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
HPTN 035 was a randomized, quadruple-masked, phase 2 prevention trial evaluating vaginal gels intended to prevent HIV infection in women. The study used a factorial design, included 3,101 participants, and had four study arms.
| Feature | HPTN 035 |
|---|---|
| Official title | BufferGel and PRO 2000/5: Vaginal Gels to Prevent HIV Infection in Women |
| Phase | Phase 2 |
| Condition | HIV Infections |
| Primary purpose | Prevention |
| Allocation | Randomized |
| Design model | Factorial |
| Masking | Quadruple |
| Enrollment | 3,101 |
| Number of arms | 4 |
| Lead sponsor | National Institute of Allergy and Infectious Diseases (NIAID) |
| Sponsor type | NIH |
| ClinicalTrials.gov | NCT00074425 |
2. Clinical Question
The central clinical question was whether vaginal gels containing BufferGel or 0.5% PRO 2000/5 could contribute to prevention of HIV infection in women while maintaining an acceptable safety profile.
Population
Women enrolled in a phase 2 HIV prevention trial.
Interventions
BufferGel and PRO 2000/5 Gel were evaluated as vaginal-gel interventions.
Comparator
Placebo gel was included among the study interventions.
Primary questions
Does the intervention strategy maintain safety, and does it prevent HIV infection as measured by seroconversion?
Statistically, the trial therefore combines two different kinds of primary evidence. One endpoint concerns safety, defined through specified genital and systemic findings. The other concerns an objectively defined infectious outcome: HIV infection measured by seroconversion.
3. Trial Design
The registry classifies HPTN 035 as randomized, factorial, quadruple-masked, and intended for prevention. These design features determine how the treatment comparison should be interpreted.
Four-arm factorial structure
The registry reports four arms while listing BufferGel, PRO 2000/5 Gel, and placebo gel among the interventions. In a factorial trial, the number of arms can exceed the number of named intervention products because treatment factors may be combined into distinct randomized groups.
The statistical consequence is important: the analysis should respect the prespecified factorial structure rather than automatically treating the four arms as four unrelated pairwise experiments. Depending on the exact factorial specification and estimand, investigators may estimate main effects of factors, assess interaction between factors, or make prespecified arm-to-arm comparisons.
4. Study Timeline
Study start
The registry records January 2005 as the study start date.
Primary completion
The registry records September 2008 as the primary completion date.
Registry status
The study status is recorded as completed.
The study period is therefore bounded by a January 2005 start and a September 2008 primary completion date. The registry does not provide a more detailed sequence of enrollment milestones or interim statistical looks in the information reported here.
5. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Safety of BufferGel and 0.5% PRO 2000/5 Gel | Safety of BufferGel and 0.5% PRO 2000/5 Gel (P), as assessed by deep epithelial disruption, other genital symptoms, or other systemic symptoms | Throughout study |
| HIV infection | HIV infection as measured by seroconversion | Throughout study |
Safety endpoint
The first primary endpoint is not a single laboratory measurement. It is a safety construct assessed through deep epithelial disruption, other genital symptoms, or other systemic symptoms. This matters statistically because a composite or multidimensional safety definition requires the analysis to make clear what counts as an event and how multiple observations within the same participant are handled.
HIV infection endpoint
The second primary endpoint is HIV infection measured by seroconversion. In a prevention trial, seroconversion is naturally a time-dependent outcome: participants can remain infection-free for different lengths of follow-up, and some participants may complete observation without seroconversion.
The registry specifies the time frame for both primary endpoints as throughout study. It does not report numerical results for either endpoint in the information available here.
6. Statistical Methodology
No formal statistical analyses are posted to ClinicalTrials.gov for HPTN 035 in the registry record represented here. The appropriate statistical framework can nevertheless be explained from the endpoint definitions and the factorial randomized design.
Analysis of HIV seroconversion
For an endpoint defined by HIV infection through seroconversion and observed throughout a study, a standard analysis would treat the outcome as a time-to-event endpoint when the timing of seroconversion and the amount of follow-up vary among participants.
A Kaplan-Meier estimator can describe the probability of remaining free of seroconversion over time. A log-rank test can compare the event-time distributions between randomized groups, while a Cox proportional-hazards model can estimate a hazard ratio.
Here, T represents time to HIV seroconversion. The survival function can be interpreted as the probability of remaining without the event beyond time t.
The key point is that time-to-event methods use the information contained in the timing of infection and censoring. A simple proportion infected can answer a different question and can be sensitive to unequal follow-up.
Analysis of the safety endpoint
The safety endpoint is defined through deep epithelial disruption, genital symptoms, or systemic symptoms. If the endpoint is summarized as whether a participant experienced at least one qualifying event during the study, a risk-based comparison can be appropriate. Depending on the exact event structure, investigators could report event proportions, risk ratios, risk differences, or time-to-first-event measures.
If participants can experience multiple qualifying safety events, a recurrent-event framework may be appropriate. The correct choice depends on the prespecified statistical analysis plan and the exact event-level data. The registry definition alone does not establish which of these approaches was used.
Factorial analysis
The factorial design creates an additional statistical consideration. If the randomized factors are independently assigned, the primary analysis can often estimate the main effect of each factor efficiently within the same trial. However, this relies on the factorial structure and its assumptions.
An interaction term can be used to test whether the effect of one intervention differs according to the level of the other intervention. That is distinct from simply comparing four arms independently. The interaction question is scientifically important because a factorial design is not merely a compact way to create four treatment groups; it is a design intended to evaluate intervention effects within a structured combination of factors.
A factorial interaction asks whether the effect of one intervention changes according to the presence or level of the other intervention. The precise model depends on the outcome scale and prespecified estimand.
Randomization as the basis for inference
Because allocation was randomized, the treatment comparison is anchored to assignment rather than to an observational comparison of people who happened to receive different gels. This is particularly important in prevention research, where behavioral and baseline differences can otherwise influence infection risk.
Randomization does not guarantee identical groups in every characteristic. Its statistical role is to create a mechanism under which, on average over repeated randomizations, treatment assignment is independent of baseline and potential-outcome characteristics, subject to the trial's design and conduct.
Quadruple masking
The quadruple-masked designation indicates that the trial incorporated masking at multiple levels. From a statistical perspective, masking is valuable because it can reduce differential behavior, assessment, management, or reporting associated with knowledge of treatment assignment.
Masking does not replace randomization, and it does not eliminate all sources of bias. It is one component of the trial's overall design for protecting the validity of the comparison.
7. Planned Analysis
The registry identifies two primary outcomes that are to be assessed throughout the study: safety of BufferGel and 0.5% PRO 2000/5 Gel, and HIV infection measured by seroconversion. Results have not been posted on ClinicalTrials.gov.
Planned analysis of HIV seroconversion
For seroconversion, the natural statistical representation is the time from study entry or randomization to HIV infection, with participants who remain uninfected at the end of observation treated as censored observations. Kaplan-Meier methods can estimate infection-free survival, and a log-rank comparison can assess differences between randomized groups.
A Cox model can express the relative event rate through a hazard ratio. For example, a hazard ratio below 1 would indicate a lower estimated instantaneous rate of seroconversion in one group relative to another under the model. That is a relative time-to-event measure, not a direct percentage reduction in the cumulative probability of infection at every time point.
Planned analysis of safety
For the safety endpoint, an analysis would first need to define the unit of analysis. If the question is whether each participant experienced at least one qualifying safety event throughout the study, each participant contributes a binary outcome. If the timing of the first event is important, time-to-first-event methods can be used. If repeated events are clinically relevant, recurrent-event methods may be preferable.
The statistical interpretation also depends on whether the safety outcome is considered as a composite. A participant meeting any one of several qualifying criteria may count as an endpoint event even if the individual components have different clinical meanings. Reporting the components separately is therefore useful when available, because a composite result can conceal differences among its constituent event types.
Planned factorial comparisons
For a factorial design, the principal analysis should reflect the prespecified factorial estimands. Main effects and interactions answer different questions. If an interaction is present, a single overall main-effect estimate can be difficult to interpret because the effect of one factor depends on the level of the other.
8. Statistical Methods Explained
Why does the factorial design matter?
A factorial design allows more than one intervention dimension to be evaluated within the same randomized trial. Rather than conducting separate trials for every possible intervention combination, the factorial structure can evaluate effects jointly, provided the assumptions behind the design are appropriate.
For HPTN 035, the registry explicitly identifies the design as factorial and reports four arms. The important statistical question is therefore not simply whether four observed groups differ. It is how the randomized factors define the comparisons of scientific interest.
Why is seroconversion naturally a time-to-event endpoint?
Participants can experience HIV infection at different times, while others may remain infection-free through the end of their observation. A time-to-event analysis preserves information about when infection occurs and can appropriately handle participants who do not experience seroconversion during observed follow-up.
This differs from an analysis that simply divides the number infected by the number enrolled. Such a proportion ignores the timing of events and may not account adequately for unequal observation periods.
What does a hazard ratio mean for HIV infection?
A hazard ratio compares the estimated instantaneous event rates between two groups over the modeled follow-up. A value below 1 indicates a lower modeled hazard in the numerator group; a value above 1 indicates a higher modeled hazard.
The hazard ratio should not be interpreted as a universal relative-risk reduction. For example, a hazard ratio of 0.50 would not necessarily mean that exactly half as many participants become infected by every fixed time point. The cumulative incidence depends on the entire event-time distribution.
Why is censoring important?
Censoring occurs when the exact event time is not observed because follow-up ends before the event occurs or for another prespecified reason. In a survival analysis, a censored participant still contributes information up to the censoring time.
The validity of standard survival methods depends on assumptions about censoring. In particular, informative differences in the likelihood of remaining under observation can create bias if they are not handled appropriately.
How is a safety endpoint different from seroconversion?
Seroconversion is a defined infectious event. The safety endpoint is broader, incorporating deep epithelial disruption, other genital symptoms, or other systemic symptoms. The analysis therefore has to account for how multiple types of safety observations are classified and whether the primary safety question concerns any event, the first event, or repeated events.
Why does interaction matter in a factorial trial?
Suppose intervention A has one effect when intervention B is absent and another effect when B is present. That difference is an interaction. If no interaction is present, the main effect of A may provide a useful summary across levels of B. If an interaction is substantial, the overall main effect can obscure the fact that the effect of A depends on B.
Why does randomization matter more than baseline numerical balance?
Randomization provides the mechanism for causal inference. It does not require every measured baseline characteristic to have exactly the same numerical distribution in every randomized group. Small differences can arise by chance. The central statistical protection comes from the randomized assignment process rather than from achieving identical descriptive tables.
9. Interpreting the Two Primary Endpoints
Safety
The safety endpoint asks whether BufferGel and 0.5% PRO 2000/5 Gel are associated with specified genital or systemic safety findings throughout the study.
HIV infection
The HIV endpoint asks whether infection occurs, with seroconversion serving as the measurement of HIV infection.
These endpoints should be interpreted separately because they address different dimensions of a prevention intervention. An intervention could have a favorable infection endpoint but unacceptable safety, or could appear safe without demonstrating prevention efficacy. The trial therefore needs evidence on both dimensions.
The two endpoints also differ statistically. Safety may involve multiple event types and repeated observations, whereas seroconversion is naturally represented as a first infectious event over time. A single statistical summary should not be expected to answer both questions adequately.
10. Factorial Design: Statistical Interpretation
Factorial trials are especially useful when investigators want to study multiple intervention components efficiently. The four-arm structure of HPTN 035 reflects that design, but the statistical interpretation depends on the underlying factor assignments.
| Factorial concept | Statistical meaning |
|---|---|
| Main effect | The average effect associated with one randomized intervention factor across levels of another factor, when the factorial assumptions support that interpretation. |
| Interaction | The extent to which the effect of one intervention changes according to the level or presence of another intervention. |
| Four-arm comparison | A comparison among the randomized combinations. It is not automatically equivalent to four independent pairwise trials. |
| Efficiency | Multiple intervention questions can potentially be addressed within one randomized trial population rather than requiring separate trials. |
One of the principal statistical advantages of a factorial design is efficiency. If the effects of the intervention factors can be estimated without substantial interaction, information from multiple arms can contribute to the estimation of a main effect.
The principal caution is that efficiency is not the same as validity under every circumstance. Strong interaction between factors can change the interpretation of main effects. For that reason, factorial analysis should distinguish the prespecified main effects from interaction assessment.
11. Missing Data, Follow-Up, and Censoring
The primary endpoints are defined as occurring throughout the study, so the amount and pattern of participant follow-up are important to their statistical interpretation. However, the registry information represented here does not report a specific missing-data or imputation strategy.
For a seroconversion endpoint analyzed as time-to-event, participants who remain HIV-free when follow-up ends can contribute censored observations. Censoring is not equivalent to a negative infection test at every later time point. It means that the event was not observed during the available period of observation.
A survival analysis uses partial follow-up information rather than requiring every participant to have the same observation duration. This is one reason time-to-event analysis is often useful for prevention trials in which infection can occur at different times.
For safety, missing assessments can have a different meaning. If a participant does not have complete information on symptoms or epithelial findings, the consequences depend on the endpoint definition and the timing of the missing observation. A complete analysis would therefore require the prespecified rules governing evaluability and missing safety information.
12. Multiplicity and Multiple Comparisons
Factorial designs can involve several scientifically relevant comparisons. Once multiple hypotheses are tested, the interpretation of nominal statistical significance becomes more complicated because the probability of obtaining at least one apparently positive result can increase with the number of comparisons.
The registry information reported here does not specify an alpha-allocation or multiplicity-adjustment procedure for HPTN 035. Consequently, the exact confirmatory hierarchy among possible factorial comparisons cannot be established from the registry record alone.
Main effects
A prespecified main effect can be a primary scientific comparison in a factorial trial, but its interpretation depends on the factorial structure and interaction assumptions.
Interactions
Interaction tests address whether treatment effects differ across factor levels. They are different hypotheses from the main effects themselves.
Four-arm contrasts
Multiple arm-to-arm comparisons can create a multiplicity problem if each is tested independently at the same nominal significance level.
Safety components
Multiple safety event types can also generate several descriptive or inferential comparisons, which should be interpreted according to their prespecified role.
13. Interim Analysis
The registry information represented here does not report an interim-analysis plan, interim efficacy boundary, alpha-spending method, or stopping rule.
That distinction matters because an interim analysis changes the statistical interpretation of an observed p-value unless the repeated-look structure is incorporated into the design. If investigators repeatedly inspect accumulating efficacy data and stop when a nominal threshold is crossed, the overall type I error can exceed the nominal single-analysis level.
A prespecified group-sequential framework can address this by controlling the cumulative probability of a false-positive conclusion across planned looks. Common approaches include alpha-spending procedures and boundary-based methods. The specific procedure for HPTN 035 is not reported in the information available here.
14. Non-Inferiority, Bayesian Methods, and Other Design Features
| Method or feature | Registry-supported interpretation |
|---|---|
| Non-inferiority margin | Not reported in the trial information represented here. |
| Bayesian analysis | Not reported in the trial information represented here. |
| Interim efficacy analysis | Not reported in the trial information represented here. |
| Missing-data/imputation method | Not reported in the trial information represented here. |
| Stratification factors | Not reported in the trial information represented here. |
| Crossover | Not reported in the trial information represented here. |
These distinctions are statistically important because the absence of a reported feature is not evidence that the feature was absent from the full protocol or statistical analysis plan. It means that the ClinicalTrials.gov information represented here does not establish that methodological detail.
15. Limitations
- No posted statistical analyses: the ClinicalTrials.gov record represented here does not provide formal statistical analysis results for the two registered primary endpoints.
- No numerical endpoint results: the registry information does not report event counts, effect estimates, confidence intervals, or p-values for the safety or HIV seroconversion endpoints.
- Four-arm allocation counts are not reported here: the total enrollment is 3,101 and the number of arms is four, but exact enrollment in each arm is not reported in the information available here.
- Factorial details are incomplete: the registry identifies a factorial design but does not provide the complete factor-by-arm allocation structure in the information represented here.
- Analysis population definitions are not reported: the registry information does not specify the exact intention-to-treat, per-protocol, or safety populations that would be used for formal inference.
- Missing-data strategy is not reported: no imputation method or censoring rule beyond the endpoint definitions is specified in the information represented here.
- Multiplicity strategy is not reported: the statistical handling of multiple factorial comparisons or component safety analyses is not specified.
- Interim-monitoring plan is not reported: the registry information does not establish whether interim efficacy analyses occurred or what stopping boundaries were used.
- Generalizability: the trial enrolled 3,101 women in a defined randomized prevention study. The extent to which its results apply to populations outside the study depends on eligibility criteria and study conduct that are not fully described in the information represented here.
16. Why This Trial Matters Statistically
HPTN 035 is a useful teaching example because its design brings together several foundational ideas in clinical-trial statistics: randomization, factorial allocation, masking, prevention endpoints, time-to-event analysis, safety assessment, and the distinction between an endpoint definition and its statistical estimator.
| Concept | How it appears in HPTN 035 |
|---|---|
| Randomization | The study is classified as randomized, providing the foundation for treatment-effect inference. |
| Factorial design | The registry identifies a factorial design with four arms. |
| Masking | The study is classified as quadruple-masked. |
| Prevention research | The primary purpose is prevention of HIV infection. |
| Time-to-event analysis | HIV infection measured by seroconversion can naturally be analyzed according to time until infection. |
| Safety analysis | Safety is defined through deep epithelial disruption, genital symptoms, and systemic symptoms. |
| Composite endpoint interpretation | The safety definition includes multiple types of qualifying findings, which can have different clinical meanings. |
| Interaction | Factorial designs require consideration of whether one intervention's effect depends on another intervention factor. |
| Censoring | Participants who do not seroconvert during observed follow-up can contribute information without necessarily experiencing the event. |
| Multiplicity | Multiple factorial comparisons can create several hypotheses that require a prespecified inferential framework. |
Why the design is particularly instructive
The trial demonstrates why statistical analysis begins with the estimand and design rather than with a favorite statistical test. The same randomized study can generate different valid summaries depending on whether the question concerns safety, infection risk, time to infection, a main factorial effect, or an interaction.
For example, a proportion of participants who seroconvert answers an absolute-risk question. A hazard ratio answers a relative time-to-event question. A factorial interaction answers whether one intervention effect changes in the presence of another factor. These quantities are related, but they are not interchangeable.
17. Statistical Concepts in This Trial
The methods most closely connected to HPTN 035 are factorial trial design and its associated analysis considerations.
18. Related Tutorials
Learn more about the methods used in this trial:
19. Related Calculators
20. Sources
- ClinicalTrials.gov: HPTN 035 (NCT00074425).
Continue through Clinical Biostats
Explore additional statistical tutorials, calculators, and clinical trial analyses.
21. Record Summary
HPTN 035 was a completed randomized phase 2 prevention trial enrolling 3,101 participants and using a four-arm factorial design with quadruple masking. The registered primary endpoints addressed two complementary questions: the safety of BufferGel and 0.5% PRO 2000/5 Gel, assessed through deep epithelial disruption, other genital symptoms, or other systemic symptoms, and HIV infection measured by seroconversion. Both endpoints were defined to be assessed throughout the study.
The statistical structure of the trial is especially important because a factorial randomized design requires attention to main effects, interactions, and the relationship between the four randomized arms. For HIV seroconversion, time-to-event methods provide a natural framework when the timing of infection and variable follow-up are relevant. For safety, the appropriate analysis depends on whether the endpoint is treated as a participant-level binary outcome, a time-to-first-event outcome, or a recurrent-event process.
The ClinicalTrials.gov record represented here does not post numerical statistical results for the primary endpoints. As a result, the most important statistical lesson from the registry record is methodological rather than numerical: the design establishes how the trial could support an unbiased comparison, while the endpoint definitions determine what must be estimated, and the statistical analysis determines how that evidence should be quantified.