This page separates reported trial results from statistical interpretation. 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
HVTN 505 was a randomized, double-masked phase 2 prevention trial evaluating an HIV-1 DNA plasmid vaccine and an HIV-1 recombinant adenoviral serotype 5 vector vaccine in HIV-uninfected, circumcised men and male-to-female transgender persons who have sex with men. The registry reports 2504 participants and two parallel study arms.
| Feature | HVTN 505 |
|---|---|
| Phase | Phase 2 |
| Condition | HIV Infections |
| Population | HIV-uninfected, circumcised men and male-to-female (MTF) transgender persons who have sex with men |
| Primary purpose | Prevention |
| Design | Randomized, double-masked, parallel |
| Enrollment | 2504 |
| Primary endpoint types | Binary; Time-to-event |
| Primary endpoints registered | 8 |
| Statistical analyses posted | 5 |
| Primary analyses with estimate + CI | 5 |
| Primary time-to-event method | Cox proportional-hazards model |
| Primary effect measure | Hazard ratio |
| Lead sponsor | National Institute of Allergy and Infectious Diseases (NIAID) |
| Sponsor type | NIH |
| ClinicalTrials.gov | NCT00865566 |
2. Clinical Question
The central statistical question was whether assignment to the HIV-1 vaccine strategy was associated with a different hazard of HIV-1 infection than assignment to placebo during the registered follow-up periods, while also evaluating participant dropout and reactogenicity.
Population
HIV-uninfected, circumcised men and male-to-female (MTF) transgender persons who have sex with men.
Intervention
DNA plasmid vaccine and HIV-1 recombinant adenoviral serotype 5 (rAD5) vector vaccine.
Comparator
DNA vaccine placebo and HIV-1 recombinant adenovirus vaccine placebo.
Primary question
How does randomized vaccine assignment relate to HIV-1 infection over the registered follow-up period, and how does assignment relate to dropout?
3. Trial Design
HIV-1 vaccine strategy
- DNA plasmid vaccine
- HIV-1 recombinant adenoviral serotype 5 (rAD5) vector vaccine
- Evaluated against placebo in the randomized comparison
Placebo strategy
- DNA vaccine placebo
- HIV-1 recombinant adenovirus vaccine placebo
- Served as the comparator in the randomized comparison
4. Trial Timeline and Unblinding
Trial start
The registry lists May 2009 as the trial start.
End of the pre-unblinding period
The registered endpoint for participant dropout prior to unblinding uses enrollment through April 22, 2013, up to the Month 24 visit.
Trial unblinding
The registry definition states that the trial was unblinded on April 23, 2013.
Primary completion
The registry lists October 6, 2017 as the primary completion date.
The distinction between pre- and post-unblinding dropout is statistically important. The registry separately defines dropout before unblinding and dropout after unblinding, allowing the time periods to be analyzed separately rather than treating all dropout events as occurring under the same masking conditions.
5. Registered Primary Endpoints
| Endpoint | Type | Time frame | Definition / registry description |
|---|---|---|---|
| Participant Dropout Through Month 48 | Binary | Enrollment through Month 48 visit | For participants remaining uninfected, dropout is assessed only for primary follow-up, i.e. clinic visits. Protocol versions 4 and earlier included 24 months of primary follow-up, and versions 5 and later included 48 months. Participants who were found to be HIV-1 infected are analyzed as non-dropouts. |
| Participant Dropout Prior to Unblinding | Binary | Enrollment until the date of dropout, through April 22, 2013 (up to Month 24 visit) | For participants remaining uninfected, dropout is assessed only for primary follow-up, i.e. clinic visits. The trial was unblinded on April 23, 2013. |
| Participant Dropout After Unblinding | Binary | April 23, 2013 through trial closure (up to Month 48 visit) | For participants remaining uninfected, dropout is assessed only for primary follow-up, i.e. clinic visits. Participants may terminate early after completing primary follow-up. |
| HIV-1 Infections Diagnosed After Day 0 Including All Available Follow-up Through the Maximum Month 48 Visit | Time-to-event | Enrollment through Month 48 visit | An event is defined as HIV-1 infection. Participants are censored if they dropped out early or completed the trial. HIV-1 diagnosis date is the date of the earliest specimen collection yielding a positive HIV test. Participants remaining uninfected were censored at the latest specimen collection with a negative HIV-1 test. |
| HIV-1 Infections Diagnosed After Day 0 Through the Month 24 Visit | Time-to-event | Enrollment through Month 24 visit | An event is defined as HIV-1 infection. Participants remaining uninfected through the Month 24 visit were censored. HIV-1 diagnosis date is defined as the date of the earliest specimen collection at or prior to Month 24 which yielded a positive HIV test. |
| Number of Participants Experiencing Local Reactogenicity: Pain and/or Tenderness | Binary | Through 3 days post each study vaccination, and to resolution for events present at the third day post vaccination | For each sign or symptom, the maximum observed grade for each participant over all vaccinations and post-vaccination assessments is used. |
| Number of Participants Experiencing Local Reactogenicity: Erythema and/or Induration | Binary | Through 3 days post each study vaccination, and to resolution for events present at the third day post vaccination | For each sign or symptom, the maximum observed grade for each participant over all vaccinations and post-vaccination assessments is used. |
| Number of Participants Experiencing Systemic Reactogenicity | Binary | Through 3 days post each study vaccination, and to resolution for events present at the third day post vaccination | For each sign or symptom, the maximum observed grade for each participant over all vaccinations and post-vaccination assessments is used. |
The registry therefore contains eight registered primary endpoints: five represented in the posted formal statistical analyses and three reactogenicity endpoints for which the ClinicalTrials.gov record does not include a formal statistical-analysis record.
6. Statistical Methodology
Cox proportional-hazards model
All five posted formal primary analyses use a Cox proportional-hazards model. The model estimates a relative hazard associated with treatment assignment while accommodating different lengths of follow-up and right-censored observations.
For a binary treatment indicator, the hazard ratio is exp(β). In this trial's reported analyses, the hazard ratio is defined as vaccine divided by placebo.
Hazard ratio
A hazard ratio compares the modeled instantaneous event rate between treatment groups over time. An HR below 1 indicates a lower estimated hazard in the vaccine group; an HR above 1 indicates a higher estimated hazard. The HR is not an absolute risk difference and does not directly state how many participants experienced an event.
The registry specifies vaccine / placebo for the reported dropout and HIV-1 infection analyses.
Time-to-event analysis and censoring
The HIV-1 infection endpoints explicitly define infection as the event and describe censoring for participants who remain uninfected, drop out early, or complete the trial. This allows participants to contribute follow-up information without requiring every participant to experience HIV-1 infection during observation.
Covariate adjustment
For the HIV-1 infection endpoint including all available follow-up through the maximum Month 48 visit, the Cox model was adjusted for age, behavioral risk score, square of behavioral risk score, race, and BMI. Including both the behavioral risk score and its square allows the fitted relationship with the hazard to accommodate curvature rather than forcing a strictly linear effect of that score.
Modified intention-to-treat populations
The infection analyses were conducted in a Modified Intent-to-Treat Cohort. The Month 48 dropout analysis used the MITT population, while the post-unblinding dropout analysis used MITT participants who were HIV-uninfected and on-study as of April 23, 2013. The population definitions therefore differ across analyses and should not be silently treated as identical analysis sets.
Score test
The registry identifies a score test in the analysis information for the dropout and Month 24 infection analyses. In a Cox model, a score test evaluates evidence against a null treatment coefficient without requiring the fitted treatment coefficient to be the sole basis of the test statistic.
7. Results: Participant Dropout Through Month 48
The first posted formal primary analysis evaluates participant dropout through the Month 48 visit in the MITT population.
Hazard ratio for dropout
95% CI: 0.63–0.84 · P < 0.001
Effect measure: vaccine / placebo
An HR of 0.73 means that the fitted Cox model estimated the instantaneous hazard of dropout in the vaccine group at approximately 73% of the corresponding hazard in the placebo group, or a 27% lower estimated hazard under that model.
It does not mean that 27% fewer participants necessarily dropped out, nor does it give the absolute probability of dropout. The HR summarizes a time-to-event comparison and depends on the analysis population, censoring rules, and model assumptions.
The 95% CI of 0.63–0.84 describes uncertainty around the estimated hazard ratio. It does not describe the range of dropout hazards experienced by individual participants.
The P-value of <0.001 addresses evidence against the model's null treatment effect under the reported testing framework. A P-value does not measure the size or clinical importance of the estimated effect; the HR and confidence interval provide the effect-size information.
Because the endpoint is a Cox time-to-event analysis, interpretation also depends on the proportional-hazards assumption. The ClinicalTrials.gov record does not report a separate assessment of that assumption.
8. Results: Participant Dropout Prior to Unblinding
The second posted dropout analysis covers enrollment through April 22, 2013, corresponding to the period before the trial was unblinded on April 23, 2013.
Hazard ratio for dropout before unblinding
95% CI: 0.59–1 · P = 0.05
Effect measure: vaccine / placebo
An HR of 0.77 corresponds to an estimated dropout hazard approximately 23% lower in the vaccine group than the placebo group under the fitted Cox model.
The 95% CI extends from 0.59 to 1. This interval is important because it includes the null value of 1 at its upper boundary, indicating substantial uncertainty about the exact relative dropout hazard in this period.
The reported P = 0.05 is a statement about statistical evidence under the specified test; it is not a measure of the magnitude of the association. The HR and its confidence interval should be read alongside the P-value.
This analysis is also temporally distinct from the post-unblinding analysis. The treatment groups were not observed under identical masking conditions after April 23, 2013, so the two dropout analyses answer different questions about different periods.
9. Results: Participant Dropout After Unblinding
The third posted dropout analysis begins on April 23, 2013 and continues through trial closure, up to the Month 48 visit. The analysis population is restricted to MITT participants who were HIV-uninfected and on-study as of April 23, 2013.
Hazard ratio for dropout after unblinding
95% CI: 0.60–0.84 · P < 0.001
Effect measure: vaccine / placebo
An HR of 0.71 means that the estimated instantaneous hazard of dropout was approximately 71% as high in the vaccine group as in the placebo group, corresponding to a 29% lower estimated hazard under the fitted model.
The 95% CI of 0.60–0.84 quantifies uncertainty around the estimated HR. It does not imply that the true individual-level dropout reduction lies somewhere between 16% and 40%, because the confidence interval concerns the population-level model parameter rather than individual treatment effects.
The P-value of <0.001 indicates strong statistical evidence against the model's null treatment association under the reported analysis. It should not be interpreted as the probability that the null hypothesis is true, nor as a measure of how large the dropout difference is.
The post-unblinding analysis has a specifically defined population and calendar period. That restriction is important when interpreting it alongside the overall Month 48 dropout analysis.
10. Results: HIV-1 Infections Through Maximum Month 48 Follow-up
The most directly relevant efficacy analysis evaluates HIV-1 infections diagnosed after Day 0 using all available follow-up through the maximum Month 48 visit. The analysis used a Modified Intent-to-Treat Cohort and a Cox proportional-hazards model adjusted for age, behavioral risk score, the square of behavioral risk score, race, and BMI.
Hazard ratio for HIV-1 infection
95% CI: 0.73–1.42 · P = 0.903
Effect measure: vaccine / placebo
An HR of 1.02 corresponds to an estimated instantaneous HIV-1 infection hazard approximately 2% higher in the vaccine group than in the placebo group under the fitted Cox model. That small numerical difference is the point estimate, not an estimate that the vaccine caused a 2% increase in every participant's infection risk.
The 95% CI of 0.73–1.42 spans values below and above 1. The interval therefore expresses uncertainty compatible with both a lower and a higher hazard relative to placebo within the range represented by the model and sampling framework.
The P-value of 0.903 indicates little statistical evidence against the null treatment association under the reported test. It does not prove that the vaccine and placebo hazards are exactly identical, and it does not measure the precision or magnitude of the effect. Those questions are better addressed using the HR and its confidence interval.
The analysis is adjusted for prespecified covariates listed in the registry data. The adjusted HR should therefore be interpreted as the treatment-assignment association from that fitted model, rather than as an unadjusted crude hazard ratio.
As with other Cox analyses, the interpretation relies on the proportional-hazards framework. The ClinicalTrials.gov record does not report a separate diagnostic for proportional hazards.
11. Results: HIV-1 Infections Through Month 24
A separate primary infection analysis restricts follow-up to the Month 24 visit. Participants remaining uninfected through Month 24 were censored according to the registered endpoint definition.
Hazard ratio for HIV-1 infection through Month 24
95% CI: 0.71–1.58 · P = 0.783
Effect measure: vaccine / placebo
An HR of 1.06 means that the fitted model estimated the instantaneous infection hazard in the vaccine group at approximately 106% of the placebo-group hazard during the registered Month 24 analysis period.
The 95% CI of 0.71–1.58 is relatively broad compared with the point estimate. It includes 1 and therefore encompasses both lower and higher infection hazards for the vaccine group relative to placebo.
The reported P = 0.783 does not provide evidence against the null treatment association under the reported test. Importantly, a nonsignificant P-value is not evidence that the two hazards are exactly equal; the confidence interval communicates the range of model estimates that remains compatible with the observed data.
This endpoint is also shorter in follow-up than the maximum Month 48 analysis. Results from the two time windows should therefore be treated as separate estimates rather than as duplicate measurements of one identical endpoint.
12. Summary of the Five Posted Primary Analyses
| Primary endpoint | Analysis population | Method | HR | 95% CI | P-value |
|---|---|---|---|---|---|
| Participant Dropout Through Month 48 | MITT population | Cox proportional-hazards model | 0.73 | 0.63–0.84 | <0.001 |
| Participant Dropout Prior to Unblinding | MITT population | Cox proportional-hazards model | 0.77 | 0.59–1 | 0.05 |
| Participant Dropout After Unblinding | MITT population participants who were HIV-uninfected and on-study as of April 23, 2013 | Cox proportional-hazards model | 0.71 | 0.60–0.84 | <0.001 |
| HIV-1 Infections Diagnosed After Day 0 Including All Available Follow-up Through the Maximum Month 48 Visit | Modified Intent-to-Treat Cohort | Cox proportional-hazards model | 1.02 | 0.73–1.42 | 0.903 |
| HIV-1 Infections Diagnosed After Day 0 Through the Month 24 Visit | Modified Intent-to-Treat Cohort | Cox proportional-hazards model | 1.06 | 0.71–1.58 | 0.783 |
The statistical pattern is different for the two classes of time-to-event endpoints. The reported dropout HRs are below 1, whereas the reported HIV-1 infection HRs are close to 1. These are different outcomes and should not be collapsed into a single overall treatment effect.
13. Reactogenicity Endpoints
Three additional registered primary endpoints concern local and systemic reactogenicity. The ClinicalTrials.gov record confirms that results were posted for these endpoints, but it does not include formal statistical analyses for them.
| Endpoint | Assessment period | Registered endpoint approach | Formal analysis reported |
|---|---|---|---|
| Pain and/or tenderness | Through 3 days post each study vaccination, and to resolution for events present at the third day post vaccination | Maximum observed grade for each participant over all vaccinations and post-vaccination assessments | No |
| Erythema and/or induration | Through 3 days post each study vaccination, and to resolution for events present at the third day post vaccination | Maximum observed grade for each participant over all vaccinations and post-vaccination assessments | No |
| Systemic reactogenicity | Through 3 days post each study vaccination, and to resolution for events present at the third day post vaccination | Maximum observed grade for each participant over all vaccinations and post-vaccination assessments | No |
For binary reactogenicity endpoints, a comparison of proportions or a model for binary outcomes would ordinarily be considered, depending on the prespecified analysis plan and the exact definition of the event. Because the ClinicalTrials.gov record does not identify a formal statistical method or provide arm-specific reactogenicity estimates, this page does not construct a comparison that is not reported in the ClinicalTrials.gov record.
14. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized arm as affected participants divided by participants at risk.
| Safety measure | Vaccine | Placebo |
|---|---|---|
| Serious adverse events, affected / at risk | 77 / 1253 | 95 / 1251 |
The denominators are also important: 77 of 1253 participants in the vaccine arm and 95 of 1251 in the placebo arm were reported as affected by serious adverse events. The ClinicalTrials.gov record does not provide a formal statistical comparison for this safety endpoint, so the counts should be read descriptively rather than converted into an unreported hypothesis test.
15. Statistical Methods Explained
Why use a Cox proportional-hazards model for HIV-1 infection?
HIV-1 infection is a time-to-event outcome because participants can become infected at different times during follow-up, while others remain uninfected when observation ends. Cox regression allows the analysis to use the timing of infection and the available follow-up while handling censoring.
What does an HIV-1 infection HR of 1.02 mean?
An HR of 1.02 is a relative hazard estimate with vaccine as the numerator and placebo as the denominator. It corresponds to an estimated instantaneous infection hazard approximately 2% higher in the vaccine group under the fitted model. It is not a 2% absolute increase in infection probability.
Why is the confidence interval important?
The HR is only a point estimate. The 95% confidence interval communicates statistical uncertainty around that estimate. For the maximum Month 48 infection analysis, the interval is 0.73–1.42, so the observed data are compatible with a range of relative hazards on both sides of the null value of 1.
Why does the P-value not measure effect size?
A P-value addresses the compatibility of the observed evidence with a specified null hypothesis under the testing framework. It does not quantify the size of the treatment effect. An HR of 1.02 and an HR of 0.73 are effect estimates; their corresponding P-values answer a different statistical question.
Why separate the Month 24 and Month 48 infection analyses?
The registered endpoints define different observation windows and censoring rules. The Month 24 analysis stops follow-up at the Month 24 visit, whereas the other infection analysis uses all available follow-up through the maximum Month 48 visit. The resulting HRs therefore summarize different periods of observation.
Why does unblinding matter for dropout analysis?
Dropout behavior can be considered in relation to the information available to participants and investigators. HVTN 505 was unblinded on April 23, 2013, and the registry separately defines dropout before and after that date. The separate Cox analyses preserve that distinction.
Why does the analysis population matter?
The infection analyses use a Modified Intent-to-Treat Cohort, whereas some dropout analyses use the MITT population and the post-unblinding analysis further restricts participants to those who were HIV-uninfected and on-study as of April 23, 2013. An HR is meaningful only in the population and time window to which it applies.
16. Confidence Intervals and the Null Value
For a hazard ratio, the conventional null value is 1. An HR of 1 represents equal modeled hazards between vaccine and placebo at the level represented by the treatment coefficient.
| Endpoint | HR | 95% CI | Relationship to HR = 1 |
|---|---|---|---|
| Dropout through Month 48 | 0.73 | 0.63–0.84 | Interval entirely below 1 |
| Dropout prior to unblinding | 0.77 | 0.59–1 | Upper boundary reaches 1 |
| Dropout after unblinding | 0.71 | 0.60–0.84 | Interval entirely below 1 |
| HIV-1 infection through maximum Month 48 follow-up | 1.02 | 0.73–1.42 | Interval spans 1 |
| HIV-1 infection through Month 24 | 1.06 | 0.71–1.58 | Interval spans 1 |
This table illustrates why a point estimate should never be interpreted without its uncertainty interval. The dropout estimates are below 1, while both HIV-1 infection estimates have confidence intervals spanning the null value.
17. Intention-to-Treat and Modified Intention-to-Treat Analysis
Randomized clinical trials derive an important statistical advantage from maintaining treatment assignment as the basis for comparison. The HVTN 505 registry identifies the MITT population for the primary HIV-1 infection analyses and for the overall Month 48 dropout analysis.
Why analyze by assignment?
Randomization establishes the treatment groups before outcome information is observed. Preserving assignment in the analysis helps retain that randomized comparison rather than replacing it with a comparison based on later treatment behavior.
Why the "modified" population matters
MITT is not necessarily identical to an all-randomized ITT population. The exact definition must come from the trial's protocol or statistical-analysis documentation. The ClinicalTrials.gov record identifies the population but does not provide a more detailed definition.
The post-unblinding dropout analysis illustrates an additional principle: analysis populations can be defined according to the state of participants at a specific calendar time. Those restrictions can be statistically appropriate for the question being asked, but they mean the estimate should not be generalized automatically to every randomized participant.
18. Censoring and Time-to-Event Endpoints
The HIV-1 infection endpoints contain explicit censoring rules. For the maximum Month 48 analysis, participants are censored if they drop out early or complete the trial, while participants who remain uninfected are censored at their latest specimen collection with a negative HIV-1 test.
Here, di represents events at a particular event time and ni represents participants at risk immediately before that time.
Although the registry-reported formal-analysis records identify Cox regression rather than a Kaplan-Meier analysis field, Kaplan-Meier estimation is the standard descriptive companion to Cox analysis for time-to-event outcomes. It can show how the estimated event-free proportion changes over follow-up, whereas the Cox model summarizes the relative hazard associated with treatment assignment.
19. Covariate Adjustment in the Month 48 Infection Analysis
The maximum Month 48 HIV-1 infection analysis adjusted the Cox model for five listed covariate terms: age, behavioral risk score, square of behavioral risk score, race, and BMI.
| Adjustment variable | Role in reported model |
|---|---|
| Age | Covariate adjustment |
| Behavioral risk score | Covariate adjustment |
| Square of behavioral risk score | Allows a nonlinear relationship with the modeled hazard |
| Race | Covariate adjustment |
| BMI | Covariate adjustment |
The inclusion of the squared behavioral risk score is particularly instructive. If a continuous predictor has a nonlinear association with outcome, adding its square allows the fitted relationship to bend rather than requiring a straight-line effect on the model's log-hazard scale.
20. What the HIV-1 Infection Hazard Ratios Do — and Do Not — Mean
The maximum Month 48 infection analysis produced an HR of 1.02. In the reported Cox model, this is the estimated vaccine-to-placebo hazard ratio for HIV-1 infection. It is close to the null value of 1.
It does not mean that 1.02% of participants became infected, that infection probability increased by exactly 2%, or that every participant experienced the same relative effect.
The 95% CI of 0.73–1.42 indicates uncertainty around the estimated hazard ratio. Because the interval crosses 1, the estimate is compatible with both a lower and a higher modeled infection hazard for vaccine relative to placebo.
The reported P = 0.903 addresses statistical evidence against the null treatment association. It does not tell us the probability that the vaccine and placebo are equivalent, nor does it quantify how clinically meaningful an HR of 1.02 would be.
21. Multiplicity and Multiple Primary Endpoints
HVTN 505 has eight registered primary endpoints, while five formal statistical analyses are included in the ClinicalTrials.gov record. This creates an important distinction between the complete endpoint registry and the subset for which formal model-based results are reported.
| Endpoint group | Number registered | Formal statistical analysis reported |
|---|---|---|
| Dropout endpoints | 3 | 3 |
| HIV-1 infection endpoints | 2 | 2 |
| Reactogenicity endpoints | 3 | 0 |
| Total | 8 | 5 |
Multiplicity is relevant whenever several hypotheses or endpoints are evaluated. The ClinicalTrials.gov record identifies the number of registered endpoints and the five posted formal analyses, but it does not provide an alpha-allocation or multiplicity-adjustment strategy. Consequently, this page does not assign a familywise-error interpretation to the reported P-values beyond the information explicitly reported.
22. Blinding, Randomization, and Causal Interpretation
Randomization and double masking are central design features of HVTN 505. Random assignment creates the basis for comparing outcomes across treatment groups, while masking can reduce the influence of knowledge of assignment on participant behavior, investigator behavior, assessment, and follow-up.
Randomization
The registry classifies allocation as RANDOMIZED. The treatment comparison is therefore defined by assigned intervention rather than by an observational exposure comparison.
Double masking
The registry classifies masking as DOUBLE. The ClinicalTrials.gov record does not specify the exact parties included in the masking designation.
Parallel design
The registry classifies the design model as PARALLEL, meaning the two randomized groups form separate treatment pathways rather than sequential crossover periods.
Prevention purpose
The primary purpose is registered as PREVENTION, with HIV-1 infection assessed as a time-to-event endpoint.
23. Missing Data, Censoring, and What Is Not Specified
The registry definitions provide explicit censoring rules for HIV-1 infection and dropout. They do not provide a separate missing-data or imputation method in the ClinicalTrials.gov record.
- HIV-1 infection: participants remaining uninfected are censored according to the registered specimen-based rules.
- Early dropout: participants can be censored in the maximum Month 48 HIV-1 infection analysis if they drop out early.
- Trial completion: participants can be censored when they complete the trial without an HIV-1 infection event.
- Formal imputation: no imputation method is identified in the ClinicalTrials.gov record.
- Reactogenicity: participant-level maximum grades are used over the specified vaccination-assessment period, but no formal inferential analysis is reported in the ClinicalTrials.gov record.
This distinction is important because censoring is part of the definition of a time-to-event analysis, whereas imputation is a separate strategy for handling missing observations. One should not assume that censoring rules imply a particular missing-data mechanism or imputation procedure.
24. Important Limitations and Interpretation Issues
- Analysis populations differ: the five formal analyses do not all use exactly the same analysis population. The post-unblinding dropout analysis has an additional restriction to participants who were HIV-uninfected and on-study as of April 23, 2013.
- Different follow-up windows: the HIV-1 infection analyses cover through Month 24 and through the maximum Month 48 follow-up, respectively. These are distinct time-to-event estimands.
- Unblinding: dropout is analyzed separately before and after April 23, 2013, because the trial's masking state changed at that date.
- Hazard-ratio assumptions: Cox HRs are model-based and rely on the proportional-hazards framework. The ClinicalTrials.gov record does not report a separate diagnostic for that assumption.
- Confidence intervals: a confidence interval describes uncertainty around a population-level parameter; it is not a prediction interval for individual participants.
- Multiplicity: eight primary endpoints are registered, but the ClinicalTrials.gov record does not specify an overall multiplicity-control strategy.
- Reactogenicity analyses: results were posted for the three reactogenicity endpoints, but the ClinicalTrials.gov record does not include formal statistical-analysis records for them.
- Missing-data methods: the ClinicalTrials.gov record does not specify an imputation procedure.
- Generalizability: the registered population is specifically limited to HIV-uninfected, circumcised men and male-to-female transgender persons who have sex with men; the results should therefore be interpreted in the context of that enrolled population.
25. Why This Trial Matters Statistically
HVTN 505 is a useful teaching case because it combines randomized prevention-trial design with several forms of time-to-event analysis. The same trial includes infection endpoints, dropout endpoints, a change in masking status, covariate-adjusted Cox regression, and participant-level reactogenicity definitions.
| Concept | How it appears in HVTN 505 |
|---|---|
| Randomization | The registry classifies the allocation as RANDOMIZED. |
| Double masking | The registry classifies masking as DOUBLE. |
| Parallel design | The registry classifies the design model as PARALLEL. |
| Time-to-event endpoint | HIV-1 infection is defined as an event with explicit censoring rules. |
| Cox proportional-hazards model | All five registry-reported formal statistical analyses use Cox regression. |
| Hazard ratio | The primary effect measure in the five registry-reported formal analyses. |
| Covariate adjustment | The maximum Month 48 infection analysis adjusts for age, behavioral risk score, its square, race, and BMI. |
| Modified intention-to-treat | The HIV-1 infection analyses use a Modified Intent-to-Treat Cohort. |
| Censoring | Participants remaining uninfected are censored according to registered specimen and follow-up rules. |
| Changing masking status | Dropout is separately analyzed before and after April 23, 2013, the reported unblinding date. |
| Repeated safety assessments | Reactogenicity endpoints use each participant's maximum observed grade over vaccinations and post-vaccination assessments. |
| Multiplicity | Eight primary endpoints are registered, while five formal statistical analyses are reported. |
26. Related Tutorials
Learn more about the methods used in this trial:
27. Related Statistical Calculators
28. Sources
- ClinicalTrials.gov: HVTN 505, NCT00865566.
- PubMed: PMID 20636279.
- PubMed: PMID 24099601.
Continue through the Clinical Biostats statistical pathway
Use the methods in this trial as a starting point for deeper study of randomization, survival analysis, Cox regression, hazard ratios, confidence intervals, and clinical-trial interpretation.
29. Record Summary
HVTN 505 provides a detailed example of statistical analysis in a randomized HIV-1 prevention trial. The registry reports eight primary endpoints, including HIV-1 infection, participant dropout, and reactogenicity. Five formal primary analyses use Cox proportional-hazards models with hazard ratios as the effect measure. The reported HIV-1 infection analyses produced HR estimates of 1.02 through the maximum Month 48 follow-up and 1.06 through Month 24, while the three reported dropout analyses produced HR estimates of 0.73, 0.77, and 0.71 for their respective follow-up periods.
The most useful statistical interpretation combines the point estimate with its confidence interval, P-value, analysis population, censoring rules, follow-up period, and model specification. In particular, the HIV-1 infection confidence intervals span the null hazard ratio of 1, while the dropout estimates are below 1. These estimates answer distinct questions and should not be interpreted as one composite treatment effect.