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
NAVIGATOR was a randomized, parallel-group, triple-masked phase 3 trial evaluating whether tezepelumab 210mg every 4 weeks (Q4W) reduced the annual rate of asthma exacerbations compared with placebo over 52 weeks in adults and adolescents with severe uncontrolled asthma.
| Feature | NAVIGATOR |
|---|---|
| Phase | Phase 3 |
| Condition | Asthma (severe uncontrolled; adults and adolescents) |
| Design | Randomized, parallel-group, triple-masked, placebo-controlled |
| Arms | Tezepelumab 210mg Q4W vs placebo |
| Enrollment | 1061 |
| Primary endpoints | Annual asthma exacerbation rate (Full Analysis Set); annual asthma exacerbation rate in participants with baseline eosinophils < 300 cells/uL |
| Primary analysis method | Negative binomial model; effect measure: rate ratio |
| Key secondary endpoints | Change from baseline at Week 52 in pre-BD FEV1, AQLQ(S)+12, ACQ-6 and Asthma Symptom Diary (mixed models) |
| Dates | Start 2017-11-23; primary completion 2020-09-08 |
| ClinicalTrials.gov | NCT03347279 |
| Sponsor | AstraZeneca (industry) |
2. Clinical Question
The central question was whether adding tezepelumab, a biological agent, to a patient's existing asthma management would reduce the frequency of asthma exacerbations over one year compared with placebo, both in the overall trial population and specifically in patients with lower baseline blood eosinophil counts.
Population
Adults and adolescents with severe uncontrolled asthma. A prespecified primary population was the Full Analysis Set; a second primary analysis was restricted to participants with baseline eosinophils < 300 cells/uL.
Intervention
Tezepelumab 210mg every 4 weeks (Q4W).
Comparator
Placebo, administered under triple masking.
Primary question
Does tezepelumab reduce the annual asthma exacerbation rate from randomisation to Week 52, overall and in participants with baseline eosinophils < 300 cells/uL?
The eosinophil-defined population matters for interpretation. Many asthma biologics act on eosinophilic inflammation, and patients with low eosinophil counts are the group in which such therapies have historically been least informative. Designating the < 300 cells/uL subgroup as a primary endpoint, rather than an exploratory subgroup, turns what would otherwise be a post hoc question into a prespecified confirmatory comparison.
3. Trial Design
Tezepelumab 210mg Q4W
- Tezepelumab (biological), 210mg every 4 weeks
- Treatment and follow-up through Study Week 52
- Triple-masked allocation
Placebo
- Placebo comparator
- Treatment and follow-up through Study Week 52
- Triple-masked allocation
The registry does not report the allocation ratio, stratification factors, a baseline characteristics table, or per-arm randomized counts. The per-arm figures above are the numbers at risk used for serious adverse event reporting.
4. Endpoints
Primary endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Annual Asthma Exacerbation Rate in Adult and Adolescent Patients With Uncontrolled Asthma | The annual exacerbation rate is based on unadjudicated exacerbations reported by the investigator in the eCRF. The analysis is based on the primary population (Full Analysis Set). | From randomisation to Study Week 52 |
| Annual Asthma Exacerbation Rate … in Subjects With Baseline Eosinophils < 300 Cells/uL | The annual exacerbation rate is based on unadjudicated exacerbations reported by the investigator in the eCRF. This analysis is based on subjects with baseline eosinophils < 300 cells/uL. | From randomisation to Study Week 52 |
Key secondary endpoints
| Endpoint | Unit | Time frame |
|---|---|---|
| Mean change from baseline at Week 52 in pre-dose/pre-bronchodilator (pre-BD) FEV1 | Litre | From randomisation to Study Week 52 |
| Mean change from baseline at Week 52 in AQLQ(S)+12 total score | Scale of score | From randomisation to Study Week 52 |
| Mean change from baseline at Week 52 in ACQ-6 | Scale of score | From randomisation to Study Week 52 |
| Mean change from baseline at Week 52 in Asthma Symptom Diary | Scale of score | From randomisation to Study Week 52 |
The primary endpoints are count-based rates (events per year), whereas the key secondary endpoints are continuous changes from baseline. These two data types call for different statistical models, which is why the trial uses a negative binomial model for the former and mixed models for the latter.
5. Statistical Methodology
Negative binomial regression for exacerbation counts
Exacerbations are discrete events that can occur more than once in the same patient. The natural outcome is the number of events per patient, divided by the time that patient was observed. A Poisson model would assume that the variance of the count equals its mean; in asthma, a minority of patients typically experience many exacerbations while most experience few or none, producing variance well above the mean (overdispersion). The negative binomial model adds a dispersion parameter that absorbs this extra variability, which yields more honest standard errors and confidence intervals than a Poisson model would.
Var(Yi) = μi + k·μi2 Rate ratio = exp(β1)
Here Yi is the number of exacerbations for patient i, ti is that patient's follow-up time (entered as an offset so that patients with shorter follow-up contribute proportionately), and k is the dispersion parameter. When k = 0 the model reduces to Poisson. The registry identifies the method as negative binomial but does not list the covariates included in the model.
Mixed models for repeated continuous measures
The key secondary endpoints were analysed with mixed models. For outcomes measured at multiple visits, a mixed model for repeated measures uses every available post-baseline observation to estimate the Week 52 treatment difference. The registry specifies that all subjects in the Full Analysis Set with at least one post-baseline change from baseline contribute to the analysis, while the "number analysed" reflects subjects with an observation (or, for the symptom diary, a weekly mean) at Week 52. The effect measure is a least squares (LS) mean difference: a model-adjusted difference between arms rather than a raw difference in observed means.
Under the usual missing-at-random assumption, the mixed model borrows information from earlier visits for patients whose Week 52 value is missing. The estimate therefore depends on that assumption as well as on the covariates in the model.
Analysis population
The primary exacerbation analysis was based on the Full Analysis Set, the registry's primary population. A full analysis set analysed by randomized assignment is the practical implementation of the intention-to-treat principle, which preserves the comparability created by randomization. The second primary endpoint restricts the analysis to participants with baseline eosinophils < 300 cells/uL; because eosinophil count is measured before randomization, this restriction still compares randomized groups.
Hypothesis framework
All posted analyses are superiority comparisons of tezepelumab 210mg Q4W against placebo, reported with two-sided 95% confidence intervals. The registry does not describe how the type I error was divided between the two primary endpoints or across the key secondary endpoints.
6. Primary Endpoint Results: Overall Population
For the primary annual exacerbation rate in the Full Analysis Set, the negative binomial model produced the following rate ratio (tezepelumab 210mg Q4W vs placebo):
Annual asthma exacerbation rate ratio
95% CI: 0.37–0.53 (two-sided) · P < 0.001
Negative binomial model · Full Analysis Set · Randomisation to Study Week 52
What it means. A rate ratio of 0.44 means that the model-estimated annual exacerbation rate in the tezepelumab arm was 44% of the rate in the placebo arm, i.e. a 56% lower estimated exacerbation rate over 52 weeks.
What it does not mean. It does not mean that 56% of patients stopped having exacerbations, nor that each patient's own exacerbation frequency fell by 56%. It is a ratio of group-level rates. A rate ratio also says nothing by itself about the absolute number of exacerbations avoided; that depends on the underlying placebo rate, which is needed to translate the relative effect into events per patient-year.
Precision. The 95% confidence interval of 0.37–0.53 is fairly narrow and lies entirely below 1. Values compatible with the data range from a 63% to a 47% lower rate. Even the least favourable end of the interval represents a substantial relative reduction.
The p-value. P < 0.001 indicates that data this extreme would be very unlikely if the true rate ratio were 1. It does not measure how large the effect is; that information is carried by the estimate and its interval.
Cautions. Exacerbations were unadjudicated and investigator-reported, so event definitions depend on investigator judgement, protected here by masking. The negative binomial model assumes a constant treatment ratio over the 52 weeks and depends on how follow-up time and premature discontinuation are handled. With two primary endpoints, the overall error rate depends on a multiplicity strategy that is not described in the registry.
7. Primary Endpoint Results: Baseline Eosinophils < 300 Cells/uL
Annual asthma exacerbation rate ratio, eosinophils < 300 cells/uL
95% CI: 0.46–0.75 (two-sided) · P < 0.001
Negative binomial model · Randomisation to Study Week 52
What it means. Among participants with baseline eosinophils < 300 cells/uL, the estimated annual exacerbation rate with tezepelumab was 59% of the placebo rate, a 41% lower estimated rate.
What it does not mean. The point estimate is closer to 1 than in the overall population, but that alone does not establish that tezepelumab works less well in low-eosinophil patients. The two populations overlap (the < 300 group is part of the Full Analysis Set), so the two rate ratios are not independent comparisons, and a formal difference in effect would require an interaction test across eosinophil strata, which the registry does not report.
Precision. The interval of 0.46–0.75 is wider than in the overall population, as expected from a smaller analysed group, yet its upper bound remains well below 1.
The p-value. P < 0.001 supports the conclusion that the rate ratio differs from 1 in this subgroup; it does not rank this effect as larger or smaller than the overall effect.
Cautions. Because this was a prespecified primary endpoint rather than an exploratory subgroup, the result carries confirmatory weight, subject to the trial's multiplicity plan. The eosinophil cut-off is a dichotomisation of a continuous biomarker; the result describes the group below the threshold, not a dose–response relationship with eosinophil count.
8. Key Secondary Endpoint Results
Four key secondary endpoints were analysed with mixed models as LS mean differences (tezepelumab 210mg Q4W minus placebo) in change from baseline at Week 52.
| Endpoint (change from baseline at Week 52) | Unit | LS mean difference | 95% CI | P-value |
|---|---|---|---|---|
| Pre-BD FEV1 | Litre | 0.13 | 0.08 to 0.18 | <0.001 |
| AQLQ(S)+12 total score | Scale of score | 0.33 | 0.2 to 0.47 | <0.001 |
| ACQ-6 | Scale of score | -0.33 | -0.46 to -0.2 | <0.001 |
| Asthma Symptom Diary | Scale of score | -0.11 | -0.19 to -0.04 | 0.004 |
Direction matters. For FEV1 and AQLQ(S)+12, higher values are better, so a positive difference favours tezepelumab. For ACQ-6 and the Asthma Symptom Diary, higher scores indicate worse control or more symptoms, so a negative difference favours tezepelumab. All four intervals exclude 0 in the favourable direction.
Statistical versus clinical relevance. The mixed-model differences are modest in absolute terms, for example 0.13 L for FEV1 and 0.33 points on AQLQ(S)+12 and ACQ-6. Whether differences of this size are clinically meaningful depends on established minimal important differences for each instrument, which the registry does not report. A small p-value confirms that a difference is unlikely to be zero; it does not confirm that the difference is important to patients.
Analysed populations. For these endpoints, the number analysed is the number of subjects with an observation at Week 52 (or, for the symptom diary, a weekly mean at Week 52), whereas every Full Analysis Set subject with at least one post-baseline value contributes to the model. The model-based estimate therefore uses more information than a simple Week 52 completers analysis would.
9. Safety: Serious Adverse Events
| Arm | Serious adverse events (affected / at risk) |
|---|---|
| Tezepelumab 210mg Q4W | 52 / 528 |
| Placebo | 73 / 531 |
Fewer participants in the tezepelumab arm than in the placebo arm experienced at least one serious adverse event, from arms of similar size at risk. No formal statistical comparison of serious adverse events is posted. Safety counts of this kind are descriptive: the trial was not powered to test differences in serious adverse event frequency, and a numerical imbalance can reflect the treatment, chance, or the fact that serious exacerbations themselves may be recorded as serious adverse events. The at-risk totals differ slightly from the enrollment figure, as is usual when the safety population is defined by receipt of study treatment.
10. Statistical Methods Explained
Why was a negative binomial model used instead of comparing the proportion of patients with an exacerbation?
Dichotomising into "any exacerbation vs none" discards information: a patient with five exacerbations counts the same as one with a single event. Modelling the count per unit time keeps that information and answers the clinically relevant question of how often exacerbations occur. The negative binomial form handles the overdispersion typical of exacerbation data, where a minority of patients account for many events.
What does a rate ratio of 0.44 mean in practice?
For every 100 exacerbations expected per year in a placebo-treated population, the model estimates about 44 in an otherwise comparable tezepelumab-treated population. It is a relative measure; converting it into exacerbations avoided per patient-year requires the absolute placebo rate.
Why is follow-up time part of the model?
Patients who withdraw early have less time to experience exacerbations. Including log follow-up time as an offset converts raw counts into rates, so that a patient observed for half a year is not treated as if they had the same opportunity for events as one observed for a full year.
Why is the eosinophil < 300 cells/uL result a primary endpoint rather than a subgroup finding?
Subgroup analyses are usually exploratory and prone to false positives. By prespecifying the low-eosinophil population as a primary endpoint, the trial committed in advance to testing it, which gives the result confirmatory standing within the trial's error-control plan. It remains a comparison within a subset, so its interval (0.46–0.75) is wider than the overall result.
Why were mixed models used for FEV1 and questionnaire scores?
These outcomes are continuous and measured repeatedly. A mixed model for repeated measures accounts for the correlation between visits in the same patient and uses all available post-baseline data, which lets patients with a missing Week 52 value still contribute under a missing-at-random assumption. The treatment effect is reported as an LS mean difference at Week 52.
Why can an ACQ-6 difference of -0.33 favour tezepelumab?
On ACQ-6, lower scores indicate better asthma control. A negative LS mean difference in change from baseline means scores fell more in the tezepelumab arm than in the placebo arm. Reading the sign correctly depends on knowing each instrument's direction, not on the sign alone.
11. Limitations
- Unadjudicated events: exacerbations were reported by investigators without central adjudication. Masking protects against differential reporting, but event definitions may vary across sites.
- Relative measures only: the registry analyses report rate ratios and LS mean differences. Absolute exacerbation rates per arm and numbers of events avoided are needed for a full clinical picture.
- Multiplicity: two primary endpoints and four key secondary endpoints were tested. The registry does not describe the testing hierarchy or alpha allocation, so each p-value should be read within the trial's own prespecified plan rather than as an isolated test.
- Model assumptions: the negative binomial model assumes a constant rate ratio over 52 weeks; the mixed models assume data are missing at random. Neither assumption can be fully verified from the registry.
- Biomarker threshold: the < 300 cells/uL cut-off dichotomises a continuous measure and does not describe how the effect varies across the full eosinophil range.
- Clinical meaningfulness: small p-values for secondary endpoints do not indicate whether the differences exceed minimal important differences for FEV1 or the questionnaires.
- Duration: results cover randomisation to Week 52; the registry does not report longer-term follow-up.
- Missing descriptive detail: the registry does not report baseline characteristics, allocation ratio or stratification factors, which limits assessment of balance and generalisability.
12. Why This Trial Matters Statistically
NAVIGATOR is a clear teaching example of how recurrent-event outcomes and continuous patient-reported outcomes are analysed side by side in a modern respiratory trial.
| Concept | How it appears in NAVIGATOR |
|---|---|
| Rate ratio | Primary effect measure for annual exacerbation rates (0.44 and 0.59) |
| Negative binomial regression | Models overdispersed exacerbation counts with follow-up time as exposure |
| Mixed-effects model | Repeated-measures analysis of FEV1, AQLQ(S)+12, ACQ-6 and symptom diary |
| Intention-to-treat analysis | Primary analysis in the Full Analysis Set |
| Confidence intervals | Two-sided 95% intervals quantify precision of every posted estimate |
| P-values | Superiority tests against placebo, distinct from effect size |
| Prespecified biomarker population | Eosinophils < 300 cells/uL as a second primary endpoint |
| Masking | Triple masking protects investigator-reported, unadjudicated event counts |
13. Related Tutorials
Learn more about the methods used in this trial:
14. Related Calculators
15. Sources
- ClinicalTrials.gov: NCT03347279 — Study to Evaluate Tezepelumab in Adults & Adolescents With Severe Uncontrolled Asthma.
- PubMed: PMID 41072729
- PubMed: PMID 40782846
- PubMed: PMID 40285963
- PubMed: PMID 39514041
- PubMed: PMID 39326921
Explore the methods behind the results
Connect this trial's endpoints and models to deeper statistical tutorials and practical calculators.
16. Record Summary
NAVIGATOR compared tezepelumab 210mg Q4W with placebo in 1061 enrolled adults and adolescents with severe uncontrolled asthma. The two prespecified primary analyses, modelled with negative binomial regression, estimated annual exacerbation rate ratios of 0.44 (95% CI 0.37–0.53) overall and 0.59 (95% CI 0.46–0.75) in participants with baseline eosinophils < 300 cells/uL. Mixed-model key secondary endpoints for lung function, quality of life, asthma control and symptoms all favoured tezepelumab. The most useful reading combines the relative rate reductions, their confidence intervals, the direction and clinical scale of each secondary measure, and the design features — masking, prespecified populations and model assumptions — that support causal interpretation.