This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
PURPOSE 2 is a randomized, double-blind, parallel phase 3 prevention trial evaluating HIV pre-exposure prophylaxis in people at risk for HIV infection. The registry reports an enrollment of 3292 participants, four arms, and two primary endpoints centered on HIV-1 incidence and comparison of lenacapavir with background HIV incidence.
| Feature | PURPOSE 2 |
|---|---|
| Trial name | PURPOSE 2 |
| NCT ID | NCT04925752 |
| Brief title | Study of Lenacapavir for HIV Pre-Exposure Prophylaxis in People Who Are at Risk for HIV Infection |
| Therapeutic area | Infectious Disease |
| Condition | Pre-Exposure Prophylaxis of HIV Infection |
| Phase | Phase 3 |
| Status | Active, not recruiting |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Prevention |
| Enrollment | 3292 |
| Arms | 4 |
| Lead sponsor | Gilead Sciences |
| Sponsor type | Industry |
2. Clinical Question
The statistical question is whether the HIV-1 incidence rate observed among participants receiving lenacapavir is lower than the registry-defined background HIV incidence rate, and whether HIV-1 incidence with lenacapavir is comparable to or lower than the incidence observed with F/TDF in the randomized blinded phase.
Population
People who are at risk for HIV infection and enrolled in the PURPOSE 2 phase 3 prevention trial.
Intervention
Lenacapavir, including the randomized blinded phase regimen described in the registry as LEN plus placebo-to-match F/TDF.
Comparator
For the randomized comparison, placebo LEN plus F/TDF. A separate primary comparison uses background HIV incidence estimated from the Incidence Phase.
Primary question
Is HIV-1 incidence with LEN lower than the registry-defined background HIV incidence, and does it meet the prespecified superiority criterion relative to background HIV?
The registry also reports a secondary comparison of HIV-1 incidence between LEN and F/TDF. That comparison uses both a rate-difference analysis and a Poisson-model rate-ratio analysis.
3. Trial Design
Trial arms and interventions
LEN + placebo-to-match F/TDF
- Sub-cutaneous lenacapavir
- Placebo-to-match F/TDF
Placebo LEN + F/TDF
- Placebo LEN
- F/TDF
The registry intervention list also identifies oral lenacapavir, placebo-to-match oral lenacapavir, and F/TAF for US participants only. The ClinicalTrials.gov record does not provide separate numerical results for each of those intervention descriptions, so this page does not infer additional arm-specific efficacy estimates.
4. Endpoints
The registry lists two primary endpoints and six posted outcome measures. The primary endpoint types are represented as count/rate and time-to-event in the ClinicalTrials.gov record. The formal statistical analyses posted for the primary endpoint use HIV-1 incidence per 100 person-years as the outcome unit.
| Endpoint | Time frame | Endpoint type | Registry status |
|---|---|---|---|
| Incidence Phase: Recent Infection Testing Algorithm (RITA) Estimate of the Background Human Immunodeficiency-1 Virus Infection Incidence Rate (bHIV) Per 100 Person Years (PY) | Incidence Phase Screening Visit (Day 1) | Count / rate | Primary; results posted |
| Randomized Blinded Phase: HIV-1 Incidence Reported Per 100 PY for LEN Compared to Background HIV (bHIV, Participants in All Screened Set) | Up to 149 weeks | Time-to-event | Primary; formal analysis posted |
| Randomized Blinded Phase: HIV-1 Incidence Reported Per 100 PY for LEN Compared to F/TDF | Up to 149 weeks | Count / rate | Secondary; analyses posted |
How the background HIV incidence was constructed
The registry definition states that bHIV per 100 person-years in the Incidence Phase was calculated using the Recent Infection Testing Algorithm, or RITA. The RITA incorporated HIV-1 testing results and recency assay testing results to estimate background HIV incidence. Recency assay testing was performed for participants in the All Screened Set found to have HIV-1 infection at the Incidence Phase Screening Visit.
How HIV-1 incidence was calculated in the randomized blinded phase
The registry states that HIV-1 incidence per 100 person-years for LEN was calculated as the number of participants who acquired HIV-1 divided by the relevant total duration of follow-up while participants were at risk. For participants not diagnosed with HIV-1, follow-up contributed through the available duration of follow-up; for participants diagnosed with HIV-1, follow-up contributed through confirmed HIV-1 diagnosis.
Person-years place participants with different amounts of follow-up on a common time scale. The resulting incidence measure is a rate, not simply the proportion of participants with an event.
5. Analysis Populations
The primary LEN-versus-background analyses were based on participants in the LEN group in the Full Analysis Set. The registry description states that the FAS included all randomized participants who received at least one dose of any study drug and had not been diagnosed with HIV-1.
| Analysis | Population | Role |
|---|---|---|
| LEN versus background HIV | Participants in the LEN group in the Full Analysis Set; background comparison used participants in the Incidence Phase with non-missing HIV-1 diagnosis | Primary efficacy comparison |
| LEN versus F/TDF | Full Analysis Set | Secondary randomized comparison |
This population structure matters because the primary comparison is not simply a conventional two-arm randomized comparison. One component is an estimated background incidence from the Incidence Phase, while the other is observed HIV-1 incidence during the randomized blinded phase.
6. Statistical Methodology
Wald testing
The primary formal analyses use a Wald test. A Wald test evaluates how far an estimated parameter is from a prespecified null value relative to its estimated standard error. In this trial, the null hypotheses are expressed in terms of the ratio of HIV-1 incidence with LEN to the comparison incidence.
For a rate ratio, the calculation is commonly performed on a transformed scale such as the logarithm of the rate ratio because the log scale is symmetric around a null value of zero.
Rate ratio
A rate ratio compares two incidence rates. A rate ratio of 1 represents equal rates. A value below 1 indicates a lower estimated incidence rate in the numerator group, while a value above 1 indicates a higher estimated incidence rate.
For example, a rate ratio of 0.043 corresponds to an estimated LEN incidence rate that is 4.3% of the comparison incidence rate, under the analysis framework used for the reported estimate.
Poisson regression
The secondary LEN-versus-F/TDF rate-ratio analysis used a Poisson model. Poisson regression is designed for count data and can incorporate person-time exposure when modeling event rates. Conceptually, the model relates the expected number of events to exposure time and treatment group.
The person-time component functions as an exposure or offset. The treatment coefficient can then be expressed as a rate ratio after exponentiation.
Confidence intervals
The reported primary confidence interval for the rate ratio versus background HIV is based on the delta method. The secondary Poisson-model confidence interval for the rate ratio versus F/TDF is derived from the Poisson model.
A 95% confidence interval quantifies statistical uncertainty around the estimated effect under the corresponding analysis framework. It does not mean that 95% of individual participants have effects within the interval, nor does it give the probability that the true parameter lies in the interval after the data have been observed.
Intention-to-treat principles
The analysis text identifies intention-to-treat analysis as a statistical concept for the posted comparisons. In a randomized trial, the intention-to-treat principle generally preserves assignment-based comparison rather than selectively analyzing participants according to outcomes or treatment changes. The specific Full Analysis Set definition posted on ClinicalTrials.gov for the primary analysis should nevertheless be distinguished from a generic statement that all randomized participants were analyzed.
7. Primary Results: LEN Versus Background HIV
The primary randomized blinded phase endpoint was HIV-1 incidence reported per 100 person-years for LEN compared with background HIV. The time frame was up to 149 weeks. Two formal primary hypotheses are posted for this same endpoint.
Primary Analysis 1: LEN incidence versus bHIV
Rate ratio for HIV-1 incidence
95% CI: 0.010–0.182 · P < 0.0001
Two-sided confidence interval; Wald test; superiority hypothesis.
| Element | Reported result |
|---|---|
| Outcome | HIV-1 incidence per 100 person-years |
| Comparison | LEN + placebo-to-match F/TDF vs Incidence Phase: All Screened Participants With Non-missing HIV-1 Diagnosis |
| Analysis population | LEN group in the Full Analysis Set |
| Effect measure | Rate ratio |
| Estimate | 0.043 |
| 95% CI | 0.010–0.182 |
| P-value | <0.0001 |
| Method | Wald test |
| Hypothesis | Superiority |
The rate ratio of 0.043 means that the estimated HIV-1 incidence rate in the LEN analysis group was 0.043 times the background incidence rate used for the comparison. Equivalently, 0.043 corresponds to 4.3% of the comparison rate, or an estimated 95.7% lower rate relative to that comparison rate.
This does not mean that every participant had a 95.7% lower individual risk, nor does it mean that 95.7% of participants were protected. A rate ratio is a population-level relative measure of incidence rates accumulated over person-time.
The 95% CI of 0.010–0.182 describes uncertainty around the estimated rate ratio. It is substantially below 1 throughout the interval, so the reported data are inconsistent with a rate ratio near or above 1 under the stated confidence-interval framework.
The P < 0.0001 value addresses the statistical evidence against the specified null hypothesis; it does not measure the size or clinical importance of the effect. Effect size is conveyed by the rate ratio and its confidence interval.
Because the comparator is background HIV incidence estimated from the Incidence Phase rather than a simultaneously randomized control group, the comparison has a different causal interpretation from the LEN-versus-F/TDF randomized comparison. The analysis population and construction of bHIV therefore deserve particular attention.
Primary Analysis 2: the prespecified 20% lower-incidence criterion
Rate ratio for HIV-1 incidence
95% CI: 0.010–0.182 · P < 0.0001
Two-sided confidence interval; Wald test; superiority hypothesis.
The second primary analysis uses the same reported estimate, confidence interval, and P-value but evaluates a different null hypothesis. The registry analysis notes define Null Hypothesis 02 as LEN/bHIV ≥ 0.8. The null was to be rejected if HIV-1 incidence in LEN was significantly and at least 20% lower than bHIV.
| Element | Reported result |
|---|---|
| Null hypothesis | LEN/bHIV ≥ 0.8 |
| Alternative direction | LEN/bHIV < 0.8 |
| Estimate | 0.043 |
| 95% CI | 0.010–0.182 |
| P-value | < 0.0001 |
| Method | Wald test |
| CI method | Delta method |
The important statistical distinction is that this analysis is not merely asking whether the rate ratio differs from 1. Its stated null threshold is 0.8, corresponding to the requirement that LEN incidence be at least 20% lower than the background incidence to reject the null.
The observed estimate of 0.043 is below that threshold, and the reported 95% confidence interval of 0.010–0.182 is also entirely below 0.8. The reported P < 0.0001 is the formal statistical evidence reported for the hypothesis.
This illustrates why the null hypothesis must be read before interpreting a P-value. A P-value is always calculated relative to a specified null model. The same point estimate can therefore be used to answer different statistical questions depending on whether the null value is 1, 0.8, or another prespecified threshold.
The analysis does not establish that the true individual-level probability of HIV infection is reduced by exactly a particular percentage. It establishes an estimated difference in incidence rates under the registry's rate-based analysis framework.
8. Secondary Results: LEN Versus F/TDF
The registry posts two secondary analyses of HIV-1 incidence per 100 person-years comparing LEN plus placebo-to-match F/TDF with placebo LEN plus F/TDF. Both analyses use the Full Analysis Set and a time frame of up to 149 weeks.
Rate-difference analysis
Rate difference
95% CI: -1.669 to -0.255 · P < 0.0001
Two-sided confidence interval; hybrid approach.
| Element | Reported result |
|---|---|
| Comparison | LEN + placebo-to-match F/TDF vs placebo LEN + F/TDF |
| Analysis population | Full Analysis Set |
| Effect measure | Rate difference |
| Estimate | -0.828 |
| 95% CI | -1.669 to -0.255 |
| P-value | <0.0001 |
| Method | Hybrid approach |
| CI method | Exact CI based on a hybrid approach |
| Registry null hypothesis | LEN - F/TDF ≥ 0.8/100 PY |
A rate difference is an absolute, rather than relative, comparison. The estimate of -0.828 indicates that the LEN incidence rate was estimated to be 0.828 HIV-1 events per 100 person-years lower than the F/TDF incidence rate, according to the posted analysis.
The registry describes the null hypothesis as LEN - F/TDF ≥ 0.8/100 PY, with rejection intended to establish that HIV-1 incidence in LEN was not substantially greater than F/TDF and that LEN was comparable to F/TDF under that criterion.
The negative rate difference provides an absolute rate comparison. Unlike a rate ratio, it remains expressed in the original incidence-rate scale. That makes it directly interpretable as a difference in HIV-1 events per 100 person-years.
The 95% CI of -1.669 to -0.255 lies below zero. Thus, under the posted confidence-interval framework, the estimated rate difference is below zero throughout the interval.
The P < 0.0001 value should not be read as a measure of how large the rate difference is. The magnitude is conveyed by -0.828 and its confidence interval. The P-value instead addresses the specified statistical hypothesis.
The registry calls the method a hybrid approach and specifically states that the exact confidence interval for the rate difference is based on that hybrid approach. The details reported here do not justify replacing that reported method with a simpler normal approximation.
Poisson-model rate-ratio analysis
Rate ratio
95% CI: 0.024–0.513 · P = 0.00245
Two-sided confidence interval; Poisson model; superiority hypothesis.
| Element | Reported result |
|---|---|
| Comparison | LEN + placebo-to-match F/TDF vs placebo LEN + F/TDF |
| Analysis population | Full Analysis Set |
| Effect measure | Rate ratio |
| Estimate | 0.111 |
| 95% CI | 0.024–0.513 |
| P-value | 0.00245 |
| Method | Poisson model |
| Hypothesis | Superiority |
| Registry null hypothesis | LEN vs F/TDF ≥ 1 |
The rate ratio of 0.111 means that the estimated HIV-1 incidence rate in the LEN group was 0.111 times the rate in the F/TDF group under the Poisson-model analysis. Equivalently, the estimated rate in LEN was 11.1% of the F/TDF rate, corresponding to an estimated 88.9% lower incidence rate relative to F/TDF.
Again, this is not an individual-level statement. It does not mean that each participant's personal risk was reduced by 88.9%, nor does it imply that 88.9% of participants experienced a particular treatment benefit.
The 95% CI of 0.024–0.513 quantifies uncertainty around the model-based rate ratio. Because the interval is below 1, the entire reported interval corresponds to a lower LEN incidence rate than F/TDF.
The P = 0.00245 value addresses the posted hypothesis that the LEN rate ratio is at least 1. It should be interpreted separately from the effect size. A small P-value does not tell us whether an effect is large; the rate ratio and its confidence interval provide that information.
The analysis is explicitly a Poisson-model analysis, so the model's treatment of event counts and person-time is part of the interpretation. The ClinicalTrials.gov record does not report additional model diagnostics, so no stronger claim about model adequacy should be made from the result alone.
9. Comparing Rate Ratios and Rate Differences
PURPOSE 2 is particularly useful statistically because the posted analyses express the LEN-versus-F/TDF comparison in two different effect measures.
| Measure | Reported estimate | 95% CI | What it answers |
|---|---|---|---|
| Rate difference | -0.828 | -1.669 to -0.255 | How much lower is the LEN incidence rate on the absolute rate scale? |
| Rate ratio | 0.111 | 0.024–0.513 | How does the LEN incidence rate compare proportionally with the F/TDF incidence rate? |
These are complementary rather than interchangeable statistics. The rate difference preserves the original rate scale, whereas the rate ratio expresses the relative relationship between the two rates. A complete statistical interpretation should retain the distinction rather than describing both as simply "risk reduction."
Absolute measure
The rate difference of -0.828 describes an estimated separation between the incidence rates in events per 100 person-years.
Relative measure
The rate ratio of 0.111 describes the LEN incidence rate relative to the F/TDF incidence rate.
10. Understanding the Primary Hypotheses
The posted analyses use more than one null value. This is important because the statistical meaning of a result depends on the question being tested.
| Analysis | Null hypothesis in registry data | Reported estimate |
|---|---|---|
| Primary 1 | LEN/bHIV ≥ 1 | 0.043 |
| Primary 2 | LEN/bHIV ≥ 0.8 | 0.043 |
| Secondary rate ratio | LEN vs F/TDF ≥ 1 | 0.111 |
| Secondary rate difference | LEN - F/TDF ≥ 0.8/100 PY | -0.828 |
The first primary hypothesis asks whether LEN incidence is lower than background HIV incidence. The second adds a quantitative threshold: the LEN incidence must be at least 20% lower than bHIV. These are related but distinct statistical questions.
The same principle applies to the secondary analyses. A rate ratio uses a null of 1 because equal rates correspond to a ratio of 1. A rate difference uses a null on the original rate scale.
11. Statistical Methods Explained
Why use person-years rather than simply counting infections?
Participants can contribute different amounts of follow-up time. Incidence per 100 person-years accounts for this unequal exposure by using the amount of time participants were actually observed while at risk. Two groups with the same number of HIV-1 events could therefore have different incidence rates if their accumulated follow-up differed.
Why was Poisson regression used?
Poisson regression is a natural framework for modeling event counts together with exposure time. In this trial, the secondary LEN-versus-F/TDF analysis reports a Poisson-model rate ratio. The model allows the comparison to be expressed as a relative incidence rate rather than merely as a raw difference in event counts.
What does a rate ratio of 0.111 mean?
It means the estimated HIV-1 incidence rate in the LEN group was 0.111 times the F/TDF rate in the reported analysis. On a percentage scale, that corresponds to 11.1% of the comparator rate, or an estimated 88.9% lower incidence rate relative to F/TDF. It is a rate comparison, not a statement about every individual's probability of infection.
Why are both a rate difference and a rate ratio useful?
The rate difference gives an absolute separation between incidence rates, while the rate ratio gives a relative separation. Relative measures can communicate proportional differences efficiently, whereas absolute measures retain the original event-rate scale. They answer different questions and should not be substituted for one another.
Why does the null hypothesis matter?
A P-value has meaning only relative to a null hypothesis. PURPOSE 2 illustrates this directly: one primary analysis uses LEN/bHIV ≥ 1, while another uses LEN/bHIV ≥ 0.8. The latter incorporates a specific 20% lower-incidence criterion. Therefore, the estimate should always be read together with the hypothesis being tested.
What does the 95% confidence interval tell us?
The confidence interval describes statistical uncertainty around the estimated rate ratio under the stated analysis method. For the primary rate ratio, the interval is 0.010–0.182. For the secondary Poisson-model rate ratio, it is 0.024–0.513. These intervals do not describe individual treatment responses and do not mean that 95% of participants fall within those numerical ranges.
Why is a Wald test used?
A Wald test compares an estimated parameter with a specified null value after accounting for the estimated standard error. It is therefore useful for testing hypotheses about a rate ratio or another model-based parameter. In PURPOSE 2, the registry explicitly identifies the Wald test for the two primary analyses.
12. Confidence Intervals and Precision
The reported estimates are much more informative when considered together with their confidence intervals.
| Comparison | Estimate | 95% CI | Interpretive scale |
|---|---|---|---|
| LEN vs bHIV | 0.043 | 0.010–0.182 | Rate ratio |
| LEN vs F/TDF | 0.111 | 0.024–0.513 | Rate ratio |
| LEN vs F/TDF | -0.828 | -1.669 to -0.255 | Rate difference |
The interval around the LEN-versus-bHIV rate ratio spans from 0.010 to 0.182. The interval around the LEN-versus-F/TDF rate ratio spans from 0.024 to 0.513. Because rate ratios are multiplicative measures, the distance between the lower and upper limits should not be interpreted in the same way as an ordinary symmetric interval on the raw rate scale.
A point estimate is only one estimate from the observed data. The confidence interval shows how much statistical uncertainty surrounds it under the relevant method. For this reason, 0.043 with a 95% CI of 0.010–0.182 communicates substantially more information than 0.043 by itself.
The interval also helps separate statistical evidence from effect magnitude. The P-value addresses the null hypothesis, whereas the estimate and confidence interval describe the estimated magnitude and its uncertainty.
13. P-Values: Evidence Versus Effect Size
The posted analyses include P-values of <0.0001, < 0.0001, <0.0001, and 0.00245. These values belong to different hypotheses and analysis methods.
| Analysis | P-value | Question being tested |
|---|---|---|
| Primary analysis 1 | <0.0001 | LEN/bHIV ≥ 1 |
| Primary analysis 2 | < 0.0001 | LEN/bHIV ≥ 0.8 |
| Secondary rate difference | <0.0001 | LEN - F/TDF ≥ 0.8/100 PY |
| Secondary rate ratio | 0.00245 | LEN vs F/TDF ≥ 1 |
A P-value is not the probability that the null hypothesis is true. It is also not the probability that the observed treatment effect is due to chance. More precisely, it measures the compatibility of the observed data with the specified null model under the assumptions of the statistical test.
For that reason, the P-value should be read alongside the effect estimate, confidence interval, analysis population, and definition of the comparison.
14. Safety Results
The ClinicalTrials.gov record reports serious adverse events by arm for the randomized blinded phase. These are the safety figures available for this page.
| Randomized blinded phase group | Participants with serious AEs | At risk | Reported proportion |
|---|---|---|---|
| LEN + placebo-to-match | 71 | 2183 | 71/2183 |
| Placebo LEN + F/TDF | 43 | 1088 | 43/1088 |
The ClinicalTrials.gov record reports serious adverse events as affected/at-risk counts rather than as a formal hypothesis-tested treatment comparison. Accordingly, this page does not calculate an additional risk ratio, risk difference, confidence interval, or P-value from those counts.
15. What the Rate Ratio Does — and Does Not — Mean
A rate ratio of 0.043 means the estimated HIV-1 incidence rate for LEN was 0.043 times the background HIV incidence rate used in the primary comparison.
It does not mean that 4.3% of participants acquired HIV, that exactly 4.3% of individual risk remained for every participant, or that 95.7% of participants personally experienced a reduction in risk.
A rate ratio of 0.111 means that the estimated HIV-1 incidence rate in LEN was 0.111 times the F/TDF rate in the reported Poisson-model analysis.
That corresponds to an estimated 88.9% lower incidence rate relative to F/TDF, but it remains a population-level rate comparison rather than an individual-level probability statement.
The rate difference of -0.828 indicates an estimated absolute difference of -0.828 HIV-1 events per 100 person-years between LEN and F/TDF, under the posted hybrid analysis.
This is fundamentally different from the rate ratio because it is expressed on the original incidence-rate scale rather than as a proportion.
16. Why the Background HIV Comparison Requires Care
The primary comparison is statistically unusual enough to deserve separate attention. One side of the comparison is the HIV-1 incidence rate in the LEN group during the randomized blinded phase. The other is background HIV incidence estimated using RITA from the Incidence Phase.
Randomized comparison
LEN versus F/TDF compares two groups within the randomized blinded phase and therefore benefits directly from the trial's randomized allocation.
Background comparison
LEN versus bHIV compares randomized-phase LEN incidence with a background incidence estimate constructed from the Incidence Phase.
The distinction does not make the background comparison unusable; rather, it determines what the result can logically establish. A comparison against an estimated background rate relies on the way that background rate was constructed and on the comparability of the populations and observation periods underlying the two quantities.
The registry's use of RITA is therefore part of the statistical design, not merely a descriptive detail. The bHIV value is an estimated reference incidence, rather than a treatment arm randomized concurrently against LEN.
17. Non-Inferiority Logic and the 0.8 Threshold
The secondary rate-difference analysis and one of the primary rate-ratio analyses use prespecified thresholds rather than the conventional equality null alone. This is an important statistical teaching point.
For the primary analysis, the registry specifies LEN/bHIV ≥ 0.8 as Null Hypothesis 02. The stated rejection criterion is that HIV-1 incidence in LEN be significantly and at least 20% lower than bHIV.
For the secondary rate-difference analysis, the registry specifies LEN - F/TDF ≥ 0.8/100 PY as Null Hypothesis 03 and states that rejection would support the conclusion that HIV-1 incidence in LEN is not substantially greater than F/TDF, described in the registry as LEN being comparable to F/TDF.
The ClinicalTrials.gov record does not identify the overall type-I-error allocation across all four posted analyses. This page therefore does not infer a multiplicity adjustment beyond the hypotheses and methods explicitly reported.
18. Crossover, Factorial Design, and Other Design Features
The ClinicalTrials.gov record identifies the study as randomized, double masked, and parallel. They do not report a factorial design or a crossover design in the ClinicalTrials.gov record.
| Design topic | What the ClinicalTrials.gov record supports |
|---|---|
| Randomization | Yes. Allocation is listed as randomized. |
| Blinding | Yes. Masking is listed as double. |
| Parallel design | Yes. Design model is parallel. |
| Factorial design | Not reported in the ClinicalTrials.gov record. |
| Crossover | Not reported in the ClinicalTrials.gov record. |
| Interim analysis | Not reported in the ClinicalTrials.gov record. |
| Bayesian methods | Not reported in the statistical analyses posted on ClinicalTrials.gov. |
| Imputation / missing-data method | Not reported in the statistical analyses posted on ClinicalTrials.gov. |
| Stratification variables | Not reported in the ClinicalTrials.gov record. |
These omissions are important because they prevent the page from attributing design or analysis procedures that are not actually present in the ClinicalTrials.gov record.
19. Interpreting Person-Time and Censoring
The primary incidence endpoint is based on person-time at risk. This creates a different statistical structure from a simple binary endpoint such as "infected versus not infected at the end of follow-up."
Participants without an HIV-1 diagnosis contribute follow-up time while they remain at risk. Participants diagnosed with HIV-1 contribute time up to confirmed diagnosis. Consequently, the incidence calculation incorporates both the number of events and the amount of observed time at risk.
This is why a rate ratio should not be reconstructed simply by dividing the number of HIV-1 events by the number randomized. The denominator is person-time, not participant count.
The ClinicalTrials.gov record does not provide a complete participant-level censoring dataset, so this page does not attempt to reconstruct Kaplan-Meier curves or independently calculate incidence rates from unreported event and person-time totals.
20. A Worked Statistical Reading of the Primary Result
Consider the primary rate ratio of 0.043. The correct sequence of interpretation is more informative than simply saying the result was statistically significant.
- Identify the outcome: HIV-1 incidence per 100 person-years.
- Identify the comparison: LEN in the randomized blinded phase versus the background HIV incidence estimate.
- Identify the effect measure: rate ratio.
- Read the point estimate: 0.043 indicates the LEN incidence rate was estimated at 4.3% of the comparison rate.
- Read the confidence interval: 0.010–0.182 describes uncertainty around the estimated rate ratio.
- Read the hypothesis: the first analysis tests LEN/bHIV ≥ 1; the second tests LEN/bHIV ≥ 0.8.
- Read the P-value: <0.0001 provides statistical evidence against the respective stated null hypothesis.
- Apply the design context: bHIV is a RITA-based background estimate rather than a simultaneously randomized comparator arm.
This sequence prevents several common statistical errors: confusing incidence with risk, confusing a rate ratio with an individual probability, ignoring the confidence interval, and treating the background comparison as if it were identical to the randomized LEN-versus-F/TDF comparison.
21. Why This Trial Matters Statistically
PURPOSE 2 is a useful teaching example because it combines randomized clinical-trial methodology with incidence-rate analysis and a background-incidence comparator. It also demonstrates why the choice of estimand and null hypothesis matters before any P-value is interpreted.
| Concept | How it appears in PURPOSE 2 |
|---|---|
| Randomization | The trial is randomized with a parallel design. |
| Blinding | The trial is double masked, with placebo-to-match interventions listed. |
| Incidence rate | HIV-1 incidence is reported per 100 person-years. |
| Person-time | Follow-up duration contributes to the incidence-rate denominator. |
| Rate ratio | Used for the primary LEN-versus-bHIV comparison and the secondary LEN-versus-F/TDF Poisson analysis. |
| Rate difference | Used as a secondary LEN-versus-F/TDF effect measure. |
| Poisson regression | Used for the secondary rate-ratio analysis versus F/TDF. |
| Wald test | Used for the two primary analyses. |
| Confidence interval | Primary rate-ratio CI is based on the delta method; the secondary rate-ratio CI comes from the Poisson model. |
| Null hypothesis | Different analyses use different null thresholds, including 1 and 0.8. |
| Full Analysis Set | Used for the reported randomized-phase efficacy analyses. |
| Background incidence | bHIV is estimated using RITA in the Incidence Phase. |
22. Important Limitations and Interpretation Issues
- Background comparator: the primary comparison uses a RITA-based background HIV incidence estimate rather than a simultaneously randomized control group.
- Analysis-population distinction: the primary LEN analysis is described using a Full Analysis Set definition, while the background comparison uses All Screened Participants With Non-missing HIV-1 Diagnosis from the Incidence Phase.
- Rate versus risk: incidence per 100 person-years is a rate and should not automatically be described as a cumulative infection probability.
- Different effect measures: the secondary analysis reports both a rate difference and a rate ratio. These measures should not be treated as interchangeable.
- Different null hypotheses: the primary analyses use both 1 and 0.8 as null thresholds, while the secondary analyses use different rate-ratio and rate-difference nulls.
- Poisson-model assumptions: the secondary rate ratio comes from a Poisson model. The ClinicalTrials.gov record does not report additional model diagnostics or alternative model checks.
- Wald inference: the primary analyses use Wald tests. The interpretation therefore depends on the standard-error framework underlying those tests.
- Multiplicity: four statistical analyses are posted, including two primary analyses and two secondary analyses. The ClinicalTrials.gov record does not specify an overall multiplicity-adjustment procedure across all posted analyses.
- Missing-data methods: the ClinicalTrials.gov record does not report an imputation strategy. No particular imputation method should therefore be attributed to the trial.
- Subgroups: the ClinicalTrials.gov record contains no subgroup estimates, so no subgroup treatment-effect conclusions can be drawn from this record.
- Interim analysis: the ClinicalTrials.gov record does not report an interim-analysis procedure, alpha-spending method, or stopping boundary.
- Bayesian methods: no Bayesian method is reported in the statistical analyses posted on ClinicalTrials.gov.
23. What the Results Establish Statistically
Primary evidence
The registry reports a LEN-versus-bHIV rate ratio of 0.043 with a 95% CI of 0.010–0.182 and P < 0.0001 under the posted Wald analyses.
Randomized comparison
The registry reports a secondary LEN-versus-F/TDF rate ratio of 0.111 with a 95% CI of 0.024–0.513 and P = 0.00245 from a Poisson model.
Absolute comparison
The secondary LEN-versus-F/TDF rate difference is -0.828, with a 95% CI of -1.669 to -0.255 and P < 0.0001.
Safety description
Serious adverse events are reported descriptively as 71/2183 for LEN plus placebo-to-match and 43/1088 for placebo LEN plus F/TDF.
These findings should be kept on their respective statistical scales. The primary result is a comparison with estimated background incidence; the secondary results include a randomized LEN-versus-F/TDF comparison expressed both as an absolute rate difference and a relative rate ratio.
24. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
25. Related Statistical Calculators
26. Sources
- ClinicalTrials.gov: PURPOSE 2, NCT04925752.
- PubMed: PubMed record for the linked PURPOSE 2 publication (PMID 41449429).
Continue through the Clinical Biostats statistical pathway
Explore the statistical concepts that connect trial design, incidence-rate analysis, effect measures, confidence intervals, and hypothesis testing.
27. Record Summary
PURPOSE 2 provides a useful statistical case study because its efficacy results are expressed as incidence rates per 100 person-years and analyzed with several complementary effect measures. The primary LEN-versus-background comparison reports a rate ratio of 0.043 with a 95% CI of 0.010–0.182 and P < 0.0001. The randomized LEN-versus-F/TDF comparison reports a rate difference of -0.828 with a 95% CI of -1.669 to -0.255 and P < 0.0001, as well as a Poisson-model rate ratio of 0.111 with a 95% CI of 0.024–0.513 and P = 0.00245.
The most important statistical lesson is that these numbers answer different questions. A rate ratio expresses a relative incidence-rate comparison, a rate difference expresses an absolute incidence-rate comparison, a confidence interval describes uncertainty around the estimate, and a P-value evaluates evidence against a specified null hypothesis. The primary background comparison additionally requires attention to the construction of bHIV through RITA and to the distinction between a background reference population and a randomized comparator.