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
REVISIT was a randomized, single-blind, parallel-group phase 3 trial evaluating the efficacy, safety and tolerability of aztreonam-avibactam (ATM-AVI), given with metronidazole where appropriate, versus meropenem, given with colistin where appropriate, for serious infections due to Gram-negative bacteria. The primary endpoint was clinical cure at the test-of-cure (TOC) visit on Day 28, analyzed in two populations.
| Feature | REVISIT |
|---|---|
| Phase | Phase 3 |
| Conditions | Complicated intra-abdominal infection (cIAI); hospital-acquired pneumonia (HAP); ventilator-associated pneumonia (VAP) |
| Design | Randomized, parallel-group, single-masked, treatment purpose |
| Arms | 2: aztreonam-avibactam ± metronidazole; meropenem ± colistin |
| Enrollment | 422 |
| Primary endpoint | Percentage of participants with clinical cure at the TOC visit (Day 28), in the ITT and CE analysis sets |
| Primary effect measure | Difference in clinical cure rate, with a two-sided 95% CI by the unstratified Miettinen-Nurminen method |
| Status | Completed; results posted |
| ClinicalTrials.gov | NCT03329092 |
| Lead sponsor | Pfizer (industry) |
2. Clinical Question
The trial asked whether a regimen built on aztreonam-avibactam produced clinical cure rates comparable to a meropenem-based regimen in hospitalized adults with serious Gram-negative infections of the abdomen or lungs. The companion agents reflect the clinical settings: metronidazole adds anaerobic coverage in intra-abdominal infection, and colistin was available alongside meropenem.
Population
Participants with serious infections due to Gram-negative bacteria: complicated intra-abdominal infection, hospital-acquired pneumonia or ventilator-associated pneumonia.
Intervention
Aztreonam-avibactam (ATM-AVI), with or without metronidazole (MTZ).
Comparator
Meropenem (MER), with or without colistin (COL; reported by arm as colistimethate).
Primary question
How does the percentage of participants with independently adjudicated clinical cure at the Day 28 TOC visit compare between the two regimens?
3. Trial Design
Aztreonam-avibactam ± metronidazole
- Aztreonam-avibactam (ATM-AVI)
- Metronidazole (MTZ) added where indicated
- Investigational regimen
Meropenem ± colistin
- Meropenem (MER)
- Colistin (COL) / colistimethate added where indicated
- Active comparator regimen
The arm sizes shown are the numbers at risk in the serious adverse event summary (275 and 137, a total of 412), which is smaller than the 422 enrolled. The ITT analysis set, by contrast, includes all randomized participants regardless of receipt of study drug, so the efficacy and safety denominators are not the same population.
4. Analysis Populations
REVISIT is a useful example of how antibacterial trials define several nested or overlapping analysis sets, each answering a slightly different question. The registry defines them as follows.
| Analysis set | Registry definition | Role in REVISIT |
|---|---|---|
| Intent-to-treat (ITT) | All randomized participants regardless of receipt of study drug | Co-primary analysis population |
| Clinically evaluable (CE) | ITT participants who met cIAI or HAP/VAP criteria; received at least 48 hours of study treatment (or <48 hours before discontinuing for an AE); no concomitant antibiotics for baseline pathogens between first dose and TOC except protocol-allowed; no disallowed prior antibiotics; no important protocol deviations; no indeterminate outcome at TOC; no monomicrobial infections due to non-eligible pathogens and not only Gram-positive pathogens | Co-primary analysis population |
| Microbiological ITT (micro-ITT) | Subset of ITT with at least one Gram-negative baseline pathogen from an adequate specimen obtained before study treatment | Secondary analysis |
| Microbiologically evaluable (ME) | Participants included in both the micro-ITT and CE sets, with at least one etiologic pathogen from an adequate baseline culture regardless of susceptibility to study agents | Secondary analysis |
The ITT set protects the comparison created by randomization. The CE set removes participants whose outcome could be distorted by protocol deviations, inadequate exposure or confounding antibiotics, at the cost of excluding people after randomization. Reporting both, and seeing whether they agree, is the conventional approach in antibacterial trials because each set is biased in a different direction when something goes wrong.
5. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary: clinical cure at TOC, ITT set | Improvement in baseline signs and symptoms such that, after study treatment, no further antimicrobial treatment for the index infection (cIAI or HAP/VAP) was required; for cIAI, additionally no unplanned drainage or surgical intervention since the initial procedure. Determined by the Independent Clinical Adjudication Committee. 95% CI based on Jeffreys' method. | TOC visit (Day 28) |
| Primary: clinical cure at TOC, CE set | Same definition and adjudication, analyzed in the CE analysis set. | TOC visit (Day 28) |
| Secondary: clinical cure at TOC, micro-ITT and ME sets | Clinical cure in the microbiologically defined populations | TOC visit (Day 28) |
| Secondary: clinical cure at TOC by type of infection | cIAI and HAP/VAP separately, in the ITT and CE sets | TOC visit (Day 28) |
| Other prespecified: clinical cure at end of treatment (EOT) | Clinical cure in the ITT, micro-ITT, CE and ME sets, and by type of infection | EOT visit (within 24 hours after last infusion) |
Clinical cure is a binary endpoint: each participant is classified as cured or not at a fixed visit. That structure determines the whole analytic approach. Within each arm the natural summary is a proportion with a binomial confidence interval (Jeffreys' interval in REVISIT), and between arms the natural summary is the difference in proportions with a confidence interval designed for two independent binomial samples (Miettinen-Nurminen).
6. Primary Result: Clinical Cure at TOC, ITT Analysis Set
Difference in clinical cure rate (ATM-AVI ± MTZ vs MER ± COL)
95% CI: −6.6 to 12.4 · two-sided, unstratified Miettinen-Nurminen
ITT analysis set: all randomized participants regardless of receipt of study drug
The registry lists the comparison as aztreonam-avibactam ± metronidazole versus meropenem ± colistimethate. Read in that order, the point estimate of 2.7 percentage points indicates a numerically higher clinical cure rate with the aztreonam-avibactam regimen. The registry does not state a formal hypothesis (superiority or non-inferiority) or a p-value for this comparison.
What the estimate means. A difference of 2.7 is measured in percentage points on the absolute scale: among ITT participants, the estimated proportion with adjudicated clinical cure at Day 28 was 2.7 points higher in the aztreonam-avibactam arm than in the meropenem arm. It is an absolute risk difference, not a relative improvement of 2.7%.
What it does not mean. It does not show that aztreonam-avibactam is superior. The 95% CI runs from −6.6 to 12.4 and includes zero, so the data are compatible with a cure rate up to 6.6 points lower and up to 12.4 points higher with aztreonam-avibactam. Equally, the absence of a statistically significant difference is not by itself evidence that the regimens are equivalent.
Precision. The interval spans about 19 percentage points. That width reflects the sample size and the unequal arm sizes; an interval this wide can accommodate differences that clinicians would consider meaningful in either direction.
P-values and effect size. No p-value is posted. Even if one were, it would describe compatibility with a null difference of zero, not the size of the difference. The confidence interval is the more informative summary here because it conveys both direction and magnitude.
Cautions. Without a stated non-inferiority margin, the lower bound of −6.6 cannot be formally judged against a prespecified threshold on this page. The ITT set includes participants who may not have received study drug, which tends to pull differences toward zero; the CE analysis below provides the complementary per-protocol-style view.
7. Primary Result: Clinical Cure at TOC, CE Analysis Set
Difference in clinical cure rate (ATM-AVI ± MTZ vs MER ± COL)
95% CI: −7.0 to 13.2 · two-sided, unstratified Miettinen-Nurminen
CE analysis set: evaluable participants meeting exposure, protocol-adherence and outcome-determinacy criteria
The CE analysis produces the same point estimate as the ITT analysis, 2.7 percentage points, with a slightly wider confidence interval (−7.0 to 13.2). The agreement between the two co-primary populations is the main structural finding of the trial.
What the estimate means. Among participants who met the clinically evaluable criteria, the estimated clinical cure proportion at Day 28 was 2.7 percentage points higher with aztreonam-avibactam.
What it does not mean. The CE set is defined after randomization. Participants are excluded for reasons that can be related to treatment (for example, discontinuation, indeterminate outcome or need for other antibiotics), so the CE comparison no longer carries the full protection of randomization. It answers "what happens among participants who received the treatment as intended," not "what happens when the regimen is assigned."
Precision. The interval (−7.0 to 13.2) is wider than in the ITT set, consistent with the smaller number of participants retained after the evaluability criteria. It again includes zero.
P-values and effect size. No p-value is posted; the interval shows that the observed difference is small relative to the uncertainty around it.
Cautions. Concordance between ITT and CE results is reassuring because the two sets are vulnerable to different biases. It does not remove the need to judge the lower confidence bound against a clinically justified margin, which the registry does not report.
8. Secondary Results at TOC
The secondary endpoints repeat the same difference-in-proportions analysis in the microbiologically defined populations and separately by type of infection. Every analysis used the unstratified Miettinen-Nurminen method for the 95% CI.
| Analysis | Role | Difference in cure rate | 95% CI |
|---|---|---|---|
| ITT set | Primary | 2.7 | −6.6 to 12.4 |
| CE set | Primary | 2.7 | −7.0 to 13.2 |
| Micro-ITT set | Secondary | 0.5 | −10.2 to 12.1 |
| ME set | Secondary | 2.6 | −8.4 to 14.7 |
| ITT set, cIAI | Secondary | 2.4 | −7.4 to 13.0 |
| ITT set, HAP/VAP | Secondary | 4.3 | −15.5 to 23.1 |
| CE set, cIAI | Secondary | 5.6 | −4.0 to 16.6 |
| CE set, HAP/VAP | Secondary | −7.9 | −31.9 to 17.3 |
The vertical line marks a difference of zero. Points to the right indicate a numerically higher cure rate with aztreonam-avibactam. Every interval crosses zero.
Reading the infection-type results
The cIAI analyses closely track the overall results (2.4 in ITT, 5.6 in CE), which suggests that intra-abdominal infections make up much of the overall evidence. The HAP/VAP analyses are far less precise: the ITT interval spans −15.5 to 23.1, and the CE interval spans −31.9 to 17.3. The CE HAP/VAP point estimate is negative (−7.9), while the ITT HAP/VAP estimate is positive (4.3).
9. Clinical Cure at End of Treatment
Clinical cure was also assessed at the end-of-treatment (EOT) visit, within 24 hours after the last infusion, as an other prespecified outcome. EOT captures the response while patients are still close to therapy; TOC, at Day 28, additionally requires that the cure be sustained after antibiotics stop.
| Analysis at EOT | Difference in cure rate | 95% CI |
|---|---|---|
| ITT set | 2.3 | −6.2 to 11.5 |
| Micro-ITT set | −1.2 | −10.7 to 9.4 |
| CE set | 0.3 | −8.3 to 9.9 |
| ME set | 1 | −8.5 to 12.0 |
| ITT set, cIAI | 1.9 | −6.9 to 11.9 |
| ITT set, HAP/VAP | 3.9 | −15.2 to 23.4 |
| CE set, cIAI | 3.1 | −5.2 to 13.2 |
| CE set, HAP/VAP | −10.4 | −32.6 to 14.9 |
The EOT pattern mirrors TOC: point estimates cluster near zero in the overall and cIAI analyses, the HAP/VAP intervals are wide, and the CE HAP/VAP estimate is again negative. Because EOT and TOC are measured on the same participants, their agreement is expected and should not be counted as independent confirmation.
10. Safety: Serious Adverse Events
The registry reports serious adverse events (SAEs) by arm as the number of participants affected over the number at risk.
| Arm | Participants with ≥1 SAE | At risk |
|---|---|---|
| Aztreonam-avibactam ± metronidazole | 53 | 275 |
| Meropenem ± colistimethate | 25 | 137 |
Because the arms are of unequal size, raw counts (53 versus 25) should not be compared directly; the comparison must be made on proportions of the number at risk. The registry does not post a formal statistical comparison of SAE rates, which is typical for safety data: these summaries are descriptive, and trials are rarely powered to detect differences in individual adverse event categories. A difference-in-proportions interval, such as a Miettinen-Nurminen or Newcombe interval, would be the natural way to express uncertainty if one were required.
11. Statistical Methodology
Within-arm proportions: Jeffreys' interval
The registry states that the 95% CI for the clinical cure percentage in each arm was based on Jeffreys' method. The Jeffreys interval is a Bayesian-motivated interval for a single binomial proportion, taken from the quantiles of a Beta posterior under the Jeffreys prior. It has good frequentist coverage and, unlike the simple Wald interval, behaves well when the proportion is near 0 or 1 or the sample is small.
where x is the number of participants cured and n the number analyzed in that arm.
Between-arm comparison: Miettinen-Nurminen interval
Every posted between-arm analysis used the unstratified Miettinen-Nurminen method to construct the two-sided 95% CI for the difference in clinical cure rates. This is a score-based interval: it inverts a score test of H0: p1 − p2 = δ over a grid of δ values, using variances estimated under each constrained null rather than from the observed proportions.
p̃1 and p̃2 are maximum-likelihood estimates constrained so that p̃1 − p̃2 = δ; N = n1 + n2. The 95% CI is the set of δ for which |Z(δ)| ≤ 1.96.
"Unstratified" means that the interval was computed without weighting across strata such as infection type or region. In a trial that mixes cIAI and HAP/VAP participants, a stratified version (for example, with Cochran-Mantel-Haenszel weights) is a common alternative; the registry specifies the unstratified form for all posted analyses.
Analysis sets
Efficacy was analyzed in the ITT and CE sets as co-primary populations, with the micro-ITT and ME sets as secondary populations. Safety was summarized in the participants at risk in each arm (275 and 137).
Adjudicated endpoint
Clinical cure was determined by an Independent Clinical Adjudication Committee. Central adjudication standardizes the application of a clinical-judgment endpoint across sites and, in a trial that is not fully blinded, reduces the risk that knowledge of treatment influences outcome classification.
Hypothesis framework
The registry lists the hypothesis for each posted analysis as "other / not stated" and does not report a non-inferiority margin, p-values or a multiplicity strategy. The posted results are therefore best read as estimates with confidence intervals rather than as formal hypothesis tests.
12. Statistical Methods Explained
Why is the effect reported as a difference in percentages rather than a ratio?
For a binary outcome such as clinical cure, the absolute difference is directly interpretable: a difference of 2.7 means 2.7 more participants cured per 100 treated. Antibacterial trials conventionally use the risk difference because non-inferiority margins in this field are specified on the absolute scale, and because cure rates tend to be high enough that relative measures compress real differences.
Why use the Miettinen-Nurminen interval instead of a simple Wald interval?
The Wald interval (difference ± 1.96 × standard error) uses the observed proportions to estimate variance and can have poor coverage when proportions are extreme or groups are small, as in the HAP/VAP analyses here. The Miettinen-Nurminen score interval re-estimates the variance under each candidate difference, which gives coverage much closer to the nominal 95% and avoids intervals that extend beyond logically possible values.
Why do the ITT and CE analyses both have a difference of 2.7 but different intervals?
The CE set is a subset of the ITT set. Removing non-evaluable participants can leave the difference unchanged while reducing the sample size, which widens the interval (−7.0 to 13.2 versus −6.6 to 12.4). The fact that the point estimate did not move suggests that exclusions were not heavily concentrated in one arm in a way that changed the comparison.
Why are the HAP/VAP intervals so much wider than the cIAI intervals?
The width of a confidence interval for a difference in proportions shrinks roughly with the square root of the sample size in each arm. Intervals of roughly 39 to 49 percentage points in HAP/VAP, compared with about 20 in cIAI, indicate that far fewer participants contributed to the pneumonia analyses. Precision, not a different treatment effect, is the most direct explanation for the unstable point estimates in that group.
If the confidence interval includes zero, does that mean the treatments are equivalent?
No. An interval including zero shows that the trial did not detect a difference; it does not show there is none. Equivalence or non-inferiority requires that the relevant confidence bound lie within a prespecified margin. For the ITT primary analysis, the data are compatible with a cure rate as much as 6.6 points lower with aztreonam-avibactam, and whether that is acceptable depends on a clinically justified margin that the registry does not state.
Why report both Jeffreys and Miettinen-Nurminen intervals?
They answer different questions. The Jeffreys interval describes uncertainty in the cure rate within one arm; the Miettinen-Nurminen interval describes uncertainty in the difference between arms. The between-arm interval cannot be recovered by checking whether two within-arm intervals overlap, a common error in reading trial results.
13. Limitations and Interpretation Issues
- No stated hypothesis or margin: the registry does not report a non-inferiority margin or p-values, so the confidence intervals cannot be formally judged against a prespecified decision threshold from the registry alone.
- Precision: even the primary intervals span roughly 19 to 20 percentage points, which leaves room for clinically relevant differences in either direction.
- Post-randomization exclusions: the CE and ME sets exclude participants on criteria assessed after randomization, which can introduce bias if exclusion is related to treatment.
- Infection-type analyses: the HAP/VAP estimates are imprecise and change sign between analysis sets; no interaction test is reported, so effect modification by infection type cannot be established.
- Multiplicity: sixteen between-arm analyses are posted across populations, infection types and visits, without a reported multiplicity adjustment. Secondary and other prespecified analyses should be read as supportive.
- Masking: single masking leaves some potential for assessment bias, partly mitigated by independent adjudication of the primary endpoint.
- Heterogeneous regimens: both arms allow an optional companion agent (metronidazole or colistin), so each arm represents a treatment strategy rather than a single drug.
- Safety denominators: SAE counts are descriptive, with arms of unequal size, and are not accompanied by a posted statistical comparison.
14. Why This Trial Matters Statistically
REVISIT is a clear teaching case for the analysis of binary endpoints in antibacterial trials, where the design question is usually whether a new agent performs comparably to an established one rather than whether it is better.
| Concept | How it appears in REVISIT |
|---|---|
| Difference in proportions | All efficacy comparisons are absolute differences in clinical cure rates |
| Score-based confidence intervals | Unstratified Miettinen-Nurminen intervals for every between-arm difference |
| Single-proportion intervals | Jeffreys' method for the cure rate within each arm |
| Intention-to-treat | ITT set of all randomized participants as a co-primary population |
| Evaluable-population analysis | CE and ME sets as sensitivity-style complements to ITT |
| Confidence intervals vs p-values | Results posted as estimates and intervals without p-values |
| Subgroup precision | Wide, sign-changing HAP/VAP intervals versus stable cIAI intervals |
| Repeated assessment | EOT and TOC measured on the same participants |
| Independent adjudication | Committee-determined clinical cure in a single-masked trial |
15. Related Tutorials
Learn more about the methods used in this trial:
16. Related Calculators
17. Sources
- ClinicalTrials.gov: NCT03329092: Aztreonam-Avibactam ± Metronidazole Versus Meropenem ± Colistin for Serious Gram-Negative Infections.
- PubMed: PMID 41269522.
- PubMed: PMID 40530972.
- PubMed: PMID 39389071.
- PubMed: PMID 38252170.
Explore more trial statistics
Each trial page connects its endpoints and methods to tutorials and calculators so the statistical reasoning can be followed step by step.
18. Record Summary
REVISIT compared aztreonam-avibactam ± metronidazole with meropenem ± colistin in 422 participants with complicated intra-abdominal infection or hospital-acquired/ventilator-associated pneumonia. In both co-primary analysis sets, the difference in adjudicated clinical cure at Day 28 was 2.7 percentage points (ITT 95% CI −6.6 to 12.4; CE 95% CI −7.0 to 13.2). Secondary and end-of-treatment analyses were broadly consistent, with every interval including zero and markedly lower precision in the HAP/VAP subgroup. The most useful reading combines the absolute differences, the width and location of the Miettinen-Nurminen intervals, the agreement between ITT and evaluable populations, and an explicit recognition that equivalence or non-inferiority must be judged against a clinically justified margin.