A Study of Baricitinib (LY3009104) in Participants With COVID-19: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study of Baricitinib (LY3009104) in Participants With COVID-19 (NCT04421027), a randomized phase 3 study. The primary outcome is Percentage of Participants Who Die or Require Non-Invasive Ventilation/High-Flow Oxygen or Invasive Mechanical Ventilation (Including Extracorporeal Membrane Oxygenation [ECMO]).

2. Clinical Question

3. Trial Design

The registry describes NCT04421027 as a randomized, double-blind, placebo-controlled, parallel-group Phase 3 study of baricitinib in patients with COVID-19 infection. Eli Lilly and Company is listed as sponsor. The study status is completed, and the registry reports enrollment of 1525. The recorded study start was 2020-06-12, and the primary completion date was 2021-02-12.

Participants were assigned to baricitinib plus standard of care or placebo plus standard of care. The baricitinib regimen is described as 4 milligrams, administered orally every day as two 2 mg tablets. The placebo group received two placebo tablets orally every day. The registry record identifies double masking but does not report who specifically was masked or the procedures used to maintain masking. Allocation details beyond randomized assignment are not reported in the registry record.

The primary endpoint is a percentage measured over Day 1 to Day 28. Its population is described as all participants randomly assigned to study intervention, except participants with missing baseline ordinal scale values. The arm denominators shown for this endpoint are 756 for placebo plus standard of care and 762 for baricitinib plus standard of care. These outcome denominators are not interchangeable with the registry's enrollment figure, nor with denominators for other outcomes.

The record lists 21 outcome measures and 23 analyses. It also identifies secondary outcomes involving ordinal-scale status or improvement, ventilator-free days, time to recovery, and duration of hospitalization. The original sample-size justification, detailed randomization scheme, planned interim analyses, and full prespecified statistical analysis plan are not reported in the registry record. Accordingly, the reported design features can be described, but they do not establish the rationale behind the chosen enrollment or analysis procedures.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Treatment Period

Reasons for not completing this period:

Follow-Up Period

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: All participants randomly assigned to study intervention. Participants with missing baseline ordinal scale values were excluded.

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

Population: All participants randomly assigned to study intervention. Participants were analyzed according to the intervention to which they were assigned.

6. Primary Endpoint

The registry lists the primary outcome as Percentage of Participants Who Die or Require Non-Invasive Ventilation/High-Flow Oxygen or Invasive Mechanical Ventilation (Including Extracorporeal Membrane Oxygenation [ECMO]), measured over Day 1 to Day 28.

The primary endpoint is the percentage of participants who, between Day 1 and Day 28, died or required non-invasive ventilation or high-flow oxygen, or invasive mechanical ventilation, including extracorporeal membrane oxygenation. It combines death with several forms of respiratory support into one binary outcome: a participant either met at least one listed condition during the timeframe or did not. The record describes this as a percentage and gives a 95% confidence interval for each arm.

As a general statistical rationale, a composite endpoint can capture clinically consequential events across a range of illness progression. It can also make the outcome broader than any one component. A composite percentage alone does not show how much each component contributed, whether the components occurred with similar frequency, or whether the observed difference was driven by one component. Component-specific primary results are not reported in the registry record.

The posted percentages were 30.5 for placebo plus standard of care and 27.8 for baricitinib plus standard of care. Those arm-level percentages describe the observed frequency of the combined endpoint, not the probability that any individual participant will experience an event. Interpretation of the comparison also depends on its uncertainty interval, the analysis population, and the fact that participants with missing baseline ordinal scale values were excluded. The endpoint is therefore informative about this defined composite and timeframe, but does not represent every aspect of recovery or illness severity.

7. Results Posted in the Registry

• Baricitinib + SOC: 27.8% (212/762) versus 30.6% (231/756) on control; risk difference -2.7 percentage points (95% CI -7.3 to 1.8), risk ratio 0.91 (0.78–1.07); unadjusted two-proportion z-test p = 0.241.

8. Secondary Efficacy Endpoints

The registry also posts results for secondary outcomes. For each outcome below, the endpoint type, the test applied to the posted summary statistics, the result and its interpretation are given in turn. Secondary endpoints are usually not adjusted for multiple testing, so they support rather than confirm the primary finding.

Percentage of Participants With at Least 1-Point Improvement on National Institute of Allergy and Infectious Diseases Ordinal Scale (NIAID-OS) or Live Discharge From Hospital

• Endpoint type: Binary: the percentage of participants with the event (counts rebuilt from the posted percentages and group sizes).
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Baricitinib + SOC: 65.0% (495/762) versus 63.5% (480/756) on control; risk difference 1.5 percentage points (95% CI -3.4 to 6.3), risk ratio 1.02 (0.95–1.10); unadjusted two-proportion z-test p = 0.551.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Baricitinib + SOC and the control. The interval gives the range of effects that remain compatible with the data.

Number of Ventilator-Free Days

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard errors.
• Result:
  • Baricitinib + SOC: mean 24.50 versus 23.70 on control; difference 0.80 (95% CI -0.28 to 1.88), standardized effect d = 0.07; z-test p = 0.147.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Baricitinib + SOC and the control. The interval gives the range of effects that remain compatible with the data.

Time to Recovery

• Endpoint type: Time to event: compared as a hazard ratio.
• Test: Hazard ratio as reported by the sponsor (Log Rank); it was not recalculated.
• Result:
  • Placebo + SOC vs Baricitinib + SOC: hazard ratio 1.11 (95% CI 0.99–1.24), reported p = 0.145 (as reported in the registry).
• Interpretation:
  • The 95% confidence interval includes no difference (a ratio of 1), so these data do not show a statistically significant difference between Placebo + SOC vs Baricitinib + SOC and the control. The interval gives the range of effects that remain compatible with the data.

The secondary outcomes address several related dimensions: ordinal-scale improvement or clinical status, ventilator-free days, time to recovery, and hospitalization duration. Taken together, the posted comparisons do not provide a consistently precise pattern of separation between groups. The available estimates generally have uncertainty intervals that include their respective null values or come with p-values that do not establish a clear difference. These outcomes complement the primary composite by describing aspects of recovery and care, but they are not interchangeable with the primary endpoint.

The secondary outcomes also differ in their measurement and analysis. Some are binary percentages at a specified assessment time, some are continuous or count-like summaries, and time to recovery is analyzed as a time-to-event outcome. The meaning of an estimate and its uncertainty therefore depends on the particular outcome and method. Results from one secondary measure should not be treated as confirmation of a different outcome merely because both relate to clinical status.

Unadjusted secondary comparisons carry less weight when many outcomes are examined because each separate test creates an opportunity for a chance finding. A formal multiplicity strategy can address this issue, but one is not reported in the registry record. Without that information, isolated secondary p-values are best treated cautiously and should not be used to override the primary analysis. The registry record does not report results for every listed secondary outcome in the details available for interpretation, so conclusions should be limited to the posted analyses rather than generalized across all measures.

Overall, the secondary evidence offers context about recovery and clinical status but does not change the primary statistical reading: the primary estimate favors a lower observed event percentage in the baricitinib group, while its confidence interval includes the null. The secondary results do not remove that uncertainty or establish a definitive effect on other aspects of illness.

Secondary outcomes registered for this trial, for reference (only those above were recalculated):

• Percentage of Participants With at Least 1-Point Improvement on National Institute of Allergy and Infectious Diseases Ordinal Scale (NIAID-OS) or Live Discharge From Hospital
• Number of Ventilator-Free Days
• Time to Recovery
• Percentage of Participants at Each Clinical Status Using the National Institute of Allergy and Infectious Diseases Ordinal Scale (NIAID-OS) at Day 4
• Percentage of Participants at Each Clinical Status Using the National Institute of Allergy and Infectious Diseases Ordinal Scale (NIAID-OS) at Day 7
• Percentage of Participants at Each Clinical Status Using the National Institute of Allergy and Infectious Diseases Ordinal Scale (NIAID-OS) at Day 10
• Percentage of Participants at Each Clinical Status Using the National Institute of Allergy and Infectious Diseases Ordinal Scale (NIAID-OS) at Day 14
• Duration of Hospitalization

9. Statistical Interpretation

10. Statistical Methods Explained

What does the primary comparison estimate?

The primary outcome is binary at the participant level: the event is the occurrence of any condition listed in the composite during Day 1 to Day 28. The arm percentages estimate the proportion meeting that definition in each group. The recalculated risk difference compares those proportions on an absolute scale, with the baricitinib group subtracted from the placebo group. A value below zero therefore corresponds to a lower observed proportion in the baricitinib group. This is an unadjusted comparison reconstructed from posted summary counts and denominators; it is not the same as the registry's regression analysis.

How should a confidence interval be read?

General statistical explanation: a confidence interval represents uncertainty about an estimated quantity under the assumptions of the method. For a risk difference, zero is the value representing no difference between groups. For an odds ratio, one is the null value. An interval spanning its relevant null value is compatible with no difference as well as with effects in other directions. It does not mean that every value in the interval is equally plausible, and it is not a guarantee that the true effect lies inside the interval.

The primary risk-difference interval is -0.073 to 0.018, and it spans zero. The sponsor's reported odds-ratio interval is 0.67 to 1.08, and it spans one. These are estimates on different scales and come from different analyses, so their values should not be directly compared as if they were the same measure.

What does the p-value contribute?

General statistical explanation: a p-value measures how unusual data at least as incompatible with a specified null model would be, if that model and the analysis assumptions were correct. It is not the probability that the null hypothesis is true, nor a measure of the size or clinical importance of an effect. The recalculated unadjusted p-value for the primary risk difference is 0.241; the reported logistic-regression p-value is 0.180. Their differing values reflect differing analysis approaches and do not change the central point that the reported primary confidence intervals include their null values.

Why do the analyses use different methods?

The registry reports logistic regression for the primary endpoint, with an odds ratio as the parameter. Logistic regression models the odds of a binary outcome. The posted summary counts also allow a simple unadjusted risk-difference recalculation. The registry record does not report the regression covariates or enough model detail to reproduce that analysis fully. A recalculation from aggregate counts cannot substitute for the sponsor's model-based result.

For ventilator-free days, the registry reports an analysis of variance (ANOVA) and a least-squares mean difference. ANOVA compares group means under its model assumptions; the least-squares means and their difference are model-based summaries. The posted means are 23.7 and 24.5, with a reported sponsor difference of 0.75 and p = 0.059. The recalculated unadjusted mean difference is 0.8, with an interval from -0.281 to 1.881. The record does not report all ANOVA model details.

For time to recovery, the registry reports a log-rank analysis and hazard ratio. General statistical explanation: time-to-event methods account for both the time until an event and participants whose event time is not observed during follow-up, subject to the method's assumptions. A hazard ratio compares event rates over time; interpretation depends on which group is in the numerator and how the event is defined. The registry lists the comparison as placebo plus standard of care versus baricitinib plus standard of care and reports a hazard ratio of 1.11, with a 95% confidence interval from 0.99 to 1.24 and p = 0.145. Details about censoring and other time-to-event assumptions are not reported in the registry record.

Why can over-dispersion matter?

General statistical explanation: over-dispersion occurs when observed variability is greater than a model expects. If a model assumes less variability than the data contain, standard errors can be too small and confidence intervals too narrow, making results appear more precise than warranted. This is a general consideration for models of counts or clustered outcomes; the registry record does not report an over-dispersion assessment for this trial. It should not be assumed that over-dispersion occurred or that it affected the posted analyses.

What is not specified?

The registry record does not report the original sample-size justification, full prespecified analysis plan, handling of all missing data beyond the stated exclusion for missing baseline ordinal scale values, or any interim-analysis plan. It also does not report a multiplicity adjustment for the secondary comparisons. These omissions limit independent assessment of whether every analysis followed the original protocol, but do not invalidate the reported estimates. They define what can and cannot be concluded from the registry record.

11. Confidence Intervals and Precision

The primary arm estimates are percentages with 95% confidence intervals: 30.5, with an interval from 27.2 to 33.8, for placebo plus standard of care; and 27.8, with an interval from 24.6 to 31.0, for baricitinib plus standard of care. Each arm interval describes uncertainty around its own percentage. The intervals overlap, but overlap between two separate arm intervals is not itself the formal test of the difference. The direct comparison is more appropriately read from the risk-difference interval and the sponsor's model-based odds-ratio interval.

The recalculated primary risk difference is -0.027, with an interval from -0.073 to 0.018. The interval spans zero, so the posted summary data do not pin down the direction of the difference. It permits a negative difference as well as a positive one. The sponsor's odds ratio of 0.85 has a 95% confidence interval from 0.67 to 1.08, which spans the odds-ratio null value of 1. Neither interval supports a precise claim of a difference in one direction.

General statistical explanation: interval width reflects precision as well as the scale of the outcome. A wider interval allows a broader range of values compatible with the data and method; a narrower interval indicates a more precise estimate under those assumptions. Precision is not the same as validity, and a narrow interval would not remove concerns about outcome definition, missing data, or model assumptions.

The registry reports confidence intervals for the primary endpoint but does not report a clinically important difference threshold. Consequently, the interval cannot be classified from this record alone as ruling out every clinically meaningful effect. It is more accurate to describe the estimate and its uncertainty than to treat a non-significant p-value as evidence of equivalence or no effect.

12. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The planning illustrations translate an assumed target effect into a projected sample requirement; they are not counts of participants observed in this trial. For non-time-to-event outcomes, the figures are participants per arm. For the time-to-recovery planning illustration, the figures are total events, as indicated by the Schoenfeld approach. The original trial's own design assumptions and sample-size justification are not reported in the registry record.

The stated planning assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. The target effect is set as a stated fraction of the observed effect. For time-to-event planning, the calculation is based on total events. For the primary endpoint, planning for 100% of the observed effect gives 4,340 participants per arm for 80% power and 5,809 participants per arm for 90% power. Planning for 75% of the observed effect gives 7,769 participants per arm for 80% power and 10,399 participants per arm for 90% power. Planning for 50% of the observed effect gives 17,597 participants per arm for 80% power and 23,556 participants per arm for 90% power.

In plain language, the 80% power column is the projected per-arm sample size for an 80% chance of detecting the chosen effect under the stated assumptions, if that effect is real and the model is appropriate. The 90% column uses a higher detection probability and therefore generally requires more participants per arm. Smaller target effects require larger samples because they are harder to distinguish from random variation.

These figures are planning illustrations based on the observed effect, not observed power. They do not describe the trial's original planning process, and they should not be interpreted as evidence that the trial was or was not adequately powered.

13. How the Sample Size Is Derived

Power is the probability that a study of a given size rejects the null hypothesis when a specific true effect exists. It depends on four things: the effect to detect, the variability of the outcome, the significance level (here a two-sided α of 0.05, so z_1−α/2 = 1.960) and the desired power (z_1−β = 0.842 for 80% and 1.282 for 90%).

Formula

n is the number of participants in each arm, p₀ and p₁ are the control and treatment proportions, and p̄ is their average.

Worked calculation for this trial

1. Control proportion p₀ = 0.306; treatment proportion p₁ = 0.278; average p̄ = 0.292.
2. First term: z_1−α/2 × √[2p̄(1−p̄)] = 1.96 × 0.6429 = 1.2601.
3. Second term: z_1−β × √[p₀(1−p₀) + p₁(1−p₁)] = 0.842 × 0.6427 = 0.5409.
4. n = (1.2601 + 0.5409)² / (-0.0273)² = 4,339.4, rounded up to 4,340 participants per arm for 80% power.

The smaller-effect rows in the table above repeat the same calculation with the effect multiplied by 0.75 and 0.5 (for ratios, applied to the logarithm of the ratio).

R code

alpha <- 0.05
z_a   <- qnorm(1 - alpha / 2)          # two-sided
p0 <- 0.305555555555556                          # control proportion
p1 <- 0.278215223097113                          # treatment proportion
n_per_arm <- function(power, frac = 1) {
  q1 <- p0 + (p1 - p0) * frac               # smaller effect = fraction of the observed difference
  pbar <- (p0 + q1) / 2
  z_b <- qnorm(power)
  ceiling(round((z_a * sqrt(2 * pbar * (1 - pbar)) + z_b * sqrt(p0 * (1 - p0) + q1 * (1 - q1)))^2 / (q1 - p0)^2, 6))
}
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: n per arm = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(n_per_arm(0.80, f), big.mark = ","), format(n_per_arm(0.90, f), big.mark = ",")))

# Check against R's built-in function (unrounded n per arm)
cat("power.prop.test:", round(power.prop.test(p1 = p0, p2 = p1, power = 0.80)$n, 1), "\n")

# Simulation: how often does the z-test reject at the 80%-power sample size?
set.seed(4421027)
n    <- n_per_arm(0.80)
reps <- 20000
x0 <- rbinom(reps, n, p0); x1 <- rbinom(reps, n, p1)
ph <- (x0 + x1) / (2 * n)
z  <- ((x1 - x0) / n) / sqrt(ph * (1 - ph) * 2 / n)
cat(sprintf("Simulated power at n = %s per arm: %.3f (%d replicates)\n", format(n, big.mark = ","), mean(abs(z) > z_a, na.rm = TRUE), reps))

Output from running the code

100% of observed effect: n per arm = 4,340 (80% power), 5,809 (90% power)
 75% of observed effect: n per arm = 7,769 (80% power), 10,399 (90% power)
 50% of observed effect: n per arm = 17,597 (80% power), 23,556 (90% power)
power.prop.test: 4339.4 
Simulated power at n = 4,340 per arm: 0.801 (20000 replicates)

The simulation draws many pairs of event counts at the planned sample size and counts how often the two-proportion z-test rejects the null. It should land close to the target power; at very small required sample sizes it can fall below it, because the normal approximation is rough there.

14. Analyses Reported by the Sponsor

The registry's own primary analysis is a logistic regression, reported as an odds ratio of 0.85, with a 95% confidence interval from 0.67 to 1.08 and p = 0.180. The simple recalculation from the posted arm summaries gives a risk difference of -0.027, with a 95% confidence interval from -0.073 to 0.018 and an unadjusted p-value of 0.241. The estimates answer related questions on different scales: the logistic regression reports an odds ratio, while the recalculation compares event proportions directly.

The sponsor's model-based analysis should be used as the primary reported analysis when describing the registry result. The recalculation is a transparent, unadjusted check based on the posted summary statistics; it does not reproduce the regression model. The registry record does not report the full model specification, including any covariates, so the reasons for the numerical difference cannot be determined from the record alone.

Both analyses have confidence intervals that include their respective null values. The estimates therefore convey uncertainty rather than a clear separation between groups. The registry record does not identify a discrepancy flag, so there is no stated conflict requiring the recalculation to supersede or invalidate the sponsor's result. The two estimates should remain clearly distinguished by method and effect scale.

15. Multiplicity, Missing Data and Analysis Populations

The registry record lists 21 outcome measures and 23 analyses, including multiple secondary outcomes and assessments at different times. When many hypotheses are tested, the chance of at least one apparently positive result can increase even if no true differences exist. General statistical explanation: multiplicity procedures can control that risk by adjusting significance thresholds or confidence intervals, or by organizing tests in a prespecified hierarchy. The registry record does not report a multiplicity adjustment or a testing hierarchy. The number of analyses alone does not establish that a particular adjustment was required or omitted from the original analysis plan; it means the record does not permit that assessment.

The primary analysis population is described as all participants randomly assigned to study intervention, with participants missing baseline ordinal scale values excluded. This makes the stated population different from a literal count of everyone enrolled or assigned if baseline ordinal data were missing. The record does not give the number excluded for missing baseline values, describe further missing-outcome handling for the primary endpoint, or explain whether any imputation was used. Those details are not reported in the registry record.

The arm denominators vary across outcomes. For example, the primary outcome has denominators of 756 and 762, while other measures have their own listed denominators. Outcome-specific denominators are not evidence by themselves of why data are unavailable. The registry record does not report reasons for these differences or a detailed accounting of missing observations by group and outcome.

Randomized assignment is an important design feature, but interpretation of the posted analysis also depends on how participants were included, how missing observations were handled, and whether the reported methods followed a prespecified plan. The record confirms randomization and the stated exclusion for missing baseline ordinal scale values, but does not report enough information to assess all of those issues. No interim-analysis plan or detailed statistical analysis plan is reported. These gaps warrant caution in interpreting multiple secondary results and in judging the full analysis process; they do not, by themselves, show that the reported results are biased.

16. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 5% reporting threshold.

Most frequent serious adverse event terms

Serious events with the highest arm-level rate: acute respiratory failure (3.9%), septic shock (3.2%), covid-19 pneumonia (2.8%), respiratory failure (2.3%), pulmonary embolism (1.6%).

The safety record lists serious adverse events for the placebo group and the 4 Milligrams (mg) Baricitinib group, as well as separate follow-up groups. In the placebo group, 135 participants were reported with a serious adverse event among 752 at risk; in the 4 Milligrams (mg) Baricitinib group, 110 were reported among 750 at risk. Follow-up entries report 12 among 613 for placebo follow-up and 18 among 645 for 4mg Baricitinib follow-up. Other adverse events are recorded as 0 in each listed group. The record also lists deaths as 104, 65, 16, and 18 across those respective groups.

The registry lists acute respiratory failure, septic shock, Covid-19 pneumonia, respiratory failure, pulmonary embolism, and Covid-19 among the most frequent serious adverse event terms, with percentages of 3.9%, 3.2%, 2.8%, 2.3%, 1.6%, and 1.3%, respectively. These terms and frequencies describe reported events; they do not establish that study treatment caused an event. The registry record does not provide enough detail to determine event relatedness, timing, overlap between categories, or a formal between-group safety comparison. Follow-up groups should be read as separately reported entries, not combined with the main treatment groups.

17. What the Result Does, and Does Not, Mean

The primary result describes the percentage of participants meeting a particular composite definition over Day 1 to Day 28. That definition includes death and specified respiratory-support requirements. The observed percentage was lower in the baricitinib plus standard-of-care group than in the placebo plus standard-of-care group, but the risk-difference interval includes zero and the reported odds-ratio interval includes one. The statistical evidence therefore does not establish a clear difference between the groups on this endpoint.

This interpretation is not equivalent to saying the treatments are identical. A non-significant comparison can reflect uncertainty, and the confidence interval permits a range of effects in either direction. The registry record does not report a threshold for clinical importance, so it cannot establish whether particular values within the interval would matter clinically. Statistical significance, effect magnitude, precision, and clinical relevance are separate considerations.

The endpoint does not capture every possible aspect of COVID-19, recovery, or quality of life. It combines several events, and the registry result does not provide the component-specific primary comparisons needed to determine which events account for the overall pattern. Secondary measures add information about selected aspects of clinical status and recovery, but they are not substitutes for the primary outcome and are subject to uncertainty and multiplicity concerns.

These results should be understood within the enrolled hospitalized population, the recorded treatment regimens, standard-of-care comparator, analysis population, and follow-up timeframe. The registry record does not establish how the findings apply to other populations, treatment regimens, or periods beyond the stated assessment. Group-level percentages are not individual predictions, and event counts do not establish that treatment caused or prevented particular events. The appropriate conclusion is bounded: the posted data show a numerically lower primary-event percentage with baricitinib plus standard of care, but the comparative estimate remains imprecise enough that a clear difference is not established.

18. Why This Trial Matters Statistically

This trial illustrates why an observed difference and a demonstrated difference are not the same thing. The arm percentages point in one direction, but the direct confidence intervals include the null values. Statistical interpretation therefore requires reading the point estimate together with its uncertainty rather than treating the observed ranking of groups as conclusive.

It also illustrates why effect scale matters. A risk difference communicates an absolute comparison between proportions, while an odds ratio communicates a relative comparison of odds. The registry's logistic-regression estimate and the unadjusted summary-data recalculation are related, but they are not interchangeable. Knowing which analysis is sponsor-reported and which is a simple recalculation prevents a model-based result from being mistaken for a direct subtraction of percentages.

The composite endpoint highlights a common design trade-off. Combining death with respiratory-support requirements creates a single outcome spanning several forms of deterioration, but the overall result does not show the contribution of each component. A useful reading therefore respects the stated endpoint while avoiding claims about specific components that are not separately reported.

Finally, the record shows why secondary findings and planning illustrations need context. Multiple outcomes may offer a broader picture but make unadjusted isolated results less persuasive. Sample-size projections based on observed effects are planning tools, not proof of the trial's original design rationale or a measure of observed power. By separating design details, sponsor analyses, recalculations, uncertainty, and unreported information, the results can be interpreted without overstating what they establish.

19. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.
• Participant counts were reconstructed from posted percentages and arm sizes, so small rounding differences are possible.

The registry record leaves several trial-specific uncertainties. It reports randomization and double masking but does not describe the sequence-generation method, allocation concealment, who was masked, or how masking was maintained. The original sample-size justification and full statistical analysis plan are not reported, so the rationale for enrollment and the prespecified analysis choices cannot be evaluated from this record.

The primary analysis excludes participants with missing baseline ordinal scale values, but the number excluded and the handling of other missing outcome data are not reported. The arm denominators differ by outcome, and the record does not explain all denominator differences. This limits assessment of the potential effect of missingness on each comparison.

The primary outcome is a composite, and component-specific primary results are not reported. The posted result therefore cannot show whether death, non-invasive ventilation or high-flow oxygen, or invasive mechanical ventilation including ECMO contributed most to the comparison. The record also does not report a clinically important difference threshold, limiting interpretation of the practical importance of values compatible with the confidence interval.

There are numerous listed outcomes and analyses, but a multiplicity strategy is not reported. Secondary comparisons should consequently be interpreted cautiously. Detailed model specifications are also unavailable for the logistic regression and ANOVA, and the registry record does not provide all assumptions and implementation details for the time-to-recovery analysis. These limitations restrict independent reproduction and evaluation of the analyses; they do not prove that the trial methods were inappropriate or that the findings are biased.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: A Study of Baricitinib (LY3009104) in Participants With COVID-19 — NCT04421027.
• Registry record last updated 2022-07-28; status last verified 2022-07.
• Results first posted 2022-03-24.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

23. Record Summary

NCT04421027 is a completed randomized, double-blind, placebo-controlled, parallel-group Phase 3 study of baricitinib plus standard of care versus placebo plus standard of care in hospitalized participants with COVID-19. The registry reports enrollment of 1525 and a primary outcome covering Day 1 to Day 28: death or specified respiratory-support requirements. The observed primary percentages were 27.8 for baricitinib plus standard of care and 30.5 for placebo plus standard of care.

The unadjusted risk-difference interval includes zero, and the sponsor-reported odds-ratio interval includes one. The posted primary results therefore do not establish a clear group difference. Secondary measures address recovery and clinical status but do not eliminate the primary uncertainty, and multiplicity handling is not reported. The registry record also does not report the original sample-size justification, full statistical analysis plan, or complete details of missing-data handling.