← Clinical Trials
HIV-1 Phase 3 Switch Study NCT01815736

GS-US-292-0109: Complete Statistical Analysis of E/C/F/TAF in Virologically Suppressed HIV-1

An independent statistical review of the randomized phase 3 GS-US-292-0109 trial evaluating a switch from a TDF-containing combination regimen to a TAF-containing fixed dose combination in virologically suppressed, HIV-1-positive participants.

Completed  ·  Enrollment 1443  ·  March 27, 2013 – March 16, 2015
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results and trial characteristics on this page are restricted to the ClinicalTrials.gov record for NCT01815736.

Registry context: 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

GS-US-292-0109 was a randomized, parallel-group, open-label phase 3 trial evaluating switching from a TDF-containing combination regimen to E/C/F/TAF in virologically suppressed, HIV-1-positive participants. The primary registered endpoint was the percentage of participants with HIV-1 RNA < 50 copies/mL at Week 48.

1443
Enrolled
Phase 3
2
Arms
Parallel design
48
Primary week
HIV-1 RNA < 50 copies/mL
96
Later week
Virologic and CD4 analyses
FeatureGS-US-292-0109
Trial nameGS-US-292-0109
ClinicalTrials.gov identifierNCT01815736
PhasePhase 3
StatusCompleted
Therapeutic areaInfectious Disease
ConditionsHIV; HIV Infections
AllocationRandomized
Design modelParallel
MaskingNone
Primary purposeTreatment
Enrollment1443
Lead sponsorGilead Sciences
Sponsor typeIndustry
StartMarch 27, 2013
Primary completionMarch 16, 2015

2. Clinical Question

The registered clinical question was whether virologically suppressed, HIV-1-positive participants could switch from a TDF-containing combination regimen to E/C/F/TAF while maintaining virologic suppression. The primary statistical question was framed as a non-inferiority comparison at Week 48, with a prespecified 12% margin.

Population

Virologically suppressed, HIV-1-positive participants.

Intervention

E/C/F/TAF, a TAF-containing fixed dose combination.

Comparator

Stay on Baseline Treatment Regimen (SBR).

Primary question

At Week 48, is the proportion with HIV-1 RNA < 50 copies/mL in the E/C/F/TAF group no more than 12 percentage points worse than the SBR group?

3. Trial Design

01
Randomize1443 enrolled
02
Two armsE/C/F/TAF vs SBR
03
Week 48Primary virologic endpoint
04
Safety / biomarkersBMD, creatinine, CD4, symptoms
05
Week 96Longer-term virologic and CD4 assessment
ARM A

E/C/F/TAF

  • Switch to E/C/F/TAF.
  • TAF-containing fixed dose combination.
  • Primary comparison at Week 48.
  • Serious adverse events were reported for the randomized phase.
ARM B

Stay on Baseline Treatment Regimen

  • Remain on the baseline treatment regimen.
  • Served as the randomized comparator for the primary endpoint.
  • Virologic, laboratory, bone mineral density, CD4 and symptom outcomes were compared with E/C/F/TAF.
Open-label design: The registry classifies masking as none. Therefore, the treatment assignment was not masked, an important design distinction when interpreting outcomes involving participant-reported symptoms.

4. Trial Timeline

March 27, 2013

Trial start

The registered study start date was March 27, 2013.

Week 48

Primary efficacy assessment

The registered primary endpoint was the percentage of participants with HIV-1 RNA < 50 copies/mL at Week 48, analyzed using the snapshot algorithm.

Week 96

Longer-term assessments

The registry contains Week 96 analyses of HIV-1 RNA < 50 copies/mL, HIV-1 RNA < 20 copies/mL, and change from baseline in CD4 cell count.

March 16, 2015

Primary completion

The registered primary completion date was March 16, 2015.

5. Primary Endpoint

EndpointDefinition / time frameStatistical analysis
Percentage of Participants With HIV-1 RNA < 50 Copies/mL at Week 48 Week 48. The percentage achieving HIV-1 RNA < 50 copies/mL was analyzed using the snapshot algorithm, which defines virologic response using viral load at the predefined time point within an allowed window together with study-drug discontinuation status. Cochran-Mantel-Haenszel; difference in percentages; non-inferiority framework

The endpoint is binary: each participant is classified according to the registered snapshot definition. That makes the principal treatment contrast a difference in proportions, rather than a mean difference or a time-to-event hazard ratio.

6. Statistical Methodology

Snapshot algorithm

The Week 48 primary endpoint used the snapshot algorithm. Rather than treating viral load as a continuous longitudinal measurement, the analysis classifies each participant's Week 48 virologic response using the viral load at the predefined assessment window and study-drug discontinuation status.

Binary treatment effect
Risk difference = P(E/C/F/TAF response) − P(SBR response)

A positive difference means that the observed percentage with HIV-1 RNA below the specified threshold was higher in the E/C/F/TAF group than in the SBR group.

Cochran-Mantel-Haenszel test

The primary comparison used the Cochran-Mantel-Haenszel method. The registry states that the difference in percentages and its confidence interval were calculated using the Mantel-Haenszel proportion adjusted by the prior treatment regimen.

Non-inferiority framework

The registry specifies a 12% non-inferiority margin. In the primary analysis, the null hypothesis was that the E/C/F/TAF group was at least 12% worse than the SBR group. The alternative hypothesis was that E/C/F/TAF was less than 12% worse than SBR.

Non-inferiority logic
Non-inferiority is supported when the lower confidence bound is above −12 percentage points.

The relevant comparison is therefore the confidence interval against the prespecified margin, not simply whether a conventional superiority p-value is below 0.05.

ANOVA

Several continuous outcomes were analyzed with ANOVA. For the BMD and CD4 analyses, the reported effect measure was the difference in least squares means. The BMD models included treatment and prior treatment regimen as fixed effects.

ANCOVA and covariate adjustment

Change from baseline in serum creatinine was analyzed using ANCOVA. The model included study treatment and prior treatment regimen as fixed effects and baseline serum creatinine as a covariate. This uses baseline information directly in the model when estimating the adjusted treatment-group difference.

Wilcoxon / Mann-Whitney analysis

The change from baseline in the overall EFV-related Symptom Assessment Score was analyzed with the Wilcoxon (Mann-Whitney) method. This is a nonparametric approach and does not require the same distributional assumptions as a conventional linear-model comparison of means.

7. Primary Results: HIV-1 RNA < 50 Copies/mL at Week 48

NDA Data Cut

Difference in percentages

2.7 percentage points

95.01% CI: −0.3 to 5.6   ·   P = 0.051

Analysis: Cochran-Mantel-Haenszel, adjusted by prior treatment regimen

Clinical Biostats interpretation

The estimated difference was 2.7 percentage points, meaning the observed percentage of participants meeting the HIV-1 RNA < 50 copies/mL snapshot definition was estimated to be 2.7 percentage points higher with E/C/F/TAF than with SBR in this analysis.

The estimate does not mean that an individual participant had a 2.7% greater probability of response, nor does it establish a treatment effect of exactly 2.7 percentage points. It is a group-level estimated risk difference.

The 95.01% confidence interval of −0.3 to 5.6 percentage points describes statistical uncertainty around the estimated difference. Importantly for the non-inferiority question, the lower bound of −0.3 is above the prespecified −12 percentage-point margin.

The reported P = 0.051 should not be interpreted as the probability that the treatment is ineffective, nor as a measure of the size of the effect. In this non-inferiority framework, the prespecified margin and confidence interval provide the key logic for assessing whether the treatment could be at least 12 percentage points worse.

The analysis was based on the Full Analysis Set, defined as randomized participants who received at least one dose of study drug, with the NDA data cut comprising participants through the data cut for the E/C/F/TAF NDA. The prior-treatment adjustment is part of the reported analysis and should not be ignored when interpreting the estimate.

All Participants

Difference in percentages

4.1 percentage points

95% CI: 1.6 to 6.7   ·   P < 0.001

Analysis: Cochran-Mantel-Haenszel, adjusted by prior treatment regimen

Clinical Biostats interpretation

The all-participants analysis estimated a 4.1 percentage-point difference in favor of E/C/F/TAF. The entire 95% confidence interval, from 1.6 to 6.7 percentage points, is above zero and is also well above the −12 percentage-point non-inferiority margin.

The estimate is a difference between group-level percentages, not a statement that every participant experienced the same change. The confidence interval expresses precision around the treatment-group comparison rather than the range of responses for individual patients.

The reported P < 0.001 indicates strong statistical evidence against the stated null hypothesis in the reported analysis. It does not quantify the clinical importance of a 4.1 percentage-point difference, nor does it mean that the probability of the null hypothesis being true is less than 0.1%.

Because this is a snapshot-based binary endpoint, the result depends on the registered classification rule, including viral-load status within the allowed Week 48 window and study-drug discontinuation status.

Why the two primary estimates differ: The registry reports two primary analyses using different data-cut descriptions. The NDA Data Cut gives a difference of 2.7 with a 95.01% CI of −0.3 to 5.6, while the All Participants analysis gives a difference of 4.1 with a 95% CI of 1.6 to 6.7. They should not be silently combined into a single estimate.

8. Secondary Virologic Results

HIV-1 RNA < 50 Copies/mL at Week 96

Difference in percentages

3.7 percentage points

95% CI: 0.4 to 7.0   ·   P = 0.017

Analysis: Cochran-Mantel-Haenszel, adjusted by prior treatment regimen

The Week 96 analysis retained the same non-inferiority framework and compared E/C/F/TAF with SBR in the Full Analysis Set. The confidence interval is entirely above zero and its lower bound is above the −12 percentage-point non-inferiority margin.

HIV-1 RNA < 20 Copies/mL at Week 48

AnalysisDifference in percentages95% CIP-value
NDA Data Cut1.8−1.7 to 5.30.29
All Participants3.20.1 to 6.30.031

Both analyses used the Cochran-Mantel-Haenszel method and a non-inferiority hypothesis with the same 12% margin. The difference between the two estimates again illustrates why data-cut definitions matter. The NDA Data Cut confidence interval crosses zero, while the All Participants confidence interval begins at 0.1 percentage points. Neither p-value should be interpreted as an effect-size measure.

HIV-1 RNA < 20 Copies/mL at Week 96

Difference in percentages

5.3 percentage points

95% CI: 1.6 to 9.0   ·   P = 0.003

Analysis: Cochran-Mantel-Haenszel

The Week 96 result provides a longer-term binary virologic comparison using the stricter HIV-1 RNA < 20 copies/mL threshold. The reported confidence interval remains above both zero and the non-inferiority boundary of −12 percentage points.

9. Bone Mineral Density Results

Hip BMD at Week 48

AnalysisDifference in least squares means95% CIP-value
NDA Data Cut2.0781.697 to 2.459<0.001
All Participants1.8071.488 to 2.126<0.001

The endpoint was percent change from baseline in hip bone mineral density at Week 48. The reported analysis used ANOVA, with treatment and prior treatment regimen as fixed effects. The effect measure was the difference in least squares means.

Clinical Biostats interpretation

The NDA Data Cut estimated a 2.078 difference in least squares means, while the All Participants analysis estimated 1.807. The confidence intervals are relatively narrow compared with their point estimates and do not cross zero. These are adjusted model-based mean differences, not simple differences between two unadjusted sample means.

Spine BMD at Week 48

AnalysisDifference in least squares means95% CIP-value
NDA Data Cut1.9701.551 to 2.390<0.001
All Participants2.0001.549 to 2.452<0.001

Spine BMD was also expressed as percent change from baseline at Week 48. The same ANOVA structure was reported, including treatment and prior treatment regimen as fixed effects.

Clinical Biostats interpretation

The spine results show positive adjusted differences in least squares means in both reported analyses. The confidence intervals exclude zero. The p-values provide evidence against the superiority null hypothesis specified for these analyses, but they do not describe the magnitude or practical importance of the BMD difference. The magnitude is described by the least-squares-mean difference and its confidence interval.

10. Serum Creatinine Results

Change From Baseline in Serum Creatinine at Week 48

AnalysisDifference in least squares means (mg/dL)95% CIP-value
NDA Data Cut−0.05−0.07 to −0.03<0.001
All Participants−0.04−0.05 to −0.02<0.001

The serum-creatinine analysis used ANCOVA in the Safety Analysis Set, defined as randomized participants who received at least one dose of study drug, excluding participants with prior treatment of EFV/FTC/TDF. The model included study treatment and prior treatment regimen as fixed effects and baseline serum creatinine as a covariate.

Clinical Biostats interpretation

The negative estimates indicate that the adjusted change from baseline in serum creatinine was lower in the E/C/F/TAF group than in the SBR group under the reported model. For the NDA Data Cut, the estimated difference was −0.05 mg/dL; for All Participants, it was −0.04 mg/dL.

The confidence intervals quantify uncertainty around these adjusted mean differences. They do not indicate the range of serum-creatinine changes that individual participants experienced. Because ANCOVA incorporates baseline serum creatinine as a covariate, the result is not simply an unadjusted subtraction of two observed group means.

11. CD4 Cell Count Results

Week 48

AnalysisDifference in least squares means (cells/uL)95% CIP-value
NDA Data Cut6−14 to 260.56
All Participants11−8 to 290.26

The Week 48 CD4 analyses used ANOVA. The model included treatment and prior treatment regimen as fixed effects. The Full Analysis Set with available data was used.

Clinical Biostats interpretation

The estimated differences were 6 cells/uL and 11 cells/uL for the NDA Data Cut and All Participants analyses, respectively. Both confidence intervals include zero. The reported p-values, 0.56 and 0.26, therefore do not provide evidence against the superiority null hypothesis specified for these analyses.

This does not mean that the treatment groups are mathematically identical. Rather, the estimated differences are compatible with a range of values that includes no difference, given the analysis and available data.

Week 96

Difference in least squares means

18 cells/uL

95% CI: −2 to 38   ·   P = 0.074

Analysis: ANOVA

The Week 96 estimate was 18 cells/uL, with a 95% confidence interval from −2 to 38 cells/uL. The interval includes zero, and the reported p-value was 0.074.

12. EFV-Related Symptom Assessment

The registry reports change from baseline in the overall EFV-related Symptom Assessment Score at Week 48. This endpoint was analyzed using the Wilcoxon (Mann-Whitney) method in participants in the EFV-Related Symptom Analysis Set with available data.

AnalysisMethodP-valueHypothesis type
NDA Data CutWilcoxon (Mann-Whitney)<0.001Superiority
All ParticipantsWilcoxon (Mann-Whitney)<0.001Superiority
Clinical Biostats interpretation

The registry reports p-values but no effect estimate or confidence interval for this endpoint. Therefore, the statistical evidence can be described only at the level reported: both the NDA Data Cut and All Participants analyses produced P < 0.001.

A p-value alone does not tell us the magnitude of the difference between groups. The Wilcoxon / Mann-Whitney analysis is a rank-based comparison, so a statistically small p-value should not be converted into a numerical treatment effect without an appropriate effect estimate.

The open-label design is also relevant for a symptom assessment because participant-reported outcomes can be influenced by awareness of treatment assignment. That is a design consideration, not evidence that the reported result is biased in a particular direction.

13. Safety Results

The registry reports serious adverse events separately for the randomized and extension phases. These figures are presented as affected participants over participants at risk.

Phase / treatment groupSerious adverse eventsAffected / at risk
Randomized Phase: E/C/F/TAFSerious adverse events79 / 959
Randomized Phase: Stay on Baseline TreatSerious adverse events39 / 477
Extension Phase: E/C/F/TAF From E/C/F/TASerious adverse events27 / 905
Extension Phase: E/C/F/TAF From SBRSerious adverse events22 / 424
Interpretation of safety denominators: The registry gives different denominators for the randomized and extension phases. The randomized-phase figures and extension-phase figures should therefore not be treated as if they came from the same analysis population or follow-up period.

The safety results also illustrate why an adverse-event comparison should retain its analysis population and phase. A numerator without its corresponding at-risk denominator is not a complete measure, and an extension-phase safety count cannot automatically be compared with a randomized-phase count as though exposure time and eligibility were identical.

14. Statistical Methods Explained

Why was a Cochran-Mantel-Haenszel analysis used?

The primary endpoint is binary: participants either meet the HIV-1 RNA threshold under the snapshot definition or they do not. The Cochran-Mantel-Haenszel approach provides a way to compare treatment groups while adjusting the proportion estimate for prior treatment regimen, as specified in the registry analysis notes.

Why is non-inferiority judged against the margin rather than simply the p-value?

The prespecified non-inferiority margin was 12 percentage points. The question is whether E/C/F/TAF could be at least 12 percentage points worse than SBR. Consequently, the lower confidence bound is compared with −12. A conventional p-value against zero answers a different question and is not, by itself, the defining criterion for non-inferiority.

What does a risk difference of 4.1 mean?

A risk difference of 4.1 means that the estimated percentage of participants meeting the binary virologic endpoint was 4.1 percentage points higher in the E/C/F/TAF group than in the SBR group in the reported All Participants primary analysis. It does not mean a 4.1-fold increase, a 4.1% increase in an individual's response probability, or a 4.1% reduction in viral load.

Why does the confidence interval matter?

The confidence interval shows the precision of the estimated treatment difference. For the All Participants primary analysis, the interval was 1.6 to 6.7 percentage points. For the NDA Data Cut, it was −0.3 to 5.6. The difference between those intervals illustrates that the precise data-cut definition can affect the estimated effect and its uncertainty.

Why use ANCOVA for serum creatinine?

ANCOVA incorporates baseline serum creatinine as a covariate while also accounting for treatment and prior treatment regimen. This can provide an adjusted comparison of Week 48 change rather than relying only on an unadjusted comparison of observed changes.

Why use ANOVA for BMD and CD4?

The registry reports ANOVA models for the percent change in BMD and change in CD4 cell count. For these analyses, the treatment effect was summarized as a difference in least squares means. The model therefore estimates adjusted group means under the specified fixed-effect structure rather than merely comparing raw arithmetic means.

What does the Wilcoxon / Mann-Whitney test tell us?

The Wilcoxon / Mann-Whitney method is a nonparametric rank-based comparison. For the EFV-related symptom score, the registry reports P < 0.001 but does not report an effect estimate or confidence interval. The p-value therefore indicates statistical evidence under the specified test but does not establish how large the between-group difference was.

15. Covariate Adjustment and Prior Treatment Regimen

Prior treatment regimen appears repeatedly in the registered statistical methods. For the primary virologic analyses, the Mantel-Haenszel proportion was adjusted by prior treatment regimen. For BMD analyses, treatment and prior treatment regimen were fixed effects in the ANOVA model. For serum creatinine, treatment and prior treatment regimen were fixed effects and baseline serum creatinine was an additional covariate.

Endpoint familyAdjustment reported in registry
HIV-1 RNA < 50 copies/mLMantel-Haenszel proportion adjusted by prior treatment regimen
HIV-1 RNA < 20 copies/mLMantel-Haenszel proportion adjusted by prior treatment regimen
Hip and spine BMDTreatment and prior treatment regimen as fixed effects
Serum creatinineTreatment and prior treatment regimen as fixed effects; baseline serum creatinine as a covariate
CD4 cell countTreatment and prior treatment regimen as fixed effects
EFV-related symptom scoreWilcoxon (Mann-Whitney); no covariate adjustment reported in the registry-reported analysis text

Adjustment does not mean that the analysis has "removed" all baseline differences. It means that the statistical model explicitly incorporates the prespecified variables when estimating the treatment comparison. The interpretation remains conditional on the correctness and appropriateness of the reported model.

16. Intention-to-Treat Principles and Analysis Sets

The registry's primary analyses refer to a Full Analysis Set that included participants who were randomized and received at least one dose of study drug. Several secondary analyses use more specific analysis sets.

Analysis setRegistry definition / role
Full Analysis SetRandomized participants who received at least 1 dose of study drug for the primary NDA Data Cut analyses.
Full Analysis Set with available dataUsed for several Week 48 and Week 96 CD4 analyses.
Hip DXA Analysis SetParticipants who received at least 1 dose and had nonmissing baseline hip BMD, with available data.
Spine DXA Analysis SetParticipants who received at least 1 dose and had nonmissing baseline spine BMD, with available data.
Safety Analysis SetRandomized participants who received at least 1 dose of study drug; the serum-creatinine analysis additionally excluded participants with prior EFV/FTC/TDF treatment.
EFV-Related Symptom Analysis SetParticipants with available data for the overall EFV-related Symptom Assessment Score.

The distinction matters because an estimate from a DXA analysis set answers a narrower question than an estimate from the Full Analysis Set. Similarly, safety analyses generally condition on exposure, while randomized efficacy analyses preserve the treatment assignment created by randomization as their organizing principle.

17. Multiplicity and the Number of Reported Analyses

The registry contains 17 statistical analyses, including 2 analyses of the primary endpoint and multiple secondary analyses covering virologic suppression, BMD, serum creatinine, CD4 cell count and EFV-related symptoms.

Analysis familyReported analysesPrimary role
HIV-1 RNA < 50 copies/mL at Week 482Primary
Hip BMD at Week 482Secondary
Spine BMD at Week 482Secondary
Serum creatinine at Week 482Secondary
HIV-1 RNA endpoints at Weeks 48 and 964Secondary
CD4 cell count at Weeks 48 and 963Secondary
EFV-related Symptom Assessment Score at Week 482Secondary
Multiplicity caution: The ClinicalTrials.gov record identifies multiple secondary statistical analyses but do not provide a multiplicity-adjustment procedure, alpha-allocation scheme, or formal endpoint hierarchy beyond the primary endpoint designation. Accordingly, the individual secondary p-values should be interpreted as the registry reports them rather than as evidence that every secondary comparison was independently powered and error-controlled.

18. What the Primary Non-Inferiority Result Does — and Does Not — Mean

Effect estimate

The primary All Participants estimate of 4.1 percentage points is a difference in percentages of participants achieving HIV-1 RNA < 50 copies/mL at Week 48 under the snapshot algorithm. It is not a hazard ratio, odds ratio, relative risk, or continuous change in viral load.

Confidence interval

The 95% CI of 1.6 to 6.7 percentage points describes uncertainty around the estimated group difference. It does not mean that 95% of individual participant-level treatment effects lie between 1.6 and 6.7 percentage points.

Non-inferiority margin

The prespecified margin was 12 percentage points. Because the reported lower confidence bounds are above −12 in the primary analyses, the confidence intervals are inconsistent with the null hypothesis that E/C/F/TAF is at least 12 percentage points worse than SBR.

P-value

The primary NDA Data Cut reports P = 0.051, while the All Participants analysis reports P < 0.001. These values should not be treated as direct measures of effect magnitude. In particular, the NDA Data Cut's p-value slightly above 0.05 does not by itself answer the non-inferiority question because the prespecified −12 percentage-point margin is the relevant reference point.

19. Results by Endpoint Family

EndpointAnalysisEffect95% CIP-value
HIV-1 RNA < 50 copies/mL, Week 48NDA Data CutRisk difference 2.7−0.3 to 5.6 (95.01%)0.051
HIV-1 RNA < 50 copies/mL, Week 48All ParticipantsRisk difference 4.11.6 to 6.7<0.001
Hip BMD, Week 48NDA Data CutMean difference 2.0781.697 to 2.459<0.001
Hip BMD, Week 48All ParticipantsMean difference 1.8071.488 to 2.126<0.001
Spine BMD, Week 48NDA Data CutMean difference 1.9701.551 to 2.390<0.001
Spine BMD, Week 48All ParticipantsMean difference 2.0001.549 to 2.452<0.001
Serum creatinine, Week 48NDA Data CutMean difference −0.05 mg/dL−0.07 to −0.03<0.001
Serum creatinine, Week 48All ParticipantsMean difference −0.04 mg/dL−0.05 to −0.02<0.001
HIV-1 RNA < 50 copies/mL, Week 96All ParticipantsRisk difference 3.70.4 to 7.00.017
HIV-1 RNA < 20 copies/mL, Week 48NDA Data CutRisk difference 1.8−1.7 to 5.30.29
HIV-1 RNA < 20 copies/mL, Week 48All ParticipantsRisk difference 3.20.1 to 6.30.031
HIV-1 RNA < 20 copies/mL, Week 96All ParticipantsRisk difference 5.31.6 to 9.00.003
CD4 cell count, Week 48NDA Data CutMean difference 6 cells/uL−14 to 260.56
CD4 cell count, Week 48All ParticipantsMean difference 11 cells/uL−8 to 290.26
CD4 cell count, Week 96All ParticipantsMean difference 18 cells/uL−2 to 380.074
EFV-related symptom score, Week 48NDA Data CutNot reportedNot reported<0.001
EFV-related symptom score, Week 48All ParticipantsNot reportedNot reported<0.001

This table deliberately preserves the effect measures actually reported in the ClinicalTrials.gov record. Where no numerical estimate or confidence interval was posted, none is inferred from the p-value.

20. Limitations and Interpretation Issues

21. Why This Trial Matters Statistically

GS-US-292-0109 is a useful teaching case because it combines a binary non-inferiority endpoint with covariate-adjusted continuous outcomes and a nonparametric symptom analysis. It also illustrates how the same clinical comparison can require different statistical methods depending on the endpoint's measurement scale.

ConceptHow it appears in GS-US-292-0109
RandomizationRandomized allocation to E/C/F/TAF or Stay on Baseline Treatment Regimen.
Parallel designTwo parallel treatment groups.
Non-inferiorityPrimary HIV-1 RNA < 50 copies/mL endpoint with a 12% margin.
Risk differencePrimary virologic effect measure reported as difference in percentages.
Cochran-Mantel-Haenszel testPrimary binary analysis adjusted by prior treatment regimen.
Snapshot algorithmDefines Week 48 virologic response using the predefined window and discontinuation status.
ANOVAUsed for BMD and CD4 change analyses.
ANCOVAUsed for serum creatinine with baseline creatinine as a covariate.
Least squares meansUsed to express adjusted continuous treatment differences.
Wilcoxon / Mann-WhitneyUsed for the EFV-related Symptom Assessment Score.
Confidence intervalsReported for the primary and many secondary effect estimates.
Analysis populationsFull Analysis, DXA, Safety and symptom-specific sets are used for different endpoints.

22. Statistical Concepts in This Trial

Learn more about the methods used in this trial:

23. Related Statistical Calculators

24. Sources

Continue through the Clinical Biostats statistical library

Explore the underlying methods through focused tutorials and statistical calculators for clinical-trial analysis.

25. Record Summary

GS-US-292-0109 demonstrates how a randomized phase 3 switch study can require several distinct statistical frameworks. The primary endpoint was a binary Week 48 virologic outcome analyzed with the Cochran-Mantel-Haenszel method and interpreted using a prespecified 12% non-inferiority margin. Secondary analyses used ANOVA for BMD and CD4 outcomes, ANCOVA for serum creatinine with baseline adjustment, and the Wilcoxon / Mann-Whitney method for the EFV-related symptom score.

The most important statistical distinction is between effect size, precision, and hypothesis testing. The primary risk differences describe the estimated separation between treatment groups; the confidence intervals describe uncertainty around those estimates; and the p-values describe evidence under the corresponding statistical tests. For a non-inferiority endpoint, the clinically specified margin provides an additional reference point that cannot be replaced by a generic p-value threshold.

Clinical Biostats methodology: A trial-results page should distinguish the registered endpoint, the analysis population, the statistical method, the effect measure, and the hypothesis being tested. When different data cuts or analysis sets are reported, they should remain explicitly labeled rather than being silently combined.