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
OAKTREE was a randomized, double-blind, parallel phase 3 trial evaluating obeldesivir versus placebo in nonhospitalized participants with COVID-19. The registry identifies four primary endpoints spanning time-to-event and binary outcomes, with additional secondary analyses covering symptom resolution, medically attended events or death, viral load, antigen negativity, symptom relapse, and antigen rebound.
| Feature | OAKTREE |
|---|---|
| Trial name | OAKTREE |
| ClinicalTrials.gov identifier | NCT05715528 |
| Phase | Phase 3 |
| Condition | COVID-19 |
| Therapeutic area | Infectious Disease |
| Design | Randomized, double-blind, parallel |
| Primary purpose | Treatment |
| Enrollment | 2011 |
| Lead sponsor | Gilead Sciences |
| Sponsor type | Industry |
| Status | Completed |
| Trial start | 2023-02-08 |
| Primary completion | 2023-11-28 |
2. Clinical Question
The central question was whether obeldesivir differed from placebo with respect to time to COVID-19 symptom alleviation by Day 29 in nonhospitalized participants with COVID-19, under a superiority framework.
Population
Nonhospitalized participants with COVID-19 enrolled in the phase 3 OAKTREE trial.
Intervention
Obeldesivir.
Comparator
Obeldesivir placebo.
Primary question
Does obeldesivir change the time to COVID-19 symptom alleviation by Day 29 compared with placebo?
3. Trial Design
Obeldesivir
- Obeldesivir drug intervention
- Evaluated against placebo
- Included in the randomized analysis of efficacy and safety endpoints
Obeldesivir Placebo
- Placebo drug intervention
- Served as the randomized comparison group
- Included in the same endpoint framework
The registry identifies four primary endpoints: one time-to-event endpoint and three binary endpoints. The statistical analyses posted for the trial include stratified log-rank testing, log-rank testing, MMRM, and Fisher exact testing.
4. Analysis Populations
The primary time-to-event analysis used the Full Analysis Positive Set (FAPS). This population included participants, except those from one site that did not comply with study rules, who were randomized, received at least one dose of study drug, and were SARS-CoV-2 positive at baseline as confirmed by the specified central laboratory Cepheid Xpert Xpress CoV-2/Flu/RSV plus test or SARS-CoV-2 RT-qPCR test. Participants with available data were analyzed.
| Analysis set | Registry definition / role |
|---|---|
| Full Analysis Positive Set | Randomized participants who received at least 1 dose, were SARS-CoV-2 positive at baseline by the specified testing, excluding participants from one site that did not comply with study rules. |
| Virology Analysis Set | Randomized participants who received at least 1 dose and had baseline SARS-CoV-2 viral load ≥ LLOQ. |
| Antigen Analysis Set | Randomized participants who received at least 1 dose and had at least 1 SARS-CoV-2 rapid antigen test positive from Day 1 to Day 5 and SARS-CoV-2 PCR positive at baseline per the FAPS. |
| Available-data subsets | Several secondary analyses used participants in the relevant analysis set with available data; relapse analyses also specified recovery or antigen-negativity criteria. |
These population definitions matter because a treatment effect estimated in a specific analysis set applies directly to that analyzed population. In particular, the primary result is not simply a comparison of all 2011 enrolled participants; the registry specifies the FAPS criteria used for that analysis.
5. Registered Primary Endpoints
| Primary endpoint | Time frame | Type | Registry definition |
|---|---|---|---|
| Time to Coronavirus Disease 2019 (COVID-19) Symptom Alleviation by Day 29 | First dose date up to Day 29 | Time-to-event | Time to alleviation of targeted COVID-19 symptoms by Day 29. Symptom alleviation required all targeted symptoms scored moderate/severe at baseline to become mild/none and all symptoms scored mild/none at baseline to become none for at least 48 consecutive hours. |
| Percentage of Participants Experiencing Treatment-emergent Adverse Events (TEAEs) | First dose date up to Day 5 plus 30 days | Binary | TEAEs included AEs beginning on or after study-drug start through 30 days after permanent discontinuation, or AEs leading to premature study-drug discontinuation. |
| Percentage of Participants Experiencing Laboratory Abnormalities | First dose date up to Day 5 plus 30 days | Binary | Treatment-emergent laboratory abnormalities were values increasing at least 1 toxicity grade from baseline at any postbaseline time point through last dose plus 30 days. |
| Percentage of Participants Experiencing Serious Adverse Events (SAEs) and Adverse Events (AEs) Leading to Study Drug Discontinuation | First dose date up to Day 5 plus 30 days | Binary | Percentages were rounded off. |
The registry reports four primary endpoints, but the posted statistical-analysis entries contain one formal primary-endpoint analysis: the analysis of time to COVID-19 symptom alleviation by Day 29. The registry does report results for the other primary endpoints, but the posted statistical-analysis entries do not provide an estimate, confidence interval, and p-value for those endpoints in the statistical-analysis records represented here.
6. How the Primary Symptom Endpoint Was Defined
The primary efficacy endpoint was a time-to-event outcome rather than a simple percentage measured at a single visit. The event was symptom alleviation, and the registry specified that the relevant symptom improvement had to persist for at least 48 consecutive hours.
The targeted symptoms were:
- Stuffy or runny nose
- Sore throat
- Shortness of breath
- Cough
- Low energy or tiredness
- Muscle or body aches
- Headache
- Chills or shivering
- Feeling hot or feverish
For participants who completed Day 29 or discontinued before Day 29 without symptom alleviation, censoring was used when the specified conditions were met. The registry describes the censoring time as the last date/time at which symptom alleviation was assessed or Day 28, whichever occurred first, relative to the first dose date/time.
7. Primary Result: Time to COVID-19 Symptom Alleviation by Day 29
The posted primary analysis compared obeldesivir with placebo in the Full Analysis Positive Set using a stratified log-rank test. The registry states that the hazard ratio and two-sided 95% confidence interval were estimated using a Cox regression model with the randomization stratification factor as a covariate.
Hazard ratio for symptom alleviation
95% CI: 0.997–1.211 · P = 0.0681
Analysis: stratified log-rank test; Cox regression used for the hazard ratio and confidence interval.
| Primary endpoint | Obeldesivir vs placebo | 95% CI | P-value | Analysis |
|---|---|---|---|---|
| Time to COVID-19 symptom alleviation by Day 29 | HR 1.099 | 0.997–1.211 | 0.0681 | Stratified log-rank test; Cox regression with randomization stratification factor as covariate |
The hazard ratio of 1.099 is a relative time-to-event measure. In the fitted analysis, a hazard ratio above 1 indicates a higher estimated instantaneous rate of reaching the symptom-alleviation event in the obeldesivir group than in the placebo group. In practical terms, the point estimate is compatible with faster symptom alleviation under the modeled comparison.
The estimate does not mean that participants were 9.9% more likely to recover by Day 29. A hazard ratio describes the relative event rate over follow-up; it is not the same as a risk ratio, an absolute percentage-point difference, or a probability that an individual participant will benefit.
The two-sided 95% confidence interval, 0.997–1.211, describes uncertainty around the estimated hazard ratio. Its lower boundary is very close to 1, while the interval extends above 1. The interval therefore communicates that the exact magnitude of the relative event-rate difference is uncertain.
The P-value of 0.0681 measures the compatibility of the observed test statistic with the null hypothesis under the specified testing framework. It does not measure the size of the treatment effect, the probability that the treatment works, or the clinical importance of the result.
Because this is a time-to-event analysis, interpretation also depends on censoring and on the assumptions underlying the Cox model used to estimate the hazard ratio. The registry does not provide a proportional-hazards diagnostic in the posted statistical-analysis record, so the HR should be understood as the model-based summary reported for this endpoint.
8. Secondary Time-to-Event Results
Time to COVID-19 Symptom Resolution by Day 29
This secondary endpoint was analyzed from Day 1 up to Day 29 in participants in the Full Analysis Positive Set with available data. The registry reports a stratified log-rank test and a Cox-model hazard ratio using the randomization stratification factor as a covariate.
Hazard ratio for symptom resolution
95% CI: 0.935–1.147 · P = 0.5558
The hazard ratio of 1.036 is close to 1, so the estimated relative event rate for symptom resolution was close to that of the placebo comparison under the reported model. The 95% CI of 0.935–1.147 spans 1 and provides a range of plausible hazard-ratio values on both sides of the null value.
The P-value of 0.5558 is not an effect-size measure. It indicates that the observed test statistic was not unusual under the null hypothesis used for this comparison. It should not be converted into a statement about the probability that obeldesivir is effective or ineffective.
COVID-19 Related Medically Attended Visits or All-cause Death by Day 29
The registry analyzed this endpoint with a log-rank test in the Full Analysis Positive Set. The hazard ratio and two-sided confidence interval were estimated using Cox regression.
Hazard ratio for medically attended visit or death
95% CI: 0.045–5.464 · P = 0.5581
The point estimate of 0.495 corresponds to a lower estimated instantaneous event rate in the obeldesivir group under the fitted Cox model. However, the confidence interval is extremely wide: 0.045–5.464. That width is the most important statistical feature of this estimate because it indicates substantial uncertainty about the magnitude and even the direction of the underlying relative effect.
A hazard ratio near 0.5 should therefore not be interpreted in isolation. The P-value of 0.5581 does not provide evidence about the size of the effect; it is the result of the specified log-rank test. The combination of a wide confidence interval and the reported P-value shows why an isolated point estimate can be misleading when an endpoint has relatively limited information.
Time to Antigen Negativity
The Antigen Analysis Set was used for this endpoint. Participants needed to be randomized, receive at least one dose, have at least one positive rapid antigen test from Day 1 to Day 5, and have a positive baseline SARS-CoV-2 PCR according to the registry definition.
Hazard ratio for antigen negativity
95% CI: 1.032–1.256 · P = 0.0015
The hazard ratio of 1.138 indicates a higher estimated instantaneous rate of reaching antigen negativity in the obeldesivir group under the reported Cox model. Unlike the medically attended visit/death endpoint, the 95% CI of 1.032–1.256 lies above 1, indicating that the reported estimate is on the side of the null associated with faster antigen negativity in the obeldesivir group.
The P-value of 0.0015 is the result of the specified stratified log-rank analysis. It should not be read as a measure of the size of the effect. The magnitude of the estimated effect is described by the hazard ratio and its confidence interval.
Antigen negativity is also biologically and clinically different from symptom alleviation. A statistical effect on one endpoint does not automatically establish the same effect on another endpoint measuring a different aspect of COVID-19.
9. Secondary Virology Result: Change From Baseline in SARS-CoV-2 Viral Load
The registry analyzed change from baseline in SARS-CoV-2 nasal swab viral load at Day 5 in the Virology Analysis Set using an MMRM. Baseline viral load and randomization strata were included as covariates.
Least-squares mean difference
95% CI: -0.03–0.08 · P = 0.4254
Outcome unit: log10 copies/mL
The reported least-squares mean difference of 0.02 is the model-adjusted comparison of change from baseline at Day 5, expressed in log10 copies/mL. Because the confidence interval ranges from -0.03 to 0.08, the estimated difference is compatible with effects on either side of zero under the model.
The MMRM framework is useful for repeated or longitudinal measurements because it can model the outcome across visits while accounting for baseline covariates and the within-participant correlation of repeated observations. Here, the registry specifically identifies baseline viral load and randomization strata as covariates.
The P-value of 0.4254 does not quantify the magnitude of the viral-load difference. The estimate and confidence interval provide the more direct description of the modeled treatment contrast.
10. Secondary Categorical Results
Moderate Relapse of COVID-19 Symptoms by Day 29
The registry analyzed this binary endpoint using Fisher's exact test among participants in the Full Analysis Positive Set who had short symptom recovery and available data.
Fisher exact test · P = 0.6469
Fisher's exact test is appropriate for a two-group comparison of a binary outcome, particularly when cell counts may be small enough that a large-sample approximation is undesirable. The registry reports the P-value but does not provide an effect estimate and confidence interval for this analysis in the posted statistical-analysis entry.
Viral Antigen Rebound
This binary endpoint was evaluated among participants from the Antigen Analysis Set who had achieved antigen negativity. The registry again used Fisher's exact test.
Fisher exact test · P = 0.6978
The analysis addresses the frequency of antigen rebound among the specified subset rather than the time to antigen negativity itself. That distinction is important: the denominator is defined by prior antigen negativity, whereas the time-to-negativity analysis considers the transition to negativity over follow-up.
Relapse of COVID-19 Symptoms by Day 29
The registry analyzed relapse using Fisher's exact test among participants in the Full Analysis Positive Set with available data.
Fisher exact test · P = 0.5707
Again, the P-value is a test result rather than an effect-size measure. Without an accompanying risk difference, risk ratio, odds ratio, or confidence interval in the posted analysis entry, the P-value alone does not describe the magnitude of any between-group difference.
11. Statistical Methodology
Stratified log-rank test
The primary endpoint and several secondary time-to-event endpoints were analyzed using stratified log-rank methods. The registry states that the randomization stratification factor was used as the stratum for the relevant analyses.
A log-rank test compares the observed pattern of event times between treatment groups while accounting for the ordering of events over follow-up. Stratification allows comparisons to be made within strata and then combined, reducing the influence of the stratification factor on the treatment comparison.
For symptom-related endpoints, "survival" is better understood as remaining free of the event being analyzed. For symptom alleviation, the event is symptom alleviation; therefore, the event hazard concerns reaching alleviation rather than experiencing death.
Cox regression for hazard ratios
The registry reports that hazard ratios and two-sided 95% confidence intervals were estimated using Cox regression, with the randomization stratification factor included as a covariate for the relevant analyses.
The direction depends on the definition of the event. For symptom alleviation or antigen negativity, the event is a favorable transition; for an endpoint involving a medically attended visit or death, the event represents an unfavorable clinical outcome.
This distinction is essential. A hazard ratio of 1.138 for time to antigen negativity and a hazard ratio of 0.495 for medically attended visit or death cannot be interpreted with the same verbal shorthand simply because both are greater than or less than 1. The meaning depends on what event is being modeled.
MMRM
The change-from-baseline viral-load endpoint was analyzed using an MMRM. The registry specifies baseline viral load and randomization strata as covariates.
An MMRM is designed for repeated measurements in which each participant may contribute multiple observations over time. The model can estimate adjusted mean differences while accounting for correlation among repeated observations from the same participant. The treatment estimate is therefore a model-based comparison rather than a simple subtraction of two unadjusted group means.
Fisher exact test
Fisher's exact test was used for the binary relapse and rebound endpoints. The test evaluates the association between treatment assignment and a binary outcome using the observed contingency table.
Unlike a hazard ratio, Fisher's exact test does not inherently provide a time-to-event effect estimate. Its P-value answers a different statistical question from the Cox-model hazard ratios used for the time-to-event outcomes.
Multiple endpoint interpretation
OAKTREE has four registered primary endpoints and multiple secondary outcomes. Consequently, each result needs to be interpreted in the context of its endpoint role and prespecified hypothesis framework. A small P-value on one secondary endpoint does not, by itself, establish that all endpoints are positive, and a P-value from one endpoint should not be treated as interchangeable with the P-value from another endpoint.
12. Statistical Methods Explained
Why was a stratified log-rank test used for the primary endpoint?
The primary outcome is a time-to-event endpoint: the time from first dose to sustained symptom alleviation. A stratified log-rank test compares event-time distributions while incorporating the randomization stratification factor. This is preferable to reducing the endpoint to a single binary indicator because participants who reach symptom alleviation at different times contribute different amounts of information.
What does a hazard ratio of 1.099 mean for symptom alleviation?
A hazard ratio of 1.099 means that the fitted model estimated a higher instantaneous rate of reaching symptom alleviation in the obeldesivir group than in the placebo group. It does not mean that exactly 9.9% more participants recovered, nor does it mean that every participant had a 9.9% improvement in recovery probability.
Why is the direction of a hazard ratio endpoint-dependent?
For time to symptom alleviation, reaching the event is favorable, so a hazard ratio above 1 corresponds to faster occurrence of that event. For an endpoint defined as medically attended visit or death, reaching the event is unfavorable, so a hazard ratio below 1 corresponds to a lower estimated event rate. The event definition must therefore always be read alongside the HR.
Why is the confidence interval important?
A point estimate such as 1.099 is only one estimate of the treatment contrast. The 95% CI of 0.997–1.211 shows the uncertainty surrounding that estimate under the specified model and sampling framework. The much wider interval for medically attended visits or death, 0.045–5.464, illustrates how an apparently notable point estimate can be statistically imprecise.
Why was MMRM used for viral load?
Viral load is a continuous measurement and can be observed repeatedly over time. MMRM provides a framework for modeling repeated continuous outcomes while accounting for within-participant correlation and incorporating baseline viral load and randomization strata as covariates. The resulting least-squares mean difference is therefore adjusted rather than a simple unadjusted difference.
Why use Fisher's exact test for relapse and rebound?
Relapse and rebound are binary outcomes in the posted analyses. Fisher's exact test provides an exact test of association between treatment and the binary outcome without relying on the same large-sample approximation used by a conventional chi-square test. The registry reports Fisher's exact testing for the specified analyses.
13. Primary Endpoint Interpretation in Context
The primary symptom-alleviation analysis deserves careful interpretation because the estimated hazard ratio was above 1 while the two-sided P-value was 0.0681. The point estimate and confidence interval should be considered together rather than treating either number as a standalone conclusion.
Effect estimate
HR 1.099 indicates a higher estimated rate of reaching symptom alleviation in the obeldesivir group under the reported Cox model.
Precision
The 95% CI of 0.997–1.211 indicates that the precise magnitude of the relative difference is uncertain.
Hypothesis test
The two-sided P-value of 0.0681 is the result of the prespecified statistical comparison; it is not a measure of treatment-effect magnitude.
Endpoint meaning
The event is sustained symptom alleviation, so the HR describes the rate of reaching that favorable event rather than the rate of an adverse event.
The distinction between the hazard ratio and the P-value is especially important in time-to-event analyses. A P-value answers a hypothesis-testing question, while the HR provides an estimate of relative event rate and the confidence interval describes uncertainty around that estimate. Reporting all three gives a more complete statistical description than any one alone.
14. Safety
The registry reports serious adverse events by randomized arm using affected participants over those at risk:
| Safety outcome | Affected / at risk |
|---|---|
| Obeldesivir | 2 / 979 |
| Placebo | 4 / 976 |
These figures correspond to serious adverse events by arm as reported in the registry. The numerator is the number of affected participants and the denominator is the number at risk in the corresponding arm.
The registered primary safety endpoints also include treatment-emergent adverse events, laboratory abnormalities, and serious adverse events or adverse events leading to study-drug discontinuation. The registry defines the relevant time windows, but the posted statistical-analysis entries represented here do not provide estimate-and-confidence-interval analyses for those binary primary safety endpoints.
15. Primary Endpoint Versus Secondary Endpoints
| Endpoint | Role | Method | Effect measure | Reported result |
|---|---|---|---|---|
| Time to COVID-19 symptom alleviation by Day 29 | Primary | Stratified log-rank; Cox regression | HR | 1.099 (95% CI 0.997–1.211), P = 0.0681 |
| Time to COVID-19 symptom resolution by Day 29 | Secondary | Stratified log-rank; Cox regression | HR | 1.036 (95% CI 0.935–1.147), P = 0.5558 |
| COVID-19 related medically attended visits or all-cause death by Day 29 | Secondary | Log-rank; Cox regression | HR | 0.495 (95% CI 0.045–5.464), P = 0.5581 |
| Change from baseline in SARS-CoV-2 nasal swab viral load at Day 5 | Secondary | MMRM | Least-squares mean difference | 0.02 (95% CI -0.03–0.08), P = 0.4254 |
| Time to antigen negativity | Secondary | Stratified log-rank; Cox regression | HR | 1.138 (95% CI 1.032–1.256), P = 0.0015 |
| Moderate relapse of COVID-19 symptoms by Day 29 | Secondary | Fisher exact test | P-value | P = 0.6469 |
| Viral antigen rebound | Secondary | Fisher exact test | P-value | P = 0.6978 |
| Relapse of COVID-19 symptoms by Day 29 | Secondary | Fisher exact test | P-value | P = 0.5707 |
The table illustrates why trial interpretation should remain endpoint-specific. The analyses do not all measure the same biological or clinical process, use the same population, or have the same statistical estimand. A time-to-event HR, an adjusted mean difference, and a Fisher exact-test P-value are different statistical objects.
16. Understanding the Antigen-Negativity Result
The time-to-antigen-negativity result is statistically distinct from the primary symptom endpoint. Its hazard ratio was 1.138, with a 95% CI of 1.032–1.256 and P = 0.0015.
The HR above 1 indicates a higher estimated rate of reaching antigen negativity in the obeldesivir group. Because antigen negativity is the event, the analysis is asking how rapidly participants transition to that state.
The confidence interval provides an estimated range around the relative event rate. Unlike the primary symptom-alleviation estimate, this interval lies above 1, so the reported model-based estimate is consistently on the same side of the null across the interval.
This result should not automatically be translated into a claim about symptom benefit. Antigen status and symptom status are related but distinct endpoints, and the trial separately analyzed them for that reason.
17. Understanding the Viral-Load Analysis
The Day 5 viral-load analysis differs fundamentally from the time-to-event analyses because viral load is a continuous quantitative measurement. The registry reports a least-squares mean difference of 0.02 log10 copies/mL, with a 95% CI of -0.03–0.08 and P = 0.4254.
The registry identifies baseline viral load and randomization strata as covariates in the MMRM. The reported estimate therefore represents an adjusted comparison of change rather than a raw difference in observed group means.
The confidence interval crosses zero, which is the null value for a mean difference. The result therefore illustrates why a continuous-outcome analysis should be read through the estimate, its uncertainty interval, and its hypothesis test together rather than through the P-value alone.
18. Censoring and Time-to-Event Analysis
The primary endpoint uses censoring for participants who reach the end of the Day 29 observation window or discontinue before Day 29 without symptom alleviation under the registry's specified conditions. Censoring allows those participants to contribute information for the period during which their event status was observed.
This is one of the central advantages of Kaplan-Meier and related survival-analysis methods. A participant does not have to experience the event for the analysis to use the participant's available follow-up. The statistical model can distinguish an observed event from an observation that ends before the event occurs.
For a symptom-alleviation endpoint, an event means that the specified sustained symptom improvement was observed. A censored observation contributes follow-up until the registry-defined censoring point.
However, censoring is not simply a technical convenience. The validity of a time-to-event analysis depends on assumptions about how censoring relates to the underlying event process. The registry describes the censoring rule for the primary endpoint but does not provide a detailed diagnostic of censoring assumptions in the posted statistical-analysis entry.
19. Confidence Intervals and Statistical Precision
The OAKTREE results illustrate three very different levels of statistical precision.
| Endpoint | Estimate | 95% CI | Width / interpretation |
|---|---|---|---|
| Symptom alleviation | HR 1.099 | 0.997–1.211 | Relatively concentrated around the point estimate; lower boundary close to 1. |
| Medically attended visit or death | HR 0.495 | 0.045–5.464 | Very wide interval; substantial uncertainty about the treatment effect. |
| Viral load | 0.02 | -0.03–0.08 | Continuous-outcome difference with uncertainty on both sides of zero. |
| Antigen negativity | HR 1.138 | 1.032–1.256 | Interval lies above 1 for the reported relative event-rate measure. |
The medically attended visit/death endpoint is especially instructive. Its point estimate is below 1, but the confidence interval ranges from 0.045 to 5.464. That interval is compatible with a substantial reduction, little difference, or a substantial increase in the modeled event rate. The point estimate alone would conceal that uncertainty.
20. P-values: What They Do and Do Not Tell You
The posted analyses contain P-values ranging from 0.0015 for time to antigen negativity to 0.6978 for viral antigen rebound. These numbers should be interpreted in the context of the corresponding endpoint, analysis population, test, and hypothesis.
P-value is not effect size
A P-value does not tell you how large a treatment effect is. Effect size is described by measures such as HRs or mean differences.
P-value is not clinical importance
A statistical test does not determine whether an effect is clinically meaningful. Clinical interpretation requires understanding the endpoint and magnitude.
P-value is not probability of benefit
A P-value is not the probability that obeldesivir works or that the null hypothesis is true.
Endpoint matters
P = 0.0015 for antigen negativity and P = 0.0681 for symptom alleviation answer different statistical questions about different outcomes.
21. Limitations
- Different endpoints measure different phenomena: symptom alleviation, symptom resolution, medically attended events or death, viral load, antigen negativity, relapse, and rebound should not be treated as interchangeable measures of one outcome.
- Analysis populations differ: the primary endpoint uses the FAPS, whereas viral-load and antigen endpoints use specifically defined analysis sets.
- Subset analyses can reduce information: relapse and rebound analyses apply additional eligibility conditions, so their denominators differ conceptually from the full randomized population.
- Wide uncertainty for some outcomes: the 95% CI for medically attended visits or death spans 0.045–5.464, demonstrating substantial statistical imprecision for that hazard-ratio estimate.
- Hazard-ratio assumptions: Cox regression produces a model-based HR. Interpretation depends on the suitability of the proportional-hazards framework over the relevant follow-up.
- Multiple endpoints: the trial includes four registered primary endpoints and multiple secondary outcomes. Each result must be interpreted according to its prespecified statistical role rather than treating every P-value as an equivalent confirmatory test.
- Available-data analyses: several endpoints specify that participants with available data were analyzed. The resulting estimand therefore depends on the defined analysis population and observed data.
- Binary primary safety endpoints: the registry lists several binary safety outcomes as primary endpoints, but the posted statistical-analysis entries represented here do not provide estimate-and-confidence-interval analyses for those endpoints.
22. Why This Trial Matters Statistically
OAKTREE is a useful statistical case study because it combines several common clinical-trial methods within one randomized phase 3 study. The primary endpoint is a time-to-event outcome, while the trial also uses longitudinal continuous-outcome modeling and exact testing for binary outcomes.
| Concept | How it appears in OAKTREE |
|---|---|
| Randomization | Participants were randomized to obeldesivir or placebo. |
| Double blinding | The registry identifies the study as double-blind. |
| Time-to-event analysis | Primary symptom alleviation and several secondary endpoints were analyzed as time-to-event outcomes. |
| Stratified log-rank test | Used for the primary symptom endpoint and specified secondary time-to-event analyses. |
| Hazard ratio | Used to summarize relative event rates for time-to-event endpoints. |
| Cox regression | Used to estimate hazard ratios and two-sided 95% confidence intervals. |
| MMRM | Used for change from baseline in Day 5 SARS-CoV-2 nasal swab viral load. |
| Covariate adjustment | Baseline viral load and randomization strata were used as covariates in the viral-load MMRM; randomization stratification was incorporated into relevant Cox analyses. |
| Fisher exact test | Used for relapse and rebound binary endpoints. |
| Multiple endpoints | Four registered primary endpoints plus multiple secondary outcomes require endpoint-specific interpretation. |
23. What the Primary Result Means Statistically
The primary HR of 1.099 represents the estimated relative instantaneous rate of reaching the defined symptom-alleviation event in the obeldesivir group compared with placebo under the reported Cox model.
The 95% CI of 0.997–1.211 quantifies uncertainty around the estimated HR. It is not a range containing 95% of individual patient outcomes and does not describe the variability of treatment effects across individual participants.
The P-value of 0.0681 is the result of the reported two-sided stratified log-rank hypothesis test. It is not the probability that the treatment has no effect, and it does not quantify the clinical magnitude of the estimated difference.
The event requires sustained improvement for at least 48 consecutive hours across the targeted symptoms according to their baseline severity. Consequently, the statistical event is more specific than simply reporting whether a participant felt better at a single assessment.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Calculators
26. Sources
- ClinicalTrials.gov: OAKTREE, NCT05715528.
- PubMed: PMID 41475511.
- PubMed: PMID 40675167.
Continue through the Clinical Biostats knowledge graph
Connect the trial's endpoints and statistical methods to deeper tutorials and practical statistical tools.
27. Record Summary
OAKTREE provides a useful example of how a randomized phase 3 trial can evaluate a clinical question through several complementary statistical frameworks. Its primary endpoint, time to COVID-19 symptom alleviation by Day 29, was analyzed with a stratified log-rank test and a Cox model, producing an HR of 1.099 with a two-sided 95% CI of 0.997–1.211 and P = 0.0681. Secondary analyses extended the statistical assessment to symptom resolution, medically attended visits or death, viral load, antigen negativity, symptom relapse, and antigen rebound.
The trial also demonstrates why statistical interpretation should not rely on P-values alone. Hazard ratios describe relative event rates, confidence intervals describe uncertainty around estimated effects, MMRM produces adjusted longitudinal comparisons for continuous outcomes, and Fisher's exact test addresses specified binary outcomes. Reading those methods together with the endpoint definitions and analysis populations provides a more complete statistical understanding of the evidence.