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
ATTACC was a randomized, parallel, unmasked phase 2/3 treatment trial conducted in patients with COVID-19 and pneumonia. The registry identifies heparin as the intervention and defines the primary endpoint as days alive and free of organ support through day 21.
| Feature | ATTACC |
|---|---|
| Trial acronym | ATTACC |
| Full title | Antithrombotic Therapy to Ameliorate Complications of COVID-19 |
| NCT ID | NCT04372589 |
| Phase | Phase 2/3 |
| Status | Completed |
| Condition | COVID-19; Pneumonia |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1200 |
| Number of arms | 2 |
| Intervention | Heparin |
| Lead sponsor | University of Manitoba |
| Sponsor type | OTHER |
2. Clinical Question
The central statistical question is whether randomized assignment to the trial's heparin intervention affects the amount of time that participants remain alive and free of organ support during the first 21 days after trial entry.
Population
Participants enrolled with COVID-19 and pneumonia.
Intervention
The registry identifies heparin as the drug intervention.
Comparator
The trial has 2 randomized arms. The registry fields summarized here do not specify the comparator regimen.
Primary question
Does randomized treatment assignment affect days alive and free of organ support through day 21?
3. Trial Design
Trial chronology
Study start
The registry lists May 20, 2020 as the trial start date.
Primary completion
The registry lists May 17, 2021 as the primary completion date.
Completed
The ClinicalTrials.gov record lists the study status as completed.
4. Trial Arms and Treatment Structure
Heparin
- Drug intervention identified in the registry.
- Part of a 2-arm randomized parallel design.
- Primary outcome assessed through day 21.
Comparator arm
- The registry identifies a second randomized arm.
- The comparator treatment is not specified in the registry fields summarized here.
- The primary endpoint compares the randomized treatment groups.
5. Primary Endpoint
| Endpoint | Time frame | Registry definition |
|---|---|---|
| Mortality and days free of organ support | 21 days | The primary endpoint in the trial is days alive and free of organ support at day 21. This endpoint is defined as the number of days that a patient is alive and free of organ support through the first 21 days after trial entry. Organ support is defined as receipt of invasive or non-invasive mechanical ventilation, high flow nasal oxygen (>30 L/min), vasopressor therapy, or ECMO support. Death at any time (including beyond 21 days) during the index hospital stay is assigned the worst possible score of -1. |
What the endpoint combines
The endpoint is not simply a mortality proportion and it is not simply a binary indicator of organ support. It combines survival with the amount of time spent alive without qualifying organ support during the first 21 days after trial entry.
Alive without support
A participant contributes days during which the participant is alive and does not receive one of the specified forms of organ support.
Organ support
Organ support includes invasive or non-invasive mechanical ventilation, high flow nasal oxygen (>30 L/min), vasopressor therapy, or ECMO support.
Death
Death at any time during the index hospital stay, including death beyond day 21, is assigned the worst possible score of -1.
Time horizon
The primary endpoint is evaluated through the first 21 days after trial entry.
6. Statistical Methodology
Composite and ordinal-style outcome structure
The ATTACC primary endpoint is a particularly important example of a clinical outcome that incorporates both survival and organ-support status. Rather than reducing every participant to a simple alive/dead indicator at one fixed time, the endpoint records the number of days a participant is alive and free of organ support through day 21, while assigning death the worst possible score of -1.
This structure preserves more information about the clinical course than a binary endpoint can provide. Two participants can both be alive at day 21 but have different endpoint values if one spent substantially more time receiving organ support.
Bounded follow-up
The endpoint has a fixed 21-day observation window. This makes the estimand explicitly tied to the first 21 days after trial entry rather than to unrestricted long-term follow-up.
The registry additionally specifies that death at any time during the index hospital stay, including death beyond day 21, receives the worst possible score of -1.
Why the endpoint is statistically different from survival analysis
A conventional survival endpoint records the time until an event such as death. The ATTACC primary endpoint instead assigns a score representing time alive without organ support and gives death a separate worst possible value. The statistical analysis therefore needs to respect the endpoint's ordered and bounded structure rather than automatically treating it as an ordinary time-to-death variable.
Randomized treatment comparison
Because allocation is randomized, the primary comparison can be framed as a comparison of the endpoint distribution between the two randomized arms. The key statistical principle is to preserve the randomized treatment assignment when estimating the treatment effect.
7. Statistical Methods Explained
Why use days alive and free of organ support?
A binary outcome such as survival at a particular day would distinguish participants who died from those who survived, but it would not fully describe the burden of organ support among survivors. A days-free-of-support endpoint incorporates both dimensions into a single outcome.
Why is death assigned a score of -1?
The registry explicitly assigns death the worst possible score of -1, including death beyond day 21 during the index hospital stay. This creates a clear ordering in which death is worse than remaining alive but requiring organ support.
Why is the endpoint bounded at 21 days?
A fixed 21-day window gives the endpoint a common observation horizon. Every randomized participant is evaluated against the same prespecified time frame rather than allowing longer follow-up to contribute additional endpoint days for some participants.
Why does organ support need a precise definition?
Composite outcomes become difficult to interpret if their components are ambiguous. The registry specifies four categories of organ support: invasive or non-invasive mechanical ventilation, high flow nasal oxygen above 30 L/min, vasopressor therapy, and ECMO support. These definitions establish which clinical states count as requiring organ support.
Why should this endpoint not automatically be analyzed as a hazard ratio?
A hazard ratio is naturally suited to a time-to-event endpoint such as time to death. ATTACC's primary endpoint is instead a numerical score representing days alive and free of organ support, with death assigned a worst score. Treating it automatically as a conventional survival endpoint could change the statistical question being answered.
What does randomization contribute statistically?
Randomization provides the basis for comparing outcomes between treatment assignments while reducing systematic differences in measured and unmeasured baseline characteristics in expectation. The validity of the causal comparison then depends on preserving the randomized groups in the primary analysis and on appropriate handling of subsequent observations.
8. Planned Analysis
The primary statistical analysis would ordinarily compare the distribution of the prespecified days-alive-and-free-of-organ-support outcome between the randomized treatment groups. Because the outcome is an ordered numerical measure with a specifically defined worst score for death, an analysis based on the endpoint's distribution is generally more appropriate than automatically applying a standard binary test or an ordinary time-to-event model.
For this type of endpoint, a rank-based comparison can be considered when the primary scientific question concerns whether one randomized group tends to have better endpoint values than the other. Another possible framework is a regression model designed for bounded or ordinal outcomes, provided its assumptions match the prespecified analysis plan. The exact confirmatory method should be taken from the trial's statistical analysis plan rather than inferred solely from the endpoint definition.
Clinical interpretation should also recognize the endpoint's special death rule. A participant who dies receives -1 regardless of whether the death occurs within or beyond day 21 during the index hospital stay. Consequently, the endpoint deliberately places mortality at the bottom of the outcome scale rather than treating death simply as a missing observation after day 21.
9. Understanding the Primary Endpoint Statistically
| Clinical state | Contribution to the endpoint | Statistical meaning |
|---|---|---|
| Alive and free of organ support | Counts toward days alive and free of organ support through day 21 | Higher values indicate more time alive without qualifying organ support |
| Receiving specified organ support | Does not count as a day free of organ support | Reduces the number of days accumulated as support-free |
| Death during the index hospital stay | Assigned -1 | Worst possible endpoint value under the registry definition |
This structure means that the primary endpoint contains more information than a simple day-21 survival indicator. It can distinguish different clinical trajectories among participants who remain alive at day 21, while also placing death at the extreme unfavorable end of the scale.
The treatment effect for this endpoint should be interpreted as a difference in the distribution of a prespecified clinical score, not automatically as a reduction in the hazard of death. A favorable shift in the endpoint would represent more days alive without qualifying organ support, but the exact magnitude and uncertainty of such a shift require the formal statistical result.
10. What a Formal Analysis Would Need to Establish
A rigorous analysis of the primary endpoint would address several related questions.
- Location of the distributions: Do randomized participants assigned to the intervention tend to have higher endpoint values?
- Magnitude: How large is the between-group difference on the prespecified endpoint scale?
- Uncertainty: What confidence interval or other uncertainty measure surrounds the treatment-effect estimate?
- Statistical evidence: Does the prespecified hypothesis test provide evidence against the null hypothesis under the trial's error-control framework?
- Clinical meaning: Does a difference in days alive and free of organ support represent a clinically important improvement, considering both mortality and organ-support burden?
These questions should be kept distinct. Statistical significance, effect magnitude, precision, and clinical importance are related but are not interchangeable concepts.
11. Why the Death Rule Matters
The unusual feature of the ATTACC primary endpoint is the explicit assignment of a score of -1 to death at any time during the index hospital stay, including death beyond day 21.
Worst endpoint value
Death during the index hospital stay receives the worst possible score under the registry definition.
12. Organ Support as a Statistical Component
The registry defines organ support using four categories:
| Organ-support category | Included in the primary endpoint definition |
|---|---|
| Invasive mechanical ventilation | Yes |
| Non-invasive mechanical ventilation | Yes |
| High flow nasal oxygen | Yes, when >30 L/min |
| Vasopressor therapy | Yes |
| ECMO support | Yes |
The endpoint therefore captures several clinically different forms of advanced support within one outcome definition. This is statistically efficient when the scientific objective is to summarize freedom from organ support, but it also means that the final treatment effect can reflect differences across multiple component pathways.
13. Composite Endpoint Interpretation
Composite or multi-component endpoints require careful interpretation because a treatment effect can arise through different components. In ATTACC, the endpoint integrates mortality and days without organ support. A difference between treatment groups could therefore reflect differences in survival, differences in the amount of time requiring organ support among survivors, or both.
Mortality component
Death receives the worst possible endpoint value of -1.
Support component
Time receiving specified organ support reduces the number of days counted as alive and free of support.
Survival without support
Days alive without qualifying organ support contribute positively to the endpoint.
Overall distribution
The primary comparison considers the resulting endpoint across the randomized groups.
This is one reason that a single summary statistic should be accompanied by an explanation of what the endpoint actually measures. A numerical treatment effect cannot be interpreted properly without understanding the clinical states represented by the outcome scale.
14. Randomization and Causal Interpretation
The randomized allocation is central to the statistical interpretation of ATTACC. Before treatment is administered, randomization is intended to make the treatment groups comparable in expectation with respect to characteristics that could affect the endpoint.
The individual participant has only one observed treatment trajectory, so the causal treatment effect is estimated through the contrast between randomized groups rather than by observing both treatments in the same participant.
This framework is particularly important when the endpoint incorporates rapidly changing clinical status. Because treatment assignment occurs before the subsequent clinical course unfolds, the randomized comparison can evaluate the effect of assignment without requiring baseline prognostic factors to be perfectly balanced numerically.
15. Missing Data and Follow-Up Considerations
The primary endpoint is defined over a fixed 21-day period, which creates a specific missing-data question: how should the endpoint be handled when complete information about survival or organ-support status is unavailable during part of the observation window?
The registry definition itself specifies the endpoint but does not provide a missing-data or imputation strategy in the record summarized here. A formal analysis would therefore need to follow the prespecified statistical analysis plan for incomplete observations rather than selecting an imputation rule after examining the results.
16. Multiplicity and Multiple Endpoints
The registry identifies one registered primary endpoint: mortality and days free of organ support at 21 days. The endpoint itself combines survival and organ-support information into a single primary outcome.
When a trial has multiple confirmatory endpoints or multiple primary hypotheses, statistical multiplicity can increase the probability of a false-positive finding unless the testing strategy accounts for it. In the registry information summarized here, no additional primary endpoint or multiplicity-adjustment procedure is specified.
| Statistical feature | ATTACC registry information |
|---|---|
| Registered primary endpoint | Mortality and days free of organ support |
| Primary time frame | 21 days |
| Additional registered primary endpoints | None identified in the endpoint information summarized here |
| Posted statistical analysis | None |
17. Interim Analysis and Adaptive Features
The registry information summarized here identifies ATTACC as a phase 2/3 randomized trial but does not report an interim-analysis procedure, alpha-spending strategy, stopping boundary, or adaptive decision rule.
These design features can materially affect the interpretation of a confirmatory clinical trial. If interim efficacy or futility analyses are performed, the nominal significance level of an individual analysis cannot necessarily be interpreted as though it came from a single fixed final analysis.
For ATTACC, the appropriate statistical interpretation of any interim or adaptive procedure would require the prespecified statistical analysis documentation describing information times, stopping rules, and error control.
18. Stratification and Covariate Adjustment
The registry information summarized here identifies randomized allocation but does not specify randomization strata or a prespecified covariate-adjustment model for the primary endpoint.
That distinction matters because stratified analyses and covariate-adjusted analyses can improve statistical efficiency when the adjustment variables are prespecified and appropriately modeled. They should not, however, be confused with post hoc adjustment selected after observing treatment-group differences.
19. Bayesian Methods
No Bayesian statistical method is identified in the registry information summarized here. A Bayesian analysis would require specification of a prior distribution, likelihood, posterior estimand, and decision rule or interval interpretation.
Those elements are distinct from a conventional frequentist confidence interval and p-value. Without a specified Bayesian analysis plan, the primary endpoint should not be described using posterior probabilities or Bayesian decision thresholds.
20. Non-Inferiority and Superiority Logic
The registry information summarized here does not specify a non-inferiority margin. The primary endpoint is registered as mortality and days free of organ support rather than as a non-inferiority endpoint with a prespecified margin.
In a non-inferiority trial, the interpretation depends critically on a clinically justified margin established before the results are examined. Without such a margin, a conventional comparison cannot be relabeled as a non-inferiority analysis simply because the observed difference is small.
21. Crossover and Treatment Switching
The registry information summarized here does not report a crossover or treatment-switching procedure.
If treatment switching occurs after randomization, an intention-to-treat comparison generally remains focused on the effect of assignment to the randomized strategy. Additional analyses can sometimes estimate the effect of treatment actually received, but those analyses require different assumptions and may be more vulnerable to post-randomization confounding.
22. Results Status
Accordingly, the statistical interpretation available from the registry centers on the design and endpoint rather than an observed treatment-effect estimate. A completed trial and a statistically analyzed trial record are not synonymous: completion describes the study's status, while a posted analysis provides the numerical evidence needed to quantify the treatment comparison.
23. Limitations
- No posted primary result: the ClinicalTrials.gov record does not provide a numerical treatment-effect estimate, confidence interval, or p-value for the registered primary endpoint.
- Comparator specification: the registry information summarized here identifies heparin as the intervention and confirms 2 randomized arms, but does not specify the comparator regimen.
- Endpoint complexity: the primary endpoint combines mortality with days alive and free of several forms of organ support, so its interpretation requires attention to the complete outcome definition.
- Fixed time horizon: the primary endpoint is explicitly defined through day 21, so it should not be interpreted as a measure of long-term clinical outcome.
- Death coding: death during the index hospital stay receives a score of -1, including death beyond day 21. This special rule is essential to interpreting the endpoint.
- Missing-data strategy: the registry information summarized here does not specify an imputation or missing-data procedure.
- Analysis specification: the registry information summarized here does not identify the exact confirmatory statistical test or model for the primary endpoint.
- Interim monitoring: no interim-analysis or alpha-spending strategy is identified in the registry information summarized here.
- Stratification: no randomization strata or prespecified covariate-adjustment variables are identified in the registry information summarized here.
24. Why This Trial Matters Statistically
ATTACC provides a useful statistical teaching example because its primary endpoint is designed to capture more of the acute clinical course than a simple mortality endpoint. It combines survival and organ-support burden within a fixed 21-day window and gives death the worst possible score.
| Concept | How it appears in ATTACC |
|---|---|
| Randomization | Participants are assigned using a randomized allocation. |
| Parallel-group design | The registry identifies a parallel design with 2 arms. |
| Fixed follow-up window | The primary endpoint is assessed through 21 days after trial entry. |
| Composite clinical outcome | Mortality and days free of organ support are incorporated into one endpoint. |
| Ordered clinical outcome | More days alive and free of organ support correspond to better endpoint values, with death assigned the worst score. |
| Endpoint definition | Organ support has four explicitly defined clinical categories. |
| Missing-data considerations | Incomplete observation of the 21-day clinical course requires a prespecified analytical strategy. |
| Phase 2/3 design | The registry identifies the study as phase 2/3. |
| Unmasked treatment | The registry identifies masking as none. |
25. A Deeper Statistical Reading of the Endpoint
The endpoint can be viewed as a clinical utility-like scale in which the investigator has specified an ordering of outcomes. Death is placed at the bottom with a value of -1. Survival with organ support occupies an intermediate clinical state because those days do not accumulate as days free of support. Survival without qualifying organ support contributes positively through day 21.
This design has an important advantage: it avoids forcing every participant into a single binary outcome. A participant who survives but requires prolonged organ support is clinically distinguishable from a participant who survives with little or no organ-support requirement.
The tradeoff is that the numerical scale itself embodies a clinical judgment about how these states should be ordered. Assigning death the value -1 is an explicit design decision, while the 21-day limit defines the period over which support-free survival is counted. The statistical result therefore has to be interpreted as a treatment effect on this particular prespecified outcome, not as a generic measure of "overall clinical benefit."
When an endpoint is constructed from multiple clinical states, the endpoint definition is part of the statistical model of the clinical question. Understanding exactly how death, organ support, and survival contribute to the score is necessary before interpreting any treatment-effect estimate.
26. Sources
- ClinicalTrials.gov: NCT04372589 — Antithrombotic Therapy to Ameliorate Complications of COVID-19 (ATTACC).
Continue through the Clinical Biostats trial-analysis library
Explore additional clinical trial analyses and statistical methods across the Clinical Biostats site.
27. Record Summary
ATTACC is a completed randomized phase 2/3 parallel-group treatment trial with 1200 participants and 2 arms. Its registered primary endpoint is mortality and days free of organ support at 21 days, defined as the number of days a participant is alive and free of invasive or non-invasive mechanical ventilation, high flow nasal oxygen (>30 L/min), vasopressor therapy, or ECMO support through the first 21 days after trial entry. Death at any time during the index hospital stay, including beyond day 21, is assigned the worst possible score of -1.
Statistically, the trial illustrates why endpoint construction matters. The primary outcome is neither a conventional binary mortality endpoint nor a conventional time-to-death endpoint. It is a bounded clinical outcome that integrates survival and organ-support burden, making the definition of the outcome central to the interpretation of any eventual treatment comparison.