← Clinical Trials
Tuberculosis Prevention Phase 3 Completed NCT01404312

ACTG A5279: Complete Statistical Analysis of Rifapentine-Isoniazid in TB Prevention

An independent statistical analysis of the randomized phase 3 ACTG A5279 trial comparing a rifapentine-plus-isoniazid regimen with an isoniazid regimen for tuberculosis prevention in participants with HIV infection.

BRIEF TB  ·  Randomized parallel-group design  ·  Prevention
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results are restricted to the ClinicalTrials.gov trial data posted on ClinicalTrials.gov for ACTG A5279. Where the ClinicalTrials.gov record does not provide a result or design detail, that information is not reconstructed from external publications.

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

ACTG A5279, also known as the Brief Rifapentine-Isoniazid Evaluation for TB Prevention (BRIEF TB), was a randomized phase 3 prevention trial comparing two tuberculosis-prevention regimens in a population with HIV infection. The registry reports 3000 enrolled participants, two arms, and a primary endpoint based on the incidence of active tuberculosis or specified deaths.

3000
Enrolled
Registry enrollment
2
Arms
Parallel design
3
Phase
Phase 3
1
Primary endpoint
Count / rate
FeatureACTG A5279
Brief titleBrief Rifapentine-Isoniazid Evaluation for TB Prevention (BRIEF TB)
PhasePhase 3
StatusCOMPLETED
ConditionsTuberculosis; HIV Infections
AllocationRANDOMIZED
Design modelPARALLEL
MaskingNONE
Primary purposePREVENTION
Enrollment3000.0
Arms2
Start2012-05-23
Primary completion2017-11-14
Lead sponsorNational Institute of Allergy and Infectious Diseases (NIAID)
Sponsor typeNIH
ClinicalTrials.govNCT01404312

2. Clinical Question

The statistical question was whether the rifapentine-plus-isoniazid regimen could be shown to be non-inferior to the isoniazid regimen for the registered primary endpoint, while the registry also reports several secondary endpoints under superiority hypotheses.

Population

Participants enrolled in a phase 3 tuberculosis-prevention trial involving the conditions tuberculosis and HIV infections. The ClinicalTrials.gov record does not provide a more detailed baseline population description.

Intervention

Rifapentine (RPT) plus isoniazid (INH), with pyridoxine (Vitamin B6) listed as a dietary supplement.

Comparator

Isoniazid (INH) regimen.

Primary question

Is the incidence rate for the composite tuberculosis/death endpoint sufficiently close to the comparator rate that the RPT-plus-INH regimen satisfies the prespecified non-inferiority criterion?

3. Trial Design

01
Randomize 3000.0 enrolled
02
Arm A RPT Plus INH
03
Arm B INH regimen
04
Follow Primary endpoint through follow-up
05
Analyze Rate and secondary outcomes
ARM A

RPT Plus INH Regimen

  • Rifapentine (RPT)
  • Isoniazid (INH)
  • Pyridoxine (Vitamin B6)
ARM B

INH Regimen

  • Isoniazid (INH)
  • Pyridoxine (Vitamin B6) is listed among the trial interventions

The ClinicalTrials.gov record identifies the study as randomized, parallel-group, and unmasked. It does not report a factorial design, crossover procedure, or treatment switching. Those features therefore are not incorporated into the statistical interpretation.

4. Enrollment, Timing, and Analysis Populations

CharacteristicRegistry information
Enrollment3000.0
Start date2012-05-23
Primary completion2017-11-14
Primary efficacy analysis populationAll participants who started study treatment
Secondary safety analysis populationAll participants who started study treatment

The distinction between the enrollment figure and the efficacy analysis population is important. The primary analysis is explicitly defined as including all participants who started study treatment. The ClinicalTrials.gov record does not provide a separate intention-to-treat count, per-protocol count, or modified intention-to-treat count.

5. Primary Endpoint

EndpointRegistry definition / time frameEndpoint typeAnalysis
Incidence of First Diagnosis of Active Tuberculosis, Death Related to Tuberculosis, or Death From Unknown Cause From entry to occurrence of event, up to end of follow-up 3 years after last participant enrolled (median follow-up time: 3.3 years) Count / rate Cochran-Mantel-Haenszel test

The registry defines the endpoint as an incidence rate measured in events per 100 person-years. The registry-reported definition states that the incidence rate was estimated and that a 95.1% confidence interval was used to account for interim analysis of the primary efficacy outcome.

6. Primary Result: Tuberculosis or Specified Death Endpoint

Incidence rate difference

-0.0231

95.1% CI: -0.346 to 0.300   ·   two-sided

Effect measure: Arm A incidence rate − Arm B incidence rate

The reported estimate is the incidence rate in the RPT-plus-INH arm minus the incidence rate in the INH arm. Because the estimate is negative, the observed standardized incidence rate was lower in Arm A under the direction specified by the registry analysis notes.

Primary analysis featureReported value
Analysis populationAll participants who started study treatment
Groups comparedRPT Plus INH Regimen (Arm A) vs INH Regimen (Arm B)
MethodCochran-Mantel-Haenszel test
Effect measureIncidence Rate Difference
Estimate-0.0231
Confidence interval95.1% CI -0.346 to 0.300
HypothesisNon-inferiority
Non-inferiority margin1.25 events per 100 person-years

Clinical Biostats interpretation

What the estimate means

The incidence rate difference of -0.0231 events per 100 person-years means that the estimated incidence rate in Arm A minus the estimated incidence rate in Arm B was -0.0231 events per 100 person-years. The registry explicitly defines a negative estimate as favoring Arm A.

It does not mean that the probability of tuberculosis or death was reduced by 0.0231, nor does it mean that 0.0231% of participants were prevented from experiencing the endpoint. The measure is a difference in incidence rates expressed per 100 person-years.

What the confidence interval says

The 95.1% confidence interval extends from -0.346 to 0.300 events per 100 person-years. This interval describes statistical uncertainty around the estimated incidence-rate difference. It includes zero, so the interval is compatible with a small rate difference in either direction under an ordinary difference-in-rates interpretation.

For a non-inferiority trial, however, the relevant question is not simply whether the interval excludes zero. The key comparison is whether the unfavorable side of the interval crosses the prespecified non-inferiority margin.

Why the non-inferiority margin matters

The registry specifies a non-inferiority margin of 1.25 events per 100 person-years. The upper confidence limit of 0.300 is below that margin. Thus, using the reported estimate and confidence interval together with the registry's stated non-inferiority framework, the reported interval is consistent with satisfying the prespecified non-inferiority criterion.

The registry also states that sample-size determination assumed a primary endpoint rate of 2.0/100 person-years, used a one-sided 0.025 alpha level, and targeted at least 90% power. The resulting required sample size was approximately 2500.

Do not interpret this as a superiority claim. The primary hypothesis is explicitly labeled non-inferiority. A negative point estimate favors Arm A according to the registry's direction convention, but the non-inferiority analysis is answering whether the difference is sufficiently far from the prespecified unfavorable margin. It is a different statistical question from demonstrating superiority.

7. Secondary Endpoint Results

The registry contains five secondary statistical analyses in addition to the primary analysis. These address serious adverse events, targeted adverse events, toxicity-related drug management, non-TB deaths, and deaths from any cause.

Serious Adverse Events

Risk difference for any serious adverse event

-0.016

95% CI: -0.035 to 0.002   ·   P = 0.073

Fisher exact test; superiority hypothesis

ArmParticipants with serious adverse eventsParticipants at risk
RPT Plus INH Regimen (Arm A)401496
INH Regimen (Arm B)681504

The reported risk difference is defined as the proportion in Arm A minus the proportion in Arm B. The estimate of -0.016 therefore describes a lower observed proportion with an SAE in Arm A under that direction convention. The two-sided 95% confidence interval ranges from -0.035 to 0.002, and the reported superiority p-value is 0.073.

Targeted Adverse Events

Risk difference for any targeted adverse event

-.0058

95% CI: -0.019 to 0.007   ·   P = 0.405

Fisher exact test; superiority hypothesis

The registry defines the estimate as the proportion of participants with a targeted adverse event in Arm A minus the corresponding proportion in Arm B. The confidence interval crosses zero, and the reported two-sided p-value is 0.405.

Toxicity-Related Study Drug Management

Proportional-odds analysis

OR 2.093

95% CI: 1.315 to 3.332   ·   two-sided

Proportional odds (ordinal) regression; superiority hypothesis

The endpoint records the number of participants in each category of an ordered categorical variable indicating the most stringent level of study-drug management due to toxicity required during treatment. The registry describes the analysis as a proportional-odds model with covariate adjustment.

The analysis note states that the odds ratio is expressed as the odds of being in a higher category of toxicity-related management for Arm B compared with Arm A. Under that reported direction, an odds ratio of 2.093 indicates higher modeled odds of being in a more stringent management category for Arm B relative to Arm A.

Multiplicity: the registry analysis notes explicitly state that this analysis was not adjusted for multiple comparisons. The confidence interval should therefore be interpreted as the interval reported for this individual analysis, rather than as evidence that a separate familywise error adjustment was performed across the secondary endpoints.

Death Due to a Non-TB Event

Fine-Gray competing-risk hazard ratio

1.396

95% CI: 0.762 to 2.559   ·   P = 0.2802

Fine-Gray model; two-sided superiority hypothesis

This time-to-event analysis examined cumulative incidence of death due to a non-TB event. The registry specifies a Fine-Gray competing-risk analysis, treating TB-related deaths as competing risks and other deaths, including deaths of unknown cause, as the event of interest.

The registry defines the hazard ratio as Arm B hazard divided by Arm A hazard, so an HR greater than 1 favors Arm A according to the analysis notes. The point estimate of 1.396 therefore points in that direction, but the 95% confidence interval is wide and includes 1. The reported two-sided p-value is 0.2802.

Death From Any Cause

Overall survival comparison

P = 0.3078

Log-rank test; two-sided superiority hypothesis

The registry defines this endpoint as the cumulative incidence of death from any cause, measured from entry to occurrence of the event through the stated follow-up period. The analysis used a log-rank test, with the null hypothesis that the survival curve in Arm A equals the survival curve in Arm B.

Secondary endpointMethodEffect measure / result
Participants with one or more serious adverse events Fisher exact test RD -0.016; 95% CI -0.035 to 0.002; P = 0.073
Participants with a targeted adverse event Fisher exact test RD -.0058; 95% CI -0.019 to 0.007; P = 0.405
Ordered toxicity-related drug-management category Proportional odds regression OR 2.093; 95% CI 1.315 to 3.332
Cumulative incidence of death due to a non-TB event Fine-Gray competing-risks model HR 1.396; 95% CI 0.762 to 2.559; P = 0.2802
Cumulative incidence of death from any cause Log-rank test P = 0.3078

8. Safety Results

The ClinicalTrials.gov record provides serious adverse event counts by treatment arm. Both the count and the denominator are reported directly in the trial data.

Safety measureRPT Plus INH RegimenINH Regimen
Participants with one or more serious adverse events40 / 149668 / 1504
AnalysisFisher exact test
Risk difference-0.016 (95% CI -0.035 to 0.002)
P-value0.073

The serious-adverse-event analysis asks whether the proportion of participants experiencing any SAE differed between the two randomized regimens. Fisher's exact test was used, and the registry specifies a superiority hypothesis for this endpoint.

Safety interpretation

The observed SAE counts were 40 of 1496 in Arm A and 68 of 1504 in Arm B. The reported risk difference of -0.016 is the Arm A proportion minus the Arm B proportion. The confidence interval extends from -0.035 to 0.002, so the reported uncertainty includes both a modest lower risk in Arm A and a small difference in the opposite direction.

The p-value of 0.073 is not a measure of the magnitude or clinical importance of the safety difference. It describes the compatibility of the observed data with the specified null hypothesis under the test procedure.

9. Statistical Methodology

Cochran-Mantel-Haenszel analysis

The primary analysis is reported as a Cochran-Mantel-Haenszel test. The analysis notes state that the Mantel-Haenszel method was used for estimating the standardized incidence rate in each arm and the incidence rate difference.

Primary effect measure
Incidence Rate Difference = Incidence Rate in Arm A − Incidence Rate in Arm B

The registry reports the resulting estimate in events per 100 person-years. Negative values favor Arm A according to the registry-reported analysis notes.

Fisher exact test

Fisher's exact test was used for the two binary adverse-event endpoints. The method is designed for comparing categorical outcomes between two groups and is particularly useful when exact inference is desired rather than relying on a large-sample approximation.

Proportional-odds regression

The toxicity-management endpoint is ordered categorical rather than simply a two-level outcome. Proportional-odds regression uses the ordering of the categories and models the odds of being at or above progressively higher levels. The registry reports an odds ratio and identifies covariate adjustment as an analysis concept.

Interpretation of the reported odds ratio
OR = Odds of being in a higher category for Arm B compared with Arm A

The analysis note gives this direction explicitly. An OR above 1 therefore corresponds to higher modeled odds of being in a more stringent toxicity-management category for Arm B.

Fine-Gray competing-risks model

For non-TB deaths, the registry specifies a Fine-Gray competing-risk analysis. This is important because a participant who experiences a competing event can no longer experience the event of interest in the same way as a participant who remains event-free.

The registry specifically treats TB-related deaths as competing risks, while other deaths, including deaths of unknown cause, are the event of interest. The reported hazard ratio is defined as Arm B hazard divided by Arm A hazard.

Log-rank test

The all-cause mortality endpoint uses the log-rank test to compare survival curves between the two randomized groups. This is a time-to-event method that incorporates the timing of events and accommodates right-censoring.

Kaplan-Meier estimation and time-to-event endpoints

Kaplan-Meier estimation is the standard descriptive framework for displaying survival over time when right-censored observations are present. The ClinicalTrials.gov record identifies death from any cause as a time-to-event endpoint and specifies a log-rank comparison, but it does not separately list Kaplan-Meier estimation as a posted statistical-analysis method for this endpoint.

10. Non-Inferiority Design

The primary endpoint is explicitly analyzed under a non-inferiority hypothesis. The registry specifies a non-inferiority margin of 1.25 events per 100 person-years.

The margin

The margin defines how much worse the RPT-plus-INH regimen could be, on the incidence-rate-difference scale, while still satisfying the trial's non-inferiority criterion.

The observed interval

The reported 95.1% confidence interval runs from -0.346 to 0.300 events per 100 person-years, with the upper limit below the 1.25-event margin.

Design assumptions

Sample-size determination assumed a primary endpoint rate of 2.0/100 person-years, a one-sided 0.025 alpha level, and at least 90% power.

Required enrollment

The registry states that these assumptions required a sample size of approximately 2500; the trial's reported enrollment was 3000.0.

Why the confidence interval is central

Non-inferiority is fundamentally about ruling out an unacceptable loss of efficacy, not necessarily about proving that the two treatments have identical effects. For an incidence-rate difference defined as Arm A minus Arm B, the unfavorable direction is a sufficiently positive difference. The prespecified margin establishes the boundary that the confidence interval must not cross under the trial's non-inferiority framework.

Here, the reported upper confidence limit is 0.300, while the non-inferiority margin is 1.25 events per 100 person-years. The upper limit therefore remains below the stated margin.

Non-inferiority is not equivalence. A non-inferiority result does not establish that the two regimens have exactly the same incidence rate. It establishes the conclusion permitted by the prespecified non-inferiority margin and analysis framework.

11. Interim Analysis and Confidence-Interval Level

The primary endpoint definition states that a 95.1% confidence interval was used to account for an interim analysis of the primary efficacy outcome. The registry therefore provides an important clue about the relationship between interim monitoring and the final inferential threshold.

Primary inferential quantities
95.1% two-sided CI  ·  one-sided 0.025 alpha for sample-size determination

The confidence interval level is reported by the registry as 95.1%. The ClinicalTrials.gov record does not provide the complete interim-monitoring schedule, stopping boundary, or alpha-spending function, so those details are not inferred here.

This distinction matters because a trial that permits an interim efficacy assessment cannot generally be interpreted as though the data were examined only once at the end. Adjusting the confidence level is one way a prespecified design can account for interim examination of the primary outcome.

12. Multiplicity

The trial has one registered primary endpoint and multiple secondary endpoints. The ClinicalTrials.gov record explicitly states that the proportional-odds toxicity-management analysis was not adjusted for multiple comparisons.

Analysis familyRoleMultiplicity information in the ClinicalTrials.gov record
Active TB / TB-related or unknown-cause death endpointPrimary95.1% CI used to account for interim analysis
Serious adverse eventsSecondaryNo multiple-comparison adjustment reported in the ClinicalTrials.gov record
Targeted adverse eventsSecondaryNo multiple-comparison adjustment reported in the ClinicalTrials.gov record
Toxicity-related drug managementSecondaryAnalysis note explicitly says not adjusted for multiple comparisons
Non-TB deathSecondaryNo multiple-comparison adjustment reported in the ClinicalTrials.gov record
All-cause deathSecondaryNo multiple-comparison adjustment reported in the ClinicalTrials.gov record

A nominal p-value from a secondary analysis should therefore not automatically be interpreted as though the trial had allocated a separate familywise error budget to that endpoint. The registry's explicit note concerning the toxicity-management analysis is especially important when interpreting its confidence interval and p-value.

13. Statistical Methods Explained

Why is the primary endpoint expressed per 100 person-years?

The primary endpoint is an incidence-rate outcome. Person-time allows participants to contribute follow-up even when their observation periods differ. Expressing the rate per 100 person-years provides a common scale for comparing the two randomized groups.

Why is the primary analysis a non-inferiority analysis?

The registry identifies the primary hypothesis as non-inferiority and provides a specific margin of 1.25 events per 100 person-years. The statistical objective is therefore to determine whether the RPT-plus-INH incidence rate is not unacceptably higher than the comparator rate according to that prespecified boundary.

Why does the confidence interval cross zero without preventing non-inferiority?

The estimate is a difference, so zero represents no rate difference. The confidence interval extends across zero, meaning the data are compatible with both a small negative and a small positive difference. Non-inferiority uses a different boundary: the prespecified unfavorable margin of 1.25 events per 100 person-years. The entire reported interval remains below that margin.

What does the odds ratio of 2.093 mean?

For the ordered toxicity-management endpoint, the registry defines the odds ratio as the odds of being in a higher category for Arm B compared with Arm A. An OR of 2.093 therefore represents higher modeled odds of being in a more stringent category for Arm B under that definition. It is not a risk ratio and should not be read as saying that the probability of an event was 2.093 times higher.

Why was Fisher's exact test used for adverse events?

The two adverse-event endpoints are binary comparisons between treatment groups. Fisher's exact test provides an exact test of the association between treatment assignment and the binary outcome rather than relying on a large-sample approximation.

Why is a Fine-Gray model used for non-TB death?

Non-TB death is analyzed in the presence of competing risks. A TB-related death changes the participant's ability to experience the non-TB death endpoint. The Fine-Gray framework directly addresses the cumulative-incidence setting specified in the registry analysis.

Why does a p-value not measure effect size?

A p-value describes how compatible the observed data are with a specified null hypothesis under the statistical test. It does not quantify how large or clinically important an effect is. The effect estimate and its confidence interval are needed to understand magnitude and precision.

14. Understanding the Primary Estimate

Estimate

The incidence-rate difference of -0.0231 events per 100 person-years is a standardized difference between the two randomized groups. The negative sign indicates that the estimated rate was lower in Arm A under the registry's stated direction convention.

Precision

The 95.1% confidence interval of -0.346 to 0.300 events per 100 person-years shows that the point estimate is not known exactly. The width of the interval reflects uncertainty in estimating the underlying difference from the trial data.

Non-inferiority logic

The relevant comparison is between the upper confidence limit, 0.300, and the non-inferiority margin, 1.25 events per 100 person-years. Because the upper limit is below the margin, the reported result is consistent with the registry's non-inferiority framework.

What the result does not mean

The result does not establish that the two regimens have identical risks, does not provide an individual patient's probability of developing tuberculosis, and does not convert the incidence-rate difference into a percentage reduction in risk.

15. Interpreting the Secondary Time-to-Event Analyses

Fine-Gray hazard ratio

The non-TB death analysis reports an HR of 1.396, with a 95% CI of 0.762 to 2.559 and P = 0.2802. Because the registry defines the ratio as Arm B hazard divided by Arm A hazard, values above 1 favor Arm A according to the registry-reported analysis convention.

The confidence interval is broad relative to the point estimate and crosses 1. The p-value is 0.2802. These quantities should be considered together rather than treating the point estimate alone as a definitive description of the treatment difference.

Log-rank comparison of all-cause death

The all-cause death endpoint was analyzed using the log-rank test, with the null hypothesis that the survival curve in Arm A equals the survival curve in Arm B. The reported p-value was 0.3078.

No median survival is reported in the ClinicalTrials.gov record. The ClinicalTrials.gov record provides the log-rank p-value but does not provide median survival estimates, survival probabilities at selected time points, or a Kaplan-Meier event table. Those quantities are therefore not added here.

16. Covariate Adjustment and Stratification

The ClinicalTrials.gov record identifies covariate adjustment as an analysis concept associated with the proportional-odds regression for toxicity-related drug management. It does not provide the covariates or their parameter estimates.

The primary analysis is described as a Mantel-Haenszel analysis used to estimate standardized incidence rates and their difference. The ClinicalTrials.gov record does not identify specific stratification variables. Consequently, no stratification factors are asserted for ACTG A5279 beyond the standardized-analysis description provided by the registry.

17. Censoring and Time-to-Event Interpretation

Several endpoints are defined from entry to the occurrence of an event during follow-up. Time-to-event analyses differ from simple binary analyses because both the occurrence and timing of events contribute information, and participants who have not experienced the event by the end of observation can contribute censored follow-up.

All-cause death

Analyzed with a log-rank test comparing the survival curves between Arm A and Arm B.

Non-TB death

Analyzed with a Fine-Gray competing-risks model because TB-related deaths are treated as competing risks.

Primary endpoint

Reported as an incidence rate in events per 100 person-years rather than as a median event time.

Follow-up

The registered primary endpoint is followed through 3 years after the last participant was enrolled.

The ClinicalTrials.gov record does not specify a particular censoring rule, missing-data imputation method, or sensitivity analysis for censoring. These topics are therefore not assigned a trial-specific method beyond the endpoint and analysis descriptions actually reported.

18. Safety and Efficacy Use Different Statistical Questions

One useful feature of ACTG A5279 as a statistical teaching example is the contrast between its primary non-inferiority analysis and its secondary superiority analyses.

QuestionEndpointEffect / testHypothesis
Is the incidence rate acceptably close to the comparator?Active TB / specified death endpointIncidence rate differenceNon-inferiority
Do SAE proportions differ?Any serious adverse eventRisk difference; Fisher exactSuperiority
Do targeted AE proportions differ?Any targeted adverse eventRisk difference; Fisher exactSuperiority
Do toxicity-management categories differ?Ordered toxicity-management variableOdds ratio; proportional odds regressionSuperiority
Do non-TB death rates differ in competing-risk analysis?Non-TB deathHazard ratio; Fine-Gray modelSuperiority
Do all-cause survival curves differ?Death from any causeLog-rank p-valueSuperiority

These are not interchangeable questions. A non-inferiority incidence-rate analysis, a binary risk-difference analysis, an ordinal regression, and competing-risk survival analysis each target a different statistical estimand.

19. Limitations

20. Why This Trial Matters Statistically

ACTG A5279 is a useful statistical case study because it places several distinct clinical-trial methods within the same randomized comparison. The primary question is non-inferiority on an incidence-rate scale, while secondary analyses move across binary outcomes, ordinal regression, competing risks, and conventional time-to-event comparison.

ConceptHow it appears in ACTG A5279
RandomizationRandomized, parallel-group phase 3 design
Non-inferiorityPrimary hypothesis with a margin of 1.25 events per 100 person-years
Incidence ratesPrimary endpoint expressed as events per 100 person-years
Cochran-Mantel-HaenszelPrimary analysis method for standardized incidence rate and incidence rate difference
Confidence intervals95.1% CI for the primary efficacy outcome and 95% CIs for secondary effect estimates
Fisher exact testBinary serious-adverse-event and targeted-adverse-event comparisons
Risk differenceSecondary binary safety effect measure
Proportional odds regressionOrdered toxicity-related drug-management outcome
Fine-Gray modelNon-TB death with TB-related deaths as competing risks
Log-rank testAll-cause death time-to-event comparison
Interim analysis95.1% CI used for the primary efficacy outcome to account for interim analysis
MultiplicityToxicity-management analysis explicitly reported as not adjusted for multiple comparisons

21. A Practical Reading Strategy for This Trial

A statistically disciplined reading of ACTG A5279 starts with the estimand rather than the p-value.

  1. Identify the endpoint. The primary outcome is an incidence-rate endpoint combining first active tuberculosis diagnosis with specified deaths.
  2. Identify the effect measure. The primary comparison is an incidence-rate difference in events per 100 person-years.
  3. Check the direction. The estimate is Arm A minus Arm B, with a negative value favoring Arm A.
  4. Find the non-inferiority margin. The prespecified margin is 1.25 events per 100 person-years.
  5. Read the confidence interval against the margin. The upper limit is 0.300, below the 1.25 margin.
  6. Separate secondary endpoints. Their hypotheses are labeled superiority and use different estimands and methods.
  7. Account for competing risks. The non-TB death analysis specifically treats TB-related deaths as competing events.
  8. Check multiplicity. At least one secondary analysis is explicitly reported without adjustment for multiple comparisons.

22. What the P-Values Do — and Do Not — Mean

Primary endpoint

The registry-reported primary result is framed as non-inferiority, so the central inferential comparison is the confidence interval with the 1.25 events-per-100-person-years margin rather than a conventional superiority p-value.

Serious adverse events

P = 0.073 describes the Fisher-exact test under the stated equality null hypothesis. It does not measure the size of the SAE difference.

Targeted adverse events

P = 0.405 is the reported two-sided Fisher-exact result. It does not imply that the two proportions are mathematically identical.

Non-TB death

P = 0.2802 accompanies the Fine-Gray hazard-ratio analysis. It does not describe the size of the competing-risk effect by itself.

The same principle applies across all endpoints: a p-value is evidence relative to a null hypothesis and statistical model. It is not a universal measure of treatment effect, clinical importance, or precision.

23. Overall Statistical Interpretation

Primary endpoint

The primary incidence-rate difference was -0.0231 events per 100 person-years, with a 95.1% CI of -0.346 to 0.300. Because the upper confidence limit of 0.300 is below the prespecified non-inferiority margin of 1.25 events per 100 person-years, the reported result is consistent with the trial's stated non-inferiority criterion.

Safety endpoints

The reported SAE risk difference was -0.016, with a 95% CI of -0.035 to 0.002 and P = 0.073. The targeted-adverse-event risk difference was -.0058, with a 95% CI of -0.019 to 0.007 and P = 0.405.

Ordinal toxicity management

The proportional-odds analysis reported an OR of 2.093 with a 95% CI of 1.315 to 3.332, using the registry's convention of Arm B relative to Arm A for higher toxicity-management categories. The analysis note states that it was not adjusted for multiple comparisons.

Time-to-event secondary endpoints

The Fine-Gray analysis of non-TB death reported HR 1.396, 95% CI 0.762 to 2.559, P = 0.2802. The log-rank analysis of all-cause death reported P = 0.3078. These endpoints address different time-to-event questions and should not be collapsed into a single effect measure.

24. Related Tutorials

Learn more about the methods used in this trial:

25. Related Statistical Calculators

Explore calculators aligned with the main estimands and methods in this trial:

26. Sources

The numerical results and trial-design facts on this page are restricted to the ClinicalTrials.gov record. The PubMed records are provided as source links identified in that trial data; no additional numerical results from those publications have been incorporated.

Continue through the Clinical Biostats trial-analysis library

Use the related tutorials and calculators to study the estimands, tests, confidence intervals, and survival methods illustrated by this trial.

27. Record Summary

ACTG A5279 provides a compact but methodologically diverse example of randomized clinical-trial statistics. Its primary endpoint uses an incidence-rate difference and a prespecified non-inferiority margin, with a 95.1% confidence interval accounting for interim analysis. Secondary endpoints then illustrate Fisher's exact testing and risk differences for binary safety outcomes, proportional-odds regression for an ordered toxicity-management outcome, Fine-Gray competing-risk analysis for non-TB death, and a log-rank test for all-cause death.

The most important statistical distinction is between non-inferiority and superiority. The primary result is interpreted against a 1.25 events-per-100-person-years margin, whereas the secondary analyses are framed under superiority hypotheses. Reading the effect measure, direction convention, confidence interval, hypothesis type, and analysis population together provides a more informative interpretation than focusing on p-values alone.

Clinical Biostats methodology: A trial-results page should not merely repeat the registry record. The goal is to reconstruct the statistical story of the trial while clearly separating reported evidence from educational interpretation, preserving the original estimand and avoiding unsupported reconstruction of missing results.