This page separates reported trial information from statistical interpretation. The ClinicalTrials.gov record contains three posted primary outcome measures, but no entries in the posted results.
1. Trial at a Glance
HEART-FID was a randomized, double-blind, parallel-group phase 3 trial evaluating ferric carboxymaltose versus placebo for heart failure with iron deficiency. The ClinicalTrials.gov record reports enrollment of 3065 participants, two treatment arms, and three primary endpoints.
| Feature | HEART-FID |
|---|---|
| Phase | Phase 3 |
| Conditions | Heart Failure; Iron-deficiency |
| Design | Randomized, parallel assignment, double-blind |
| Primary purpose | Treatment |
| Enrollment | 3065 |
| Interventions | Ferric Carboxymaltose; Placebo |
| Primary endpoints | Number of Deaths; Number of Hospitalizations for Heart Failure; Change in 6MWT Distance |
| Trial status | COMPLETED |
| Start | 2017-03-15 |
| Primary completion | 2023-02-02 |
| Lead sponsor | American Regent, Inc. |
| ClinicalTrials.gov | NCT03037931 |
2. Clinical Question
The trial evaluates ferric carboxymaltose against placebo in participants with heart failure and iron deficiency. The registered primary outcomes address three clinically distinct dimensions: death, hospitalization for heart failure, and change in walking distance measured by the six-minute walk test.
Population
Participants in the trial with the registered conditions of heart failure and iron-deficiency.
Intervention
Ferric Carboxymaltose.
Comparator
Placebo.
Primary question
How do the randomized treatment groups compare on death, hospitalization for heart failure, and change in six-minute walk distance over the registered time frames?
3. Trial Design
Ferric Carboxymaltose
- Drug intervention.
- 1532 participants were at risk for the reported serious-adverse-event measure.
Placebo
- Drug/placebo intervention.
- 1533 participants were at risk for the reported serious-adverse-event measure.
The two serious-adverse-event denominators sum to 3065, matching the reported total enrollment. This provides a useful internal consistency check on the arm-level safety denominators reported in the ClinicalTrials.gov record.
4. Endpoints
| Primary endpoint | Time frame | Registry definition | Endpoint type |
|---|---|---|---|
| Number of Deaths | 1 year | The number of participants who died out of the total treated group. | Binary |
| Number of Hospitalizations for Heart Failure | 1 year | The number of participants who were hospitalized for heart failure out of the total treated group. | Binary |
| Change in 6MWT (Six Minute Walk Test) Distance | 6 months | The change in meters walked at baseline compared to 6 months later. | Other / unclear |
Three endpoints, three statistical structures
The endpoint definitions imply different natural statistical representations. The first two are binary outcomes over a fixed one-year time frame. The third is a continuous change-from-baseline outcome measured in meters at six months.
Death at 1 year
A participant either meets the registered event definition or does not. A treatment comparison can therefore be expressed through event proportions, a risk difference, a risk ratio, or an odds ratio.
HF hospitalization at 1 year
This is also represented in the registry as a binary endpoint over one year. The appropriate analysis depends on the prespecified estimand and whether event timing or recurrent hospitalization is part of the analysis.
6MWT change
The endpoint is the change in meters walked from baseline to six months. A continuous-outcome analysis can compare mean change, adjusted mean change, or another prespecified summary.
Why the distinction matters
A percentage of participants with an event and a mean change in meters are different estimands. They should not be combined into one generic effect measure without an explicit statistical framework.
5. Results
ClinicalTrials.gov reports results for all three registered primary endpoints, but the ClinicalTrials.gov record contains no formal statistical analyses.
Number of Deaths
Registered primary endpoint
The number of participants who died out of the total treated group.
No formal statistical estimate, confidence interval, or p-value is present in the posted results.
For a binary one-year mortality endpoint, a conventional analysis would compare the proportion of participants who died in each randomized group. Depending on the prespecified estimand, useful measures could include an absolute risk difference, relative risk, or odds ratio, with a confidence interval and an appropriate hypothesis test. If exact event timing and censoring were central to the analysis instead, a time-to-event approach could be considered. The ClinicalTrials.gov record does not identify which formal comparison was used.
The registry definition establishes what was counted: participants who died during the one-year endpoint period. It does not, in the ClinicalTrials.gov record, establish the magnitude of the difference between treatment groups.
A confidence interval would be needed to describe the precision of a reported treatment-effect estimate. A p-value, if reported, would address evidence against a specified null hypothesis under the chosen analysis; it would not measure the size or clinical importance of the effect.
Without the formal analysis entry, it would also be inappropriate to assume a particular denominator, missing-data rule, continuity correction, or hypothesis-testing framework beyond the registry's definition.
Number of Hospitalizations for Heart Failure
Registered primary endpoint
The number of participants who were hospitalized for heart failure out of the total treated group.
No formal statistical estimate, confidence interval, or p-value is present in the posted results.
For the registry-defined binary endpoint, a standard analysis would compare the proportion of participants experiencing a hospitalization for heart failure by one year. However, heart-failure hospitalization can also be represented as a time-to-first-event or recurrent-event outcome in other trial frameworks. Those approaches answer different questions. Because the registry-reported statistical-analysis field is empty, the formal estimand and analysis cannot be reconstructed from the ClinicalTrials.gov record alone.
The endpoint is defined as the number of participants who were hospitalized for heart failure out of the total treated group. That wording does not specify whether the formal analysis emphasized first hospitalization, timing, recurrent events, or a composite estimand.
An effect estimate should therefore not be reverse-engineered from the endpoint label. If a risk difference were reported, its confidence interval would describe uncertainty around that difference. If a hazard ratio were used, its interpretation would instead depend on the time-to-event model and its assumptions.
Change in 6MWT Distance
Registered primary endpoint
Change in meters walked at baseline compared to 6 months later.
No formal statistical estimate, confidence interval, or p-value is present in the posted results.
A continuous change-from-baseline endpoint is commonly analyzed by comparing mean change between randomized groups or by using an analysis of covariance that adjusts the six-month value for baseline walking distance and treatment assignment. Other methods may be appropriate depending on the prespecified estimand, distributional assumptions, repeated measurements, and missing-data strategy. The ClinicalTrials.gov record does not identify the formal method used.
The estimand is expressed in meters of change, not as a percentage of participants. A positive between-group difference would represent greater improvement in walking distance for one group under the chosen direction of subtraction.
The precision of a continuous treatment effect would normally be conveyed by a confidence interval around the between-group difference or adjusted mean difference. A p-value would address the null hypothesis specified for that comparison, not the size or practical importance of a difference.
Baseline imbalance, missing six-month measurements, non-normality, and the prespecified direction of the treatment contrast can all affect how the result should be interpreted. None of those details are in the posted results provided here.
6. Statistical Methodology
The ClinicalTrials.gov record identifies the study as randomized, parallel, double-blind, and treatment-focused. It also identifies three primary endpoints but provides no statistical-analysis entries. The methodology below therefore distinguishes what the registry establishes from what would ordinarily be considered for these endpoint types.
| Endpoint | Registry type | Typical statistical framework | What cannot be established from the ClinicalTrials.gov record |
|---|---|---|---|
| Number of Deaths | Binary | Two-group comparison of one-year event proportions; risk difference, relative risk, or odds ratio may be used depending on the prespecified estimand. | Exact effect measure, confidence interval method, p-value, and missing-data rule. |
| Number of Hospitalizations for Heart Failure | Binary | Two-group comparison of one-year event proportions, or a time-to-event/recurrent-event approach if prespecified. | Whether the formal analysis treated hospitalization as binary, time-to-first event, recurrent events, or another estimand. |
| Change in 6MWT Distance | Other / unclear | Comparison of six-month change between randomized groups; ANCOVA is one common approach when baseline measurement is available. | Exact model, covariates, estimand, distributional assumptions, and missing-data strategy. |
Randomization and causal comparison
Randomization is the central design feature supporting an unbiased comparison of treatment assignments, assuming the randomized groups are analyzed according to their assigned treatment and the trial's conduct and analysis preserve the intended comparison. Randomization does not make every observed variable identical between groups; chance imbalance remains possible.
Blinding
The trial is described as double-blind. Blinding can reduce the influence of knowledge of treatment assignment on participant behavior, outcome assessment, reporting, and other aspects of trial conduct. Its value depends on whether the masking procedure remains credible throughout the study.
Binary endpoint analysis
For a fixed one-year binary outcome, the basic data structure is a two-by-two table: treatment group versus whether the participant experienced the registered event. Several effect measures are possible, and each answers a somewhat different question.
The risk difference expresses the absolute difference in event probability between groups. Its interpretation depends on the direction of the subtraction and the endpoint definition.
Continuous change-from-baseline analysis
The six-minute walk endpoint differs fundamentally from the binary endpoints because its measurement scale is continuous. An analysis based on the difference in mean change can be clinically interpretable in meters. An ANCOVA approach can improve precision when baseline distance is strongly associated with the six-month outcome.
Missing data
Change-from-baseline analyses require both baseline and six-month information for the relevant estimand, unless a model-based approach accommodates incomplete follow-up. The ClinicalTrials.gov record does not state which imputation, likelihood-based, or other missing-data method was used.
7. Statistical Methods Explained
Why does randomization matter statistically?
Randomization creates treatment groups through an assignment mechanism rather than by allowing participants or investigators to choose treatment. In principle, this balances both measured and unmeasured prognostic factors in expectation. The resulting between-group comparison can therefore be interpreted as a comparison of randomized treatment strategies, subject to trial adherence and analysis assumptions.
Why is blinding important?
When participants and investigators do not know the assigned treatment, expectations about treatment can have less opportunity to influence behavior, reporting, concomitant care, or outcome assessment. Double blinding therefore addresses a source of bias that randomization alone does not eliminate.
How would a one-year death endpoint normally be analyzed?
If the estimand is simply whether death occurred by one year, each participant contributes an event indicator and the treatment groups can be compared using an appropriate binary-outcome method. The resulting effect could be expressed as an absolute risk difference or a relative measure. If timing of death and censoring are central to the estimand, survival analysis may instead be appropriate.
How is heart-failure hospitalization different from mortality?
Both are registered as binary endpoints over one year, but hospitalization can occur more than once and can have a clinically meaningful timing structure. A simple binary endpoint asks whether the participant experienced the specified event during the period. A time-to-first-event analysis asks when the first event occurred. A recurrent-event analysis asks about the burden of repeated events. These are different estimands and should not be treated as interchangeable.
Why might ANCOVA be used for six-minute walk change?
When baseline walking distance is measured, an ANCOVA model can compare six-month walking distance between randomized groups while adjusting for baseline distance. This can account for prognostic baseline variation and may provide a more precise treatment comparison than analyzing raw change alone. Whether ANCOVA was actually prespecified or used in HEART-FID is not established by the ClinicalTrials.gov record.
Why should a p-value not be treated as an effect size?
A p-value is a probability calculation under a specified null hypothesis and statistical model. It does not state how large the treatment effect is. Effect estimates and confidence intervals are needed to describe magnitude and precision. A small p-value can occur with a modest effect in a large sample, while an important effect can be estimated imprecisely in a smaller sample.
8. Analysis Populations and Denominators
The ClinicalTrials.gov record provides total enrollment of 3065 and serious-adverse-event denominators of 1532 for ferric carboxymaltose and 1533 for placebo. They do not provide separate named efficacy analysis populations such as intention-to-treat, per-protocol, or modified intention-to-treat populations.
| Population / denominator | Ferric Carboxymaltose | Placebo | What the ClinicalTrials.gov record establishes |
|---|---|---|---|
| Total trial enrollment | 3065 | Total participants enrolled. | |
| Serious adverse-event denominator | 1532 | 1533 | Participants at risk for the registry-reported serious-AE counts. |
| Formal efficacy analysis population | Not reported | No named efficacy analysis population is provided in the ClinicalTrials.gov record. | |
9. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm. These counts are affected participants divided by the registry-reported number at risk.
| Safety measure | Ferric Carboxymaltose | Placebo |
|---|---|---|
| Serious adverse events, affected / at risk | 413 / 1532 | 401 / 1533 |
The ClinicalTrials.gov record establishes the affected counts and their denominators. It does not supply a formal between-group hypothesis test or confidence interval for serious adverse events.
The serious-adverse-event counts are close in absolute number, but an appropriate comparison should use the corresponding denominators rather than comparing raw counts alone.
For a binary safety endpoint, an analyst could report event proportions and an absolute risk difference or relative measure, together with its confidence interval. The appropriate method depends on the prespecified safety estimand and analysis population.
The serious-AE data should also be kept separate from the primary efficacy endpoints. Safety and efficacy address different outcomes and should not be collapsed into a single numerical conclusion.
10. Multiplicity and the Three Primary Endpoints
HEART-FID has three registered primary endpoints: two binary outcomes at one year and a change-from-baseline six-minute walk outcome at six months. Multiple primary endpoints create a statistical question beyond the analysis of any single endpoint: how should the overall type I error and the interpretation of the collection of tests be controlled?
| Endpoint | Time frame | Primary endpoint status | Formal multiplicity method reported? |
|---|---|---|---|
| Number of Deaths | 1 year | Yes | No |
| Number of Hospitalizations for Heart Failure | 1 year | Yes | No |
| Change in 6MWT Distance | 6 months | Yes | No |
The absence of a registry-reported statistical-analysis entry means the multiplicity procedure cannot be reconstructed from the ClinicalTrials.gov record. Possible prespecified approaches in a multi-endpoint trial include a hierarchical testing strategy, alpha allocation, a composite or hierarchical estimand, or another familywise error-control method. Those are methodological possibilities, not claims about the method used in HEART-FID.
11. Hierarchical Thinking About the Endpoints
The three registered endpoints differ in clinical type and measurement scale. This makes the trial a useful example of why endpoint hierarchy matters in clinical-trial statistics.
Mortality
A binary one-year endpoint centered on death, an objectively defined clinical event.
Hospitalization
A binary one-year endpoint centered on hospitalization for heart failure.
Functional capacity
A continuous six-month endpoint based on change in six-minute walk distance.
Interpretation
These outcomes represent different dimensions of clinical status and cannot be reduced to one effect measure without an explicitly defined statistical estimand.
12. What a Formal Analysis Would Need to Specify
Because the registry-reported statistical-analysis field is empty, several important elements of a complete statistical analysis remain unspecified in the ClinicalTrials.gov record. These should be obtained from the prespecified statistical analysis plan or an authoritative detailed results source before making claims about formal hypothesis testing.
| Statistical element | Why it matters |
|---|---|
| Estimand | Defines precisely what treatment effect is being estimated and for which population, treatment strategy, outcome, and time frame. |
| Analysis population | Determines which participants contribute to the primary efficacy comparison. |
| Effect measure | Determines whether the result is expressed as a risk difference, relative risk, odds ratio, mean difference, or another quantity. |
| Confidence interval method | Determines how statistical uncertainty around the estimate is quantified. |
| Hypothesis test | Defines the null hypothesis and the test used to evaluate evidence against it. |
| Multiplicity control | Determines how the three primary endpoints are handled within the overall type I error framework. |
| Missing-data strategy | Defines how incomplete endpoint observations affect the analysis, particularly for the six-month 6MWT outcome. |
| Sensitivity analyses | Evaluate whether conclusions depend materially on modeling or missing-data assumptions. |
13. Missing Data and Censoring
Missing data are particularly relevant to the six-minute walk endpoint because the registered outcome requires a comparison between baseline and six months. A participant without a usable six-month measurement cannot simply be treated as having a particular amount of change without an explicit rule.
For the one-year binary endpoints, the analysis also needs a defined approach to participants whose outcome status cannot be established at the relevant time point. If the analysis instead uses time-to-event information, censoring becomes an explicit component of the statistical model.
14. Interim Analysis, Crossover, Factorial Design, and Bayesian Methods
the ClinicalTrials.gov record does not provide information supporting a specific interim-analysis procedure, crossover design, factorial design, or Bayesian primary analysis.
| Design topic | Information reported | Interpretation |
|---|---|---|
| Interim analysis | Not reported | No interim-testing method should be inferred. |
| Crossover | Not reported | The trial should not be described as a crossover study. |
| Factorial design | Not reported | The registered model is parallel assignment, not a reported factorial design. |
| Bayesian methods | Not reported | No Bayesian primary analysis should be assumed. |
| Stratification | Not reported | No stratification factors are identified in the ClinicalTrials.gov record. |
This distinction is important because the absence of a field in the ClinicalTrials.gov record is not evidence that a statistical feature could never have been used elsewhere in the trial documentation. It means only that the feature cannot be responsibly asserted from the ClinicalTrials.gov record.
15. Trial Timeline
Trial start
The ClinicalTrials.gov record identifies 2017-03-15 as the study start date.
Randomized treatment study
The study is registered as a phase 3 randomized, parallel, double-blind treatment trial with two intervention arms.
Enrollment
the ClinicalTrials.gov record reports actual enrollment of 3065 participants.
Primary completion
The ClinicalTrials.gov record identifies 2023-02-02 as the primary completion date.
Study status
the ClinicalTrials.gov record reports the study as completed.
16. Why This Trial Matters Statistically
HEART-FID is a useful teaching example because its registered primary endpoints span different statistical data structures within one randomized trial. That creates an opportunity to distinguish binary-event analysis from continuous change-from-baseline analysis while also considering the consequences of having multiple primary endpoints.
| Concept | How it appears in HEART-FID |
|---|---|
| Randomization | Randomized allocation in a two-arm parallel design. |
| Blinding | Double-blind trial design. |
| Binary endpoints | Number of deaths and number of hospitalizations for heart failure at 1 year. |
| Continuous endpoint | Change in six-minute walk distance at 6 months. |
| Multiple primary endpoints | Three registered primary endpoints require a defined inferential framework. |
| Effect measures | Binary and continuous endpoints can require different measures of treatment effect. |
| Missing data | Especially relevant to a six-month change-from-baseline outcome. |
| Safety analysis | Serious adverse events are reported with arm-specific affected and at-risk counts. |
| Analysis populations | Safety denominators are reported, while named formal efficacy populations are not. |
17. Statistical Methods Explained: Reading the Trial Correctly
A binary endpoint records whether the participant experienced the specified outcome within the defined period. For HEART-FID, the registry defines both death and hospitalization for heart failure this way at one year. The primary result would normally compare the resulting event proportions between randomized groups using a prespecified effect measure.
Change in six-minute walk distance is measured on a continuous scale. Instead of counting participants with an event, the analyst compares the amount of change in meters between treatment groups. The resulting treatment effect is naturally expressed in meters or an adjusted difference in meters.
Statistical methods should follow the estimand and outcome structure. A binary event, a time-to-event outcome, and a continuous change-from-baseline outcome contain different information and require different assumptions. A single generic analysis cannot automatically serve all three.
When several primary hypotheses are tested, the probability of at least one false-positive finding can increase if each hypothesis is tested independently at the same nominal level. A prespecified multiplicity strategy determines how the evidence should be interpreted collectively.
A confidence interval places the estimated treatment effect in an uncertainty framework. It provides information that a p-value alone does not: the plausible precision and direction of the estimated effect under the selected statistical model.
Raw counts do not account for the number of participants exposed to the risk of an event. The registry-reported serious-adverse-event results therefore report both affected participants and the corresponding at-risk denominators: 413/1532 for ferric carboxymaltose and 401/1533 for placebo.
18. Serious Adverse Events: Statistical Interpretation
The registry-reported serious-adverse-event results can be viewed as two binomial proportions: affected participants among those at risk in each treatment arm.
A formal analysis would compare these proportions using a prespecified effect measure and confidence-interval method. The ClinicalTrials.gov record does not report that formal comparison.
It is useful to distinguish a descriptive safety table from a formal safety hypothesis test. The former tells the reader how many affected participants were observed relative to the at-risk population. The latter requires a specified null hypothesis, effect measure, analysis population, and inferential procedure.
19. Limitations
- No registry-reported statistical analyses: the trial data identify three posted primary endpoints but contain no formal statistical analyses. Formal estimates, confidence intervals, and p-values therefore cannot be reconstructed from the ClinicalTrials.gov record alone.
- Endpoint structure: the two one-year primary endpoints are labeled binary, while the six-month 6MWT endpoint is classified as other / unclear. The precise estimands and models are not reported.
- Multiplicity: three primary endpoints require an explicit inferential framework, but the ClinicalTrials.gov record does not identify the multiplicity-control method.
- Analysis populations: the ClinicalTrials.gov record identifies arm-specific safety denominators but do not provide named formal efficacy analysis populations.
- Missing data: no imputation or sensitivity-analysis strategy is posted on ClinicalTrials.gov for the primary outcomes.
- Statistical model assumptions: the ClinicalTrials.gov record does not establish whether any time-to-event model, continuous-outcome model, or particular binary regression model was used.
- Safety interpretation: serious-adverse-event counts are reported, but no formal between-group safety comparison is provided in the registry-reported statistical-analysis field.
20. What the Registry Data Do Establish
Despite the absence of formal statistical-analysis entries, the ClinicalTrials.gov record contains several important design facts that are sufficient for a structured statistical interpretation.
Large randomized population
The trial enrolled 3065 participants, providing a substantial randomized population for the registered primary outcomes.
Two-arm comparison
The intervention structure consists of ferric carboxymaltose and placebo in a parallel-group randomized design.
Double blinding
The double-blind structure is directly identified in the ClinicalTrials.gov record and is relevant to reducing expectation-related bias.
Different outcome scales
The primary endpoints span binary clinical events and a continuous functional measurement, making endpoint-specific analysis important.
21. A Statistical Reading of the Trial
The most important statistical lesson from HEART-FID is that a clinical trial is not defined by one universal statistical test. The randomized design establishes the comparison structure, but the endpoint determines what must be estimated.
For the one-year death endpoint, the central quantity would ordinarily be a comparison of event risk between randomized groups. For one-year hospitalization, the same binary framework may be appropriate if the estimand is simply whether a participant was hospitalized, although other event-based frameworks could answer different questions. For six-minute walk distance, the natural treatment effect is a difference in functional change rather than a ratio of event hazards.
Those distinctions matter because the same numerical result can have different meanings depending on its estimand. A risk difference answers an absolute probability question. A risk ratio answers a relative probability question. A mean difference answers a continuous-outcome question. A hazard ratio answers a time-to-event model question. They should not be substituted for one another without justification.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Statistical Calculators
24. Sources
- ClinicalTrials.gov: NCT03037931.
- PubMed: PubMed record 41137844.
- PubMed: PubMed record 39890013.
- PubMed: PubMed record 38250799.
- PubMed: PubMed record 37632463.
- PubMed: PubMed record 37598795.
The PubMed links above are provided as source-navigation links only. The numerical trial facts and statistical claims on this page are restricted to the ClinicalTrials.gov record.
Continue through Clinical Biostats
Use the broader tutorial and calculator collections to explore randomized trial design, endpoint selection, and statistical analysis methods.
25. Record Summary
HEART-FID is a completed randomized, double-blind, parallel-group phase 3 trial with 3065 enrolled participants and two treatment arms: ferric carboxymaltose and placebo. Its three registered primary endpoints span one-year death, one-year hospitalization for heart failure, and six-month change in six-minute walk distance.
The ClinicalTrials.gov record also provide arm-specific serious-adverse-event counts of 413/1532 for ferric carboxymaltose and 401/1533 for placebo. These data support descriptive safety interpretation, but the registry-reported statistical-analysis field contains no formal treatment-effect estimates, confidence intervals, or p-values.
Statistically, the trial is therefore most useful as an example of how randomized design, blinding, binary endpoints, continuous change-from-baseline outcomes, multiplicity, analysis populations, and missing-data considerations fit together. The appropriate statistical interpretation begins with the registered estimand and endpoint definition rather than assuming a particular model or significance test.