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
SILCAAT was a randomized, parallel-group, phase 3 treatment trial in HIV infection comparing Proleukin plus antiretroviral therapy with antiretroviral therapy alone in patients with low CD4+ counts. The registry reports a primary time-to-event analysis using a stratified Cox proportional-hazards model and a longitudinal mixed-effects analysis for CD4+ cell count.
| Feature | SILCAAT |
|---|---|
| Phase | Phase 3 |
| Condition | HIV Infection |
| Brief title | Interleukin-2 Plus Antiretroviral Therapy for HIV-Infected Patients With Low CD4+ Counts |
| Design | Randomized, parallel-group |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1695 |
| Intervention | Proleukin |
| Comparator | Antiretroviral therapy only |
| Lead sponsor | University of Minnesota |
| Study start | March 2001 |
| Primary completion | January 2007 |
| ClinicalTrials.gov | NCT00013611 |
2. Clinical Question
The central statistical question was whether adding Proleukin to antiretroviral therapy changed the time to a new or recurrent disease progression event or death compared with antiretroviral therapy alone in HIV-infected patients with low CD4+ counts.
Population
HIV-infected patients with low CD4+ counts participating in the randomized SILCAAT study.
Intervention
Proleukin plus antiretroviral therapy.
Comparator
Antiretroviral therapy only.
Primary question
Does the addition of Proleukin alter the occurrence of fatal or non-fatal AIDS-related opportunistic disease or death from any cause?
3. Trial Design
Proleukin Plus Antiretroviral Therapy
- Proleukin
- Antiretroviral therapy
Antiretroviral Therapy Only
- Antiretroviral therapy
4. Primary Endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| New or Recurrent Disease Progression Events, as Defined, or Death. | Number of participants with fatal or non-fatal AIDS-related opportunistic disease or death from any cause. AIDS-related opportunistic disease includes the CDC Category C 1993 definition plus Aspergillosis, invasive; Bartonellosis; Chagas disease of the CNS; Herpes zoster, multidermal; Leishmaniasis, visceral; Lymphoma, Hodgkin's; Microsporidiosis (> 1 month's duration); Nocardiosis; Penicillium marneffti, disseminated; Pneumocystis carinii, extrapulmonary; and Rhodococcus equi disease. | From randomization to date last known to be alive or November 15, 2008, whichever is earlier |
The registry categorizes the registered primary endpoint as binary, because the outcome measure records whether participants experienced a qualifying event. The posted formal analysis, however, treats the endpoint as a time-to-event outcome and uses a Cox proportional-hazards model. That distinction matters: the analysis incorporates the timing of events and censoring rather than reducing follow-up to a simple event/no-event comparison.
5. Statistical Methodology
Cox proportional-hazards model
The primary analysis used a Cox proportional-hazards model to compare the rate of the primary event over follow-up. The registry reports the model as a stratified analysis, with stratification by CD4+ count stratum and country of randomization.
For a treatment indicator, the exponentiated coefficient, exp(β), is the estimated hazard ratio. The model compares the instantaneous event rate between randomized groups while accounting for the specified stratification.
Stratification
The primary analysis was stratified by CD4+ count stratum, defined as 50–199 or 200–299, and by country of randomization. Stratification allows the baseline hazard to differ across these strata while estimating a common treatment hazard ratio across the strata.
Mixed-effects model for CD4+ cell count
CD4+ cell count was measured every 4 months from randomization through the date last known to be alive or November 15, 2008, whichever was earliest. The registry reports a mixed models analysis with the effect measure expressed as a mean difference in final values.
Repeated measurements from the same participant are correlated, so a mixed-effects framework is appropriate for longitudinal data because it can represent within-participant dependence while estimating the treatment-group difference over repeated observations.
Superiority hypothesis
The registry identifies the hypothesis type as superiority. For the primary endpoint, the null hypothesis was that event rates in the two groups were equal. The study was designed with 80%, two-sided alpha=.05, assuming a 28% reduction in the hazard of opportunistic disease or death.
Primary time-to-event analysis
Cox proportional-hazards regression with stratification by CD4+ count stratum and country of randomization.
Longitudinal analysis
Mixed-effects modeling of CD4+ cell count measured every 4 months.
6. Primary Result: Disease Progression or Death
The primary analysis compared Proleukin plus antiretroviral therapy with antiretroviral therapy only from randomization through the date last known to be alive or November 15, 2008, whichever was earlier.
Hazard ratio for new or recurrent disease progression events or death
95% CI: 0.70–1.18 · P = 0.47
Stratified Cox proportional-hazards model
| Primary endpoint | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| New or Recurrent Disease Progression Events, as Defined, or Death. | Hazard ratio | 0.91 | 0.70–1.18 | 0.47 |
The estimated hazard ratio of 0.91 corresponds to an estimated instantaneous event rate about 9% lower in the Proleukin-plus-antiretroviral-therapy group than in the antiretroviral-therapy-only group, under the Cox model and its proportional-hazards interpretation.
The hazard ratio does not mean that 9% fewer participants experienced the endpoint, nor does it describe an individual's probability of remaining event-free. It is a relative time-to-event measure comparing the modeled event rates between the randomized groups.
The 95% confidence interval of 0.70–1.18 describes statistical uncertainty around the estimated hazard ratio. Because the interval includes 1, the data are compatible with both a lower and a higher event hazard for the Proleukin group under the model.
The P = 0.47 value addresses the statistical evidence against the null hypothesis of equal event rates. It does not measure the magnitude or clinical importance of the estimated treatment effect. The effect size is described by the hazard ratio and its confidence interval.
Interpretation also depends on the Cox proportional-hazards framework, censoring, and the prespecified stratification. The registry does not provide enough additional information here to establish whether the proportional-hazards assumption was formally assessed.
7. Secondary Time-to-Event Results
All-Cause Mortality
Hazard ratio for all-cause mortality
95% CI: 0.77–1.44 · P = 0.73
Stratified Cox proportional-hazards model
An estimated hazard ratio of 1.06 means that the modeled instantaneous rate of death was estimated to be 6% higher in the Proleukin-plus-antiretroviral-therapy group than in the antiretroviral-therapy-only group. This is a relative hazard estimate, not a statement that 6% more participants died.
The 95% confidence interval of 0.77–1.44 spans 1 and is sufficiently broad to include both lower and higher mortality hazards. The P = 0.73 value describes the evidence against equal hazards; it does not quantify the size of the mortality effect.
New or Recurrent Disease Progression Events
Hazard ratio for disease progression events
95% CI: 0.51–1.06 · P = 0.10
Stratified Cox proportional-hazards model
An estimated hazard ratio of 0.73 corresponds to an estimated instantaneous progression-event rate about 27% lower in the Proleukin-plus-antiretroviral-therapy group under the fitted Cox model.
The 95% confidence interval of 0.51–1.06 indicates substantial uncertainty around that estimate and includes 1. The P = 0.10 value should not be interpreted as a measure of effect size; the hazard ratio and confidence interval provide the information about magnitude and precision.
Because this is a secondary analysis, it should also be interpreted in the context of the overall trial design rather than as an isolated numerical finding.
Grade 4 Clinical Events
Hazard ratio for grade 4 clinical events
95% CI: 0.90–1.34 · P = 0.35
Stratified Cox proportional-hazards model
The registry defines this outcome as the number of participants experiencing at least one grade 4 event and analyzes the time to such an event using Cox regression.
The estimated hazard ratio of 1.10 corresponds to an estimated instantaneous rate of grade 4 events 10% higher in the Proleukin-plus-antiretroviral-therapy group. The 95% confidence interval of 0.90–1.34 includes 1, and the P = 0.35 value does not measure the size of the observed effect.
The endpoint is defined around participants experiencing at least one grade 4 event, so it should not be interpreted as a comparison of the total number of grade 4 events.
8. Secondary Longitudinal Result: CD4+ Cell Count
CD4+ cell count was assessed every 4 months from randomization through the date last known to be alive or November 15, 2008, whichever was earliest. The registry reports a mixed-effects model analysis using final values as the effect measure.
Mean difference in final CD4+ cell count
95% CI: 40.3–65.7 · P < 0.001
Outcome unit: cells per cubic mm
| Endpoint | Method | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|
| CD4+ Cell Count | Mixed Models Analysis | Mean Difference (Final Values) | 53.0 cells per cubic mm | 40.3–65.7 | < 0.001 |
The estimated mean difference of 53.0 cells per cubic mm indicates that the modeled final CD4+ cell count was higher by 53.0 cells per cubic mm in the Proleukin-plus-antiretroviral-therapy group compared with the antiretroviral-therapy-only group.
This is a difference in a continuous outcome, not a hazard ratio. It does not mean that every participant experienced an increase of exactly 53.0 cells per cubic mm.
The 95% confidence interval of 40.3–65.7 describes uncertainty around the estimated mean difference. The interval remains above zero, which is consistent with a positive between-group difference under the reported model.
The P < 0.001 value describes evidence against a null mean difference of zero. It does not indicate that the magnitude of the difference is necessarily large or clinically important; those are separate questions requiring clinical context.
9. Safety Results
The registry reports serious adverse events by randomized arm as affected participants over participants at risk.
| Safety measure | Proleukin Plus Antiretroviral Therapy | Antiretroviral Therapy Only |
|---|---|---|
| Serious adverse events | 203/849 | 186/846 |
The reported denominators are different from the overall randomized enrollment of 1695. Accordingly, the serious-adverse-event counts should be interpreted using the affected/at-risk values reported by the registry rather than substituting the overall enrollment as the denominator.
10. Statistical Methods Explained
Why was a Cox proportional-hazards model used for the primary endpoint?
The primary outcome concerns whether participants experience disease progression or death during follow-up, and the registry reports the endpoint's time frame from randomization through the date last known to be alive or November 15, 2008, whichever was earlier. A Cox model is designed for this type of time-to-event outcome because it uses both event occurrence and event timing while accommodating right-censored follow-up.
What does a hazard ratio of 0.91 mean?
A hazard ratio of 0.91 means that the fitted model estimates an instantaneous event rate approximately 9% lower in the Proleukin-plus-antiretroviral-therapy group than in the antiretroviral-therapy-only group. The interpretation is conditional on the Cox model and should not be converted into a percentage of participants who avoid the event.
Why is the confidence interval important?
The point estimate alone gives only one estimate of the treatment effect. The 95% confidence interval of 0.70–1.18 shows the uncertainty around the primary hazard-ratio estimate. Because it includes 1, the data are compatible with a range of treatment effects that includes no difference in hazard.
Why does the P-value not measure effect size?
The P-value evaluates the statistical evidence against a specified null hypothesis. It does not tell us how large the treatment effect is. In SILCAAT, the primary effect size is described by the hazard ratio of 0.91 and its 95% confidence interval of 0.70–1.18, while P = 0.47 summarizes the evidence against equal event rates.
Why was stratification used?
The primary analysis was stratified by CD4+ count stratum and country of randomization. Stratification permits the baseline event hazard to differ across these strata while estimating the treatment comparison within the stratified Cox framework.
Why use a mixed-effects model for CD4+ cell count?
CD4+ cell count was measured repeatedly every 4 months. Measurements from the same participant are related, so treating all observations as independent would not adequately represent the longitudinal structure. A mixed-effects model provides a framework for analyzing repeated measurements while accounting for within-participant correlation.
11. Understanding the Primary Analysis More Deeply
The registry reports a two-sided superiority analysis with alpha=.05 and a design assumption of a 28% reduction in the hazard of opportunistic disease or death.
Relative effect versus absolute outcome
The primary result is expressed as a hazard ratio rather than as an absolute difference in the proportion of participants experiencing the endpoint. A hazard ratio describes a relative difference in the modeled instantaneous event rate over time. An absolute risk difference would answer a different question and would require corresponding event-probability estimates.
The role of censoring
The primary time frame ends at the date last known to be alive or November 15, 2008, whichever is earlier. In a time-to-event analysis, participants who do not experience the event before their available follow-up can contribute information up to their censoring time. This is one reason a Cox model can use follow-up information from participants whose observation periods differ.
Why the primary endpoint description matters
The endpoint combines fatal and non-fatal AIDS-related opportunistic disease with death from any cause. Its interpretation therefore differs from an endpoint restricted to mortality alone. The secondary all-cause mortality analysis separates mortality from the broader primary composite outcome.
Primary endpoint versus secondary endpoints
The primary endpoint is the prespecified disease-progression-or-death outcome. All-cause mortality, new or recurrent disease progression events, grade 4 clinical events, and CD4+ cell count are reported as secondary outcomes. Their estimates answer related but distinct questions and should not be treated as interchangeable measures.
12. Stratification and the Cox Model
The registry specifies stratification by two variables: CD4+ count stratum and country of randomization. CD4+ count was divided into the 50–199 and 200–299 strata for this analysis.
| Stratification factor | Reported categories | Role in analysis |
|---|---|---|
| CD4+ count stratum | 50–199 or 200–299 | Stratification of Cox analysis |
| Country of randomization | Country-specific strata | Stratification of Cox analysis |
Stratification is different from simply adjusting for a covariate with a single regression coefficient. In a stratified Cox model, the baseline hazard can vary across strata while the treatment effect is estimated across the stratified comparison.
A hazard ratio from a stratified Cox model is not simply a crude ratio of overall event rates. It is a model-based summary of the treatment comparison within the specified survival-analysis framework.
13. Interpreting the Secondary Findings Together
| Endpoint | Effect estimate | 95% CI | P-value | Analysis |
|---|---|---|---|---|
| New or recurrent disease progression events or death | HR 0.91 | 0.70–1.18 | 0.47 | Cox proportional-hazards model |
| All-cause mortality | HR 1.06 | 0.77–1.44 | 0.73 | Cox proportional-hazards model |
| New or recurrent disease progression events | HR 0.73 | 0.51–1.06 | 0.10 | Cox proportional-hazards model |
| Grade 4 clinical events | HR 1.10 | 0.90–1.34 | 0.35 | Cox proportional-hazards model |
| CD4+ cell count | Mean difference 53.0 | 40.3–65.7 | < 0.001 | Mixed-effects model |
These outcomes illustrate why a clinical trial should not be reduced to a single P-value. The primary endpoint is a time-to-event outcome with an estimated hazard ratio of 0.91, while the CD4+ analysis is a continuous longitudinal outcome with a mean difference of 53.0 cells per cubic mm. They use different estimands and different statistical models.
The disease-progression secondary analysis has a lower hazard-ratio estimate of 0.73, whereas the all-cause mortality estimate is 1.06. These are distinct endpoints and should be interpreted according to their own definitions and confidence intervals rather than combined into a single treatment-effect summary.
14. Planned Statistical Logic and Trial Design
The registry's design description provides the statistical assumptions underlying the primary hypothesis test. The planned analysis was not framed as a non-inferiority comparison: the hypothesis type was superiority, and the null hypothesis was equal event rates between the groups.
15. What the Primary Result Does — and Does Not — Establish
What the HR describes
The relative difference in modeled instantaneous event rates between the two randomized groups under the stratified Cox model.
What the HR does not describe
It is not a percentage of participants benefiting, a relative risk, an absolute risk difference, or a probability for an individual participant.
What the CI describes
Statistical uncertainty around the estimated hazard ratio under the model and sampling framework.
What the P-value describes
Evidence against the specified null hypothesis; it does not measure clinical magnitude or importance.
16. Limitations and Interpretation Issues
- Composite primary endpoint: the primary endpoint combines fatal and non-fatal AIDS-related opportunistic disease with death from any cause. Different components can have different clinical meanings.
- Time-to-event modeling: the primary analysis depends on the Cox proportional-hazards framework. The registry does not report a formal assessment of the proportional-hazards assumption.
- Confidence-interval width: the primary 95% confidence interval extends from 0.70 to 1.18, indicating uncertainty around the point estimate.
- Secondary endpoints: secondary analyses answer different questions and should not be interpreted as replacements for the prespecified primary endpoint.
- Longitudinal outcome: CD4+ cell count was repeatedly measured, making the mixed-effects framework appropriate for the correlation structure of repeated observations.
- Safety denominators: serious adverse events are reported using affected/at-risk counts of 203/849 and 186/846 rather than the overall enrollment denominator.
- Limited methodological detail: the registry identifies the principal statistical methods but does not provide all details that would ordinarily be found in a full statistical analysis plan, such as complete covariance specifications for the mixed-effects model or a detailed missing-data strategy.
17. Why This Trial Matters Statistically
SILCAAT is a useful statistical teaching case because it combines two fundamentally different analysis problems within the same randomized trial: time-to-event outcomes and repeated continuous measurements.
| Concept | How it appears in SILCAAT |
|---|---|
| Randomization | Participants were randomly allocated to two parallel treatment groups. |
| Superiority testing | The registry specifies a superiority hypothesis with a two-sided alpha=.05. |
| Time-to-event endpoint | The primary endpoint and several secondary endpoints were analyzed over follow-up. |
| Hazard ratio | The primary and secondary time-to-event analyses use hazard ratios. |
| Cox proportional-hazards model | The registry reports Cox regression for the time-to-event analyses. |
| Stratified analysis | The Cox analyses were stratified by CD4+ count stratum and country of randomization. |
| Confidence interval | 95% two-sided confidence intervals accompany the reported effect estimates. |
| P-value | P-values quantify statistical evidence against the corresponding null hypotheses. |
| Mixed-effects model | Repeated CD4+ cell-count measurements were analyzed using mixed models. |
| Different estimands | Hazard ratios for event outcomes and a mean difference for CD4+ cell count answer different statistical questions. |
18. Related Tutorials
Learn more about the methods used in this trial:
19. Related Calculators
20. Sources
- ClinicalTrials.gov: NCT00013611.
- PubMed: PMID 24870935.
- PubMed: PMID 19828532.
Continue with Clinical Biostats
Explore statistical tutorials and calculators that provide deeper explanations of the methods used in randomized clinical trials.
21. Record Summary
SILCAAT provides a clear example of how different clinical endpoints require different statistical estimands and models. The primary disease-progression-or-death endpoint was analyzed with a stratified Cox proportional-hazards model and produced a hazard ratio of 0.91 with a 95% confidence interval of 0.70–1.18 and P = 0.47. Secondary time-to-event analyses used the same general Cox framework, while repeated CD4+ cell-count measurements were analyzed using a mixed-effects model and produced a reported mean difference of 53.0 cells per cubic mm with a 95% confidence interval of 40.3–65.7 and P < 0.001.