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 cilexetil against placebo in type 1 diabetes. The registered primary endpoint assessed the number of participants with a 2-step or greater increase in the Early Treatment Diabetic Retinopathy Study (ETDRS) severity scale over 5 years.
| Feature | DIRECT |
|---|---|
| Trial name | Diabetic Retinopathy Candesartan Trials |
| Phase | Phase 3 |
| Condition | Type 1 Diabetes |
| Design | Randomized, parallel-group |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 5238 |
| Intervention | Candesartan cilexetil |
| Comparator | Placebo |
| Primary endpoint type | Registered as binary; analyzed as a time-to-event endpoint |
| Primary statistical method | Log-rank test, reported as generalized |
| Effect measure | Hazard ratio |
| Hypothesis | Superiority |
| Trial status | Completed |
| Start | 2001-06 |
| Primary completion | 2008-04 |
| Lead sponsor | AstraZeneca |
2. Clinical Question
The primary clinical question was whether treatment with candesartan cilexetil differed from placebo in the time to a clinically meaningful progression in diabetic retinopathy, defined in the registry as a 2-step or greater increase in the ETDRS severity scale.
Population
Participants with type 1 diabetes enrolled in the phase 3 DIRECT trial.
Intervention
Candesartan cilexetil.
Comparator
Placebo.
Primary question
Does candesartan differ from placebo in the time to a 2-step or greater increase in ETDRS severity scale?
3. Trial Design
Candesartan
- Candesartan cilexetil
- Randomized treatment assignment
- Primary comparison based on time to the registered retinopathy endpoint
Placebo
- Placebo
- Randomized treatment assignment
- Primary comparison based on time to the registered retinopathy endpoint
The design is important statistically because randomization establishes the treatment comparison before outcomes are observed, while quadruple masking reduces opportunities for knowledge of treatment assignment to influence trial conduct or assessment.
4. Primary Endpoint
| Endpoint | Registry definition | Analysis |
|---|---|---|
| 2-step or greater increase in ETDRS severity scale | Number of participants with a 2-step or greater increase in Early Treatment Diabetic Retinopathy Study severity scale. | Generalized log-rank test; hazard ratio as the effect measure. |
Time frame
The endpoint was assessed from baseline to the end of study, defined as 5 years, with visits after a half year, one year, and thereafter one visit per year.
What constituted a 2-step increase?
The registry defines two steps as either a 1-step change in each eye or a 2-step change in one eye only. The ETDRS scale has 11 steps, from 1 through 11, where a score of 1 represents no retinopathy and a score of 11 represents proliferative retinopathy.
5. Statistical Methodology
Intention-to-treat analysis
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 based on later treatment behavior.
Generalized log-rank test
The registry reports a generalized log-rank test for the difference between treatments. The log-rank framework is designed for comparing time-to-event distributions between randomized groups while accounting for the fact that not every participant necessarily experiences the event during follow-up.
Conceptually, the log-rank test compares the observed pattern of events with the pattern expected if the treatment groups had the same event-time distribution.
Hazard ratio
The effect measure reported for the primary analysis was the hazard ratio. A hazard ratio compares the estimated instantaneous event rates between treatment groups over the analyzed time-to-event follow-up.
A hazard ratio is a relative time-to-event measure. It is not an absolute risk difference, does not state the proportion of participants who experienced benefit, and does not mean that every participant experienced the same proportional change in risk.
Why time-to-event analysis is appropriate
The endpoint is evaluated over a 5-year period with repeated scheduled visits. Participants can therefore differ in when the endpoint is first observed, and some participants may complete observation without experiencing the endpoint. A time-to-event framework preserves information about event timing and accommodates censoring.
6. Primary Result
The posted primary analysis compared candesartan with placebo using a generalized log-rank method in the intention-to-treat population. The reported hazard ratio was 0.821, with a two-sided 95% confidence interval of 0.673 to 1.001 and a p-value of 0.0508.
Hazard ratio for the primary retinopathy endpoint
95% CI: 0.673–1.001 · P = 0.0508
Candesartan vs placebo · Generalized log-rank analysis · Intention-to-treat population
| Primary endpoint | Candesartan vs placebo |
|---|---|
| Effect measure | Hazard ratio |
| Hazard ratio | 0.821 |
| 95% confidence interval | 0.673–1.001 |
| P-value | 0.0508 |
| Hypothesis type | Superiority |
| Analysis method | Generalized log-rank test |
| Analysis population | Intention-to-treat population |
The estimated hazard ratio of 0.821 corresponds to an estimated instantaneous event rate approximately 17.9% lower with candesartan than with placebo under the fitted time-to-event comparison. This is a relative interpretation of the estimated hazard, not a statement that 17.9% of participants avoided retinopathy progression.
The result does not establish that every participant experienced a 17.9% reduction in risk, nor does it provide an absolute reduction in the probability of a 2-step ETDRS progression.
The two-sided 95% confidence interval of 0.673–1.001 describes uncertainty around the estimated hazard ratio. Because the interval extends slightly above 1, it includes the possibility of no difference under the conventional hazard-ratio reference point.
The p-value of 0.0508 measures the compatibility of the observed test statistic with the null hypothesis under the specified testing framework. It is not a measure of the magnitude of the treatment effect, the probability that the treatment works, or the probability that the null hypothesis is true.
The interpretation also depends on the time-to-event model and censoring framework. In particular, a single hazard ratio is most straightforward to interpret when the relative hazards are reasonably stable over time. The registry does not provide additional proportional-hazards diagnostics in the posted analysis.
7. Reading the Confidence Interval
The confidence interval is particularly informative for this result because it sits close to the conventional null value of 1.00. The point estimate is below 1, while the upper confidence limit is 1.001.
Point estimate
The hazard ratio of 0.821 is the single estimated relative treatment effect from the posted analysis.
Lower confidence limit
The lower limit of 0.673 represents a materially smaller estimated hazard than the point estimate, illustrating the uncertainty surrounding the estimate.
Upper confidence limit
The upper limit of 1.001 is essentially at the conventional null value of 1, so the interval does not exclude no difference.
Precision
The interval gives a more complete description of uncertainty than the point estimate alone. It should be considered alongside the study design and endpoint definition.
A confidence interval should not be interpreted as a range in which the true effect has a fixed probability of lying for this particular realized interval. Rather, it is produced by a procedure designed to achieve the stated coverage probability across repeated applications under its statistical assumptions.
8. What the P-value Does — and Does Not — Mean
The primary p-value was 0.0508. The analysis was specified as a superiority comparison and used a two-sided confidence interval. The p-value should therefore be interpreted within that prespecified hypothesis-testing framework.
The hazard ratio communicates the estimated relative treatment effect; the confidence interval communicates uncertainty around that estimate; and the p-value describes evidence against the null hypothesis under the specified statistical model.
Because the reported p-value is close to 0.05, the confidence interval is especially useful for interpretation. The interval communicates that the estimate is compatible with a range of effects that includes the conventional no-difference value.
9. Safety Results
The registry reports serious adverse events by treatment arm using affected participants and the corresponding reported at-risk counts.
| Serious adverse event measure | Affected / at risk |
|---|---|
| Candesartan | 102 / 710 |
| Placebo | 112 / 710 |
These serious-adverse-event figures are separate from the primary efficacy analysis. The efficacy comparison is based on the randomized intention-to-treat population, whereas the registry presents serious adverse events using the reported affected and at-risk counts shown above.
10. Statistical Methods Explained
Why was a log-rank test used?
The primary endpoint was observed over time rather than at a single fixed assessment. A log-rank test is designed to compare the timing of events between treatment groups while incorporating follow-up information and censoring. That makes it appropriate for a primary endpoint represented as time to progression in the ETDRS severity scale.
What does a hazard ratio of 0.821 mean?
A hazard ratio of 0.821 means that the estimated instantaneous event rate for the candesartan group was approximately 82.1% of that for the placebo group under the reported analysis. Equivalently, 1 minus 0.821 is 0.179, so the point estimate corresponds to an approximately 17.9% lower estimated hazard.
This does not mean that the probability of progression was exactly 17.9% lower, because a hazard ratio and an absolute event probability are different quantities.
Why does the confidence interval extend to 1.001?
The confidence interval of 0.673–1.001 reflects uncertainty around the estimated hazard ratio. A hazard ratio of 1 represents equal instantaneous event rates between groups. Because the upper confidence limit is 1.001, the interval reaches slightly above that reference value.
Why is the endpoint described as binary but analyzed as time-to-event?
The registered outcome counts participants who experience a defined event: a 2-step or greater increase in ETDRS severity. That makes the eventual outcome naturally describable as an event/no-event measure. The posted statistical analysis adds the timing of that event, treating it as a time-to-event endpoint and using a generalized log-rank test.
Why does intention-to-treat analysis matter?
The posted analysis includes all randomized patients with any post-randomization data and preserves their randomized treatment assignment. This maintains the treatment comparison created by randomization and avoids redefining the primary comparison based on subsequent treatment behavior.
What does the ETDRS definition add to the analysis?
The endpoint is not simply a nonspecific clinical deterioration. It specifies a 2-step or greater increase on an 11-step severity scale, with two steps defined as either a 1-step change in each eye or a 2-step change in one eye. That explicit definition establishes what constitutes an event before the statistical comparison is made.
What should be concluded from a p-value of 0.0508?
The p-value provides evidence against the null hypothesis according to the stated testing framework, but it should not be treated as a binary label that summarizes the entire trial. The hazard ratio, its confidence interval, the endpoint definition, the intention-to-treat population, and the time-to-event analysis all contribute to the statistical interpretation.
11. Why the Time-to-Event Framework Matters
The trial followed the primary endpoint from baseline through 5 years, with scheduled assessments after a half year, one year, and annually thereafter. This creates a longitudinal structure in which both whether and when an endpoint occurs can contain information.
| Component | Statistical role |
|---|---|
| Event | A 2-step or greater increase in ETDRS severity scale. |
| Time origin | Baseline. |
| Follow-up window | Baseline to end of study, defined as 5 years. |
| Scheduled assessment | After a half year, one year, and thereafter one visit per year. |
| Comparison | Candesartan versus placebo. |
| Statistical test | Generalized log-rank test. |
| Effect measure | Hazard ratio. |
This framework is different from simply comparing the percentage of participants who had progressed by the end of 5 years. A fixed-time binary analysis discards information about when the event occurred, while a time-to-event analysis incorporates the ordering and timing of events as well as censored follow-up.
12. Censoring and Follow-Up
Time-to-event methods are designed for settings in which some participants may not experience the event during observed follow-up. Such participants can still contribute information through the period during which their event status is observed.
Censoring does not mean that a participant is treated as having the event at the time observation stops. Instead, the participant contributes follow-up information up to the point at which the event status is no longer observed under the analysis framework.
The validity of a time-to-event analysis depends on assumptions about the censoring mechanism and the statistical model. The registry identifies the log-rank method and intention-to-treat population but does not provide additional posted detail about censoring assumptions or imputation procedures.
13. Statistical Interpretation of the Primary Result
The hazard ratio of 0.821 is below 1, so the point estimate favors a lower estimated instantaneous rate of the primary retinopathy event in the candesartan group relative to placebo.
The registry does not provide an absolute 5-year event probability or an absolute risk difference in the posted statistical analysis. Therefore, the hazard ratio should not be translated into an absolute number of events prevented.
The 95% confidence interval of 0.673–1.001 is important because it indicates substantial uncertainty around the point estimate and reaches the conventional no-difference value.
The two-sided p-value of 0.0508 is close to the conventional 0.05 threshold. The appropriate interpretation is the full statistical result rather than reducing the trial to a threshold-based label.
The posted analysis does not provide additional estimates that would establish the absolute probability of progression, median time to progression, subgroup-specific treatment effects, or detailed model diagnostics.
14. Limitations and Interpretation Issues
- Endpoint timing: the primary endpoint is defined through scheduled visits, while the posted analysis uses a time-to-event framework. The exact handling of event timing between visits is not further described in the posted analysis.
- Confidence interval: the upper confidence limit of 1.001 means the reported 95% interval reaches the conventional null value for a hazard ratio.
- Proportional hazards: the registry reports a hazard ratio but does not provide posted diagnostics establishing whether proportional-hazards assumptions are satisfied over the complete follow-up period.
- Censoring: the registry identifies a survival-analysis method but does not provide additional posted detail about censoring assumptions or imputation procedures.
- Subgroups: no subgroup-specific primary endpoint estimates are included in the posted statistical analysis.
- Multiplicity: the posted analysis does not provide additional information about multiplicity adjustments beyond identifying the primary hypothesis as superiority.
- Safety analysis: serious adverse events are reported by affected and at-risk counts, but the posted statistical analysis does not provide a comparative hypothesis test or effect estimate for those safety data.
15. Why This Trial Matters Statistically
DIRECT is a useful teaching case because the primary endpoint combines a clinically defined ordinal-scale change with a time-to-event statistical framework. The trial therefore illustrates why endpoint definition, follow-up structure, analysis population, test statistic, effect measure, confidence interval, and p-value need to be interpreted together.
| Concept | How it appears in DIRECT |
|---|---|
| Randomization | Participants were randomized in a parallel-group design. |
| Blinding | The trial used quadruple masking. |
| Intention-to-treat analysis | The posted primary analysis included all randomized patients with any post-randomization data. |
| Time-to-event endpoint | The primary retinopathy progression event was analyzed over 5 years. |
| Log-rank test | The treatment groups were compared using a generalized log-rank test. |
| Hazard ratio | The primary treatment effect was reported as HR 0.821. |
| Confidence interval | The 95% CI was 0.673–1.001. |
| P-value | The two-sided p-value was 0.0508. |
| Endpoint definition | A 2-step increase meant a 1-step change in each eye or a 2-step change in one eye. |
| Longitudinal assessment | Visits occurred after a half year, one year, and thereafter annually through the 5-year endpoint window. |
16. Related Tutorials
Learn more about the methods used in this trial:
17. Related Statistical Calculators
18. Sources
- ClinicalTrials.gov: DIRECT, NCT00252733.
- PubMed: PMID 38682786.
- PubMed: PMID 21309844.
- PubMed: PMID 21225239.
- PubMed: PMID 19451554.
- PubMed: PMID 18823656.
Continue with Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical-trial analyses covering study design, endpoints, survival analysis, and statistical inference.
19. Record Summary
DIRECT provides a clear example of how a clinical endpoint defined as progression on an ordinal disease-severity scale can be analyzed as a time-to-event outcome. The randomized, parallel-group, quadruple-masked design compared candesartan cilexetil with placebo, while the primary analysis used the intention-to-treat population and a generalized log-rank test.
The primary result was a hazard ratio of 0.821, with a two-sided 95% confidence interval of 0.673–1.001 and a p-value of 0.0508. The point estimate corresponds to an approximately 17.9% lower estimated instantaneous event rate with candesartan relative to placebo, but the confidence interval reaches the conventional null value. The result therefore illustrates why the hazard ratio, confidence interval, p-value, endpoint definition, and time-to-event framework should be interpreted together rather than reducing the analysis to a single numerical threshold.