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
BARI-COVID was a phase 2/3, open-label, non-randomized crossover study of baricitinib in symptomatic patients infected by COVID-19. The registered primary endpoint evaluates the percentage of patients requiring transfer to intensive care compared with the rate of transfers observed in controls.
| Feature | BARI-COVID |
|---|---|
| Acronym | BARI-COVID |
| Phase | Phase 2/3 |
| Study title | Baricitinib in Symptomatic Patients Infected by COVID-19: an Open-label, Pilot Study. |
| Allocation | Non-randomized |
| Design model | Crossover |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 200 |
| Number of arms | 2 |
| Intervention | Baricitinib |
| Lead sponsor | Hospital of Prato |
| Status | Unknown |
| Start date | 2020-05-16 |
| Primary completion date | 2020-06-30 |
| ClinicalTrials.gov | NCT04320277 |
2. Clinical Question
The registered clinical question concerns whether patients receiving baricitinib have a different percentage requiring transfer to the intensive care unit compared with controls during the specified 2-week assessment period.
Population
Symptomatic patients infected by COVID-19, as described in the study title.
Intervention
Baricitinib, recorded as the study drug intervention.
Comparator
Controls, against whom the rate of ICU transfers was to be compared.
Primary question
What percentage of patients requires transfer to ICU, and how does that percentage compare with the rate observed in controls over 2 weeks?
3. Trial Design
Baricitinib
- Baricitinib is the registered drug intervention.
- The study is described as open-label.
- The primary endpoint evaluates ICU transfer during 2 weeks.
Controls
- The registry describes a control group for the primary comparison.
- The primary endpoint compares ICU-transfer percentage with the rate observed in controls.
- The registry does not provide additional control-treatment details in the information summarized here.
4. Crossover Design and Its Statistical Consequences
The registry identifies BARI-COVID as a crossover study model. In a crossover design, individuals may contribute observations under more than one treatment condition or sequence. The defining statistical feature is therefore that observations can be linked within individuals rather than treating every observation as independent.
For a conventional crossover trial, an analysis may account for treatment, period, sequence, and within-person correlation. The exact model depends on how the crossover was implemented, the endpoint structure, and the timing of treatment changes. Those details are not specified in the registry information summarized here, so the appropriate analysis cannot be reduced to a single universally correct model.
When two observations come from the same person, their covariance can be important. Ignoring that dependence can produce standard errors that do not correctly represent the information in the data.
There is an additional issue here because the primary endpoint is an ICU-transfer percentage measured over 2 weeks. A crossover analysis is most naturally useful when outcomes can meaningfully be compared across treatment periods. An ICU transfer is also a clinical event that can change what happens later in a patient's course. For such an endpoint, period effects, carryover effects, and the timing of crossover would need to be considered explicitly in a full statistical analysis.
5. Primary Endpoint
| Endpoint | Definition | Time frame | Registry description |
|---|---|---|---|
| Percentage of patients requiring transfer to ICU | The percentage of patients requiring transfer to ICU as compared with the rate of transfers observed in controls. | 2 weeks | The percentage of ICU admission in patients and controls will be compared for statistical difference. |
This endpoint is a binary outcome summarized as a percentage: whether a patient requires ICU transfer during the specified 2-week period. The comparison is therefore fundamentally different from a continuous endpoint such as a laboratory measurement or a time-to-event endpoint such as overall survival.
Primary statistical contrast
Baricitinib group versus controls over 2 weeks
The registry specifies a comparison for statistical difference but does not post a numerical statistical analysis.
6. Planned Analysis
The registry specifies that the percentage of ICU admission in patients and controls will be compared for statistical difference over 2 weeks. Results have not been posted on ClinicalTrials.gov.
How this endpoint is typically analysed
For a binary outcome, the most direct analysis begins with the number and percentage of patients experiencing the event in each comparison group. A two-group comparison can then use a chi-square test when its large-sample assumptions are appropriate, or a Fisher exact test when expected cell counts are small.
An effect estimate is usually more informative than a p-value alone. Depending on the design and prespecified estimand, useful measures can include an absolute risk difference, risk ratio, or odds ratio, each accompanied by a confidence interval.
| Measure | What it answers |
|---|---|
| Risk difference | How many percentage points the ICU-transfer probability differs between groups. |
| Risk ratio | How the probability of ICU transfer compares proportionally between groups. |
| Odds ratio | How the odds of ICU transfer compare between groups. |
| P-value | How compatible the observed data are with the null hypothesis specified by the statistical test. |
| 95% confidence interval | The statistical uncertainty around an estimated treatment-group contrast under the chosen analysis framework. |
7. Statistical Methodology
Binary endpoint analysis
The primary endpoint can be represented for each patient as a binary indicator:
The group-level percentage is the proportion of patients with Y = 1. The registry specifically defines the primary endpoint as the percentage requiring transfer to ICU.
Risk difference
A risk difference compares the event probabilities directly:
A risk difference is expressed in probability units, or percentage points when the probabilities are expressed as percentages.
This measure is especially intuitive for a clinical event such as ICU transfer because it directly describes the difference in event probability between the groups.
Risk ratio
A risk ratio instead compares the two probabilities multiplicatively:
A risk ratio of 1 represents equal event probabilities. Values below 1 indicate a lower observed event probability in the numerator group, while values above 1 indicate a higher observed event probability.
Odds ratio
An odds ratio compares odds rather than probabilities:
Odds ratios can differ materially from risk ratios when the event is not rare, so the measure should be identified explicitly rather than described simply as a relative risk.
Confidence intervals
A confidence interval places an uncertainty range around an estimated effect. For example, a confidence interval for a risk difference describes the precision of the estimated difference under the statistical model and sampling framework used to construct it.
Adjustment for a non-randomized comparison
Because allocation is recorded as non-randomized, an adjusted analysis could be considered if clinically important baseline variables were available. Depending on the available data and prespecified estimand, possible approaches include logistic regression or other regression methods for binary outcomes.
Adjustment, however, does not automatically eliminate confounding. It can address measured covariates included in an appropriate model, while unmeasured or inadequately measured differences between treatment groups can remain.
8. Statistical Methods Explained
Why is ICU transfer treated as a binary endpoint?
The registered endpoint asks whether patients require transfer to ICU during a specified 2-week period. That naturally creates an event indicator for each patient: ICU transfer occurred or it did not occur during the defined time frame. The corresponding group-level statistic is therefore a percentage.
Why might a risk difference be useful?
The risk difference communicates an absolute change in probability. If two groups differed by a specified number of percentage points, that difference would directly describe how much more or less common ICU transfer was in one group than the other. It does not depend on translating a relative effect into an absolute effect without knowing the underlying control risk.
Why might a risk ratio also be reported?
A risk ratio provides a proportional comparison. It answers a different question from the risk difference: rather than asking how many percentage points separate the groups, it asks how many times as large one event probability is relative to the other.
Why does a p-value not measure the size of an effect?
A p-value addresses compatibility with a specified null hypothesis under the chosen statistical model. It is influenced by sample size, variability, and the observed effect. A small p-value does not by itself indicate a large clinical effect, while a larger p-value does not establish that the groups are clinically identical.
Why is non-randomization important?
Randomization is a mechanism for balancing both measured and unmeasured characteristics probabilistically between treatment groups. With non-randomized allocation, group differences can arise before treatment. A statistical association between baricitinib exposure and ICU transfer therefore has a different causal interpretation from the corresponding association in a randomized trial.
Why does the crossover label matter?
Crossover designs create within-person dependence and can introduce period and carryover considerations. If the outcome is affected by what happened during an earlier treatment period, simply comparing observations as though they came from independent groups can give an inappropriate analysis.
9. Interpreting a Statistical Difference
The registry states that the percentage of ICU admission in patients and controls will be compared for statistical difference. A statistical test would evaluate a prespecified null hypothesis about the comparison, while an effect estimate would quantify the magnitude and direction of the observed difference.
The primary endpoint concerns a clinically consequential event: transfer to an intensive care unit during 2 weeks. The most useful interpretation would therefore place any relative measure alongside the underlying event percentages and an absolute difference.
Because the registry identifies allocation as non-randomized, an observed treatment-group difference should be distinguished from a causal treatment effect. Differences in patient characteristics, disease severity, timing, or other factors could contribute to the comparison.
10. Missing Data and Endpoint Definition
The registry defines the primary time frame as 2 weeks. For a binary endpoint with a fixed assessment window, a complete analysis needs a clear rule for patients whose ICU-transfer status at the end of that window is unavailable.
Common approaches include defining an explicit estimand and handling missing outcome information through prespecified assumptions or sensitivity analyses. The appropriate method depends on why outcomes are missing and whether missingness is related to treatment, disease severity, or other factors.
Complete outcome
A patient with a known ICU-transfer status during the 2-week window can contribute directly to the binary endpoint analysis.
Missing outcome
An unavailable ICU-transfer status requires an explicit analysis rule rather than silently treating the missing value as either an ICU transfer or a non-event.
The registry information summarized here does not specify a missing-data or imputation method. The distinction is important because the handling of missing endpoint information can affect both the estimated event percentages and the uncertainty around them.
11. Analysis Population
The registry reports an enrollment of 200 and identifies two arms, but the information summarized here does not provide separate arm-level enrollment counts or define an analysis population for the primary endpoint.
| Population concept | Statistical role |
|---|---|
| Enrolled population | The registry records 200 participants as the study enrollment. |
| Primary endpoint population | A complete analysis would need a prespecified definition of which enrolled patients contribute to the 2-week ICU-transfer comparison. |
| Crossover population | Because the design is identified as crossover, the analysis would need to specify how patients changing treatment condition contribute to the endpoint. |
For a non-randomized crossover study, the definition of the analysis population is particularly important. Excluding patients after treatment begins can introduce selection effects if the reasons for exclusion are related to prognosis or treatment response.
12. Multiplicity and Multiple Comparisons
The registry information summarized here identifies one registered primary endpoint: the percentage of patients requiring transfer to ICU compared with controls over 2 weeks.
Multiplicity becomes important if additional endpoints, repeated analyses, subgroup comparisons, or multiple time points are formally tested. Testing several hypotheses without an appropriate error-control strategy can increase the probability of obtaining a statistically significant result by chance alone.
| Issue | Statistical implication |
|---|---|
| Single primary endpoint | The principal hypothesis concerns ICU transfer during 2 weeks. |
| Additional endpoints | Formal testing of multiple outcomes may require an explicit multiplicity strategy. |
| Subgroups | Separate subgroup tests can produce unstable estimates and additional false-positive opportunities. |
| Repeated looks at the data | Repeated significance testing can alter type I error unless the monitoring plan accounts for it. |
13. Interim Analysis
The registry information summarized here does not identify an interim statistical analysis plan, alpha-spending method, efficacy boundary, or formal stopping rule.
That distinction matters because a study can be monitored while data accumulate, but repeated formal hypothesis testing requires an appropriate statistical framework. Without such a framework, a nominal p-value from an unplanned repeated look may not have the same interpretation as a p-value from a single prespecified analysis.
14. Non-Inferiority, Equivalence, and Superiority
The registered primary endpoint is described as a comparison for statistical difference in ICU admission between patients and controls. The registry information summarized here does not specify a non-inferiority margin or an equivalence margin.
That means the statistical interpretation should not substitute non-inferiority logic for the registered comparison. In a non-inferiority trial, the central question is whether an effect remains within a prespecified clinically acceptable margin. A conventional difference test asks a different question about evidence against equality.
The choice of hypothesis depends on the clinical objective and prespecified design, not simply on which statistical test is easiest to calculate.
15. Bayesian Methods
The registry information summarized here does not identify a Bayesian analysis, prior distribution, posterior probability, Bayesian decision rule, or credible interval.
A Bayesian analysis could theoretically be constructed for a binary ICU-transfer endpoint, but that would require an explicitly specified prior and estimand. A posterior probability is not interchangeable with a conventional frequentist p-value, and the two frameworks should not be mixed without stating the underlying assumptions.
16. Safety and Other Outcomes
The registry information summarized here identifies the primary ICU-transfer endpoint but does not provide numerical serious-adverse-event results or other safety results by arm.
For a clinical trial evaluating an intervention in symptomatic patients, safety is an important separate dimension of interpretation. A complete safety analysis would ordinarily describe the number of patients exposed, adverse-event definitions, serious adverse events, treatment discontinuations, and the time period used for each safety measure.
17. Study Timing and Registry Status
Study start
The registry records 2020-05-16 as the study start date.
Primary completion date
The registry records 2020-06-30 as the primary completion date.
Unknown
The registry status is recorded as unknown.
The short interval between the recorded start date and primary completion date places the registered primary endpoint in a clearly defined early study period. The endpoint's 2-week time frame also provides a fixed window for the intended ICU-transfer comparison.
18. What a Complete Statistical Report Would Show
For this particular endpoint, a statistically informative report would begin with the number of patients in each analysis group, the number requiring ICU transfer, and the corresponding percentages. It would then provide an absolute and relative effect estimate with confidence intervals, together with the prespecified hypothesis test.
| Result component | Why it matters |
|---|---|
| Patients analysed per group | Defines the denominator behind each percentage. |
| ICU transfers per group | Shows the underlying event counts. |
| ICU-transfer percentage | Directly reports the registered primary endpoint. |
| Risk difference | Provides an absolute treatment-group contrast. |
| Risk ratio or odds ratio | Provides a relative treatment-group contrast. |
| 95% confidence interval | Shows statistical precision around the effect estimate. |
| Hypothesis-test result | Addresses the prespecified question of statistical difference. |
| Analysis population | Clarifies which patients contribute to the comparison. |
For BARI-COVID, the crossover and non-randomized features would add another layer: the report would need to make clear whether the comparison is between treatment groups, treatment periods, or another prespecified unit of analysis.
19. Limitations
- Non-randomized allocation: treatment assignment was not randomized, so baseline differences between groups can confound the association between baricitinib and ICU transfer.
- Crossover design: crossover creates within-person dependence and makes period, sequence, and carryover effects potentially important.
- Endpoint timing: the primary endpoint is restricted to a 2-week window, so it addresses ICU transfer during that interval rather than longer-term clinical outcomes.
- Analysis population: the registry information summarized here does not define the final analysis population for the primary endpoint.
- Arm-level enrollment: total enrollment is reported as 200, but separate arm-level enrollment counts are not included in the available registry information.
- Missing-data strategy: the registry information summarized here does not specify how missing ICU-transfer status would be handled.
- Unreported statistical results: no formal statistical analyses are posted to ClinicalTrials.gov for the primary endpoint.
- Limited registry detail: the information summarized here does not specify the crossover sequence, period structure, treatment timing, or statistical model used for the crossover design.
- Status uncertainty: the registry currently records the study status as unknown.
20. Why This Trial Matters Statistically
BARI-COVID is a useful teaching case because several fundamental clinical-trial concepts appear simultaneously in a relatively compact design. The primary outcome is straightforward to describe as a binary event, but the study's non-randomized allocation and crossover model make interpretation more complex than a simple two-group percentage comparison.
| Concept | How it appears in BARI-COVID |
|---|---|
| Binary endpoint | ICU transfer during a 2-week period can be represented as an event/non-event outcome. |
| Absolute effect | A risk difference can describe the difference in ICU-transfer probability between groups. |
| Relative effect | A risk ratio or odds ratio can express the treatment-group contrast proportionally. |
| Confidence interval | Quantifies uncertainty around an estimated effect. |
| Non-randomized allocation | Requires caution when interpreting an observed group difference causally. |
| Crossover design | Requires attention to within-person dependence, period effects, sequence, and potential carryover. |
| Fixed time frame | The primary endpoint is defined over 2 weeks. |
| Multiplicity | Additional formal comparisons would require attention to type I error. |
| Missing data | The analysis needs an explicit rule for unavailable 2-week ICU-transfer status. |
21. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
22. Related Statistical Calculators
23. Sources
- ClinicalTrials.gov: BARI-COVID, NCT04320277.
- PubMed: PubMed record.
Continue through the Clinical Biostats knowledge graph
Explore clinical-trial design and statistical methods through related tutorials and calculators.
24. Record Summary
BARI-COVID is registered as a phase 2/3, open-label, non-randomized crossover study with 200 participants and two arms. Its registered primary endpoint is the percentage of patients requiring transfer to ICU compared with the rate of transfers observed in controls over 2 weeks. Statistically, the endpoint is naturally expressed as a binary event rate, with risk difference, risk ratio, or odds ratio providing complementary measures of effect and confidence intervals providing information about precision.
The study design is central to interpretation. Non-randomized allocation means that an observed difference between treatment groups cannot automatically be interpreted as a causal treatment effect, while the crossover designation means that within-person dependence and period-related effects may need to be incorporated into the analysis. The registry identifies the primary comparison but does not provide a posted formal statistical analysis or numerical primary endpoint results.