← Clinical Trials
HIV-1 Infection Phase 3 Non-Inferiority NCT01780506

GS-US-292-0104: Complete Statistical Analysis of E/C/F/TAF in HIV-1 Infection

An independent statistical analysis of the randomized phase 3 GS-US-292-0104 trial evaluating the safety and efficacy of E/C/F/TAF versus E/C/F/TDF in HIV-1 positive, antiretroviral treatment-naive adults, with emphasis on the Week 48 virologic response endpoint and its non-inferiority analysis.

Trial status: Completed  ·  Enrollment: 872  ·  Sponsor: Gilead Sciences
Scope of this record

This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. The numerical results on this page are restricted to the ClinicalTrials.gov record.

1. Trial at a Glance

GS-US-292-0104 was a randomized, double-blind, parallel phase 3 trial comparing E/C/F/TAF with E/C/F/TDF in HIV-1 positive, antiretroviral treatment-naive adults. The registered primary endpoint was the percentage of participants with HIV-1 RNA < 50 copies/mL at Week 48, analyzed using the snapshot algorithm. The posted formal comparison used a Cochran-Mantel-Haenszel analysis and reported a risk difference with a two-sided 95.002% confidence interval.

872
Enrollment
Phase 3
2
Primary groups
E/C/F/TAF vs E/C/F/TDF
0.5
Risk difference
Percentage-point difference
0.78
P-value
Posted primary analysis
FeatureGS-US-292-0104
Trial nameGS-US-292-0104
ClinicalTrials.gov identifierNCT01780506
PhasePhase 3
StatusCompleted
Therapeutic areaInfectious Disease
ConditionsHIV; HIV Infections
Enrollment872
AllocationRandomized
Design modelParallel
MaskingDouble
Primary purposeTreatment
StartDecember 26, 2012
Primary completionAugust 26, 2014
Lead sponsorGilead Sciences
Sponsor typeIndustry

2. Clinical Question

The central statistical question was whether E/C/F/TAF was sufficiently similar to E/C/F/TDF with respect to virologic response at Week 48 to satisfy the prespecified non-inferiority criterion. The primary endpoint measured the percentage of participants achieving HIV-1 RNA < 50 copies/mL at Week 48.

Population

HIV-1 positive, antiretroviral treatment-naive adults.

Intervention

E/C/F/TAF, with E/C/F/TAF placebo also listed among the trial interventions.

Comparator

E/C/F/TDF, with E/C/F/TDF placebo also listed among the trial interventions.

Primary question

Is the E/C/F/TAF group less than 12% worse than the E/C/F/TDF group with respect to the Week 48 virologic response rate?

3. Trial Design

01
Randomize 872 enrolled
02
Parallel groups Randomized allocation
03
Double-blind Active and placebo interventions
04
Week 48 Virologic response
05
Compare CMH risk difference
PRIMARY COMPARISON · E/C/F/TAF

E/C/F/TAF

  • Randomized treatment group.
  • Double-blind trial design.
  • Evaluated for HIV-1 RNA < 50 copies/mL at Week 48.
  • Primary analysis used the Full Analysis Set.
PRIMARY COMPARISON · E/C/F/TDF

E/C/F/TDF

  • Randomized comparator group.
  • Double-blind trial design.
  • Evaluated for HIV-1 RNA < 50 copies/mL at Week 48.
  • Primary analysis used the Full Analysis Set.

The registry data list three arms and four intervention entries: E/C/F/TAF, E/C/F/TDF, E/C/F/TDF placebo, and E/C/F/TAF placebo. The formal statistical analysis in the ClinicalTrials.gov record compares E/C/F/TAF directly with E/C/F/TDF.

4. Analysis Population and Primary Endpoint

The posted primary analysis used the Full Analysis Set, defined in the trial data as participants who were randomized and received at least one dose of study drug. This is important because the analyzed population is not described simply as everyone enrolled: the formal comparison is based on randomized participants who also received study drug.

FeatureRegistered / reported specification
Primary endpointPercentage of Participants With HIV-1 RNA < 50 Copies/mL at Week 48
Time frameWeek 48
Endpoint typeBinary
Endpoint definitionThe percentage of participants achieving HIV-1 RNA < 50 copies/mL at Week 48 was analyzed using the snapshot algorithm, which defines a patient's virologic response status using only the viral load at the predefined time point within an allowed window of time, along with study drug discontinuation status.
Analysis populationFull Analysis Set: participants who were randomized and received at least 1 dose of study drug.
Groups comparedE/C/F/TAF vs E/C/F/TDF
MethodCochran-Mantel-Haenszel test
Effect measureDifference in percentages; normalized as risk difference
Hypothesis typeNon-inferiority or equivalence

5. What the Snapshot Algorithm Means

The primary endpoint was not simply a laboratory measurement recorded at any time during follow-up. The registry definition specifies the snapshot algorithm, which classifies virologic response using the viral load at the predefined Week 48 time point within an allowed window together with study-drug discontinuation status.

This distinction matters statistically. A binary endpoint constructed by an algorithm can classify participants differently from an analysis that uses every viral-load measurement longitudinally. The analysis therefore asks a specific question: what percentage of participants meet the protocol-defined virologic response classification at the Week 48 assessment window?

Primary endpoint structure
Response = HIV-1 RNA < 50 copies/mL at Week 48 under the snapshot algorithm

The endpoint is binary: each participant contributes to the response comparison according to the prespecified snapshot classification rather than contributing a continuous viral-load trajectory.

6. Primary Result

The ClinicalTrials.gov record contains one formal statistical analysis, corresponding to the primary endpoint. The reported comparison was between E/C/F/TAF and E/C/F/TDF in the Full Analysis Set.

Difference in percentages

0.5

95.002% CI: -3.0 to 4.0   ·   P = 0.78

Cochran-Mantel-Haenszel analysis adjusted by baseline HIV-1 RNA and region stratum.

Primary analysis componentReported result
EndpointPercentage of Participants With HIV-1 RNA < 50 Copies/mL at Week 48
ComparisonE/C/F/TAF vs E/C/F/TDF
Analysis populationFull Analysis Set
MethodCochran-Mantel-Haenszel
Effect measureDifference in percentages / risk difference
Estimate0.5
Confidence interval95.002% CI -3.0 to 4.0
CI typeTwo-sided
P-value0.78
HypothesisNon-inferiority or equivalence
AdjustmentMantel-Haenszel proportions adjusted by baseline HIV-1 RNA and region stratum
Clinical Biostats interpretation

The reported risk difference of 0.5 means that the estimated difference in the Week 48 virologic-response percentages was 0.5 percentage points for E/C/F/TAF relative to E/C/F/TDF under the reported analysis.

The estimate does not mean that 0.5% of participants responded, nor does it mean that an individual participant had a 0.5% greater probability of response. It is a between-group difference in the percentage meeting the prespecified binary endpoint.

The two-sided 95.002% confidence interval extends from -3.0 to 4.0. Thus, the observed estimate is compatible with a modest difference in either direction within the uncertainty represented by the reported interval.

The P-value of 0.78 is evidence about the compatibility of the observed data with the relevant null hypothesis under the specified test; it is not a measure of the magnitude or clinical importance of the treatment difference. A large or small P-value should not be substituted for the effect estimate and its confidence interval.

For this trial, the central issue is non-inferiority rather than simply asking whether a conventional superiority test produced a small P-value. The reported non-inferiority margin was 12%, so the confidence interval must be interpreted relative to that prespecified margin.

7. The Non-Inferiority Question

The registry-reported analysis text states the hypotheses explicitly. The null hypothesis was that the E/C/F/TAF group was ≥ 12% worse than the E/C/F/TDF group with respect to the percentage of participants achieving HIV-1 RNA < 50 copies/mL at Week 48. The alternative hypothesis was that E/C/F/TAF was < 12% worse than E/C/F/TDF.

Null hypothesis

The E/C/F/TAF group is at least 12% worse than the E/C/F/TDF group for the primary virologic-response endpoint.

Alternative hypothesis

The E/C/F/TAF group is less than 12% worse than the E/C/F/TDF group for the primary virologic-response endpoint.

Confidence-interval logic
Lower 95.002% CI = -3.0  >  -12% non-inferiority boundary

The reported lower confidence limit of -3.0 is above the -12% boundary specified by the non-inferiority hypothesis. On the registry-reported analysis, the confidence interval therefore excludes differences as unfavorable as the prespecified non-inferiority margin.

This is the key statistical distinction between a non-inferiority analysis and an ordinary superiority test. The question is not whether the estimated difference is statistically distinguishable from zero. Instead, the question is whether the uncertainty around the difference excludes a loss large enough to cross the prespecified non-inferiority margin.

Do not use P = 0.78 to judge non-inferiority. The reported P-value is not the decision rule for the non-inferiority margin described in the registry-reported analysis. The margin is evaluated using the treatment-effect estimate and its confidence interval. Here, the reported lower confidence limit of -3.0 is above the -12% boundary.

8. Why the Cochran-Mantel-Haenszel Test Was Used

The primary endpoint is binary: each participant is classified according to whether the Week 48 snapshot criterion is met. The Cochran-Mantel-Haenszel framework is designed for comparing categorical outcomes while accounting for prespecified strata.

In this analysis, the Mantel-Haenszel proportions were adjusted by baseline HIV-1 RNA and region stratum. This allows the treatment comparison to account for those stratification variables rather than treating every participant as though the comparison were completely unstratified.

Conceptual structure
Stratified treatment comparison  →  baseline HIV-1 RNA + region stratum  →  adjusted proportions  →  risk difference

The registry reports the resulting effect as a difference in percentages, with the effect measure described as a risk difference.

9. Statistical Methods Explained

Why is a binary endpoint analyzed differently from a continuous measurement?

The Week 48 primary endpoint classifies participants into a virologic-response outcome based on the snapshot algorithm. Because the outcome is binary, the treatment comparison focuses on percentages and their difference rather than a mean or median viral-load value.

What does a risk difference of 0.5 mean?

A risk difference of 0.5 represents a 0.5-percentage-point difference in the proportion meeting the primary response definition between E/C/F/TAF and E/C/F/TDF under the reported analysis. It is an absolute rather than relative effect measure.

Why does the confidence interval matter more than the P-value for non-inferiority?

Non-inferiority is defined relative to a clinically specified margin. The relevant question is whether the confidence interval extends beyond that margin. Here, the reported interval is -3.0 to 4.0, while the non-inferiority boundary is -12%.

Why adjust for baseline HIV-1 RNA and region?

The reported Mantel-Haenszel analysis uses these variables as strata. Stratification permits the comparison to account for differences across the prespecified baseline HIV-1 RNA and region strata when estimating the treatment difference.

Does P = 0.78 mean the treatments are equivalent?

No. A P-value does not establish equivalence or non-inferiority by itself. The non-inferiority conclusion is tied to the prespecified margin and the confidence interval. A P-value of 0.78 should therefore not be translated into a statement that the two treatments have exactly the same effect.

Why is the Full Analysis Set important?

The formal analysis was conducted in participants who were randomized and received at least one dose of study drug. Identifying the analysis population matters because treatment-effect estimates depend on which participants contribute data and how their outcomes are classified.

What does the snapshot algorithm contribute to the analysis?

It fixes the primary endpoint to a predefined Week 48 assessment framework and incorporates study-drug discontinuation status. This creates a prespecified binary classification rather than allowing investigators to select an arbitrary favorable viral-load measurement from follow-up.

10. Confidence Intervals and Precision

The reported confidence interval is -3.0 to 4.0 at the 95.002% level. The interval is centered around the reported estimate of 0.5, but its width communicates that the estimate is not known with unlimited precision.

Point estimate

The estimated risk difference is 0.5 percentage points.

Lower limit

The reported lower confidence limit is -3.0 percentage points.

Upper limit

The reported upper confidence limit is 4.0 percentage points.

Non-inferiority boundary

The specified boundary corresponds to being 12% worse than the comparator.

A confidence interval is an uncertainty statement about the estimated treatment effect under the statistical model and sampling framework. It is not the range of effects that individual participants experienced, and it does not imply that every value inside the interval is equally plausible.

11. Primary Result in Statistical Context

The primary analysis provides three pieces of information that should be read together: the effect estimate, the confidence interval, and the non-inferiority hypothesis.

QuestionReported informationStatistical interpretation
What was estimated?Difference in percentages / risk differenceThe absolute difference in the proportion meeting the Week 48 response definition.
What was the estimate?0.5The estimated difference favored neither group by a large percentage-point amount.
How precise was it?95.002% CI -3.0 to 4.0The interval spans modest differences in either direction.
What was the non-inferiority boundary?12% worseThe relevant unfavorable boundary is -12 percentage points for the reported difference.
Does the CI cross that boundary?No; lower limit -3.0The lower confidence limit remains above the non-inferiority boundary.
What was the P-value?0.78It does not measure effect size and is not the non-inferiority decision rule.
Clinical Biostats interpretation

The most important feature of this analysis is the relationship between the -3.0 lower confidence limit and the -12% non-inferiority boundary. The observed estimate is 0.5, and the entire reported confidence interval remains above the prespecified threshold for unacceptable loss of efficacy.

This does not establish that E/C/F/TAF and E/C/F/TDF are identical. The interval allows for differences in either direction. Rather, the analysis addresses whether the data exclude a treatment difference large enough to violate the stated non-inferiority criterion.

The result also should not be interpreted as evidence that every individual participant would have the same virologic outcome under either treatment. The analysis concerns a group-level binary endpoint in the specified Full Analysis Set.

12. Intention-to-Treat Principles and the Analysis Population

The registry-reported analysis text identifies intention-to-treat analysis as an additional concept and defines the Full Analysis Set as participants who were randomized and received at least one dose of study drug.

Randomization is important because it creates the basis for a treatment-group comparison that is less vulnerable to systematic baseline selection than a non-randomized comparison. At the same time, the exact analysis population still matters. A reader should not silently replace the reported Full Analysis Set with a different population and assume that the resulting estimate would be identical.

Statistical distinction: "randomized" and "analyzed" are not interchangeable descriptions. The registry-reported primary analysis specifically identifies the Full Analysis Set definition, and that population should be retained when interpreting the reported risk difference and confidence interval.

13. Safety Results

The ClinicalTrials.gov record includes serious adverse event counts by arm. These data should be kept separate from the primary efficacy analysis because safety and efficacy answer different statistical questions.

Trial groupSerious adverse events affected / at riskDesign context
E/C/F/TAF73 / 435Double-blind
E/C/F/TDF65 / 432Double-blind
E/C/F/TAF to E/C/F/TAF0 / 90Open-label
E/C/F/TDF to E/C/F/TAF1 / 94Open-label

The double-blind groups provide the principal randomized treatment comparison represented in the ClinicalTrials.gov record. The open-label cohorts are listed separately and should not be combined with the double-blind groups as though they represented the same randomized comparison.

Safety interpretation: the ClinicalTrials.gov record reports serious adverse events as affected participants over participants at risk. They do not provide a formal statistical comparison, confidence interval, or P-value for these safety figures. A difference in observed counts alone should therefore not be converted into a claim of a statistically significant safety difference.

14. What the Risk Difference Does — and Does Not — Mean

Effect measure

The reported risk difference of 0.5 is an absolute comparison of the percentages meeting the Week 48 binary response endpoint. Unlike a relative risk or odds ratio, it is expressed directly as a difference in percentages.

It does not mean that treatment changes the outcome probability by exactly 0.5 percentage points for every individual participant. It is an estimate for the populations represented by the analysis.

Why zero is not the non-inferiority boundary

For a superiority analysis, a difference of zero is often the natural null value for a risk difference. For a non-inferiority analysis, the relevant question is different: how much worse can the new treatment be before the difference becomes unacceptable? Here, the registry-reported analysis defines that loss threshold as 12%.

Why the confidence interval is central

The lower limit of the reported 95.002% confidence interval is -3.0. Because this remains above the prespecified -12% non-inferiority boundary, the interval does not include differences as unfavorable as the margin.

15. Multiplicity, Interim Analysis, Crossover, and Bayesian Methods

The ClinicalTrials.gov record does not report a formal multiplicity strategy, interim analysis procedure, crossover analysis, missing-data imputation method beyond the snapshot algorithm's treatment of discontinuation status, or Bayesian analysis. These topics therefore cannot be reconstructed from the ClinicalTrials.gov record without introducing information from outside the permitted trial data.

Design topicWhat the ClinicalTrials.gov record establishes
MultiplicityNo formal multiplicity procedure is provided in the ClinicalTrials.gov record.
Interim analysisNo interim-analysis method is provided in the ClinicalTrials.gov record.
CrossoverNo crossover method or crossover analysis is provided in the ClinicalTrials.gov record.
Missing data / imputationThe primary endpoint definition specifies the snapshot algorithm and study-drug discontinuation status; no separate imputation model is provided.
StratificationThe primary analysis adjusts Mantel-Haenszel proportions by baseline HIV-1 RNA and region stratum.
Bayesian methodsNo Bayesian method is reported in the ClinicalTrials.gov record.

This distinction is important for an independent statistical analysis. Absence of a method from the ClinicalTrials.gov record is not evidence that a method was definitively absent from the full protocol or statistical analysis plan; it means that the method cannot be asserted here under the stated data rules.

16. Results Beyond the Primary Analysis

The registry record contains 24 posted outcome measures, but the ClinicalTrials.gov record contains only one formal statistical analysis, corresponding to the primary Week 48 virologic-response endpoint.

Accordingly, this page does not manufacture estimates, confidence intervals, or P-values for secondary outcomes. For a binary secondary endpoint, a categorical-data method such as a stratified comparison could be appropriate when specified by the protocol; however, the ClinicalTrials.gov record does not provide a formal secondary statistical comparison to report.

Why this matters: a list of registered outcomes is not the same thing as a complete set of formal statistical analyses. The primary analysis posted on ClinicalTrials.gov for GS-US-292-0104 can be interpreted quantitatively; unsupported secondary estimates should not be reconstructed from the existence of outcome measures alone.

17. Limitations and Interpretation Issues

18. Why This Trial Matters Statistically

GS-US-292-0104 is a useful teaching case because its primary analysis illustrates a central principle of clinical-trial statistics: non-inferiority is a margin-based inference, not simply a test for a non-significant difference.

ConceptHow it appears in GS-US-292-0104
RandomizationThe trial uses randomized allocation in a parallel design.
BlindingThe trial is double-blind and lists active and placebo interventions.
Binary endpointWeek 48 virologic response is defined as HIV-1 RNA < 50 copies/mL under the snapshot algorithm.
Non-inferiorityThe primary hypothesis specifies a 12% allowable loss relative to E/C/F/TDF.
Risk differenceThe reported treatment effect is a difference in percentages, normalized as a risk difference.
Confidence intervalThe 95.002% CI is -3.0 to 4.0.
Cochran-Mantel-Haenszel methodThe primary analysis uses a stratified categorical-data comparison.
StratificationBaseline HIV-1 RNA and region stratum are incorporated into the Mantel-Haenszel adjustment.
Intention-to-treat principleIntention-to-treat analysis is identified as a concept in the primary analysis.
Analysis populationThe Full Analysis Set consists of randomized participants who received at least one dose of study drug.
P-valuesThe reported P-value is 0.78 and must be interpreted separately from the non-inferiority margin.
Safety analysisSerious adverse events are reported separately by double-blind and open-label groups.

19. A Worked Reading of the Primary Analysis

A useful way to read the result is to proceed in the same order that a statistical reviewer would.

Step 1 · Identify the endpoint

The endpoint is the percentage of participants with HIV-1 RNA < 50 copies/mL at Week 48 using the snapshot algorithm.

Step 2 · Identify the estimand

The reported effect is the difference in percentages between E/C/F/TAF and E/C/F/TDF.

Step 3 · Read the estimate

The estimated risk difference is 0.5.

Step 4 · Read the uncertainty

The two-sided 95.002% CI is -3.0 to 4.0.

Step 5 · Apply the margin

The lower confidence limit of -3.0 remains above the -12% non-inferiority boundary.

Step 6 · Keep P separate

P = 0.78 does not replace the margin-based non-inferiority assessment.

This sequence prevents a common interpretive error: starting with the P-value and only afterward asking what the trial was actually designed to establish. In a non-inferiority trial, the margin and confidence interval are central to the statistical argument.

20. Statistical Interpretation vs Clinical Interpretation

Statistical interpretation

The reported risk difference was 0.5, with a two-sided 95.002% confidence interval of -3.0 to 4.0. The lower confidence limit remained above the -12% non-inferiority boundary specified in the analysis.

Clinical interpretation

The ClinicalTrials.gov record supports interpretation of the primary Week 48 endpoint through the stated non-inferiority framework. The page does not translate that statistical result into treatment recommendations or broader clinical claims.

21. What the Trial Design Tells Us About the Evidence

The combination of randomized allocation, double masking, a parallel design, and a prespecified binary Week 48 endpoint creates a structured comparison between the two principal treatment groups. Randomization provides the design basis for comparing treatment assignments, while blinding reduces the potential for knowledge of assignment to influence trial conduct.

The statistical method then adds another layer: rather than simply comparing two crude percentages, the reported Cochran-Mantel-Haenszel analysis adjusts the Mantel-Haenszel proportions by baseline HIV-1 RNA and region stratum. The final effect measure is a risk difference, allowing the non-inferiority margin to be expressed on the same absolute scale.

These pieces should be read together. A numerical estimate cannot be interpreted independently of the endpoint definition, analysis population, stratification, hypothesis, and margin.

22. Related Tutorials

Learn more about the methods used in this trial:

23. Related Calculators

The primary analysis also provides a natural pathway into calculations involving binary endpoints, absolute treatment effects, and non-inferiority margins.

24. Sources

Continue through Clinical Biostats

Use the related tutorials and calculators to explore the statistical concepts underlying randomized, non-inferiority, and categorical-endpoint analyses.

25. Record Summary

GS-US-292-0104 provides a clear example of how a randomized phase 3 trial can frame a binary efficacy endpoint as a non-inferiority question. The primary endpoint was the percentage of participants with HIV-1 RNA < 50 copies/mL at Week 48 under the snapshot algorithm. The formal analysis used the Full Analysis Set and a Cochran-Mantel-Haenszel method, with the Mantel-Haenszel proportions adjusted by baseline HIV-1 RNA and region stratum.

The reported effect was a 0.5 risk difference, with a two-sided 95.002% confidence interval of -3.0 to 4.0 and a P-value of 0.78. The central non-inferiority comparison is between the lower confidence limit of -3.0 and the prespecified 12% worse boundary: the confidence interval does not extend to that boundary. The statistical interpretation therefore depends on the non-inferiority framework rather than on the P-value alone.

The ClinicalTrials.gov record additionally report serious adverse events separately for double-blind E/C/F/TAF, double-blind E/C/F/TDF, and two open-label cohorts. Because the available statistical-analysis record contains one formal analysis, this page deliberately does not manufacture secondary endpoint estimates or unsupported safety comparisons.

Clinical Biostats methodology: A rigorous trial-results page should reconstruct the statistical question before interpreting the numerical result. For GS-US-292-0104, that means keeping the snapshot endpoint, Full Analysis Set, Cochran-Mantel-Haenszel adjustment, risk difference, confidence interval, and non-inferiority margin connected rather than interpreting any one statistic in isolation.