← Clinical Trials
COVID-19 Phase 2/3 Randomized NCT04960202

EPIC-HR: Complete Statistical Analysis of PF-07321332/Ritonavir in COVID-19

An independent statistical analysis of the randomized EPIC-HR trial of oral PF-07321332 300 mg plus ritonavir 100 mg compared with placebo in nonhospitalized high-risk adults with COVID-19, focusing on the registered primary hospitalization-or-death endpoint, reported secondary analyses, and the statistical methods used to analyze them.

Trial status: COMPLETED  ·  Enrollment: 2091  ·  Sponsor: Pfizer
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results on this page are restricted to the information contained in the ClinicalTrials.gov trial data posted on ClinicalTrials.gov for EPIC-HR and the linked PubMed records identified in that data.

Registry note: This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

EPIC-HR was a randomized, parallel-group, quadruple-masked phase 2/3 treatment trial in COVID-19. The registered primary endpoint was the percentage of participants with COVID-19 related hospitalization or death from any cause through Day 28, analyzed in a modified intent-to-treat population.

2091
Enrolled
Total trial enrollment
2
Arms
Parallel-group design
-6.137
Primary risk difference
95% CI -8.208 to -4.066
<0.0001
Primary P-value
Two-sided Wald / z-test
FeatureEPIC-HR
Trial nameEPIC-HR
NCT identifierNCT04960202
Therapeutic areaInfectious Disease
ConditionCOVID-19
PhasePhase 2/3
StatusCOMPLETED
AllocationRANDOMIZED
Design modelPARALLEL
MaskingQUADRUPLE
Primary purposeTREATMENT
Enrollment2091
Arms2
Start2021-07-16
Primary completion2021-12-09
Lead sponsorPfizer

2. Clinical Question

The central statistical question was whether PF-07321332 300 mg plus ritonavir 100 mg differed from placebo with respect to the registered primary endpoint: the percentage of participants with COVID-19 related hospitalization or death from any cause during the first 28 days of the study.

Population

Nonhospitalized high-risk adults with COVID-19, as represented by the EPIC-HR trial registry and its modified intent-to-treat analysis population.

Intervention

PF-07321332 300 mg plus ritonavir 100 mg.

Comparator

Placebo.

Primary question

Is the percentage of participants experiencing COVID-19 related hospitalization or death from any cause through Day 28 different between the randomized treatment groups, under a superiority framework?

3. Trial Design

01
Randomize 2091 enrolled
02
Mask Quadruple-masked
03
Treat PF-07321332 + ritonavir or placebo
04
Follow Day 1 through Day 28
05
Analyze Primary and secondary endpoints
Allocation
Randomized allocation in a parallel-group design.
Masking
QUADRUPLE masking was registered.
Primary purpose
Treatment.
Statistical profile
The registry reports Cox proportional-hazards models, logistic regression, Wald / z-tests, and other analyses.
ARM A

PF-07321332 + ritonavir

  • PF-07321332 300 mg
  • Ritonavir 100 mg
  • Compared with placebo on the registered primary endpoint
  • Secondary time-to-event and binary outcomes were also analyzed
ARM B

Placebo

  • Placebo
  • Randomized parallel comparator
  • Used as the reference group for the reported treatment comparisons
  • Included in the same Day 28 efficacy framework

4. Endpoints

The registry identifies one primary endpoint, a time-to-event endpoint operationalized as a percentage through Day 28 using Kaplan-Meier estimation. The posted statistical analysis reports the corresponding treatment comparison as a percentage difference, normalized here as a risk difference.

EndpointTime frameTypeAnalysis population
Percentage of Participants With Covid-19 Related Hospitalization or Death From Any Cause Through Day 28- Modified Intent-To-Treat (mITT) Population From Day 1 to Day 28 Time-to-event; reported result expressed as a percentage difference mITT

Registered primary endpoint definition

The registry describes the primary endpoint as the percentage of participants with COVID-19 related hospitalization or death from any cause during the first 28 days of the study. The registry definition states that this percentage was estimated using the Kaplan-Meier method, with survival probability for each time interval calculated from the number surviving divided by the number at risk.

This distinction is statistically important. Although the endpoint is displayed as a percentage, the registry describes it using a time-to-event framework. Participants who have not experienced the event by the relevant follow-up can contribute censored information rather than simply being treated as ordinary event-free observations at Day 28.

5. Analysis Populations

PopulationRegistry description / role
mITT Included all participants who were randomized and took at least one dose of study intervention, who at baseline did not receive nor were expected to receive COVID-19 therapeutic monoclonal antibody treatment, as described in the registry analysis definition.
mITT1 Included all participants who were randomized and took at least one dose of study intervention and who at baseline did not receive nor were expected to receive COVID-19 therapeutic monoclonal antibody treatment.
mITT2 Included all participants who were randomized and took at least one dose of study intervention. The registry specifies that the analyzed number represents participants evaluable for the relevant outcome.

The distinction between mITT, mITT1, and mITT2 matters because the same nominal endpoint can produce different estimates when the analysis population changes. A treatment effect is always tied to the population in which it was estimated; estimates from different populations should not be silently pooled or treated as interchangeable.

6. Statistical Methodology

Kaplan-Meier estimation

The registry's primary endpoint definition uses the Kaplan-Meier method to estimate the percentage of participants experiencing COVID-19 related hospitalization or death through Day 28. Kaplan-Meier estimation is designed for time-to-event data and allows participants with incomplete event follow-up to contribute information up to their censoring time.

Conceptual form
S(t) = ∏ti ≤ t (1 − di/ni)

where di is the number of events at time ti and ni is the number at risk immediately before that time.

For this endpoint, the reported quantity is framed as the percentage with the event rather than the survival probability itself. The underlying Kaplan-Meier framework therefore matters even though the final treatment comparison is reported as a percentage difference.

Wald / z-test and normal approximation

The primary statistical analysis is reported as a Normal approximation, normalized to a Wald / z-test. The treatment comparison is the difference in the percentages between the two randomized groups, accompanied by a two-sided 95% confidence interval and a p-value.

Primary effect measure
Risk difference = PercentagePF-07321332 + ritonavir − Percentageplacebo

A negative risk difference means that the estimated percentage experiencing the event was lower in the PF-07321332 plus ritonavir group than in the placebo group.

Cox proportional-hazards models

Several secondary time-to-event endpoints were analyzed with Cox proportional-hazards models. For time to sustained alleviation and time to sustained resolution of targeted COVID-19 signs and symptoms, the registry reports treatment and geographic region as model terms, together with prespecified baseline covariates that vary by analysis population.

For example, the mITT analysis of time to sustained alleviation included baseline SARS-CoV-2 serology status and baseline viral load, with viral load categorized as <4 log10 copies/mL versus ≥4 log10 copies/mL. Other analyses additionally included symptom onset duration and, for the mITT2 analysis, COVID-19 monoclonal-antibody treatment status.

Logistic regression

Binary secondary endpoints such as severe COVID-19 signs and symptoms and progression to worsening status were analyzed using logistic regression. The registry reports odds ratios with 95% confidence intervals and p-values. These models included treatment, geographic region, and selected baseline covariates.

Other registered analyses

The registry also reports analyses using the Breslow-Day test for the Day 1 and Day 5 resting peripheral oxygen saturation endpoint. These analyses are distinct from the primary Wald / z-test and secondary Cox and logistic-regression analyses.

7. Primary Result

The posted primary analysis compares the percentage of participants with COVID-19 related hospitalization or death from any cause through Day 28 in the mITT population. The registry reports a normal-approximation comparison of the two treatment-group percentages.

Primary risk difference

-6.137 percentage points

95% CI: -8.208 to -4.066   ·   P < 0.0001

PF-07321332 300 mg + ritonavir 100 mg vs placebo; mITT population; two-sided 95% CI.

Clinical Biostats interpretation

The estimate of -6.137 percentage points means that the estimated percentage experiencing the registered primary event was 6.137 percentage points lower in the PF-07321332 300 mg plus ritonavir 100 mg group than in the placebo group, using the analysis specified in the registry.

The estimate is a risk difference, not a relative risk, odds ratio, or hazard ratio. It therefore describes an absolute difference in the estimated event percentages between the randomized groups. It does not mean that an individual participant's risk was reduced by exactly 6.137%, nor does it provide the relative magnitude of treatment effect.

The 95% confidence interval of -8.208 to -4.066 describes the statistical uncertainty around the estimated difference under the analysis framework. Because the entire interval is below zero, the interval is consistent with a lower estimated event percentage in the intervention group relative to placebo. The interval does not describe the range of individual patient responses.

The p-value < 0.0001 measures the evidence against the null hypothesis under the reported statistical test. It is not a measure of effect size and does not say that the probability the treatment is effective is <0.0001. The size of the treatment difference is described by the risk difference and its confidence interval.

Because the endpoint is described by the registry using Kaplan-Meier estimation, censoring and the time-to-event construction are relevant to interpretation. The reported risk difference should therefore be understood in the context of the registry's Day 28 time-to-event analysis rather than treated as an unadjusted simple proportion difference reconstructed independently from the headline estimate.

8. Secondary Results: Sustained Alleviation of Symptoms

The registry reports three analyses of time to sustained alleviation of all targeted COVID-19 signs and symptoms through Day 28. All three use Cox proportional-hazards models and report hazard ratios greater than 1.

Analysis populationHazard ratio95% CIP-value
mITT 1.294 1.136–1.476 0.0003
mITT1 1.266 1.134–1.412 <0.0001
mITT2 1.258 1.131–1.400 <0.0001

For these analyses, the Cox models included treatment and geographic region as independent variables. The mITT analysis additionally adjusted for baseline SARS-CoV-2 serology status and baseline viral load. The mITT1 analysis additionally included symptom onset duration, while the mITT2 analysis additionally included COVID-19 monoclonal-antibody treatment status.

A hazard ratio above 1 is consistent with a higher estimated instantaneous rate of reaching the sustained-alleviation event in the PF-07321332 plus ritonavir group relative to placebo. The confidence intervals for all three analyses lie above 1.

9. Secondary Results: Sustained Resolution of Symptoms

The registry also reports three Cox-model analyses of time to sustained resolution of all targeted COVID-19 signs and symptoms through Day 28.

Analysis populationHazard ratio95% CIP-value
mITT 1.219 1.061–1.401 0.0053
mITT1 1.200 1.068–1.348 0.0022
mITT2 1.194 1.066–1.337 0.0021

The models used treatment and geographic region as independent variables, with baseline covariates corresponding to the relevant analysis population. For the mITT2 analysis, these included symptom onset duration, COVID-19 monoclonal-antibody treatment status, baseline SARS-CoV-2 serology status, and baseline viral load.

How to interpret the hazard ratios

A hazard ratio such as 1.219 means that, under the fitted Cox model, the estimated instantaneous rate of reaching sustained symptom resolution was higher in the PF-07321332 plus ritonavir group than in placebo. It does not mean that 21.9% more participants achieved resolution, nor does it mean that the time to resolution was reduced by 21.9%.

The distinction is important because hazard ratios describe relative event rates over time. They are not direct measures of median time, absolute probability, or average number of days saved.

10. Secondary Results: Severe COVID-19 Signs and Symptoms

Severe COVID-19 signs and symptoms through Day 28 were analyzed with logistic regression in the mITT, mITT1, and mITT2 populations.

Analysis populationOdds ratio95% CIP-value
mITT 0.871 0.652–1.162 0.3473
mITT1 0.936 0.740–1.182 0.5762
mITT2 0.969 0.773–1.213 0.7807

The registry states that the odds ratios, confidence intervals, and p-values were computed from logistic regression models. The covariates included treatment, geographic region, and population-specific baseline variables including SARS-CoV-2 serology, viral load, symptom onset duration, and COVID-19 monoclonal-antibody treatment status where applicable.

Why the odds ratio is not a risk ratio

An odds ratio of 0.871 does not mean that the probability of severe signs and symptoms was 12.9% lower. Odds and probabilities are different quantities. The odds ratio compares odds, while the risk ratio compares probabilities.

The 95% confidence interval of 0.652–1.162 includes 1, as do the corresponding intervals for the mITT1 and mITT2 analyses. The reported p-values are therefore not small under the stated analyses. That statistical observation should not be converted into a claim that the treatment has no possible effect; the confidence interval shows the range of effect estimates compatible with the statistical framework.

11. Secondary Results: Progression to Worsening Symptoms

The registry reports logistic-regression analyses for the number of participants with progression to a worsening status in one or more self-reported COVID-19 associated symptoms through Day 28.

Analysis populationOdds ratio95% CIP-value
mITT 1.088 0.836–1.416 0.5293
mITT1 1.053 0.850–1.303 0.6379
mITT2 1.046 0.848–1.290 0.6760

Each analysis used treatment and geographic region as model terms, with additional covariates according to the analysis population. The mITT2 model additionally incorporated COVID-19 monoclonal-antibody treatment status.

These estimates illustrate why the direction of an odds ratio should not be interpreted in isolation. An odds ratio above 1 can indicate higher estimated odds in the intervention group for the specified outcome, but the confidence interval and p-value are essential for understanding the uncertainty around that estimate.

12. Secondary Results: Resting Peripheral Oxygen Saturation

The registry includes analyses of the percentage of participants with a resting peripheral oxygen saturation ≥95% at Days 1 and 5. These analyses were reported using the Breslow-Day test and an odds ratio describing Day 5 versus Day 1 within each treatment group.

Analysis populationGroupOdds ratio95% CIP-value
mITT PF-07321332 300 mg + ritonavir 100 mg 19.400 7.788–48.328 0.1997
mITT Placebo 8.948 4.159–19.253 Not stated in the ClinicalTrials.gov record
mITT1 PF-07321332 300 mg + ritonavir 100 mg 20.875 10.097–43.156 0.2810
mITT1 Placebo 12.452 6.823–22.725 Not stated in the ClinicalTrials.gov record
mITT2 PF-07321332 300 mg + ritonavir 100 mg 21.119 10.412–42.837 0.2226
mITT2 Placebo 12.036 6.808–21.280 0.2342
Important interpretation point: these odds ratios are not treatment-group odds ratios comparing PF-07321332 plus ritonavir with placebo. The registry analysis notes identify them as odds ratios for Day 5 versus Day 1 within the specified group. Consequently, an odds ratio such as 19.400 should not be read as saying that the intervention group had 19.400 times the odds of the oxygen-saturation endpoint compared with placebo.

13. Safety

The ClinicalTrials.gov record reports serious adverse events by randomized treatment arm. The counts are presented as affected participants over participants at risk.

Safety measurePF-07321332 300 mg + ritonavir 100 mgPlacebo
Serious adverse events 19 / 1038 72 / 1053

The registry-derived ClinicalTrials.gov record do not provide a formal between-group statistical analysis of these serious-adverse-event counts. Accordingly, the page reports the counts without constructing an additional risk ratio, risk difference, confidence interval, or p-value.

This distinction illustrates an important principle in trial reporting: descriptive safety counts and formal hypothesis tests are different types of evidence. A descriptive difference in event counts does not by itself establish a statistically tested treatment effect.

14. Statistical Methods Explained

Why was a Wald / z-test used for the primary comparison?

The registry reports a normal-approximation analysis for the difference in the two treatment-group percentages. A Wald / z-test provides a direct hypothesis test for whether the estimated percentage difference differs from zero and is paired with a confidence interval for the magnitude of that difference.

What does a risk difference of -6.137 mean?

A risk difference of -6.137 means that the estimated percentage for the specified primary event was 6.137 percentage points lower in the PF-07321332 plus ritonavir group than in the placebo group. The sign tells us the direction; the absolute value describes the magnitude on the percentage-point scale.

Why does the confidence interval matter?

The point estimate is only one estimate from the sample. The 95% confidence interval of -8.208 to -4.066 shows the statistical uncertainty around that estimate under the reported analysis. A confidence interval also makes it easier to distinguish a precisely estimated difference from a very uncertain one.

What does a hazard ratio of 1.294 mean?

For the mITT analysis of time to sustained alleviation, a hazard ratio of 1.294 indicates a higher estimated instantaneous rate of reaching the sustained-alleviation event in the PF-07321332 plus ritonavir group relative to placebo under the fitted Cox model. It is not a 29.4% reduction in days to symptom alleviation and is not a risk ratio.

Why were covariates included in the Cox models?

The registry specifies treatment and geographic region as model terms and includes baseline characteristics such as SARS-CoV-2 serology and viral load. Covariate adjustment can account statistically for prespecified prognostic factors and can improve precision. It does not change the fact that the underlying comparison originated from randomized treatment assignment.

Why use logistic regression for severe symptoms?

The severe-signs-and-symptoms endpoint is binary: participants either meet the specified endpoint definition or they do not during the stated time frame. Logistic regression is designed for binary outcomes and expresses the treatment comparison as an odds ratio.

Why should the Day 5 versus Day 1 oxygen-saturation odds ratios not be interpreted as treatment effects?

The registry analysis notes explicitly identify these odds ratios as Day 5 versus Day 1 within the PF-07321332 plus ritonavir group or placebo group. They therefore describe a within-group time comparison rather than the randomized between-group treatment contrast.

15. Understanding the Primary Endpoint Statistically

The primary endpoint is especially useful for teaching because it combines a clinically concrete event with a time-to-event framework. A simple Day 28 proportion would ignore when an event occurred and how incomplete follow-up was handled. Kaplan-Meier estimation instead uses event times and risk sets to estimate the event probability over time.

Event occurrence

The event is COVID-19 related hospitalization or death from any cause during the first 28 days.

Time window

The registered time frame is from Day 1 to Day 28.

Estimation

The registry specifies Kaplan-Meier estimation for the percentage experiencing the event.

Comparison

The posted treatment comparison is a percentage difference analyzed using a normal approximation.

The resulting -6.137 percentage-point estimate is therefore best understood as the treatment-group contrast generated by that prespecified analysis framework. It should not be replaced by a separately reconstructed quantity from the available summary information.

16. Hazard Ratios and the Proportional-Hazards Assumption

The secondary symptom endpoints were analyzed using Cox proportional-hazards models. The Cox model estimates a relative hazard associated with treatment while allowing the underlying baseline hazard to vary over time.

Conceptual Cox model
h(t | X) = h0(t) exp(βTX)

The hazard ratio associated with a treatment indicator is exp(β), conditional on the model structure and covariates.

The phrase proportional hazards refers to the model assumption that the treatment hazard ratio is approximately constant over the relevant time scale. If that assumption is seriously violated, a single hazard ratio can become a less complete summary of how the two event processes differ over time.

The ClinicalTrials.gov record reports Cox proportional-hazards models but do not provide a formal proportional-hazards diagnostic. The hazard ratios should therefore be interpreted as the reported model-based summaries rather than as proof that proportional hazards were perfectly satisfied.

17. Covariate Adjustment

Covariate adjustment appears repeatedly in the EPIC-HR analyses. The exact covariates differ by endpoint and analysis population, which is preferable to assuming that every model used the same adjustment set.

AnalysisReported model terms / covariates
Time to sustained alleviation, mITT Treatment; geographic region; baseline SARS-CoV-2 serology status; baseline viral load (<4 log10 copies/mL vs ≥4 log10 copies/mL)
Time to sustained alleviation, mITT1 Treatment; geographic region; symptom onset duration (≤3 vs >3); baseline SARS-CoV-2 serology status; baseline viral load (<4 log10 copies/mL vs ≥4 log10 copies/mL)
Time to sustained alleviation, mITT2 Treatment; geographic region; symptom onset duration (≤3 vs >3); COVID-19 monoclonal-antibody treatment; baseline SARS-CoV-2 serology status; baseline viral load (<4 log10 copies/mL vs ≥4 log10 copies/mL)
Severe signs and symptoms, mITT Treatment; geographic region; baseline SARS-CoV-2 serology status; baseline viral load
Severe signs and symptoms, mITT1 Treatment; geographic region; symptom onset duration; baseline SARS-CoV-2 serology status; baseline viral load
Severe signs and symptoms, mITT2 Treatment; geographic region; symptom onset duration; COVID-19 monoclonal-antibody treatment; baseline SARS-CoV-2 serology status; baseline viral load

Adjustment is not equivalent to randomization. Randomization establishes the primary basis for comparing treatment groups, while regression adjustment specifies how additional measured characteristics enter the statistical model. The resulting estimate remains conditional on the chosen model specification.

18. P-values and Confidence Intervals

The EPIC-HR results provide a useful contrast between the two quantities. For the primary endpoint, the reported p-value is <0.0001, while the estimated risk difference is -6.137 percentage points with a 95% confidence interval of -8.208 to -4.066.

P-value

Addresses evidence against a specified null hypothesis under the stated statistical model and testing framework.

Confidence interval

Shows the statistical uncertainty surrounding the estimated treatment effect and communicates its plausible precision under the model.

Effect estimate

Describes the magnitude and direction of the observed treatment contrast.

Clinical meaning

Requires consideration of the endpoint definition, absolute scale, time frame, population, and uncertainty rather than the p-value alone.

A small p-value does not imply a large treatment effect, and a confidence interval is not simply a range of individual patient outcomes. Reporting both is important because they answer different statistical questions.

19. Multiplicity and Multiple Analyses

The ClinicalTrials.gov record contains 38 posted outcome measures and 20 posted statistical analyses, including one primary-endpoint analysis. The data identify superiority as the hypothesis type for the primary analysis and for most of the reported secondary comparisons, while some analyses are classified as "Other / not stated."

The presence of many secondary analyses means that individual p-values should be interpreted in the context of the overall analysis program. The ClinicalTrials.gov record does not provide a detailed multiplicity-adjustment procedure, endpoint hierarchy, alpha-allocation scheme, or interim-analysis plan sufficient to reconstruct a familywise error-control strategy.

Interpretation caution: the reported p-values should not be retroactively treated as though a specific multiplicity procedure had been used unless that procedure is documented in the ClinicalTrials.gov record. The primary endpoint is explicitly identified as a superiority analysis; the available data do not support adding an unstated multiplicity framework.

20. What the Results Do — and Do Not — Establish

Primary endpoint

The primary analysis reports a risk difference of -6.137 percentage points with a 95% CI of -8.208 to -4.066 and P < 0.0001. This is evidence of a lower estimated percentage for the registered hospitalization-or-death endpoint in the PF-07321332 plus ritonavir group under the stated analysis.

Symptom time-to-event endpoints

The reported Cox hazard ratios for sustained alleviation and sustained resolution are all above 1, with the corresponding confidence intervals above 1. These estimates describe the modeled relative rate of reaching the specified symptom endpoint, not a direct percentage change in symptom duration.

Severe symptoms and worsening

The reported logistic-regression odds ratios for severe signs and symptoms and progression to worsening symptoms have confidence intervals that include 1. These analyses do not establish a statistically clear difference under their respective reported tests.

These conclusions are deliberately endpoint-specific. A positive result on one endpoint should not automatically be generalized to every secondary outcome, and a non-significant secondary analysis should not be interpreted as proof of equivalence or absence of any treatment effect.

21. Limitations and Interpretation Issues

22. Why This Trial Matters Statistically

EPIC-HR is a useful teaching case because it combines several important clinical-trial methods within a relatively compact endpoint framework: randomized treatment allocation, quadruple masking, a time-to-event primary endpoint, Kaplan-Meier estimation, a normal-approximation treatment comparison, Cox proportional-hazards models, logistic regression, covariate adjustment, confidence intervals, odds ratios, hazard ratios, and multiple analysis populations.

ConceptHow it appears in EPIC-HR
Randomization The trial used randomized allocation in a parallel-group design.
Blinding The registry specifies quadruple masking.
Time-to-event endpoint The primary endpoint is defined over Day 1 through Day 28 and estimated using Kaplan-Meier methods.
Risk difference The primary treatment effect is reported as a percentage difference, normalized as a risk difference.
Wald / z-test The primary comparison uses a normal approximation.
Cox model Time to sustained alleviation and resolution of symptoms are analyzed with Cox proportional-hazards models.
Hazard ratio The symptom time-to-event analyses use hazard ratios as their effect measure.
Logistic regression Severe symptoms and progression to worsening are analyzed as binary outcomes.
Odds ratio Logistic-regression analyses report odds ratios with confidence intervals and p-values.
Covariate adjustment Geographic region and prespecified baseline characteristics enter several secondary models.
Analysis populations mITT, mITT1, and mITT2 analyses demonstrate how population definitions affect statistical interpretation.
Safety analysis Serious adverse events are reported as affected participants over participants at risk by arm.

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Calculators

25. Sources

Continue through Clinical Biostats

Use the trial as a starting point for deeper study of survival analysis, regression, confidence intervals, and clinical-trial methodology.

26. Record Summary

EPIC-HR provides a useful example of how a clinical-trial statistical story can extend beyond a single p-value. The primary endpoint combines a clinically important Day 28 event with Kaplan-Meier estimation and a reported percentage difference of -6.137 percentage points, with a 95% confidence interval of -8.208 to -4.066 and P < 0.0001. Secondary analyses demonstrate the use of Cox proportional-hazards models for symptom time-to-event endpoints and logistic regression for binary outcomes, while the different mITT populations show why the analysis population must always accompany an effect estimate.

The reported secondary hazard ratios for sustained symptom alleviation and resolution are above 1, whereas the reported odds ratios for severe COVID-19 signs and symptoms and worsening symptoms have confidence intervals that include 1. The oxygen-saturation analyses provide an additional methodological lesson because their reported odds ratios represent Day 5 versus Day 1 comparisons within treatment groups rather than direct randomized treatment effects. Finally, serious adverse events are reported descriptively by arm, with 19/1038 in the PF-07321332 plus ritonavir group and 72/1053 in the placebo group, without a formal comparison reported in the ClinicalTrials.gov record used for this page.

Clinical Biostats methodology: A trial-results page should distinguish the reported statistical estimate from its interpretation, identify the analysis population and endpoint definition, and avoid converting one effect measure into another. EPIC-HR illustrates why risk differences, hazard ratios, odds ratios, confidence intervals, and p-values must each be interpreted on their own statistical scale.