← Clinical Trials
Non-CF Bronchiectasis Phase 3 Terminated NCT03460704

PROMIS-II: Complete Statistical Analysis of Colistimethate Sodium in Non-Cystic Fibrosis Bronchiectasis

An independent statistical review of the randomized, quadruple-masked, placebo-controlled phase 3 PROMIS-II trial of colistimethate sodium (CMS) in patients with non-cystic fibrosis bronchiectasis and chronic lung infection, focusing on how a count endpoint — the annual rate of pulmonary exacerbations — was analysed with a negative binomial model.

Start: 2018-01-29  ·  Primary completion: 2022-03-15  ·  Sponsor: Zambon SpA  ·  Status: Terminated
About this analysis

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

PROMIS-II was a randomized, parallel-group, quadruple-masked phase 3 trial comparing colistimethate sodium with placebo in non-cystic fibrosis bronchiectasis (NCFB). Its primary question was whether CMS reduced the frequency of pulmonary exacerbations over 12 months. The registry lists the trial as terminated, with 287 participants enrolled.

287
Enrolled
2 parallel arms
1.004
Exacerbation rate ratio
95% CI 0.747–1.349
0.97889
P-value
Superiority test, primary endpoint
27 / 17
Participants with SAEs
of 152 CMS vs 135 placebo
FeaturePROMIS-II
Registry titleTrial in Non-cystic Fibrosis Bronchiectasis Patients With Chronic Lung Infections Treated With Colistimethate Sodium
PhasePhase 3
ConditionNon-cystic fibrosis bronchiectasis
DesignRandomized, parallel-group, two arms
MaskingQuadruple
Primary purposePrevention
InterventionsCMS (drug); placebo
Enrollment287
Primary endpointMean annual NCFB pulmonary exacerbation rate over 12 months
StatusTerminated; results posted
DatesStart 2018-01-29; primary completion 2022-03-15
SponsorZambon SpA (industry)
ClinicalTrials.govNCT03460704

2. Clinical Question

Bronchiectasis outside cystic fibrosis is characterised by recurrent pulmonary exacerbations, and chronic bacterial infection is thought to drive many of them. PROMIS-II asked whether regular treatment with the antibiotic colistimethate sodium, compared with placebo, would reduce how often exacerbations occur. The endpoint is therefore a count of events per unit of time, not a time to first event or a change in a continuous measurement — a choice that shapes the entire statistical analysis.

Population

Patients with non-cystic fibrosis bronchiectasis and chronic lung infection.

Intervention

Colistimethate sodium (CMS).

Comparator

Placebo, under quadruple masking.

Primary question

Is the mean annual rate of NCFB pulmonary exacerbations over 12 months lower with CMS than with placebo (superiority)?

3. Trial Design

01
Enrol287 participants
02
RandomizeTwo parallel arms, quadruple-masked
03
TreatCMS or placebo
04
FollowExacerbations counted over 12 months
05
AnalyseNegative binomial model, mITT population
ARM 1 · safety set n = 152

Colistimethate sodium (CMS)

  • Active drug: colistimethate sodium
  • Masked to participants, care providers, investigators and outcome assessors (quadruple masking)
  • Exacerbations recorded over the 12-month treatment period
ARM 2 · safety set n = 135

Placebo

  • Placebo
  • Same masking and assessment schedule
  • Exacerbations recorded over the 12-month treatment period

The arm sizes shown are the numbers at risk in the safety analysis set (SAF) reported with the serious adverse event data; together they account for all 287 enrolled participants. The number included in the efficacy analysis is not reported separately in the registry.

Early termination. The registry lists PROMIS-II as terminated. When a trial stops before its planned end, fewer participants or less follow-up time may contribute to the analysis than the design anticipated, which reduces precision. For a rate endpoint analysed with an exposure offset, participants with shorter follow-up still contribute, but proportionally less information.

4. Endpoints

EndpointRegistry definitionTime frame
Mean Annual NCFB Pulmonary Exacerbation Rate (primary)Frequency of pulmonary exacerbations per subject. An exacerbation required all of: (1) at least three of eight symptoms/signs concurrently for at least 24 hours — increased cough; increased sputum volume and/or consistency; increased sputum purulence; new or increased haemoptysis; increased wheezing; increased dyspnoea; increased fatigue/malaise; episodes of fever (temperature ≥38°C); (2) a clinical determination that systemic antibiotic therapy was required and prescribed; and (3) an episode lasting at least 24 hours overall (individual symptoms, such as fever, could be shorter).12 months

This composite definition matters statistically. Requiring symptoms and a prescription of systemic antibiotics makes the event more clinically specific, but it also introduces a clinician-judgement component. Quadruple masking protects that judgement from knowledge of treatment assignment, which is essential for an endpoint that depends partly on a prescribing decision.

The registry reports a formal statistical analysis for this primary endpoint only.

5. Primary Endpoint Results

The primary comparison was CMS versus placebo on the annual rate of NCFB pulmonary exacerbations, analysed in the modified intention-to-treat (mITT) population.

LS mean rate ratio, CMS vs placebo

1.004

95% CI (two-sided): 0.747–1.349   ·   P = 0.97889

Negative binomial model · Superiority hypothesis · mITT population · 12 months

ItemReported value
Outcome measureMean Annual NCFB Pulmonary Exacerbation Rate
UnitNumber of pulmonary exacerbations
Groups comparedCMS (colistimethate sodium) vs placebo
Statistical methodNegative binomial model
Covariates (fixed effects)Treatment; country; baseline use of stable concomitant oral macrolide therapy
OffsetLog exposure time on treatment
Effect measureLS mean rate ratio
Estimate1.004
95% CI (two-sided)0.747 to 1.349
P-value0.97889
HypothesisSuperiority
Analysis populationmITT (Full Analysis Set): all subjects who provided informed consent, were randomised and received at least 1 dose or partial dose of the investigational product

The registry does not report the group-specific adjusted exacerbation rates underlying this ratio, so the absolute number of exacerbations per patient-year in each arm cannot be described from the record.

Clinical Biostats interpretation

What the estimate means. A rate ratio of 1.004 means that, after adjustment for country and baseline oral macrolide use and accounting for each participant's time on treatment, the model-estimated exacerbation rate in the CMS group was essentially identical to that in the placebo group — a point estimate 0.4% higher with CMS. A ratio below 1 would have favoured CMS; this estimate sits almost exactly on the null value of 1.

What it does not mean. It does not mean that CMS has been shown to have no effect at all, nor that no individual participant benefited. A rate ratio is an average multiplicative contrast between groups; it says nothing about the distribution of benefit across individuals, and it is not a proportion of patients who had an exacerbation.

What the confidence interval says. The 95% CI of 0.747 to 1.349 is compatible with anything from roughly a 25% lower exacerbation rate (0.747) to roughly a 35% higher rate (1.349) with CMS. The interval is wide in both directions: the data do not rule out a moderate benefit, and they do not rule out moderate harm. Because this was a superiority trial, the interval should not be read as demonstrating equivalence — no equivalence margin was specified.

Why the p-value is not an effect size. P = 0.97889 says that data this close to “no difference” are entirely unsurprising if the true rate ratio is 1. It does not measure how large or small the true effect is, and a large p-value is not evidence that the null hypothesis is true. The confidence interval, not the p-value, carries the information about plausible effect sizes.

Cautions. The analysis population is a modified ITT set that requires at least one dose, so a small number of randomised but untreated participants may be excluded. The trial was terminated, which can reduce the information available relative to the design. The estimate is model-based: its validity depends on the negative binomial assumptions (a gamma-distributed frailty producing overdispersion) and on the exposure offset correctly scaling expected counts to time on treatment.

6. Safety: Serious Adverse Events

Serious adverse events (SAEs) were summarised in the safety analysis set.

Arm (safety set)Participants with ≥1 SAEAt risk
CMS (colistimethate sodium)27152
Placebo17135

More participants in the CMS arm experienced at least one serious adverse event (27 of 152) than in the placebo arm (17 of 135). No formal statistical comparison of SAE frequencies is posted to the registry. Safety tables in trials are typically descriptive: they are not powered to detect differences, many event categories are examined, and a numerical imbalance of this size can arise by chance. Conversely, the absence of a test does not make the imbalance irrelevant; it is a signal to read alongside the event-level detail in the full safety data.

Reading SAE counts: these figures count participants with at least one SAE, not total SAEs, and they do not account for differing exposure time. Because the efficacy analysis used exposure time as an offset, a fully parallel safety comparison would also consider time on treatment.

7. Statistical Methodology

Why a count model?

The primary outcome is the number of exacerbations each participant experiences during follow-up. Counts are non-negative integers, often skewed, with many participants having zero or one event and a few having several. Treating the count as a normally distributed continuous variable, or dichotomising it into “any exacerbation versus none”, would waste information or misrepresent its variance. A count regression model is the natural choice.

Negative binomial regression

The Poisson model is the simplest count model, but it forces the variance to equal the mean. In exacerbation trials, some patients are inherently more exacerbation-prone than others, so the variance typically exceeds the mean (overdispersion). The negative binomial model adds a dispersion parameter that allows for this extra between-patient variability. Ignoring overdispersion would make standard errors too small and confidence intervals too narrow.

Model structure
log E[Yi] = log(ti) + β0 + β1·CMSi + γ·countryi + δ·macrolidei
Var(Yi) = μi + k·μi2

where Yi is the number of exacerbations for participant i, ti is exposure time on treatment, and k is the dispersion parameter. The rate ratio for treatment is exp(β1).

The exposure offset

Participants did not all contribute the same time on treatment — particularly in a terminated trial. Including log-exposure time as an offset (a term with its coefficient fixed at 1) converts the model from one for raw counts to one for rates: events per unit of time. A participant with shorter exposure is expected to have proportionally fewer events than an otherwise identical participant followed for longer, and the offset builds that directly into the model.

Covariate adjustment

The model included country and baseline use of stable concomitant oral macrolide therapy as fixed effects. Country captures differences in practice patterns and background exacerbation rates across sites. Oral macrolides are themselves used to reduce exacerbations in bronchiectasis, so baseline macrolide use is a strong prognostic factor. Adjusting for prognostic factors in a randomized trial does not remove bias (randomization already addresses that) but improves precision and aligns the analysis with any stratification used in the design. The registry does not state whether these factors were also used as randomization strata.

LS mean rate ratio

The effect is reported as a least-squares (LS) mean rate ratio: the ratio of model-adjusted mean exacerbation rates in the two arms, evaluated at a standardised balance of the covariates. It is the count-data analogue of an adjusted mean difference, expressed on the ratio scale because the model is fitted on the log scale.

Analysis population

Efficacy was analysed in the mITT population (Full Analysis Set): all consenting, randomised participants who received at least one dose or partial dose of the investigational product. This keeps participants in the arm to which they were randomised, preserving the benefit of randomization, while excluding anyone who never started treatment.

Hypothesis framework

The analysis tested superiority with a two-sided 95% confidence interval. The null hypothesis was a rate ratio of 1; demonstrating superiority would have required an interval lying entirely below 1.

8. Statistical Methods Explained

Why was a negative binomial model used instead of a Poisson model?

Exacerbation counts vary much more between patients than a Poisson model allows, because some patients are intrinsically more prone to exacerbations. The negative binomial model accommodates this overdispersion with an extra parameter, giving more honest standard errors. A Poisson model applied to overdispersed data would produce confidence intervals that are too narrow and p-values that are too small.

What does a rate ratio of 1.004 mean?

It is the adjusted exacerbation rate with CMS divided by the adjusted rate with placebo. A value of 1.004 means the two rates were estimated to be almost exactly the same. A ratio below 1 would have indicated a lower estimated rate with CMS. The ratio is a relative measure; without the arm-specific rates, it does not convey how many exacerbations per year either group experienced.

Why is log-exposure time included as an offset?

More time on treatment means more opportunity for exacerbations. The offset scales each participant's expected count by their observed exposure, so the treatment comparison is made on rates rather than raw counts. This allows participants with incomplete follow-up to contribute appropriately instead of being dropped or treated as if they had been followed for the full 12 months.

Why adjust for country and baseline macrolide use?

Both are expected to predict exacerbation frequency. Including strong prognostic variables in the model reduces unexplained variability and tightens the confidence interval around the treatment effect. It does not change what is being estimated in a randomized comparison; it makes the estimate more precise.

Does P = 0.97889 prove that CMS does not work?

No. A non-significant result in a superiority trial means the trial did not demonstrate a difference. The confidence interval of 0.747 to 1.349 still includes reductions in exacerbation rate of around a quarter, as well as increases of around a third. “Absence of evidence is not evidence of absence” applies: showing that a treatment has no meaningful effect requires an equivalence or non-inferiority design with a prespecified margin.

What is the difference between mITT and ITT?

A strict intention-to-treat analysis includes every randomised participant. The mITT population used here adds the requirement of at least one dose or partial dose. The difference is usually small in a masked, placebo-controlled trial, because the decision not to start treatment cannot depend on knowledge of the assigned arm, but it is a departure from strict ITT that readers should note.

9. Limitations

10. Why This Trial Matters Statistically

PROMIS-II is a clear teaching example of how exacerbation-rate endpoints are handled, and of how to read a null result correctly.

ConceptHow it appears in PROMIS-II
Count endpointNumber of pulmonary exacerbations over 12 months
Negative binomial regressionPrimary analysis model allowing for overdispersion
OffsetLog exposure time on treatment converts counts to rates
Rate ratioLS mean rate ratio 1.004 as the treatment effect
Covariate adjustmentCountry and baseline oral macrolide use as fixed effects
Confidence interval0.747 to 1.349: wide, spanning benefit and harm
P-value0.97889: no evidence of a difference, but not proof of no effect
Modified ITTRandomised participants who received at least one dose
Composite event definitionSymptom criteria plus antibiotic prescription plus minimum duration
Early terminationAffects information and precision of the final estimate

Statistical interpretation

The primary superiority test did not show a difference in exacerbation rate between CMS and placebo. The point estimate lies essentially on 1, and the confidence interval is too wide to exclude modest benefit or modest harm.

Clinical interpretation

The trial provides no evidence from its primary analysis that CMS reduced exacerbation frequency over 12 months in this population. Serious adverse events were numerically more common with CMS, a finding that is descriptive rather than tested.

11. Related Tutorials

Learn more about the methods used in this trial:

12. Related Calculators

13. Sources

Continue learning with Clinical Biostats

Connect the endpoints and methods from this trial to in-depth statistical tutorials and hands-on calculators.

14. Record Summary

PROMIS-II compared colistimethate sodium with placebo in non-cystic fibrosis bronchiectasis using a count endpoint — pulmonary exacerbations over 12 months — analysed with a covariate-adjusted negative binomial model with an exposure offset. The LS mean rate ratio of 1.004 (95% CI 0.747–1.349; P = 0.97889) shows no demonstrated difference in exacerbation rate, with an interval wide enough to leave modest benefit and modest harm both compatible with the data. Serious adverse events occurred in 27 of 152 participants on CMS and 17 of 135 on placebo. The most useful reading of this trial combines the relative effect, its uncertainty interval, the analysis population and the design features, including early termination, that shape how much the result can tell us.

Clinical Biostats methodology: A trial-results page should do more than repeat headline numbers. The goal is to set out the statistical story of the trial in a standardized format while clearly separating reported results from educational interpretation.