← Clinical Trials
HIV Infections Phase 2 Prevention NCT00705679

VOICE: Complete Statistical Analysis of Tenofovir-Based HIV Prevention in Women

An independent statistical review of the randomized phase 2 VOICE trial evaluating tenofovir 1% vaginal gel, oral tenofovir disoproxil fumarate, and oral emtricitabine/tenofovir disoproxil fumarate in women at risk for sexually transmitted HIV infection.

Completed trial  ·  Enrollment 5029  ·  Start August 2009  ·  Primary completion August 2012
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results are taken only from the ClinicalTrials.gov record. The registry provides the official trial record.

Registry note: 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

VOICE was a randomized, parallel phase 2 prevention trial in women with HIV infections as the condition of interest. The trial enrolled 5029 participants and evaluated five intervention arms involving tenofovir 1% vaginal gel, oral tenofovir disoproxil fumarate, emtricitabine/tenofovir disoproxil fumarate, and corresponding placebo interventions.

5029
Enrollment
Participants
5
Arms
Parallel design
0.85
TFV Gel HR
95% CI 0.61–1.21
1.04
TDF-FTC HR
95% CI 0.73–1.49
FeatureVOICE
Trial nameVOICE
ClinicalTrials.gov identifierNCT00705679
PhasePhase 2
StatusCompleted
Therapeutic areaInfectious Disease
ConditionHIV Infections
Primary purposePrevention
AllocationRandomized
Design modelParallel
MaskingNone
Enrollment5029
Lead sponsorNational Institute of Allergy and Infectious Diseases (NIAID)
Sponsor typeNIH
StartAugust 2009
Primary completionAugust 2012

2. Clinical Question

The trial evaluated the safety and effectiveness of three active prevention approaches involving tenofovir-based products in women at risk for sexually transmitted HIV infection. The registered primary comparisons examined HIV-1 infection incidence between active products and their corresponding placebo groups.

Population

Women participating in a phase 2 randomized prevention trial for HIV infection. The registered analyses excluded randomized participants with no follow-up HIV testing or participants determined to be HIV-positive at randomization by PCR testing of stored enrollment plasma samples.

Interventions

Tenofovir 1% vaginal gel, oral tenofovir disoproxil fumarate (TDF), and oral emtricitabine/tenofovir disoproxil fumarate (FTC/TDF), with corresponding placebo interventions.

Comparator

The registered primary efficacy comparisons were TFV gel versus placebo gel, oral TDF versus oral placebo, and oral TDF-FTC versus oral placebo.

Primary question

How do the HIV-1 infection incidence rates compare between each active prevention product and its corresponding placebo arm during follow-up?

3. Trial Design

01
Randomize5029 participants
02
Five armsActive and placebo interventions
03
Follow-upUp to 30 months
04
HIV testingMonthly testing
05
AnalysisCox and Fisher exact methods
Allocation
Randomized allocation with a parallel design.
Masking
The registry records masking as none.
Primary purpose
Prevention of sexually transmitted HIV infection.
Follow-up
Primary HIV incidence endpoints were assessed for up to 30 months of follow-up.

Registered interventions

ACTIVE PRODUCTS

Three prevention approaches

  • Emtricitabine/tenofovir disoproxil fumarate
  • Tenofovir disoproxil fumarate
  • Tenofovir 1% vaginal gel
PLACEBO INTERVENTIONS

Corresponding placebo products

  • Emtricitabine/tenofovir disoproxil fumarate placebo
  • Tenofovir disoproxil fumarate placebo
  • Tenofovir placebo
Trial changes during follow-up: the oral TDF arm was discontinued for futility in September 2011. The TFV gel and gel placebo arms were discontinued for futility in November 2011. The TDF-FTC and oral placebo arms continued follow-up until the end of the study in August 2012.

4. Endpoints

The registry lists 10 primary endpoints. The efficacy endpoints use both person-years of follow-up and HIV-1 infection counts, with incidence rate as the formal time-to-event analysis. A separate primary safety endpoint evaluates Grade 2, 3, and 4 adverse events.

Registered primary endpointTime frameType
Person-years of Follow-up of Tenofovir 1% Gel and Vaginal Placebo Gel ArmsFor up to 30 months of follow-upBinary
Number of HIV-1 Infections of Tenofovir 1% Gel and Vaginal Placebo Gel ArmsFor up to 30 months of follow-upBinary
Incidence Rate of HIV-1 Infections of Tenofovir 1% Gel and Vaginal Placebo Gel ArmsFor up to 30 months of follow-upTime-to-event
Person-years of Follow-up of Oral TDF and Oral Placebo ArmsFor up to 30 months of follow-upBinary
Number of HIV-1 Infections of Oral TDF and Oral Placebo ArmsFor up to 30 months of follow-upBinary
Incidence Rate of HIV-1 Infections of Oral TDF and Oral Placebo ArmsFor up to 30 months of follow-upTime-to-event
Person-years of Follow-up of Oral TDF-FTC and Oral Placebo ArmsFor up to 30 months of follow-upBinary
Number of HIV-1 Infections of Oral TDF-FTC and Oral Placebo ArmsFor up to 30 months of follow-upBinary
Incidence Rate of HIV-1 Infections of Oral TDF-FTC and Oral Placebo ArmsFor up to 30 months of follow-upTime-to-event
Extended Safety of Daily Tenofovir 1% Gel, Oral TDF, and Oral FTC/TDF in Women at Risk for Sexually Transmitted HIV Infection Based on Occurrence of Grade 2, 3, and 4 Adverse EventsThroughout study, up to 2.5 yearsBinary

How the HIV endpoint was defined

Participants were followed for up to 30 months. Person-years measured the amount of follow-up time for each participant, from enrollment to the first HIV-positive test result if HIV infection occurred, or to the last HIV-negative follow-up test if HIV infection did not occur.

Participants were tested monthly for HIV-1. Positive rapid test results were confirmed by enzyme-linked immunosorbent assay (EIA) and subsequent Western blotting (WB).

5. Statistical Methodology

Cox proportional-hazards model

The three formal HIV incidence-rate comparisons were analyzed using regression with a Cox proportional-hazards model. The registry specifies that the Cox model was stratified by site.

Core model interpretation
HR < 1  →  lower estimated instantaneous event rate in the active-product group

For these analyses, the hazard ratio compares the estimated instantaneous rate of HIV-1 infection between the active product and corresponding placebo group. A hazard ratio of 1 represents equal estimated hazards; values below 1 indicate a lower estimated hazard in the active-product group, while values above 1 indicate a higher estimated hazard.

Stratified analysis

The Cox model was stratified by site. Stratification allows the baseline hazard to vary across sites while estimating the treatment comparison within the overall Cox framework. This is particularly relevant when randomized participants are recruited across multiple sites with potentially different underlying event rates.

Incidence rate and person-years

The registered incidence-rate endpoint was defined as the number of HIV-1 infections divided by the amount of person-years of follow-up to HIV-1 infection status, multiplied by 100. The resulting unit was cases per 100 person-years.

Registered incidence-rate definition
Incidence rate = HIV-1 infections ÷ person-years × 100

This construction accounts for differences in the amount of follow-up contributed by participants. It is therefore different from simply dividing the number of infections by the number of randomized participants.

Fisher exact test

The primary safety analysis used a two-sided Fisher exact test. The analysis population was all randomized participants under an intention-to-treat definition, and the registered comparison was oral TDF-FTC versus oral placebo.

Intention-to-treat analysis

For the primary safety analysis, the registry explicitly identifies an intention-to-treat population consisting of all randomized participants. For the HIV incidence analyses, the registry defines the analysis population as all randomized participants except those with no follow-up HIV testing or those determined to be HIV-positive at randomization by PCR testing of stored enrollment plasma.

6. Planned Statistical Framework

The registry's formal analyses show a coherent time-to-event framework for the HIV incidence endpoints: participants contribute follow-up time until HIV infection or their last HIV-negative follow-up assessment, and the active-versus-placebo comparisons are estimated using site-stratified Cox proportional-hazards models.

Why use person-years?

Participants can contribute different amounts of observation time. Person-years incorporate the time actually contributed before HIV infection or the end of observed HIV-negative follow-up.

Why use a Cox model?

The endpoint is inherently time-dependent: the analysis uses not only whether HIV infection occurred but also when infection occurred relative to follow-up.

Why stratify by site?

The registered model was stratified by site, allowing the underlying hazard structure to differ by site without requiring a single common baseline hazard.

Why Fisher exact?

The registry specifies a two-sided Fisher exact test for the binary safety comparison. Fisher's exact test evaluates the association between treatment group and the observed categorical safety outcome without relying on the large-sample approximation used by a conventional chi-square test.

7. Primary Results: Tenofovir 1% Gel

The primary incidence-rate comparison for tenofovir 1% vaginal gel versus placebo gel used a site-stratified Cox proportional-hazards model. The reported effect measure was a hazard ratio.

HIV-1 infection incidence: TFV Gel vs Placebo Gel

HR 0.85

95% CI: 0.61–1.21   ·   P = 0.37

Outcome unit: cases per 100 person-years  ·  Follow-up: for up to 30 months

Clinical Biostats interpretation

The estimated hazard ratio of 0.85 means that, under the fitted Cox model, the estimated instantaneous rate of HIV-1 infection in the TFV gel group was 85% of that in the placebo-gel group. Expressed as a relative quantity, HR 0.85 corresponds to an estimated hazard that is 15% lower than the comparator hazard.

This does not mean that 15% of participants were protected, that HIV infection was reduced by exactly 15% for every individual, or that the absolute number of infections was 15% lower. The hazard ratio is a model-based relative measure of the event rate over follow-up.

The 95% confidence interval, 0.61–1.21, spans 1.00. Thus the data are compatible with a range of relative effects that includes both a lower and a higher estimated hazard than placebo. The interval describes statistical uncertainty around the estimated hazard ratio; it is not a range within which individual treatment effects must fall.

The P = 0.37 value measures the strength of evidence against the specified null hypothesis under the statistical testing framework. It does not measure the magnitude of the effect, the probability that the treatment works, or the clinical importance of HR 0.85.

The registry states that the null hypothesis was that the active product would be no more than 25% effective. The Cox model was stratified by site. Interpretation also depends on the model's proportional-hazards assumption and on censoring and the defined analysis population.

8. Primary Results: Oral TDF

The oral TDF versus oral placebo comparison was also analyzed using a site-stratified Cox proportional-hazards model.

HIV-1 infection incidence: Oral TDF vs Oral Placebo

HR 1.49

95% CI: 0.97–2.29   ·   P = 0.07

Outcome unit: cases per 100 person-years  ·  Follow-up: for up to 30 months

Clinical Biostats interpretation

The estimated hazard ratio of 1.49 means that, under the fitted Cox model, the estimated instantaneous rate of HIV-1 infection in the oral TDF group was 1.49 times the estimated rate in the oral placebo group. In relative terms, this corresponds to a 49% higher estimated hazard.

That estimate should not be read as a 49% increase in the number of participants who became infected, because the hazard ratio is not a simple risk ratio. It incorporates the time-to-event structure and the censoring framework of the analysis.

The 95% confidence interval is 0.97–2.29. Because the interval includes 1, the estimated effect is statistically compatible with equal hazards as well as with a range of lower or higher hazards. The width of the interval also shows that the estimate is not highly precise.

The P = 0.07 value does not quantify the size or clinical importance of HR 1.49. It is a measure of evidence against the registered null hypothesis under the specified test. It should not be interpreted as a 7% probability that the null hypothesis is true.

The registry states that the oral TDF arm was discontinued for futility in September 2011. That design history is important when interpreting the final posted analysis because follow-up did not proceed identically across all trial arms.

9. Primary Results: Oral TDF-FTC

The oral TDF-FTC versus oral placebo incidence-rate comparison was analyzed using the same site-stratified Cox proportional-hazards framework.

HIV-1 infection incidence: Oral TDF-FTC vs Oral Placebo

HR 1.04

95% CI: 0.73–1.49   ·   P = 0.81

Outcome unit: cases per 100 person-years  ·  Follow-up: for up to 30 months

Clinical Biostats interpretation

The estimated hazard ratio of 1.04 means that the estimated instantaneous rate of HIV-1 infection in the oral TDF-FTC group was 1.04 times the estimated rate in the oral placebo group under the fitted Cox model. This is close to 1 on the relative-hazard scale.

HR 1.04 does not mean that the two groups had identical individual outcomes. It is a summary estimate of the relative event rate produced by the specified survival model.

The 95% confidence interval of 0.73–1.49 includes 1 and extends on both sides of equal estimated hazard. Consequently, the interval allows for both a lower and a higher hazard relative to placebo. The interval is also sufficiently broad that the point estimate alone should not be treated as a precise description of the underlying treatment effect.

The P = 0.81 value is evidence from the specified statistical test, not a measure of effect size. A high P-value does not establish that the treatment and placebo are biologically identical; it indicates that the observed data do not provide strong evidence against the specified null hypothesis within this analysis.

The registry states that the TDF-FTC and oral placebo arms continued follow-up until the end of study in August 2012. As with the other Cox analyses, interpretation requires attention to censoring, the analysis population, site stratification, and the proportional-hazards assumption.

10. Primary Safety Result

The registered extended-safety endpoint evaluated the occurrence of Grade 2, 3, and 4 adverse events throughout the study, up to 2.5 years. The registry describes elevated serum creatinine levels as the only safety outcome of concern where a significant difference was detected between an active arm and its corresponding placebo arm.

Extended safety: Oral TDF-FTC vs Oral Placebo

P = 0.004

Two-sided Fisher exact test  ·  Intention-to-treat population

Time frame: throughout study, up to 2.5 years

Clinical Biostats interpretation

The registry reports a two-sided Fisher exact test P-value of 0.004 for the registered safety comparison. Fisher's exact test evaluates the evidence for an association between randomized treatment group and the categorical safety outcome under the specified analysis.

The P-value does not tell us how large the safety difference was. No effect estimate or confidence interval for this safety comparison is provided in the ClinicalTrials.gov record, so the magnitude and precision of the difference should not be inferred from P = 0.004 alone.

The analysis population was all randomized participants under an intention-to-treat definition. The endpoint was defined around Grade 2, 3, and 4 adverse events, with elevated serum creatinine identified by the registry as the safety outcome of concern for which a significant difference was detected.

11. Serious Adverse Events by Arm

The ClinicalTrials.gov record reports serious adverse events as affected participants divided by participants at risk for each arm. These are descriptive safety counts and should not be substituted for the formal Fisher exact analysis of the registered extended-safety endpoint.

ArmSerious adverse eventsAffected / at risk
Oral TDFSerious adverse events17 / 1007
Oral TDF-FTCSerious adverse events42 / 1003
Oral PlaceboSerious adverse events57 / 1009
TFV GelSerious adverse events39 / 1007
Gel PlaceboSerious adverse events26 / 1003

The denominators shown above are the registry-reported affected/at-risk values. The registry data do not provide a separate effect estimate or confidence interval for these serious-adverse-event counts. They therefore provide a descriptive view of safety rather than a complete inferential comparison.

12. Why the Hazard Ratio Is the Central Efficacy Measure

For VOICE's formal HIV incidence comparisons, the registry reports a hazard ratio rather than a simple proportion infected. This reflects the fact that participants were followed over time and could contribute different amounts of person-years before HIV infection or the end of observed HIV-negative follow-up.

ComparisonHazard ratio95% CIP-value
TFV Gel vs Placebo Gel0.850.61–1.210.37
Oral TDF vs Oral Placebo1.490.97–2.290.07
Oral TDF-FTC vs Oral Placebo1.040.73–1.490.81

The three estimates illustrate why the confidence interval is essential. The point estimates range from below 1 to above 1, while every registry-reported 95% confidence interval includes 1. A point estimate alone therefore provides an incomplete description of the evidence.

13. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

The primary HIV incidence endpoint is time-to-event in structure. Participants were observed from enrollment until HIV infection or their last HIV-negative follow-up assessment. A Cox model is designed to compare event hazards while accommodating censored follow-up.

What does a hazard ratio of 0.85 mean?

For the TFV gel comparison, HR 0.85 means the fitted model estimated the instantaneous HIV-1 infection rate in the TFV gel group at 85% of the estimated rate in the placebo-gel group. It is a relative hazard measure, not an absolute risk reduction and not a statement about the outcome of every participant.

Why is a hazard ratio above 1 possible in a prevention trial?

A hazard ratio is estimated from observed event timing. An estimate above 1 means that the observed data produced a higher estimated event hazard in the active-product group than in the comparator group. For oral TDF, the registry-reported estimate was HR 1.49. This point estimate must be considered together with its 95% confidence interval of 0.97–2.29.

Why was the Cox model stratified by site?

The registry specifies site stratification. A stratified Cox model permits different baseline hazard functions across the strata while estimating the treatment comparison using the available within-stratum information. The result is a treatment-effect model that does not require all sites to share exactly the same underlying baseline hazard.

Why does person-years matter?

Two groups can have the same number of participants but different total amounts of observed follow-up. Person-years captures the amount of observation time contributed by participants. The registry's incidence rate uses infections divided by person-years and expresses the result per 100 person-years.

Why does P = 0.37 not mean that the treatment has a 37% chance of working?

A P-value is calculated under a specified null hypothesis and testing framework. It quantifies how unusual the observed data, or more extreme data, would be under that framework. It is not the probability that the treatment works, the probability that the null hypothesis is true, or a measure of the magnitude of the hazard ratio.

Why is the safety analysis different from the HIV incidence analysis?

The registry reports a Cox proportional-hazards model for the formal HIV incidence-rate comparisons but a two-sided Fisher exact test for the registered extended-safety comparison. The former uses time-to-event information; the latter evaluates a categorical safety outcome using the observed treatment-group counts.

14. Non-Inferiority, Superiority, and the Registered Null Hypothesis

The statistical-analysis records classify the primary HIV comparisons as superiority hypotheses. The registry analysis notes specify that the null hypothesis was that the active product would be no more than 25% effective.

Design target stated in the registry
94 events per pairwise comparison  ·  55% effectiveness target  ·  25% lower-effectiveness boundary  ·  90% power  ·  false-positive error rate 0.0025

The registry analysis notes state that the trial was designed so that 94 events per pairwise comparison were needed to detect 55% effectiveness while ruling out a lower effectiveness of 25% with 90% power and a false-positive error rate of 0.0025.

These are design assumptions and targets, not descriptions of the observed treatment effects. A power calculation specifies what the trial was designed to detect under stated assumptions; it does not establish that the assumed effect occurred.

15. Futility Discontinuations and Their Statistical Importance

September 2011

Oral TDF arm discontinued

The registry states that the oral TDF arm was discontinued for futility.

November 2011

TFV gel and gel placebo discontinued

The registry states that the TFV gel and gel placebo arms were discontinued for futility.

August 2012

Study completion

The TDF-FTC and oral placebo arms continued follow-up until the end of study in August 2012.

Futility discontinuation is statistically important because follow-up duration and information accumulation are not necessarily the same across all randomized comparisons. It also means that a final posted estimate should be interpreted in the context of the actual observation period and the prespecified or registry-described stopping framework.

Important distinction: a futility decision is not the same thing as proving that an intervention has no effect. It is a decision to stop a study arm because the accumulating evidence did not justify continuing according to the trial's futility framework. The ClinicalTrials.gov record does not provide the interim boundary, stopping statistic, or detailed interim-analysis calculations.

16. Censoring and Time-to-Event Interpretation

For the HIV endpoints, participants contributed person-time until the first HIV-positive test result if infected or until the last HIV-negative follow-up test if not infected. This creates a natural right-censoring structure for participants who remained HIV-negative through their last observed assessment.

Event

The event was HIV-1 infection, assessed through monthly testing with confirmatory EIA and Western blotting after positive rapid-test results.

Censoring

Participants who remained HIV-negative contributed follow-up through their last HIV-negative test rather than being treated as if they had experienced an HIV event.

Person-time

Follow-up time was accumulated from enrollment to HIV infection or the last HIV-negative follow-up result.

Model

The formal incidence-rate comparison used a Cox proportional-hazards model stratified by site.

17. Interpreting the Confidence Intervals

Confidence intervals are particularly informative for this trial because the three primary HIV hazard-ratio estimates have different point estimates but all have intervals that include 1.

ComparisonEstimate95% CIWhat the interval communicates
TFV Gel vs Placebo Gel0.850.61–1.21Uncertainty includes both lower and higher estimated hazards than placebo.
Oral TDF vs Oral Placebo1.490.97–2.29Uncertainty includes equal hazard and extends toward higher estimated hazard.
Oral TDF-FTC vs Oral Placebo1.040.73–1.49Uncertainty includes equal hazard and both lower and higher estimated hazards.

The interval is not a prediction interval for individual participants. It describes uncertainty around the estimated population-level treatment effect under the specified statistical model and confidence-interval construction.

18. Multiplicity and Multiple Primary Comparisons

The registry lists three formal HIV incidence-rate comparisons: TFV gel versus placebo gel, oral TDF versus oral placebo, and oral TDF-FTC versus oral placebo. The registry-reported design notes state a false-positive error rate of 0.0025 for the event-based pairwise comparison design.

Because multiple active-product comparisons were part of the trial, interpretation of individual P-values requires attention to the trial's prespecified error-control framework rather than treating each P-value as if it arose from an isolated experiment. The ClinicalTrials.gov record does not provide a complete multiplicity-adjustment table or a separate alpha allocation for every posted endpoint.

Practical interpretation: P-values should be read together with the stated hypothesis, design-level false-positive error rate, effect estimate, and confidence interval. The numerical P-value by itself does not establish either the magnitude or the clinical importance of a treatment effect.

19. Safety Analysis and Fisher Exact Testing

The extended-safety endpoint was defined as the occurrence of Grade 2, 3, and 4 adverse events throughout the study, up to 2.5 years. The registry states that this measure describes the number of participants with elevated serum creatinine levels, identified as the only safety outcome of concern where a significant difference was detected between an active arm and its corresponding placebo arm.

Fisher exact test
Treatment group × categorical safety outcome → exact two-sided probability

Fisher's exact test evaluates the observed categorical treatment-by-outcome table without requiring the same large-sample approximation as a chi-square procedure. In this trial, the registry specifies a two-sided test for the oral TDF-FTC versus oral placebo comparison.

The reported P = 0.004 indicates statistical evidence against the null hypothesis used for that Fisher exact test. It does not provide the size of the difference. Because the registry-reported statistical analysis does not report a risk ratio, odds ratio, risk difference, or confidence interval for this safety result, no magnitude of effect should be inferred from the P-value alone.

20. Analysis Populations

AnalysisPopulation specified in the registry
TFV Gel vs Placebo Gel incidence rateAll randomized participants except those with no follow-up HIV testing or those determined HIV-positive at randomization by PCR testing of stored enrollment plasma.
Oral TDF vs Oral Placebo incidence rateAll randomized participants except those with no follow-up HIV testing or those determined HIV-positive at randomization by PCR testing of stored enrollment plasma.
Oral TDF-FTC vs Oral Placebo incidence rateAll randomized participants except those with no follow-up HIV testing or those determined HIV-positive at randomization by PCR testing of stored enrollment plasma.
Extended safetyAll randomized participants; intention-to-treat analysis.

The distinction is important. The registry explicitly calls the safety population intention-to-treat, while the HIV incidence analyses apply specific exclusions related to follow-up HIV testing and HIV status at randomization.

21. What These Results Do — and Do Not — Establish

What the hazard ratio describes

A relative comparison of the estimated instantaneous HIV-1 infection rate between an active product and its corresponding placebo under the Cox model.

What it does not describe

It is not an individual-level probability, an absolute risk difference, or a percentage of participants who benefited.

What the confidence interval describes

Uncertainty around the estimated treatment effect under the statistical model and confidence-interval framework.

What the P-value describes

Evidence against the specified null hypothesis under the testing framework; it is not an effect-size measure.

22. Important Limitations and Interpretation Issues

23. Why This Trial Matters Statistically

VOICE is a useful statistical teaching case because it combines randomized prevention research, multiple active-versus-placebo comparisons, person-time measurement, time-to-event modeling, site stratification, exact categorical testing, and futility discontinuations within one trial.

ConceptHow it appears in VOICE
RandomizationParticipants were randomized in a five-arm parallel phase 2 prevention trial.
Time-to-event endpointHIV-1 infection was evaluated over follow-up of up to 30 months.
Person-yearsFollow-up time was accumulated to HIV infection or the last HIV-negative test.
Hazard ratioUsed for the three formal HIV incidence-rate comparisons.
Cox modelThe registry reports regression using a Cox proportional-hazards model.
Stratified analysisThe Cox model was stratified by site.
Confidence intervalEach of the three reported HIV hazard ratios has a two-sided 95% confidence interval.
P-valueReported for all four posted statistical analyses.
Fisher exact testUsed for the two-sided primary safety comparison.
Intention-to-treatExplicitly specified for the extended-safety analysis.
FutilityIndividual study arms were discontinued for futility during follow-up.
Multiple comparisonsThree active-product versus placebo HIV incidence comparisons were formally analyzed.

24. Related Tutorials

Learn more about the methods used in this trial:

25. Related Statistical Calculators

26. Sources

Continue through the Clinical Biostats statistical library

Explore the statistical concepts behind randomized prevention trials, time-to-event endpoints, categorical safety analyses, and treatment-effect estimation.

27. Record Summary

VOICE provides a useful example of how a randomized prevention trial can combine person-years of follow-up, time-to-event analysis, site-stratified Cox regression, hazard ratios, confidence intervals, and Fisher exact testing. The three reported HIV incidence comparisons produced hazard ratios of 0.85, 1.49, and 1.04, respectively, with two-sided 95% confidence intervals of 0.61–1.21, 0.97–2.29, and 0.73–1.49. The extended-safety analysis reported a two-sided Fisher exact P-value of 0.004. Interpretation of these results requires attention to the confidence intervals, the registered hypotheses, the analysis populations, the site-stratified Cox model, censoring, multiple comparisons, and the futility discontinuations that occurred during the trial.

Clinical Biostats methodology: A trial-results page should not merely repeat numerical outputs. The goal is to explain what each statistical estimate represents, what it does not establish, how uncertainty affects interpretation, and which design features materially shape the evidence.