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 and posted statistical results on this page are limited to the registry-reported ATLAS registry data.
1. Trial at a Glance
ATLAS was a randomized, parallel-group, open-label phase 3 study in adults with HIV-1 infection. The posted primary analysis evaluated whether the percentage of participants with plasma HIV-1 RNA ≥50 copies/mL at Week 48 with cabotegravir plus rilpivirine administered every 8 weeks was non-inferior to the every-4-weeks regimen.
| Feature | ATLAS |
|---|---|
| Trial name | ATLAS |
| ClinicalTrials.gov identifier | NCT03299049 |
| Therapeutic area | Infectious Disease |
| Condition | HIV Infections |
| Phase | Phase 3 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1049.0 |
| Arms | 4 |
| Lead sponsor | ViiV Healthcare |
| Sponsor type | Industry |
| Status | Active, not recruiting |
2. Clinical Question
The primary statistical question was whether the every-8-weeks regimen of long-acting cabotegravir plus long-acting rilpivirine maintained antiviral activity that was non-inferior to the every-4-weeks regimen over 48 weeks.
Population
HIV-1 infected adults enrolled in the ATLAS phase 3 study.
Intervention comparison
CAB LA + RPV LA administered every 8 weeks compared with CAB LA + RPV LA administered every 4 weeks.
Primary outcome
Percentage of participants with plasma HIV-1 RNA ≥50 copies/mL according to the FDA Snapshot Algorithm at Week 48.
Primary question
Is the Q8W regimen non-inferior to the Q4W regimen with respect to the Week 48 virologic endpoint?
3. Trial Design
The registry identifies four interventions: cabotegravir tablets, rilpivirine tablets, cabotegravir injectable suspension, and rilpivirine injectable suspension. The posted statistical analyses, however, define the primary comparison specifically as CAB LA + RPV LA Q8W versus CAB LA + RPV LA Q4W.
4. Endpoints
| Endpoint | Registry definition / time frame | Type |
|---|---|---|
| Primary | Percentage of Participants With Plasma Human Immunodeficiency Virus-ribonucleic Acid (HIV-RNA) ≥50 Copies Per Milliliter (c/mL) as Per Food and Drug Administration (FDA) Snapshot Algorithm at Week 48 | Binary |
| Secondary | Percentage of Participants With Plasma HIV-1 RNA <50 c/mL Using FDA Snapshot Algorithm at Week 48 | Binary |
Primary endpoint definition
The registered primary endpoint measures the percentage of participants with HIV-1 RNA ≥50 c/mL according to the FDA Snapshot Algorithm at Week 48. The registry definition states that this endpoint was assessed to demonstrate non-inferior antiviral activity of CAB LA + RPV LA Q8W compared with CAB LA + RPV LA Q4W over 48 weeks.
Secondary endpoint
The posted secondary endpoint reverses the direction of the virologic threshold: it measures the percentage of participants with plasma HIV-1 RNA <50 c/mL using the FDA Snapshot Algorithm at Week 48. This is a binary endpoint analyzed using the same broad categorical-data framework.
5. Statistical Methodology
Cochran-Mantel-Haenszel analysis
The posted primary and secondary analyses used the Cochran-Mantel-Haenszel (CMH) test. The CMH framework is designed for categorical outcomes and can produce an adjusted comparison when data are considered across strata. In the ClinicalTrials.gov record, the effect measure is an adjusted difference between the Q8W and Q4W groups.
The posted estimate is expressed in percentage-point units because the underlying outcome is the percentage of participants meeting a binary virologic definition.
Intention-to-treat analysis
The primary and secondary posted analyses use the ITT-E Population. The intention-to-treat principle analyzes participants according to their randomized treatment assignment. This preserves the treatment comparison created by randomization and avoids redefining the comparison based solely on treatment exposure or subsequent treatment behavior.
Confidence interval
The primary analysis reports a two-sided 95% confidence interval for the CMH-adjusted difference. Confidence intervals are particularly important in a non-inferiority trial because the direction and width of the interval determine whether the observed treatment difference is compatible with the prespecified non-inferiority criterion.
The point estimate is close to zero, while the confidence interval spans both a small negative difference and a small positive difference.
Non-inferiority logic
Non-inferiority does not require the estimated treatment difference to be exactly zero. Instead, the question is whether the data exclude a difference large enough to cross the prespecified non-inferiority boundary.
For the primary endpoint, the registry states that non-inferiority was concluded if the upper bound of the two-sided 95% CI for the CMH-adjusted treatment difference (Q8W minus Q4W) was less than 4%.
Primary non-inferiority comparison
Adjusted difference, Q8W − Q4W
95% CI: −0.6 to 2.2
Registry criterion: upper bound < 4%
The upper confidence limit is 2.2, which is below the stated 4% boundary. Thus, using the registry's posted decision rule, the observed confidence interval satisfies the stated non-inferiority criterion.
The estimate of 0.8 means that the estimated percentage of participants meeting the HIV-RNA ≥50 c/mL endpoint was 0.8 percentage points higher in the Q8W group than in the Q4W group, after the CMH adjustment described in the registry analysis.
That does not mean that the two regimens have identical antiviral activity, nor does it mean that every individual participant had the same outcome. The estimate is a population-level comparison of a binary endpoint.
The 95% CI of −0.6 to 2.2 describes uncertainty around the estimated treatment difference. It includes zero, so the data are compatible with a small difference in either direction. More importantly for the non-inferiority question, the entire upper side of the interval remains below the stated 4% margin.
The confidence interval is therefore doing something different from a p-value: it describes the range of treatment differences compatible with the statistical model and data under the specified confidence procedure. A p-value, when reported, addresses evidence against a null hypothesis; it is not a measure of effect size or clinical importance.
The registry analysis does not report a p-value for this primary non-inferiority comparison. The non-inferiority conclusion is instead expressed through the confidence-interval boundary.
6. Primary Results
HIV-1 RNA ≥50 c/mL at Week 48
The primary endpoint was the percentage of participants with plasma HIV-1 RNA ≥50 c/mL according to the FDA Snapshot Algorithm at Week 48. The analysis population was the ITT-E population, and the treatment groups compared were CAB LA + RPV LA Q8W versus CAB LA + RPV LA Q4W.
| Primary endpoint result | Posted analysis |
|---|---|
| Analysis population | ITT-E Population |
| Groups compared | CAB LA + RPV LA Q8W vs CAB LA + RPV LA Q4W |
| Method | Cochran-Mantel-Haenszel test |
| Effect measure | Adjusted difference |
| Direction | Q8W − Q4W |
| Estimate | 0.8 |
| 95% CI | −0.6 to 2.2 |
| CI type | Two-sided |
| Hypothesis | Non-inferiority |
| Non-inferiority boundary | Upper bound of the two-sided 95% CI less than 4% |
The primary estimate is 0.8 percentage points for Q8W minus Q4W. Because the endpoint represents the percentage of participants with HIV-1 RNA ≥50 c/mL, a positive difference means the observed percentage was higher in the Q8W group, while a negative difference would mean it was higher in the Q4W group.
The confidence interval extends from −0.6 to 2.2. That interval is relatively close to zero and, crucially for the registered non-inferiority framework, its upper bound remains below 4%.
The result therefore should not be summarized as proof that the two regimens are mathematically identical. Non-inferiority establishes compatibility with the prespecified boundary; it does not establish equivalence unless an equivalence design and corresponding two-sided margins were specified.
There is also no p-value reported for this primary analysis in the ClinicalTrials.gov record. Consequently, the statistical interpretation should remain centered on the adjusted difference and its confidence interval rather than introducing an unreported significance test.
7. Secondary Endpoint Results
HIV-1 RNA <50 c/mL at Week 48
The posted secondary endpoint was the percentage of participants with plasma HIV-1 RNA <50 c/mL using the FDA Snapshot Algorithm at Week 48. It was analyzed in the ITT-E population with the same Q8W versus Q4W comparison and a CMH-adjusted difference.
Secondary endpoint
Adjusted difference, Q8W − Q4W
95% CI: −2.1 to 3.7
Two-sided 95% confidence interval
| Secondary endpoint result | Posted analysis |
|---|---|
| Analysis population | ITT-E Population |
| Groups compared | CAB LA + RPV LA Q8W vs CAB LA + RPV LA Q4W |
| Method | Cochran-Mantel-Haenszel test |
| Effect measure | Adjusted difference |
| Direction | Q8W − Q4W |
| Estimate | 0.8 |
| 95% CI | −2.1 to 3.7 |
| CI type | Two-sided |
| Hypothesis | Non-inferiority |
| Registry-stated criterion | Upper bound of the two-sided 95% CI for the CMH adjusted treatment difference is greater than −10% |
The estimate of 0.8 indicates an adjusted difference of 0.8 percentage points in the proportion with HIV-1 RNA <50 c/mL, calculated as Q8W minus Q4W.
The 95% CI of −2.1 to 3.7 includes zero. Thus, the point estimate alone should not be interpreted as evidence that the Q8W regimen produces a higher or lower proportion of participants below the virologic threshold.
The ClinicalTrials.gov record states the non-inferiority criterion as requiring the upper bound to be greater than −10%. The reported upper bound is 3.7, which satisfies that criterion as it is written in the registry data.
Because this criterion is unusual in direction relative to the more familiar lower-bound formulation of a non-inferiority margin, the safest interpretation is to preserve the registry wording rather than silently rewrite it. The numerical result should therefore be read together with the exact registered decision rule.
8. Serious Adverse Events
The ClinicalTrials.gov record reports serious adverse events by arm for the two long-acting treatment regimens used in the statistical comparison.
| Treatment group | Participants with serious AEs / at risk |
|---|---|
| CAB LA + RPV LA Q8W | 27/522 |
| CAB LA + RPV LA Q4W | 19/523 |
These figures are presented as reported counts over the corresponding numbers at risk. They should not be treated as an efficacy endpoint or combined with the virologic analysis into a single benefit-risk statistic.
9. Statistical Methods Explained
Why was a Cochran-Mantel-Haenszel test used?
The primary and secondary endpoints are binary: each participant either meets the registered virologic definition or does not. The Cochran-Mantel-Haenszel framework is designed for categorical outcomes and provides an adjusted comparison across strata. In this trial's posted analysis, the resulting treatment effect is expressed as an adjusted difference.
What does an adjusted difference of 0.8 mean?
The treatment difference is calculated as Q8W minus Q4W. For the primary endpoint, an estimate of 0.8 means the adjusted percentage of participants with HIV-1 RNA ≥50 c/mL was 0.8 percentage points higher in Q8W than Q4W. For the secondary endpoint, the same numerical difference is applied to the percentage with HIV-1 RNA <50 c/mL.
The direction matters: the same number can describe different clinical outcomes depending on which binary endpoint is being analyzed.
Why is the confidence interval central to non-inferiority?
In a non-inferiority trial, the important question is not simply whether the observed estimate differs from zero. The question is whether the uncertainty interval crosses the prespecified amount of clinically unacceptable difference.
For the primary endpoint, the registry states that the upper limit of the two-sided 95% CI must be less than 4%. The observed interval ends at 2.2, so it remains inside that stated boundary.
Does a confidence interval containing zero mean the trial failed?
No. A confidence interval containing zero means that a zero treatment difference remains compatible with the data under the confidence procedure. In a non-inferiority framework, however, the decision criterion is based on the relationship between the confidence interval and the non-inferiority margin, not simply on whether zero lies inside the interval.
For the primary endpoint, the interval from −0.6 to 2.2 contains zero but has an upper bound below 4%. Those are two separate statistical facts.
Why is ITT-E important?
The posted efficacy analyses use the ITT-E population. An intention-to-treat analysis maintains participants within the treatment group to which they were randomized. This preserves the randomized comparison and avoids selectively redefining treatment groups after randomization.
Does the primary result prove equivalence?
No. The registry-reported analysis is explicitly described as a non-inferiority analysis. Non-inferiority and equivalence are related but distinct statistical objectives. A non-inferiority conclusion addresses whether the new regimen remains within an allowed loss of efficacy; an equivalence design generally requires demonstrating that the treatment difference remains within both a lower and an upper margin.
Why should the reported p-value not be reconstructed?
The registry-reported primary analysis does not report a p-value. Reconstructing one from the estimate and confidence interval would introduce an analysis that is not actually posted in the ClinicalTrials.gov record. The appropriate approach is to report the confidence interval and the registered non-inferiority rule directly.
10. Understanding the Primary Confidence Interval
| Quantity | Primary analysis | Interpretation |
|---|---|---|
| Point estimate | 0.8 | Adjusted Q8W minus Q4W difference |
| Lower 95% CI | −0.6 | Compatible with a small negative Q8W minus Q4W difference |
| Upper 95% CI | 2.2 | Largest end of the reported uncertainty interval |
| NI boundary | 4% | Registry-stated upper-bound criterion |
The interval −0.6 to 2.2 is more informative than the point estimate alone. It indicates that the estimated Q8W-minus-Q4W difference is near zero, while allowing for modest uncertainty in either direction.
The interval does not mean that 95% of individual patients will experience treatment differences between −0.6 and 2.2. It is an uncertainty interval for the population-level treatment-effect estimate.
The interval also does not establish that the treatments are identical. The non-inferiority conclusion depends on the prespecified margin and the position of the confidence interval relative to that margin.
A p-value, when available, summarizes evidence against a specified null hypothesis. It does not tell the reader how large or clinically important an effect is. In this ATLAS analysis, the registry provides the treatment difference and its confidence interval rather than a primary p-value, making the confidence-interval-based non-inferiority framework the appropriate focus.
11. Non-Inferiority Design in ATLAS
The primary hypothesis type is explicitly recorded as non-inferiority. The comparison therefore asks whether the Q8W regimen remains sufficiently close to the Q4W regimen on the registered virologic endpoint, according to the prespecified boundary.
What non-inferiority asks
Whether the new treatment regimen has not lost more efficacy than the amount permitted by the prespecified non-inferiority margin.
What non-inferiority does not ask
It does not by itself establish that the two regimens have exactly the same effect or that one regimen is superior.
Primary boundary
The registry states that the upper bound of the two-sided 95% CI must be less than 4%.
Observed upper bound
The primary 95% CI has an upper bound of 2.2, which is below the stated 4% boundary.
The distinction between the estimate and the decision rule is fundamental. The estimate of 0.8 is the observed adjusted difference. The confidence interval describes uncertainty around that estimate. The non-inferiority boundary provides the threshold against which the upper confidence limit is evaluated.
12. Intention-to-Treat Analysis
The posted analyses use the ITT-E Population. This is important because the analysis population is part of the definition of the statistical comparison. A treatment effect calculated in an ITT population answers a different question from an effect calculated only among participants who completed treatment exactly as planned.
The central purpose is to preserve the treatment comparison generated by randomization.
For non-inferiority studies, analysis-population considerations can be particularly important because deviations from assigned treatment can affect the ability to distinguish a truly similar treatment effect from a difference that has been obscured by study conduct. The registry-reported ATLAS data identify the ITT-E population for the posted analyses but do not provide additional per-protocol or treatment-adherence analyses. Those analyses are therefore not added here.
13. Categorical Endpoint Interpretation
Both posted statistical analyses use binary virologic endpoints. This means that the underlying data can be represented as a two-category outcome for each participant.
| Endpoint | Binary definition | Statistical comparison |
|---|---|---|
| Primary | HIV-1 RNA ≥50 c/mL at Week 48 | CMH adjusted difference, Q8W − Q4W |
| Secondary | HIV-1 RNA <50 c/mL at Week 48 | CMH adjusted difference, Q8W − Q4W |
An important statistical feature is that the direction of clinical interpretation changes with the endpoint definition. For the primary endpoint, a lower percentage with HIV-1 RNA ≥50 c/mL is generally the favorable direction of the binary outcome. For the secondary endpoint, a higher percentage with HIV-1 RNA <50 c/mL is the favorable direction. The numerical treatment difference must therefore always be read alongside the endpoint definition.
14. Primary vs Secondary Analysis
| Feature | Primary endpoint | Secondary endpoint |
|---|---|---|
| Time frame | Week 48 | Week 48 |
| Endpoint | HIV-1 RNA ≥50 c/mL | HIV-1 RNA <50 c/mL |
| Population | ITT-E | ITT-E |
| Method | CMH | CMH |
| Effect measure | Adjusted difference | Adjusted difference |
| Estimate | 0.8 | 0.8 |
| 95% CI | −0.6 to 2.2 | −2.1 to 3.7 |
| Hypothesis | Non-inferiority | Non-inferiority |
The two endpoints are complementary descriptions of virologic status at Week 48, but they should not be treated as interchangeable. The primary endpoint is defined by the threshold of ≥50 c/mL, while the secondary endpoint is defined by <50 c/mL.
15. Statistical Interpretation of the Secondary Confidence Interval
The secondary estimate is 0.8, with a two-sided 95% CI from −2.1 to 3.7. The interval is wider than the primary interval, extending farther in both directions from zero. This means the secondary treatment-effect estimate carries more statistical uncertainty in the posted analysis.
Because the effect is defined as Q8W minus Q4W, the positive point estimate corresponds to a higher estimated percentage below 50 c/mL in the Q8W group. But because the confidence interval crosses zero, the estimate should not be treated as evidence of a definitive directional difference by itself.
The registry text states that non-inferiority was concluded if the upper bound of the two-sided 95% CI was greater than −10%. The reported upper bound is 3.7. This page preserves that criterion exactly as reported in the registry rather than substituting a different rule.
16. Trial Timeline
Trial start
The registered ATLAS study began on October 27, 2017.
Primary completion
The registered primary completion date was June 6, 2019.
Active, not recruiting
The ClinicalTrials.gov record identifies the study status as ACTIVE_NOT_RECRUITING.
17. What This Analysis Does Not Establish
Not equivalence
A non-inferiority analysis should not automatically be described as an equivalence analysis. The registry-reported hypothesis type is non-inferiority.
Not individual prediction
The adjusted percentage difference is a population-level treatment estimate, not a prediction of an individual participant's response.
Not a p-value
The reported confidence interval should not be converted into an unreported p-value and presented as if it came from the registry analysis.
Not a safety hypothesis test
The registry-reported serious-adverse-event counts do not include a formal comparative statistical analysis.
18. Limitations
- Limited posted statistical detail: The ClinicalTrials.gov record contains two statistical analyses, one primary and one secondary, but do not provide a complete statistical analysis plan.
- No baseline table: Baseline demographic and clinical characteristics are not provided in the ClinicalTrials.gov record, so no baseline balance assessment is presented.
- No subgroup analyses: No subgroup estimates or interaction analyses are provided in the ClinicalTrials.gov record.
- No longitudinal efficacy results: The ClinicalTrials.gov record contains Week 48 primary and secondary analyses but do not provide later efficacy estimates or long-term outcome estimates.
- No formal safety comparison: Serious adverse-event counts are reported by arm, but the ClinicalTrials.gov record does not report a formal statistical comparison.
- No p-value for the primary analysis: The registry-reported primary analysis reports an adjusted difference and confidence interval, but no p-value.
- Non-inferiority interpretation: The registry's stated decision rules should be preserved exactly. In particular, the secondary endpoint's registry-reported wording uses an upper-bound criterion of greater than −10%, which differs from the conventional lower-bound formulation often encountered in non-inferiority reporting.
- Analysis population: The posted efficacy analyses use ITT-E. The ClinicalTrials.gov record does not provide corresponding per-protocol analyses that could be compared with the ITT-E results.
- Generalizability: The registry identifies the condition as HIV Infections and the population as HIV-1 infected adults, but the ClinicalTrials.gov record does not provide a detailed baseline characterization sufficient to evaluate applicability to specific clinical subpopulations.
19. Why This Trial Matters Statistically
ATLAS is a useful teaching example because the central statistical problem is not simply whether two percentages differ. The trial illustrates how a randomized clinical study can be framed around non-inferiority, analyzed with a categorical-data method, and interpreted primarily through the relationship between an estimated treatment difference, its confidence interval, and a prespecified margin.
| Concept | How it appears in ATLAS |
|---|---|
| Randomization | The trial is registered as randomized. |
| Parallel design | The registered design model is parallel. |
| Open-label design | The registered masking is none. |
| Binary endpoint | Both posted statistical analyses evaluate binary HIV-1 RNA outcomes. |
| ITT analysis | Both posted analyses use the ITT-E population. |
| Cochran-Mantel-Haenszel method | The posted statistical method for both analyses is the CMH test. |
| Adjusted difference | The effect measure is the adjusted difference, Q8W minus Q4W. |
| Confidence interval | Both analyses report two-sided 95% confidence intervals. |
| Non-inferiority | The primary and secondary analyses are identified as non-inferiority analyses. |
| Margin-based interpretation | The primary decision rule compares the upper CI bound with 4%. |
| Safety reporting | Serious adverse events are reported as affected participants over participants at risk by arm. |
20. A Practical Framework for Reading ATLAS
A reader can interpret the primary result in five steps.
- Identify the endpoint. The primary endpoint is the percentage of participants with HIV-1 RNA ≥50 c/mL at Week 48.
- Identify the direction of comparison. The treatment difference is explicitly Q8W minus Q4W.
- Read the point estimate. The adjusted difference is 0.8.
- Read the confidence interval. The two-sided 95% CI is −0.6 to 2.2.
- Apply the non-inferiority rule. The registry states that the upper confidence limit must be below 4%; the observed upper limit is 2.2.
This sequence prevents the common mistake of interpreting the point estimate without considering the uncertainty interval and the prespecified decision rule.
21. Related Statistical Concepts
Learn more about the methods used in this trial:
22. Related Statistical Calculators
Apply the statistical concepts behind this trial:
23. Sources
- ClinicalTrials.gov: ATLAS — NCT03299049.
- PubMed record: PMID 39052748.
- PubMed record: PMID 37452174.
- PubMed record: PMID 35508986.
- PubMed record: PMID 34648734.
- PubMed record: PMID 34261093.
Continue through Clinical Biostats
Connect this trial's statistical methods to deeper tutorials and practical statistical calculators.
24. Record Summary
ATLAS provides a clear example of a randomized phase 3 non-inferiority analysis for a binary virologic endpoint. The posted primary analysis used the ITT-E population and a Cochran-Mantel-Haenszel adjusted difference comparing CAB LA + RPV LA Q8W with CAB LA + RPV LA Q4W. The estimated difference was 0.8, with a two-sided 95% CI of −0.6 to 2.2. The registry states that non-inferiority was concluded when the upper confidence limit was less than 4%; the reported upper limit of 2.2 satisfies that stated rule.
The secondary Week 48 endpoint, defined as the percentage of participants with plasma HIV-1 RNA <50 c/mL, produced the same adjusted point estimate of 0.8, with a two-sided 95% CI of −2.1 to 3.7. Serious adverse events were reported as 27/522 for CAB LA + RPV LA Q8W and 19/523 for CAB LA + RPV LA Q4W, without a formal comparative safety analysis in the ClinicalTrials.gov record.