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
DECLARE-TIMI 58 is a completed study registered as a replication of the DECLARE Diabetes Trial in healthcare claims. The registry identifies diabetes as the condition, dapagliflozin and a DPP-4 inhibitor as the two interventions, and a composite outcome of stroke, myocardial infarction, and mortality as the primary endpoint.
| Feature | DECLARE-TIMI 58 |
|---|---|
| Study title | Replication of the DECLARE Diabetes Trial in Healthcare Claims |
| Status | COMPLETED |
| Condition | Diabetes |
| Enrollment | 49790 |
| Number of arms | 2 |
| Interventions | Dapagliflozin; DPP-4 inhibitor |
| Primary endpoint | Composite outcome of Stroke, MI, and Mortality |
| Primary endpoint time frame | Through study completion (a median of 120-140 days) |
| Lead sponsor | Brigham and Women's Hospital |
| Sponsor type | OTHER |
| NCT identifier | NCT04215523 |
2. Clinical Question
The central statistical question is whether the occurrence of the registry-defined composite outcome of stroke, MI, and mortality differs between the dapagliflozin intervention and the DPP-4 inhibitor intervention over the study period.
Population
The registry identifies the study condition as diabetes. Enrollment was 49790.
Intervention
Dapagliflozin, listed by the registry as a drug intervention.
Comparator
DPP-4 inhibitor, listed by the registry as a drug intervention.
Primary question
How does the time to the composite outcome of stroke, MI, and mortality compare between the two intervention groups?
3. Trial Design
The registry describes a completed two-arm study conducted in the context of healthcare claims. The available record identifies the two interventions and the total enrollment but does not report additional design descriptors such as randomization ratio, masking, study phase, or geographic scope.
Dapagliflozin
- Drug intervention
- Study condition: diabetes
- Compared with a DPP-4 inhibitor
DPP-4 inhibitor
- Drug intervention
- Study condition: diabetes
- Compared with dapagliflozin
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary endpoint | Composite outcome of Stroke, MI, and Mortality | Through study completion (a median of 120-140 days) |
The registry description states: “Composite outcome of MI, stroke, and mortality - Please refer to uploaded protocol for full definition due to size limitations.” Thus, the registry identifies the three components of the composite but does not provide their full operational definition in the available endpoint description.
Why a composite endpoint changes the statistical question
A composite endpoint combines multiple clinical events into one analysis target. For a time-to-event composite, the event time for each participant is generally defined by the first qualifying component event. The resulting treatment comparison therefore addresses the time to any component of the composite, rather than automatically establishing the same treatment effect for each component separately.
This distinction matters because the components can differ in frequency, clinical importance, and susceptibility to the intervention. A composite can increase the number of observed events and therefore provide more statistical information, but interpretation depends on understanding what contributes to the composite.
5. Statistical Methodology
The registry identifies a primary endpoint with a time frame extending through study completion. For an endpoint structured as time to a first occurrence of stroke, MI, or mortality, the natural statistical framework is a time-to-event analysis.
Conceptually, the composite event time is the earliest qualifying component event. The exact operational rules depend on the protocol's full endpoint definition.
Kaplan-Meier estimation
For a time-to-event endpoint, the Kaplan-Meier estimator can describe the probability of remaining free of the composite event over time. It accounts for participants whose event status is not observed for the entire study period by treating their follow-up as right-censored under the applicable censoring rules.
Here, di is the number of events at time ti and ni is the number at risk immediately before that time.
Hazard ratio
A Cox proportional-hazards model is a standard method for estimating a relative treatment effect for a time-to-event endpoint. The hazard ratio compares the instantaneous event rates between the two groups under the fitted model.
The hazard ratio is not a relative risk, an absolute risk difference, or the proportion of participants who avoid the event.
Log-rank comparison
A log-rank test is commonly used to compare survival distributions between two groups. It uses the ordering of observed event times and accounts for censoring. For a two-arm time-to-event study, it can provide a hypothesis test complementary to a hazard-ratio estimate.
Analysis population
The available registry information does not specify an intention-to-treat population, per-protocol population, as-treated population, or other formal analysis population. Those definitions can materially affect interpretation when exposure changes, treatment discontinuation, or eligibility events occur.
6. Planned Analysis
The registry states that the primary endpoint will be the composite outcome of stroke, MI, and mortality, measured through study completion, with a median time frame of 120-140 days. No statistical estimates, confidence intervals, or p-values are posted in the ClinicalTrials.gov record.
How an endpoint of this type is typically analysed
For a composite endpoint measured over time, a conventional primary analysis would generally compare the time-to-first-composite-event distributions between the two intervention groups. Kaplan-Meier methods can estimate event-free probabilities over follow-up, while a log-rank test can compare the groups. A Cox proportional-hazards model can estimate a hazard ratio and its confidence interval.
If the registry's underlying statistical analysis plan specifies adjustment variables, stratification, competing-risk methods, censoring rules, or a particular estimand, those details would determine the exact implementation. The available ClinicalTrials.gov record does not report those additional analysis specifications.
7. Understanding the Composite Outcome
The primary endpoint combines stroke, MI, and mortality. Statistically, a composite endpoint can be useful because a participant contributes an event when any qualifying component occurs. This can increase the total number of endpoint events relative to an analysis restricted to a single component.
Stroke
Stroke is one of the three named components of the registry-defined primary composite.
MI
Myocardial infarction is another named component of the primary composite.
Mortality
Mortality is the third named component of the primary composite.
Combined endpoint
The primary analysis concerns the composite rather than automatically providing separate statistical conclusions for each component.
The registry does not provide the component-specific event counts or effect estimates in the available record. Therefore, the overall composite and its individual components should not be treated as interchangeable statistical endpoints.
Why component balance matters
Suppose a composite is driven primarily by one component while the other components occur much less frequently. The overall composite estimate will then reflect the component contributing most of the events. This does not make the composite invalid, but it changes what the estimate represents.
Clinical interpretation also matters. Components should ideally be sufficiently related in clinical importance and expected direction of treatment effect for a single composite estimate to have a coherent meaning. The registry's abbreviated endpoint description does not provide enough detail to evaluate those properties.
8. Time-to-Event Analysis in a Healthcare-Claims Study
Healthcare-claims data can support large-scale observational comparisons, but the statistical structure differs from a conventional individually randomized clinical trial. The study title itself identifies this work as a replication in healthcare claims, making careful definition of exposure, eligibility, outcome ascertainment, follow-up, and censoring particularly important.
Large sample size does not by itself eliminate bias. In claims-based comparative effectiveness research, the statistical analysis must distinguish precision from validity: a very precise estimate can still be affected by systematic differences between exposure groups if the design or analysis does not adequately address them.
9. Statistical Methods Explained
What is the primary statistical object being compared?
The primary object is the time until the first occurrence of the composite endpoint consisting of stroke, MI, or mortality. This is different from simply comparing the percentage of participants who experienced an event because the timing of events and censoring can contain additional information.
Why use a composite endpoint?
Combining several outcomes can increase the number of events available for analysis. That can improve statistical efficiency when the components are clinically coherent and respond in a sufficiently related way to the intervention. The tradeoff is that the resulting estimate represents the combined endpoint, and the contribution of each component matters for interpretation.
What does a hazard ratio represent?
A hazard ratio compares instantaneous event rates between groups within a time-to-event model. An HR below 1 would indicate a lower estimated hazard in the dapagliflozin group relative to the DPP-4 inhibitor group, while an HR above 1 would indicate a higher estimated hazard.
Why is a confidence interval important?
A confidence interval communicates statistical precision around an estimated treatment effect. A narrow interval indicates greater statistical precision than a wide interval, all else equal. It does not describe the range of effects that individual participants experience.
Why does the p-value not measure effect size?
A p-value addresses the compatibility of the observed data with a specified null hypothesis under the statistical model. It does not tell the reader how large or clinically important an effect is. Effect estimates and confidence intervals are needed to describe magnitude and precision.
What does censoring mean?
Censoring occurs when a participant's event status is not observed beyond a particular point during follow-up. Time-to-event methods allow such participants to contribute information up to their censoring time, provided the assumptions behind the censoring mechanism and analysis are appropriate.
Why does the proportional-hazards assumption matter?
A Cox model's conventional hazard-ratio interpretation is based on proportional hazards: the relative hazard is assumed to be reasonably stable over time. If the hazards cross or the treatment effect changes substantially over follow-up, a single hazard ratio may summarize a more complicated pattern inadequately.
10. What a Reported Hazard Ratio Would Mean
If a future analysis reported an HR below 1 for the composite endpoint, the estimate would indicate a lower estimated instantaneous rate of first composite events in the dapagliflozin group relative to the DPP-4 inhibitor group under the fitted time-to-event model.
It would not mean that the same percentage of participants avoided stroke, MI, or mortality, and it would not imply that every component of the composite was reduced by the same amount.
A confidence interval around a hazard ratio would describe uncertainty in the estimated relative treatment effect under the specified model and sampling framework. The interval would not represent the range of individual participant responses.
When available, cumulative event probabilities or event rates at clinically relevant times complement the hazard ratio. Relative and absolute measures answer different questions and together provide a more complete statistical description of treatment differences.
11. Causal Interpretation and Claims-Based Data
The study is explicitly described as a replication of a diabetes trial in healthcare claims. That makes the distinction between association and causal treatment effect especially important when interpreting a claims-based comparison.
Healthcare claims can provide extensive longitudinal information about diagnoses, procedures, prescriptions, and healthcare utilization. They can also introduce systematic differences between people receiving different therapies. Examples include differences in underlying clinical risk, prescribing patterns, healthcare access, prior treatment, or coding practices.
The available registry information does not specify whether the analysis uses propensity-score methods, inverse-probability weighting, matching, multivariable adjustment, instrumental variables, or another strategy to address confounding. Those details are therefore important for evaluating the causal interpretation of any eventual effect estimate.
12. Missing Data, Censoring, and Follow-Up
The registry specifies that the primary endpoint is evaluated through study completion and gives a median time frame of 120-140 days. A time-to-event analysis would need explicit rules for when participants enter the risk set, when follow-up ends, and how incomplete outcome observation is handled.
| Issue | Statistical importance |
|---|---|
| Entry into follow-up | Defines when participants begin contributing person-time to the analysis. |
| Censoring | Determines how participants without an observed event at the end of available follow-up contribute information. |
| Outcome ascertainment | Claims-based definitions determine which records qualify as stroke, MI, or mortality. |
| Exposure definition | The classification of dapagliflozin and DPP-4 inhibitor exposure determines which treatment comparison is actually estimated. |
| Loss to follow-up | Can affect validity if censoring is related to future outcome risk in ways not adequately handled by the analysis. |
The registry's abbreviated endpoint description does not provide the full claims-based definitions or censoring rules. Those specifications would be essential for reproducing the statistical estimand precisely.
13. Non-Inferiority, Multiplicity, and Interim Analysis
The available ClinicalTrials.gov record does not report a non-inferiority margin, factorial design, crossover procedure, multiplicity strategy, interim efficacy boundary, or alpha-spending approach. Accordingly, the primary endpoint should be interpreted as a single registry-defined composite outcome without attributing a particular confirmatory framework to the study.
Non-inferiority margin
No non-inferiority margin is reported in the available registry information.
Multiplicity
No multiplicity adjustment or endpoint hierarchy is reported in the available registry information.
Interim analysis
No interim-analysis schedule or alpha-spending procedure is reported.
Factorial or crossover design
No factorial structure or crossover procedure is reported.
14. Safety
The registry identifies the interventions but does not provide serious adverse event counts or other arm-specific safety results in the available record. Consequently, there is no registry-posted safety estimate to compare between dapagliflozin and the DPP-4 inhibitor intervention.
| Safety measure | Dapagliflozin | DPP-4 inhibitor |
|---|---|---|
| Serious adverse events | Not reported in the available registry record | Not reported in the available registry record |
Safety and effectiveness are distinct statistical domains. Even in a large comparative study, an effectiveness estimate should not be treated as a global benefit-risk measure without corresponding safety information and an explicit framework for integrating the two.
15. Study Timeline
Study start
The registry lists 2019-07-08 as the study start date.
Primary completion
The registry lists 2021-02-18 as the primary completion date.
Registry status
The study is listed as COMPLETED in the ClinicalTrials.gov record.
16. Trial Scale and Statistical Precision
The enrollment of 49790 is substantial for a two-arm comparative study. From a statistical perspective, a large cohort can provide many opportunities to observe relatively uncommon outcomes and can narrow confidence intervals when event counts are sufficient.
Enrollment
Two-arm study comparing dapagliflozin with a DPP-4 inhibitor.
However, enrollment alone does not determine the precision of the primary endpoint. For a time-to-event analysis, information is driven strongly by the number and timing of observed events, as well as censoring and follow-up. The registry does not report the number of participants experiencing the composite endpoint.
Nor does a large sample size automatically resolve questions of confounding in claims-based comparisons. Precision and validity are different properties: precision concerns statistical uncertainty, whereas validity concerns whether the estimated quantity corresponds to the intended clinical comparison.
17. Limitations
- Abbreviated endpoint definition: the registry names stroke, MI, and mortality but refers to the uploaded protocol for the full definition.
- No posted statistical estimates: the available registry record does not provide an event count, hazard ratio, confidence interval, or p-value for the primary endpoint.
- Claims-based replication: healthcare-claims analyses depend on how exposure and clinical outcomes are identified from administrative data.
- Unknown adjustment strategy: the available record does not specify the covariates or statistical adjustment methods used to address differences between the intervention groups.
- Unknown censoring rules: the registry does not provide the complete follow-up and censoring specification needed to reproduce a time-to-event estimand.
- Composite interpretation: the three components of the primary endpoint can have different frequencies and clinical meanings, yet component-specific estimates are not reported in the available record.
- Limited design detail: randomization, masking, allocation, stratification, crossover, and interim-analysis procedures are not described in the available registry information.
- Limited safety reporting: arm-specific serious adverse event results are not posted in the available record.
- Short registered time frame: the primary endpoint is assessed through study completion, with a median time frame of 120-140 days.
18. Why This Trial Matters Statistically
DECLARE-TIMI 58 is a useful teaching case because it illustrates how a large healthcare-claims study can be organized around a clinically meaningful composite time-to-event endpoint while raising important questions about causal inference, endpoint construction, censoring, and precision.
| Concept | How it appears in DECLARE-TIMI 58 |
|---|---|
| Comparative effectiveness | Dapagliflozin is compared with a DPP-4 inhibitor in diabetes. |
| Large study population | The registry reports enrollment of 49790. |
| Two-arm comparison | The registry reports 2 arms. |
| Composite endpoint | Stroke, MI, and mortality are combined into the primary outcome. |
| Time-to-event analysis | The primary endpoint is assessed through study completion over a stated time frame. |
| Kaplan-Meier estimation | A standard approach for describing time to a composite event when censoring is present. |
| Hazard ratio | A standard relative effect measure for Cox analysis of time-to-event outcomes. |
| Confidence interval | Provides precision information around a future effect estimate. |
| Confounding | Claims-based comparisons require attention to systematic differences between exposure groups. |
| Outcome ascertainment | The statistical analysis depends on operational definitions of stroke, MI, and mortality. |
| Censoring | Follow-up rules determine how incomplete observation contributes to the time-to-event analysis. |
19. Related Statistical Tutorials
Learn more about the methods used in this trial:
20. Related Statistical Calculators
21. Sources
- ClinicalTrials.gov: NCT04215523 — DECLARE-TIMI 58.
- PubMed: PubMed record for PMID 33327727.
Continue through Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical trial analyses to connect trial design with the methods used to analyze clinical evidence.
22. Record Summary
DECLARE-TIMI 58 is registered as a completed replication of the DECLARE Diabetes Trial in healthcare claims, enrolling 49790 participants across 2 intervention arms. The primary endpoint is the registry-defined composite outcome of stroke, MI, and mortality, measured through study completion with a median time frame of 120-140 days.
Statistically, the endpoint is naturally framed as a time-to-event comparison. Kaplan-Meier estimation, log-rank testing, and Cox proportional-hazards modeling are standard approaches for this type of outcome, although the available registry information does not specify which particular statistical model or adjustment strategy was used. The claims-based setting also makes exposure definition, outcome ascertainment, censoring, and confounding central to interpretation.