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
DIRECT was a randomized, parallel-group, quadruple-masked phase 3 trial evaluating candesartan in participants with type 1 diabetes. The registered primary endpoint assessed the first occurrence of a 3-step or greater increase in the Early Treatment of Diabetic Retinopathy Study (EDTRS) severity scale over approximately 5 years.
| Feature | DIRECT |
|---|---|
| Trial name | DIRECT |
| Brief title | DIabetic Retinopathy Candesartan Trials. |
| Phase | Phase 3 |
| Status | Completed |
| Therapeutic area | Endocrinology & Metabolism |
| Condition | Type 1 Diabetes |
| Enrollment | 1,850 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Intervention | Candesartan |
| Primary endpoint type | Binary; analyzed as a time-to-event endpoint |
| Primary analysis | Generalized log-rank test |
| Effect measure | Hazard ratio |
| Hypothesis type | Superiority |
| Lead sponsor | AstraZeneca |
2. Clinical Question
The primary clinical question was whether treatment with candesartan differed from placebo with respect to the time to the first occurrence of a 3-step or greater increase in the EDTRS severity scale in participants with type 1 diabetes.
Population
Participants with type 1 diabetes enrolled in the DIRECT phase 3 trial.
Intervention
Candesartan.
Comparator
Placebo.
Primary question
Does candesartan alter the time to a first 3-step or greater increase in EDTRS severity scale compared with placebo?
3. Trial Design
Candesartan
- Intervention: candesartan
- Randomized treatment arm
- 951 participants at risk for the serious-adverse-event summary
Placebo
- Comparator treatment
- Randomized control arm
- 951 participants at risk for the serious-adverse-event summary
The registry identifies the study as randomized and parallel with quadruple masking. These design features are important statistically because randomization establishes the basis for comparing outcomes between treatment groups, while masking can reduce the influence of treatment knowledge on outcome assessment and other aspects of trial conduct.
4. Trial Timeline
Study start
August 2001.
Primary completion
February 2008.
Planned primary time frame
From baseline to the end of the study, i.e. 5 years.
Assessment schedule
Visits after a half year, one year, and thereafter one visit per year.
5. Primary Endpoint
| Endpoint | Registered definition | Time frame |
|---|---|---|
| Number of Participants With a 3-step or Greater Increase in EDTRS Severity Scale | Retinopathy progression was defined as the first occurrence of at least a 3-step increase in the ETDRS severity scale. Three steps were defined as either a 1-step change in one eye and a 2-step change in the other eye or as a 3-step change in one eye only. | From baseline to end of study, i.e. 5 years, with visits after a half year, one year and thereafter one visit per year. |
The registry describes EDTRS as a scale with 11 steps, numbered 1–11. The endpoint therefore does not simply count any movement on the scale. It identifies the first occurrence of a prespecified magnitude of worsening, using information from one or both eyes according to the registered definition.
6. Statistical Methodology
Intention-to-treat population
The posted primary analysis used the intention-to-treat population, defined in the registry as all randomized patients with any post-randomization data. Participants were therefore analyzed according to their randomized treatment assignment rather than being reclassified according to later treatment exposure.
Time-to-event analysis
The registry classifies the posted analysis as survival analysis. The endpoint was evaluated over follow-up, with the first qualifying 3-step or greater increase in the EDTRS severity scale treated as the event of interest.
Log-rank testing
The reported method was a generalized log-rank test. A log-rank framework compares the experience of two treatment groups across the follow-up period rather than reducing the entire endpoint to a single fixed-time proportion.
The reported generalized log-rank analysis evaluates the treatment groups across observed event times. The associated hazard ratio summarizes the relative event rate estimated from the time-to-event analysis.
Hazard ratio
The effect measure reported for the primary endpoint was the hazard ratio. The estimate was 1.024, with a two-sided 95% confidence interval of 0.800 to 1.312.
A hazard ratio near 1 indicates similar estimated instantaneous event rates between the groups under the fitted time-to-event analysis. An HR above 1 indicates a higher estimated hazard in the numerator group; an HR below 1 indicates a lower estimated hazard.
7. Primary Result
The registry reports one formal statistical analysis for the primary endpoint. The analysis compared candesartan with placebo in the intention-to-treat population using a generalized log-rank method.
Hazard ratio for a 3-step or greater EDTRS increase
95% CI: 0.800–1.312 · P = 0.8487
Analysis population: intention-to-treat population
Method: generalized log-rank test
| Primary endpoint | Candesartan vs placebo |
|---|---|
| Effect measure | Hazard ratio |
| Hazard ratio | 1.024 |
| 95% confidence interval | 0.800–1.312 |
| P-value | 0.8487 |
| Hypothesis | Superiority |
| Analysis method | Generalized log-rank test |
| Analysis population | Intention-to-treat: all randomized patients with any post-randomization data |
The estimated hazard ratio of 1.024 is very close to 1. In the context of this analysis, it corresponds to an estimated instantaneous rate of the qualifying retinopathy-progression event that was approximately 2.4% higher under candesartan than under placebo.
That interpretation is about the estimated relative hazard, not about the percentage of participants who progressed. A hazard ratio of 1.024 does not mean that 2.4% of participants experienced progression, nor does it mean that each individual participant had a 2.4% higher probability of progression.
The two-sided 95% confidence interval of 0.800 to 1.312 spans 1.0. Thus, the interval includes hazard-ratio values corresponding to lower, similar, and higher event hazards under candesartan relative to placebo. The width of the interval also indicates that the estimate has appreciable statistical uncertainty.
The reported P = 0.8487 is a measure of compatibility with the null hypothesis used for the statistical test; it is not a measure of the size or clinical importance of the treatment effect. A p-value should therefore be interpreted alongside the hazard ratio and its confidence interval.
Because the analysis is time-to-event based, interpretation also depends on the treatment-group event and censoring patterns and on assumptions underlying the hazard-based summary. The registry does not provide a separate assessment of the proportional-hazards assumption.
8. Understanding the Primary Endpoint Statistically
The endpoint combines two important ideas: a clinically defined threshold for retinopathy progression and the timing of the first occurrence of that threshold. This makes it different from a simple binary assessment performed at one fixed visit.
Event definition
An event occurs when the first qualifying 3-step or greater increase in the EDTRS severity scale is observed.
Timing
The analysis retains information about when the first qualifying event occurs during follow-up.
Multiple eyes
The registered definition permits the 3-step increase to arise from a combination of changes across both eyes or from one eye alone.
Censoring
Time-to-event methods can account for participants who do not experience the event during their observed follow-up.
9. Statistical Methods Explained
Why was a log-rank test used?
A log-rank test is designed for comparing time-to-event experience between treatment groups. Rather than asking only whether an event occurred by one selected calendar time, it incorporates the ordering of observed event times throughout follow-up. That is appropriate for a primary endpoint defined by the first occurrence of progression during a multi-year follow-up period.
What does a hazard ratio of 1.024 mean?
The hazard ratio compares the estimated instantaneous event rates between the two treatment groups within the time-to-event framework. An HR of 1.024 is close to 1, indicating that the estimated event hazards were similar under the fitted analysis. It does not represent a 2.4 percentage-point difference in cumulative incidence.
Why is the confidence interval important?
The 95% confidence interval of 0.800 to 1.312 shows the uncertainty surrounding the hazard-ratio estimate. Because it crosses 1, it includes both a lower estimated hazard and a higher estimated hazard for candesartan relative to placebo. The interval therefore provides information that cannot be conveyed by the point estimate alone.
What does the p-value of 0.8487 tell us?
The p-value quantifies the statistical evidence against the null hypothesis under the specified testing framework. A value of 0.8487 is not an estimate of treatment benefit, treatment harm, or the probability that the null hypothesis is true. It should be considered together with the estimated hazard ratio and confidence interval.
Why use an intention-to-treat population?
The posted analysis included all randomized patients with any post-randomization data and retained their randomized treatment assignment. This approach preserves the comparison created by randomization and avoids changing the treatment groups according to events that occur after randomization.
Why does the timing of progression matter?
Two participants could both eventually experience the same qualifying progression while having very different follow-up histories. A time-to-event analysis distinguishes these situations by incorporating the time until the first event and the time during which participants remain event-free.
What does the EDTRS definition contribute?
The registered endpoint uses a prespecified threshold rather than counting every change in severity. A qualifying event requires at least a 3-step increase, defined as either a 1-step change in one eye plus a 2-step change in the other eye or a 3-step change in one eye alone. This creates a reproducible event definition for the primary statistical comparison.
10. Confidence Interval Interpretation
The reported 95% confidence interval is 0.800 to 1.312. On the hazard-ratio scale, 1.0 represents equal estimated hazards between candesartan and placebo. Values below 1 favor a lower event hazard in the candesartan numerator group, while values above 1 represent a higher event hazard.
The interval therefore contains values on both sides of 1.0. Its lower boundary is compatible with an estimated hazard 20.0% lower than placebo, while its upper boundary is compatible with an estimated hazard 31.2% higher than placebo. These are interpretations of the limits of the estimated hazard ratio, not observed individual-patient effects.
11. P-value Interpretation
The reported p-value for the primary analysis was 0.8487. The statistical test was a generalized log-rank test for the treatment comparison.
| Quantity | What it describes | What it does not describe |
|---|---|---|
| Hazard ratio = 1.024 | Estimated relative event hazard under the time-to-event model | The proportion of participants who progressed or the absolute treatment difference |
| 95% CI = 0.800–1.312 | Statistical uncertainty around the estimated hazard ratio | The range of individual patient responses |
| P = 0.8487 | Evidence against the null hypothesis under the specified test | The probability that the treatment has no effect or the magnitude of the treatment effect |
These three quantities answer related but different questions. The hazard ratio describes the estimated relative effect, the confidence interval describes uncertainty around that estimate, and the p-value summarizes the evidence against the specified null hypothesis.
12. Safety
The registry reports serious adverse events by treatment arm. The number affected and the number at risk were reported for both groups.
| Treatment arm | Serious adverse events affected | At risk |
|---|---|---|
| Candesartan | 173 | 951 |
| Placebo | 161 | 951 |
Candesartan
Serious adverse events were reported in 173 of 951 participants in the registry's arm-level safety summary.
Placebo
Serious adverse events were reported in 161 of 951 participants in the corresponding registry summary.
The serious-adverse-event counts are descriptive safety results. They should not be interpreted as a formal comparative safety test because the registry information presented here does not provide a hazard ratio, risk ratio, confidence interval, or p-value for this safety endpoint.
13. Primary Analysis Population
The registry specifies that the primary analysis used the intention-to-treat population, defined as all randomized patients with any post-randomization data.
The principal statistical advantage of the intention-to-treat approach is that it preserves the treatment comparison created by randomization. It also avoids defining efficacy groups based on post-randomization treatment behavior.
The registry's definition is specifically tied to participants with any post-randomization data. That distinction matters because the analysis population is part of the interpretation of the reported hazard ratio and p-value.
14. Time-to-Event Analysis in DIRECT
The primary endpoint can be represented as a survival-analysis process. Each participant enters follow-up at baseline and is observed until the first qualifying retinopathy progression event or the end of the participant's observed follow-up.
The statistical analysis uses the timing of the first qualifying event rather than treating every assessment as an independent binary observation.
This framework is especially useful when participants have different amounts of observable follow-up. Someone who remains free of progression through a substantial period contributes information about remaining event-free during that period, even if a qualifying event is not ultimately observed.
15. What the Hazard Ratio Does — and Does Not — Mean
The primary hazard ratio was 1.024. Within the reported time-to-event analysis, this corresponds to an estimated event hazard under candesartan that was approximately 1.024 times the estimated hazard under placebo.
Expressed descriptively, 1.024 is approximately a 2.4% higher estimated hazard relative to placebo. This is a statement about the fitted hazard ratio, not about a 2.4% absolute increase in the number of participants with progression.
The HR does not mean that 1.024 times as many participants experienced progression. It also does not mean that an individual participant's probability of progression was 2.4% higher. Absolute risk and cumulative incidence require different quantities.
The 95% CI of 0.800–1.312 demonstrates that the estimate is uncertain. The interval includes hazard ratios below 1 and above 1, so the reported estimate should not be interpreted independently of its uncertainty.
The p-value of 0.8487 is a hypothesis-test quantity. It does not quantify how large or clinically important the treatment difference is. The hazard ratio and confidence interval are the appropriate reported quantities for describing the estimated relative effect and its precision.
16. Limitations and Interpretation Issues
- Time-to-event assumptions: the primary analysis uses a hazard-based framework. The registry does not report a separate assessment of the proportional-hazards assumption.
- Incomplete follow-up: time-to-event analysis requires appropriate handling of participants whose event status is not observed for the entire study period. The registry identifies the endpoint as a time-to-event analysis but does not provide additional censoring details in the posted statistical-analysis record.
- Single primary effect estimate: the registry posts one formal primary statistical analysis with one hazard-ratio estimate and confidence interval.
- Endpoint threshold: the primary outcome is based on a prespecified 3-step or greater change rather than any change in EDTRS severity.
- Two-eye structure: progression can be defined using changes in one eye or a combination of changes across both eyes, so the endpoint is not equivalent to a simple patient-level measurement from one eye.
- Safety interpretation: serious-adverse-event counts are reported by arm, but the posted information does not provide a formal comparative statistical analysis for those counts.
- Multiplicity: the available statistical-analysis record does not provide a multiplicity-adjustment framework for additional endpoints.
- Subgroups: the available statistical-analysis record does not provide subgroup-specific treatment-effect estimates for the primary endpoint.
17. Why This Trial Matters Statistically
DIRECT provides a useful teaching example of how a clinically defined binary progression endpoint can be analyzed as a time-to-event outcome. The statistical structure connects randomization, masking, intention-to-treat analysis, event definitions, log-rank testing, hazard ratios, and confidence intervals in a single primary analysis.
| Concept | How it appears in DIRECT |
|---|---|
| Randomization | The trial uses randomized allocation between candesartan and placebo. |
| Parallel design | The registry identifies a parallel-group design with 2 arms. |
| Blinding | The study is quadruple-masked. |
| Intention-to-treat analysis | The primary analysis population includes all randomized patients with any post-randomization data. |
| Time-to-event endpoint | The primary progression endpoint records the first qualifying event over follow-up. |
| Log-rank test | A generalized log-rank test was used for the primary treatment comparison. |
| Hazard ratio | The treatment effect was summarized with an HR of 1.024. |
| Confidence interval | The 95% CI was 0.800–1.312. |
| P-value | The reported p-value was 0.8487. |
| Superiority hypothesis | The registered hypothesis type was superiority. |
18. Statistical Interpretation of the Complete Primary Result
The primary statistical result can be read as a sequence rather than as a single number. First, randomization established the treatment comparison between candesartan and placebo. Second, the endpoint specified a clinically meaningful threshold for retinopathy progression. Third, follow-up incorporated the timing of the first qualifying event. Finally, the generalized log-rank framework compared the time-to-event experience between treatment groups.
Effect estimate
HR 1.024 indicates an estimated hazard close to that of the comparator under the reported analysis.
Precision
The 95% CI of 0.800–1.312 spans 1 and provides the principal measure of statistical uncertainty around the HR.
Hypothesis test
The generalized log-rank test produced P = 0.8487.
Analysis population
The posted analysis used the intention-to-treat population as defined in the registry.
These components should be interpreted together. The point estimate alone is insufficient, while the p-value alone does not describe the magnitude or precision of the treatment effect.
19. Endpoint Definition in Detail
The registered definition of retinopathy progression is based on the first occurrence of at least a 3-step increase in the ETDRS severity scale.
| Component | Registered definition |
|---|---|
| Scale | EDTRS severity scale with 11 steps (1–11) |
| Progression threshold | At least a 3-step increase |
| Two-eye definition | A 1-step change in one eye and a 2-step change in the other eye |
| One-eye definition | A 3-step change in one eye only |
| Event timing | First occurrence |
| Follow-up | Baseline to end of study, i.e. 5 years |
| Visits | After a half year, one year, and thereafter one visit per year |
This definition illustrates why endpoint specification is a statistical issue as well as a clinical one. Changing the progression threshold, the eye-level rule, or the timing rule would change which observations qualify as events and therefore change the statistical estimand.
20. Generalized Log-Rank Testing
The registry describes the primary test as a generalized log-rank test. The log-rank family is designed for comparing survival or event-time distributions between groups when observations may be censored.
The test accumulates information over event times rather than evaluating only a single fixed-time event proportion.
For DIRECT, this means the primary comparison uses the pattern of qualifying retinopathy progression over the study period. The resulting p-value of 0.8487 summarizes the statistical evidence from that time-to-event comparison under the specified testing framework.
21. Hazard Ratio and Clinical Meaning
A hazard ratio is inherently relative. It compares event rates between groups at the time scale used by the analysis. It is therefore useful for describing the relative treatment effect, but it does not by itself describe absolute event probability at a particular year.
| Question | Relevant quantity |
|---|---|
| How large is the estimated relative difference in event hazard? | Hazard ratio = 1.024 |
| How uncertain is that estimate? | 95% CI = 0.800–1.312 |
| What was the hypothesis-test result? | P = 0.8487 |
| Which participants were included in the posted efficacy analysis? | Intention-to-treat population |
| What statistical method was reported? | Generalized log-rank test |
This separation of questions is important for clinical-trial interpretation. A treatment effect should be described with its effect measure and uncertainty rather than reduced to a binary interpretation of statistical significance.
22. Safety Interpretation
The registry reports 173 serious-adverse-event-affected participants among 951 at risk in the candesartan arm and 161 among 951 at risk in the placebo arm.
The counts provide an arm-level description of serious adverse events. They show that serious adverse events were reported in both randomized treatment groups. Because the available record does not provide a comparative effect estimate or hypothesis-test result for this safety endpoint, the counts should be presented descriptively rather than converted into an unsupported formal treatment comparison.
Safety and efficacy also answer different statistical questions. The primary efficacy endpoint concerns time to a specified retinopathy progression event, whereas the serious-adverse-event summary describes affected participants within each treatment arm.
23. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary time-to-event analysis estimated an HR of 1.024, with a two-sided 95% CI of 0.800–1.312 and a generalized log-rank p-value of 0.8487.
Clinical interpretation
The primary endpoint was defined as the first occurrence of a 3-step or greater increase in EDTRS severity scale over approximately 5 years. The statistical result describes the treatment-group comparison for that prespecified endpoint.
The distinction is important because statistical interpretation describes what the estimated treatment effect and uncertainty mean mathematically, while clinical interpretation considers the endpoint itself and what constitutes a clinically meaningful change.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Calculators
26. Sources
- ClinicalTrials.gov: DIRECT — NCT00252720.
- PubMed: PubMed record 38682786.
- PubMed: PubMed record 21309844.
- PubMed: PubMed record 21225239.
- PubMed: PubMed record 19451554.
- PubMed: PubMed record 18823656.
Continue through the Clinical Biostats knowledge graph
Connect this trial's endpoint and statistical methods to deeper biostatistics tutorials and practical statistical tools.
27. Record Summary
DIRECT provides a clear example of a randomized clinical-trial endpoint that is clinically defined at the participant level but analyzed through a time-to-event framework. The primary endpoint was the first occurrence of a 3-step or greater increase in the EDTRS severity scale from baseline to the end of the study, with visits after a half year, one year, and thereafter one visit per year. The posted intention-to-treat analysis used a generalized log-rank test and reported a hazard ratio of 1.024, with a two-sided 95% confidence interval of 0.800–1.312 and a p-value of 0.8487.