the observed data with the null hypothesis used for the superiority comparison. It does not measure the size or clinical importance of the treatment effect. A P-value is therefore not a substitute for reporting the estimated difference and its confidence interval.
The SGRQ is a patient-reported continuous outcome, so interpretation also depends on the scale itself. The registry definition notes that a 4-unit change was established as clinically relevant; the estimated between-group difference of -0.52 is substantially smaller in absolute magnitude than that reference value, although the confidence interval also extends beyond it on the negative side.
Secondary endpoint: UCSD SOBQ at Week 12
Adjusted mean difference
95% CI: -7.27 to 1.39 · P = 0.1823
Nintedanib + placebo vs nintedanib + sildenafil
The adjusted mean difference was -2.94, with a 95% confidence interval from -7.27 to 1.39. The interval crosses zero, so the reported estimate is compatible with both a negative and a positive adjusted between-group difference under the analysis framework.
The P-value of 0.1823 is not an effect-size measure. The magnitude and uncertainty of the treatment comparison are conveyed more directly by the estimate and its confidence interval.
As with the primary endpoint, the MMRM adjustment means that this is not simply an arithmetic difference between two raw observed averages. The model accounts for treatment, visit, baseline UCSD SOBQ score, the specified covariates, and treatment-by-visit structure.
Secondary endpoint: SGRQ Total Score at Week 24
Adjusted mean difference
95% CI: -5.40 to 1.02 · P = 0.1809
Nintedanib + placebo vs nintedanib + sildenafil
The Week 24 adjusted mean difference was -2.19. The 95% confidence interval, -5.40 to 1.02, crosses zero and therefore does not establish a directionally positive or negative between-group difference at conventional two-sided significance levels.
The P-value of 0.1809 should be read alongside the confidence interval rather than as a measure of the magnitude of the effect. The estimate itself is modest relative to the SGRQ scale, while the interval reflects uncertainty around that estimate.
Secondary endpoint: UCSD SOBQ at Week 24
Adjusted mean difference
95% CI: -7.39 to 2.58 · P = 0.3421
Nintedanib + placebo vs nintedanib + sildenafil
The Week 24 adjusted mean difference was -2.41, with a 95% confidence interval from -7.39 to 2.58. The interval includes zero and is relatively broad compared with the point estimate.
The P-value of 0.3421 does not quantify how large or small the treatment difference is. The confidence interval is more informative for understanding the precision of the estimated difference and the range of effects that remain compatible with the data under the model.
7. Safety Results
The registry reports serious adverse events using the treated set and provides serious-adverse-event results by randomized treatment group. The treated-set definition includes all patients randomized to a treatment group who received at least one dose of randomized study medication.
| Safety measure | Nintedanib + Placebo | Nintedanib + Sildenafil |
|---|---|---|
| Serious adverse events, affected / at risk | 44 / 136 | 37 / 137 |
| Percentage-ratio estimate | 0.83 (95% CI 0.58–1.20) | |
| Risk-difference estimate | -5.37 percentage points (95% CI -16.25 to 5.52) | |
| Risk-difference P-value | 0.334 | |
The serious-adverse-event comparison was adjusted using the Cochran-Mantel-Haenszel method for the presence of any echocardiographic signs indicative of right heart dysfunction. The registry describes the percentage ratio as a relative-risk comparison and reports adjusted Mantel-Haenszel-type risk ratios and risk differences with 95% confidence intervals.
The reported percentage-ratio estimate of 0.83 indicates that the estimated relative frequency of serious adverse events was lower in the nintedanib-plus-placebo group than in the nintedanib-plus-sildenafil group under the adjusted analysis. A ratio of 1 would represent equal estimated risk, whereas a ratio below 1 represents a lower estimated risk in the first group of the comparison.
The corresponding 95% CI of 0.58 to 1.20 includes 1. The risk-difference estimate of -5.37 percentage points also has a confidence interval spanning zero, from -16.25 to 5.52.
These are not interchangeable measures. The ratio expresses relative difference, while the risk difference expresses an absolute percentage-point difference. Both are useful because a relative measure alone does not show the absolute scale of the event-rate difference.
8. Statistical Methods Explained
Why was an MMRM used?
The SGRQ and UCSD SOBQ endpoints were measured as continuous outcomes at multiple visits. MMRM is designed for this type of repeated-measures structure because it models the correlation among measurements from the same participant rather than treating every observation as statistically independent.
In this trial, the model included treatment and visit effects and a treatment-by-visit interaction. This permits the treatment contrast to be evaluated in the context of the longitudinal pattern rather than reducing the entire dataset to one unadjusted comparison.
Why adjust for baseline score?
Baseline adjustment can improve the statistical comparison when the baseline value is strongly related to the subsequent outcome. Here, baseline SGRQ total score was included as a covariate for the SGRQ analysis, and baseline UCSD SOBQ total score was included for the corresponding UCSD SOBQ analysis.
The resulting treatment estimate is therefore an adjusted mean difference, not merely the difference between two unadjusted group means.
What does the adjusted mean difference of -0.52 mean?
It represents the estimated difference between the two treatment groups in the modeled change from baseline in SGRQ Total Score at Week 12, with the specified covariates and repeated-measures structure taken into account.
It does not mean that every patient experienced a -0.52 change. It also does not establish that the difference is clinically meaningful. The estimate must be interpreted together with the 95% confidence interval and the scale of the SGRQ.
Why does the confidence interval matter?
The 95% confidence interval describes statistical uncertainty around the estimated treatment contrast under the model and sampling framework. For the primary endpoint, the interval was -3.33 to 2.30.
Because the interval crosses zero, the data are compatible with a range of possible treatment differences that includes no between-group difference. The interval should not be interpreted as saying that 95% of individual patients have effects inside that range.
Why does the p-value not measure effect size?
A P-value summarizes evidence against a null hypothesis under the specified statistical model. It depends on both the magnitude of the observed difference and the amount of information available. Consequently, a small P-value does not necessarily imply a large or clinically important effect, while a larger P-value does not prove that the treatment effects are exactly equal.
Why was the Cochran-Mantel-Haenszel test used for serious adverse events?
The serious-adverse-event endpoint is binary: participants either had an on-treatment serious adverse event during the specified period or they did not. The Cochran-Mantel-Haenszel framework permits comparison of treatment groups while adjusting for a categorical stratification variable. In INSTAGE, the analysis adjusted for the presence of any echocardiographic signs indicative of right heart dysfunction.
9. Interpreting the Primary Result Carefully
Estimate
The adjusted mean difference was -0.52. This is the estimated between-group contrast on the SGRQ change-from-baseline scale, not an individual-patient effect.
Precision
The 95% CI was -3.33 to 2.30. The interval spans zero, indicating uncertainty about the direction of the true treatment contrast under the model.
P-value
The two-sided P-value was 0.7191. It addresses statistical evidence under the superiority hypothesis; it is not a measure of effect magnitude.
Clinical scale
The SGRQ ranges from 0 to 100, with higher scores indicating more limitations. The registry notes 4 units as a clinically relevant minimum change.
10. Covariate Adjustment and the Role of Echocardiographic Signs
The MMRM analyses incorporated the presence of any echocardiographic signs as a model covariate. For the serious-adverse-event analysis, the corresponding categorical variable was specifically described as the presence of any echocardiographic signs indicative of right heart dysfunction, coded yes/no.
Adjustment can improve precision and account for prognostically relevant variation, but it does not transform an observational comparison into a randomized comparison. In this trial, randomization supplies the principal basis for the treatment comparison, while covariate adjustment refines the statistical estimation.
This structure was reported for the Week 12 primary SGRQ analysis.
11. Repeated Measures and Missing Data
The registry states that no imputation was planned for the MMRM analyses. This is an important methodological detail because longitudinal clinical-trial data frequently contain observations that are unavailable at some visits.
An MMRM can estimate treatment differences using the repeated observations that are available to the model, rather than replacing missing values with a single deterministic value. That approach still requires assumptions about the missing-data mechanism. The ClinicalTrials.gov record does not provide enough detail to characterize those assumptions beyond the stated absence of planned imputation.
The primary analysis notes also state that data collected after Visit 5, which was planned 12 weeks after the start of study treatment, were not used for the primary Week 12 analysis. This preserves the specified time frame of the primary endpoint rather than extending the primary comparison beyond its registered assessment window.
12. Week 12 Versus Week 24
| Endpoint | Estimate | 95% CI | P-value |
|---|---|---|---|
| SGRQ Total Score at Week 12 | -0.52 | -3.33 to 2.30 | 0.7191 |
| UCSD SOBQ at Week 12 | -2.94 | -7.27 to 1.39 | 0.1823 |
| SGRQ Total Score at Week 24 | -2.19 | -5.40 to 1.02 | 0.1809 |
| UCSD SOBQ at Week 24 | -2.41 | -7.39 to 2.58 | 0.3421 |
All four continuous efficacy comparisons were analyzed with MMRM and expressed as adjusted mean differences. The estimates are all negative when expressed as nintedanib plus placebo versus nintedanib plus sildenafil, but each corresponding confidence interval includes zero.
The Week 24 estimates should not be treated as independent replications of the Week 12 primary endpoint. They arise from the same randomized trial and longitudinal outcome framework, and the registry identifies the Week 12 SGRQ comparison as the primary endpoint.
13. Safety Analysis in Statistical Context
The safety endpoint differs fundamentally from the SGRQ and UCSD SOBQ endpoints. The patient-reported outcomes are continuous measurements, whereas serious adverse events are binary outcomes. Consequently, the effect measures also differ.
| Statistical question | Endpoint type | Reported measure |
|---|---|---|
| How does the mean change differ between groups? | Continuous | Adjusted mean difference |
| How does the relative event frequency differ? | Binary | Percentage ratio / relative risk |
| How does the absolute event frequency differ? | Binary | Risk difference |
This distinction matters because relative and absolute measures can tell different parts of the same statistical story. A risk ratio communicates proportional change, whereas a risk difference directly expresses the absolute separation between event probabilities.
14. Multiplicity and Multiple Endpoints
The registry data posted on ClinicalTrials.gov for INSTAGE identify one registered primary endpoint and four additional reported secondary endpoints, including continuous patient-reported outcomes at Weeks 12 and 24 and serious adverse events through Week 24.
The ClinicalTrials.gov record identifies all of these analyses as superiority comparisons. However, the ClinicalTrials.gov record does not provide a multiplicity-adjustment strategy or an alpha-allocation hierarchy for the secondary endpoints.
15. Blinding and Randomization
INSTAGE was randomized and double-blind. Randomization is important because it provides the foundation for comparing outcomes between treatment assignments without requiring investigators to make treatment assignments based on observed patient characteristics.
Blinding addresses a different source of bias. When participants and relevant trial personnel do not know treatment assignment, expectations about treatment can have less opportunity to influence trial conduct and outcome assessment. For patient-reported outcomes such as SGRQ and UCSD SOBQ, maintaining blinding can be particularly relevant because the outcomes depend on participants' reported health status.
These design features do not eliminate all uncertainty. They instead establish a stronger framework for interpreting the randomized comparison, after which the statistical model determines how the observed longitudinal data are summarized and compared.
16. Cochran-Mantel-Haenszel Analysis Explained
The Cochran-Mantel-Haenszel method is useful when a treatment comparison is to be adjusted for a categorical factor. Instead of calculating a single unadjusted comparison and ignoring the adjustment variable, the method combines information across strata defined by that variable.
In INSTAGE, the serious-adverse-event analysis adjusted for the presence or absence of any echocardiographic signs indicative of right heart dysfunction. The resulting adjusted risk ratio and risk difference therefore incorporate the specified categorical adjustment.
A negative risk difference means that the estimated event probability is lower in the first group under the stated direction of comparison.
The reported risk difference of -5.37 percentage points therefore indicates an estimated lower serious-adverse-event frequency in the nintedanib-plus-placebo group relative to the nintedanib-plus-sildenafil group, while its 95% confidence interval of -16.25 to 5.52 indicates substantial uncertainty around the absolute difference.
17. Trial Timeline
Trial start
INSTAGE began as a randomized, double-blind, parallel-group phase 3 study.
Study treatment and follow-up
The trial evaluated nintedanib plus placebo versus nintedanib plus sildenafil, with registered outcome assessments at Weeks 12 and 24.
Primary completion
The registry records December 19, 2017 as the primary completion date.
Statistical results posted
ClinicalTrials.gov reports five posted outcome measures and six statistical analyses, including one primary-endpoint analysis with an estimate and confidence interval.
18. What the Primary Confidence Interval Tells Us
The primary 95% confidence interval was -3.33 to 2.30. This interval describes uncertainty around the estimated adjusted mean difference of -0.52 under the specified statistical model and sampling framework.
It should not be interpreted as a prediction interval for individual patient responses, nor as a statement that there is a 95% probability that the true effect lies inside the interval.
For a mean difference, 0 represents no between-group difference in the modeled mean change. Because zero lies inside the primary confidence interval, the data do not exclude a null difference at the corresponding two-sided 95% confidence level.
The registry states that a minimum SGRQ change of 4 units was established as clinically relevant. Statistical significance and clinical relevance are different questions: the former concerns compatibility with a null hypothesis, while the latter concerns whether the magnitude of an observed or estimated difference is meaningful on the clinical scale.
19. Primary Versus Secondary Evidence
| Endpoint | Role | Model | Effect measure |
|---|---|---|---|
| SGRQ Total Score, Week 12 | Primary | MMRM | Adjusted mean difference |
| UCSD SOBQ, Week 12 | Secondary | MMRM | Adjusted mean difference |
| SGRQ Total Score, Week 24 | Secondary | MMRM | Adjusted mean difference |
| UCSD SOBQ, Week 24 | Secondary | MMRM | Adjusted mean difference |
| Serious adverse events, Week 24 | Secondary | Cochran-Mantel-Haenszel | Percentage ratio and risk difference |
The hierarchy matters. The Week 12 SGRQ endpoint is the registered primary endpoint, so it is the principal efficacy comparison around which the trial's primary statistical question is framed. The remaining analyses provide additional information but should not be silently promoted to primary status because their P-values happen to be smaller or larger.
20. Limitations
- Single primary endpoint: the ClinicalTrials.gov record identifies one registered primary endpoint, while several secondary endpoints were also analyzed. The ClinicalTrials.gov record does not specify a multiplicity-adjustment strategy for those secondary analyses.
- MMRM assumptions: interpretation of the repeated-measures model depends on its covariance structure and assumptions concerning the observed longitudinal data and missingness.
- No imputation planned: the registry explicitly reports that no imputation was planned. This does not remove the need to consider the assumptions governing incomplete longitudinal observations.
- Confidence intervals cross the null: the reported continuous-outcome confidence intervals all include zero, so the estimates should be interpreted as uncertain treatment contrasts rather than precise directional effects.
- Safety precision: the serious-adverse-event risk-ratio confidence interval extends from 0.58 to 1.20, while the risk-difference interval extends from -16.25 to 5.52 percentage points.
- Analysis populations differ conceptually: the primary analysis used the treated set, and the safety endpoint also used the treated set. The registry definition states that the treated set includes randomized patients who received at least one dose of randomized study medication.
- Registry scope: the ClinicalTrials.gov record does not provide a complete baseline characteristics table, subgroup analysis results, survival endpoints, or a detailed multiplicity plan, so those elements are not interpreted here.
21. Why This Trial Matters Statistically
INSTAGE is a useful teaching example because it combines several important methods for modern clinical-trial analysis without relying on time-to-event endpoints. Its primary outcome is a continuous patient-reported measure, and the trial uses a longitudinal MMRM with baseline covariate adjustment and an unstructured covariance matrix.
| Concept | How it appears in INSTAGE |
|---|---|
| Randomization | Participants were assigned in a randomized parallel-group design. |
| Blinding | The registry identifies the study as double-blind. |
| Continuous endpoint | The primary endpoint measures change in SGRQ Total Score. |
| MMRM | Repeated patient-reported outcomes were analyzed using a mixed model for repeated measures. |
| Covariate adjustment | Baseline score and presence of echocardiographic signs were incorporated into the MMRM. |
| Interaction terms | Treatment-by-visit and baseline-score-by-visit interactions were included. |
| Unstructured covariance | The MMRM used an unstructured covariance matrix. |
| Missing data | No imputation was planned for the MMRM analyses. |
| Confidence intervals | Reported 95% two-sided intervals quantify uncertainty around treatment estimates. |
| P-values | Two-sided P-values accompany the reported treatment comparisons. |
| Categorical analysis | Serious adverse events were analyzed using the Cochran-Mantel-Haenszel framework. |
| Risk difference | The serious-adverse-event analysis reports an adjusted risk difference of -5.37 percentage points. |
22. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
23. Related Statistical Calculators
24. Sources
- ClinicalTrials.gov: INSTAGE, NCT02802345.
- PubMed: PMID 31914963.
- PubMed: PMID 31365829.
- PubMed: PMID 30220235.
Continue through the Clinical Biostats statistical pathway
Use the related tutorials and calculators to explore the longitudinal, categorical, and inference methods represented in INSTAGE.
25. Record Summary
INSTAGE provides a useful example of how a randomized clinical trial with continuous patient-reported outcomes can be analyzed using a longitudinal statistical framework. The primary endpoint was change from baseline in SGRQ Total Score at Week 12, analyzed with an MMRM incorporating treatment, visit, treatment-by-visit interaction, baseline score, baseline-score-by-visit interaction, presence of echocardiographic signs, and an unstructured covariance matrix.
The primary adjusted mean difference was -0.52, with a two-sided 95% confidence interval of -3.33 to 2.30 and a P-value of 0.7191. The secondary continuous outcomes used the same general MMRM framework, while serious adverse events were analyzed using a Cochran-Mantel-Haenszel approach with covariate adjustment. The serious-adverse-event analysis reported both a relative measure and an absolute risk difference.
The most important statistical lesson is that the estimate, confidence interval, P-value, endpoint scale, analysis population, and model specification must be interpreted together. A P-value alone cannot describe treatment effect magnitude, and a nonsignificant superiority test should not be converted into a claim of equivalence or non-inferiority. For this trial, the longitudinal model and its covariate structure are central to understanding what the reported adjusted treatment differences actually represent.