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
RE-MODEL was a completed, randomized, double-blind, parallel phase 3 trial evaluating dabigatran etexilate 150mg or 220mg once daily versus enoxaparin 40mg once daily for prevention of thrombosis after knee surgery.
| Feature | RE-MODEL |
|---|---|
| Trial name | RE-MODEL |
| Phase | Phase 3 |
| Condition | Arthroplasty, Replacement, Knee; Thromboembolism |
| Primary purpose | Prevention |
| Design | Randomized, double-blind, parallel |
| Enrollment | 2101.0 |
| Arms | 3 |
| Primary endpoint type | Binary |
| Primary endpoint analyses posted | 2 |
| Results posted | Yes |
| ClinicalTrials.gov | NCT00168805 |
| Lead sponsor | Boehringer Ingelheim |
2. Clinical Question
The principal statistical question was whether either dabigatran etexilate dose could be shown to be non-inferior to enoxaparin for the registered composite primary endpoint of total venous thromboembolic events and all-cause mortality during the treatment period.
Population
Participants enrolled in a phase 3 prevention trial following knee replacement surgery for the conditions registered as arthroplasty, replacement, knee and thromboembolism.
Intervention
Dabigatran etexilate 220mg once daily or dabigatran etexilate 150mg once daily.
Comparator
Enoxaparin 40mg once daily.
Primary question
Is the absolute difference in the primary event rate for either dabigatran dose sufficiently small to satisfy the prespecified 9.2% non-inferiority margin?
3. Trial Design
Dabigatran etexilate 220mg
- Dabigatran etexilate
- 220mg once daily
- Evaluated against enoxaparin 40mg once daily
- Primary comparison used the Full Analysis Set
Dabigatran etexilate 150mg
- Dabigatran etexilate
- 150mg once daily
- Evaluated against enoxaparin 40mg once daily
- Primary comparison used the Full Analysis Set
Enoxaparin
- Enoxaparin 40mg once daily
- Reference group for the posted statistical comparisons
- Used for both primary non-inferiority comparisons
Randomized prevention trial
- Randomized allocation
- Parallel design
- Double masking
- Primary purpose: prevention
4. Endpoints
| Endpoint | Registry definition / assessment | Time frame |
|---|---|---|
| Primary | Number of Participants With Total Venous Thromboembolic Event and All-cause Mortality During Treatment Period. Total VTE includes both proximal and distal DVT detected by routine bilateral venography, symptomatic DVT confirmed by venous compression ultrasound, venography or autopsy, and PE confirmed by pulmonary V-Q scintigraphy, chest x-ray, pulmonary angiography, spiral CT or autopsy, together with all deaths. | First administration until 6-10 days |
| Secondary | Number of Participants With Major Venous Thromboembolic Event and Venous Thromboembolic Event-related Mortality During Treatment Period | First administration until 6-10 days |
| Secondary | Number of Participants With Proximal Deep Vein Thrombosis During Treatment Period | First administration until 6-10 days |
| Secondary | Number of Participants With Total Deep Vein Thrombosis During Treatment Period | First administration until 6-10 days |
| Secondary | Number of Participants With Symptomatic Deep Vein Thrombosis During Treatment Period | First administration until 6-10 days |
| Secondary | Number of Participants With Pulmonary Embolism During Treatment Period | First administration until 6-10 days |
| Secondary | Number of Participants Who Died During Treatment Period | First administration until 6-10 days |
| Secondary | Number of Participants With Bleeding Events (Defined According to Modified McMaster Criteria) During Treatment Period | First administration until 6-10 days |
The primary endpoint is binary: each participant is classified according to whether the specified total venous thromboembolic event or all-cause mortality occurred during the treatment period. The ClinicalTrials.gov record does not provide the component event counts by treatment arm, so the posted effect estimates are interpreted as risk differences rather than reconstructed from unreported event totals.
5. Analysis Populations
| Population | Registry role |
|---|---|
| Full Analysis Set | Primary efficacy population. For the primary endpoint, the registry describes this as all patients who had surgery and were randomised, received treatment, had an evaluable venogram for distal and proximal Deep Vein Thrombosis, or had confirmed symptomatic Deep Vein thrombosis. |
| Full Analysis Set - major | Used for the major venous thromboembolic event and venous thromboembolic event-related mortality endpoint; the registry specifies an evaluable venogram for proximal DVT or confirmed symptomatic DVT. |
| Full Analysis Set - pDVT | Used for proximal DVT; includes patients who had surgery and were randomised, received treatment, and had an evaluable venogram for proximal DVT or confirmed symptomatic DVT. |
| Full Analysis Set - tDVT | Used for total DVT; includes patients who had surgery and were randomised, received treatment, and had an evaluable venogram or confirmed symptomatic DVT. |
| Full Analysis Set - op | Used for symptomatic DVT, pulmonary embolism, and death; defined in the registry as patients who are treated and operated. |
| Treated set | Used for the bleeding-event endpoint. |
This distinction matters because the denominator can differ between an efficacy endpoint requiring an evaluable venogram and a safety endpoint evaluated in the treated set. A risk difference is meaningful only in the population and endpoint definition for which it was calculated.
6. Statistical Methodology
Risk difference
The primary efficacy comparisons use the risk difference, expressed as a percentage. Conceptually, if \(p_D\) denotes the event probability in a dabigatran group and \(p_E\) the event probability in the enoxaparin group, the effect is:
A negative value indicates a lower observed event risk in the dabigatran group than in the enoxaparin reference group; a positive value indicates a higher observed event risk.
The posted primary estimates are therefore absolute differences in event probability, not relative risks, odds ratios, or hazard ratios.
Normal approximation / Wald or z-test
For both primary comparisons, the registry reports a normal approximation, normalized here as a Wald / z-test. The analysis notes describe a normal approximation of the independent binomial distribution without stratification.
The Wald framework uses an estimated standard error to quantify how far the observed risk difference is from the value specified by the null hypothesis.
Fisher exact test
The registry uses Fisher exact tests for several secondary binary endpoints, including symptomatic DVT, pulmonary embolism, death, and bleeding events. Fisher's exact test is particularly useful for two-group comparisons of categorical outcomes when event counts can be small, because its calculation does not rely on the large-sample approximation required by a conventional chi-square test.
Intention-to-treat analysis and related registry terminology
The posted primary analyses identify intention-to-treat analysis among the other concepts in the analysis text. At the same time, the registry explicitly defines the primary Full Analysis Set using surgery, treatment, and evaluable venogram or confirmed symptomatic DVT criteria. Therefore, the most precise description is the registry's own Full Analysis Set definition rather than assuming that every randomized participant necessarily contributed to every primary binary endpoint.
Stratified analysis terminology
The primary analysis records also identify stratified analysis among the other concepts in the analysis text. However, the specific analysis notes state: "Risk difference versus Enoxaparin Normal approximation of independent binomial distribution without stratification." These two pieces of registry metadata should not be silently reconciled by inventing a stratification procedure. The reported numerical analysis is therefore described here according to the explicit analysis note: an independent-binomial normal approximation without stratification.
7. Non-Inferiority Framework
The defining statistical feature of the primary analysis is the 9.2% non-inferiority margin. The registry states that the absolute risk difference has to be below 9.2% and explicitly notes that non-inferiority can only be shown with the confidence interval.
Prespecified non-inferiority margin
Absolute risk-difference margin used for the primary non-inferiority analyses.
The upper bound of the 95% confidence interval is the critical quantity for the stated non-inferiority logic.
With the risk difference defined as dabigatran minus enoxaparin, the relevant question is whether the upper confidence limit remains below +9.2 percentage points. The margin therefore defines how much higher the dabigatran event risk could plausibly be while still satisfying the prespecified non-inferiority criterion.
220mg comparison
The 95% confidence interval extends from -7.3% to 4.6%. Its upper limit, 4.6%, is below the 9.2% margin.
150mg comparison
The 95% confidence interval extends from -3.1% to 8.7%. Its upper limit, 8.7%, is below the 9.2% margin.
8. Primary Results
Primary endpoint: dabigatran 220mg vs enoxaparin
Risk difference
95% CI: -7.3% to 4.6% · P = 0.6648
Hypothesis type: Non-inferiority or equivalence
| Primary endpoint | Dabigatran 220mg vs enoxaparin |
|---|---|
| Effect measure | Risk Difference (Percentage) |
| Estimate | -1.3 |
| 95% CI | -7.3 to 4.6 |
| P-value | 0.6648 |
| Method | Normal approximation; Wald / z-test |
| Analysis population | Full Analysis Set |
| Non-inferiority margin | 9.2% |
The estimated risk difference of -1.3% means that the estimated absolute event risk for the dabigatran 220mg group was 1.3 percentage points lower than the enoxaparin group for the registered primary endpoint. Because the effect is a risk difference, it describes an absolute difference rather than a relative percentage reduction.
The estimate alone does not establish that dabigatran 220mg is better than enoxaparin. The 95% confidence interval runs from -7.3% to 4.6%, so the data are compatible with a range extending from a lower event risk to a higher event risk for dabigatran. Importantly for the non-inferiority question, the upper confidence limit of 4.6% remains below the prespecified 9.2% margin.
The reported P = 0.6648 should not be interpreted as the probability that the treatments are equal or as a measure of the size of the treatment effect. For this non-inferiority analysis, the registry specifically states that non-inferiority can only be shown with the confidence interval. The p-value therefore does not replace the margin-based confidence-interval assessment.
The analysis is based on the registry-defined Full Analysis Set and the reported normal approximation. The registry analysis note specifies an independent-binomial calculation without stratification, even though stratified analysis is listed among the other concepts in the analysis metadata.
Primary endpoint: dabigatran 150mg vs enoxaparin
Risk difference
95% CI: -3.1% to 8.7% · P = 0.3553
Hypothesis type: Non-inferiority or equivalence
| Primary endpoint | Dabigatran 150mg vs enoxaparin |
|---|---|
| Effect measure | Risk Difference (Percentage) |
| Estimate | 2.8 |
| 95% CI | -3.1 to 8.7 |
| P-value | 0.3553 |
| Method | Normal approximation; Wald / z-test |
| Analysis population | Full Analysis Set |
| Non-inferiority margin | 9.2% |
The estimated risk difference of 2.8% means that the estimated absolute event risk was 2.8 percentage points higher with dabigatran 150mg than with enoxaparin for the registered primary endpoint.
That estimate should not be read as evidence that the true difference is exactly 2.8 percentage points. The 95% confidence interval ranges from -3.1% to 8.7%, reflecting uncertainty in both directions. Zero lies within the interval, so the interval is compatible with either a lower or higher event risk for dabigatran.
For the prespecified non-inferiority question, however, the important comparison is between the upper confidence limit of 8.7% and the 9.2% non-inferiority margin. The upper limit remains below the margin specified by the registry.
The reported P = 0.3553 is not an effect-size measure and does not by itself answer the non-inferiority question. The registry explicitly states that non-inferiority can only be shown with the confidence interval. The relevant interpretation therefore comes from the relationship between the confidence interval and the prespecified 9.2% margin.
9. Primary Results Side by Side
| Comparison | Risk difference | 95% CI | P-value | NI margin | Upper CI vs margin |
|---|---|---|---|---|---|
| Dabigatran 220mg vs enoxaparin | -1.3% | -7.3% to 4.6% | 0.6648 | 9.2% | 4.6% < 9.2% |
| Dabigatran 150mg vs enoxaparin | 2.8% | -3.1% to 8.7% | 0.3553 | 9.2% | 8.7% < 9.2% |
The two comparisons illustrate why non-inferiority analysis cannot be reduced to asking whether a confidence interval contains zero. Both intervals contain zero, but both upper bounds remain below the prespecified 9.2% margin. The registry's stated non-inferiority logic is therefore margin-based rather than a conventional superiority test against a null risk difference of zero.
10. Secondary Endpoint Results
Major venous thromboembolic event and VTE-related mortality
| Comparison | Risk difference | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Dabigatran 220mg vs enoxaparin | -1.0% | -3.1% to 1.2% | 0.3760 | Superiority |
| Dabigatran 150mg vs enoxaparin | 0.3% | -2.0% to 2.6% | 0.8151 | Superiority |
Both analyses used the Full Analysis Set - major and a normal approximation of the independent binomial distribution without stratification. The confidence intervals for both comparisons include zero, and the reported p-values are 0.3760 and 0.8151, respectively. These are superiority analyses, so the inferential question differs from the primary non-inferiority analysis: the null comparison is not evaluated against the 9.2% non-inferiority margin.
Proximal deep vein thrombosis
| Comparison | Risk difference | 95% CI | P-value | Method |
|---|---|---|---|---|
| Dabigatran 220mg vs enoxaparin | -0.8% | -2.8% to 1.3% | 0.4715 | Wald / z-test |
| Dabigatran 150mg vs enoxaparin | 0.1% | -2.1% to 2.3% | 0.9325 | Wald / z-test |
These analyses used the Full Analysis Set - pDVT. The estimated risk differences are close to zero relative to the widths of their confidence intervals. Neither comparison has a confidence interval that excludes zero.
Total deep vein thrombosis
| Comparison | Risk difference | 95% CI | P-value | Method |
|---|---|---|---|---|
| Dabigatran 220mg vs enoxaparin | -1.4% | -7.3% to 4.5% | 0.6463 | Wald / z-test |
| Dabigatran 150mg vs enoxaparin | 2.7% | -3.2% to 8.6% | 0.3740 | Wald / z-test |
The total-DVT analyses used the Full Analysis Set - tDVT. As with the other superiority analyses, the reported p-values are hypothesis-test outputs rather than measures of the magnitude or clinical importance of the observed risk differences.
Symptomatic deep vein thrombosis
| Comparison | Method | 95% CI | P-value | Analysis population |
|---|---|---|---|---|
| Dabigatran 220mg vs enoxaparin | Fisher exact test | 95% | 0.0385 | Full Analysis Set - op |
| Dabigatran 150mg vs enoxaparin | Fisher exact test | 95% | 0.1414 | Full Analysis Set - op |
The registry reports Fisher exact tests for these binary comparisons. It does not provide a risk-difference estimate in the ClinicalTrials.gov record for symptomatic DVT, so no effect estimate should be inferred from the p-values.
Pulmonary embolism
| Comparison | Method | 95% CI | P-value |
|---|---|---|---|
| Dabigatran 220mg vs enoxaparin | Fisher exact test | 95% | 1.0000 |
| Dabigatran 150mg vs enoxaparin | Fisher exact test | 95% | 1.0000 |
Both pulmonary-embolism comparisons were analyzed with Fisher exact tests in the Full Analysis Set - op. The ClinicalTrials.gov record contains no event counts or effect estimates for these comparisons.
Death during the treatment period
| Comparison | Method | 95% CI | P-value |
|---|---|---|---|
| Dabigatran 220mg vs enoxaparin | Fisher exact test | 95% | 1.0000 |
| Dabigatran 150mg vs enoxaparin | Fisher exact test | 95% | 1.0000 |
These analyses concern the registered endpoint "Number of Participants Who Died During Treatment Period" and use the Full Analysis Set - op. Again, the ClinicalTrials.gov record provides the p-values but not event counts or risk-difference estimates.
11. Safety Results
The ClinicalTrials.gov record provides serious adverse-event counts by treatment arm. These counts should be kept separate from the efficacy endpoint analyses because the safety denominator and endpoint definition are different.
| Arm | Serious adverse events affected | At risk |
|---|---|---|
| Dabigatran 220mg | 31 | 679 |
| Dabigatran 150mg | 44 | 703 |
| Enoxaparin | 43 | 694 |
The registry also posted bleeding-event analyses using the Treated set and Fisher exact tests. For dabigatran 220mg versus enoxaparin, the reported p-value was 0.8209; for dabigatran 150mg versus enoxaparin, it was 1.0000. The ClinicalTrials.gov record does not provide bleeding-event counts or risk-difference estimates, so the magnitude of the between-group difference cannot be reconstructed from the ClinicalTrials.gov record.
| Safety endpoint | Comparison | Method | P-value | Analysis population |
|---|---|---|---|---|
| Bleeding Events, defined according to Modified McMaster Criteria | Dabigatran 220mg vs enoxaparin | Fisher exact test | 0.8209 | Treated set |
| Bleeding Events, defined according to Modified McMaster Criteria | Dabigatran 150mg vs enoxaparin | Fisher exact test | 1.0000 | Treated set |
12. Secondary Analysis Summary
| Endpoint | 220mg vs enoxaparin | 150mg vs enoxaparin |
|---|---|---|
| Major VTE + VTE-related mortality | RD -1.0%; 95% CI -3.1% to 1.2%; P = 0.3760 | RD 0.3%; 95% CI -2.0% to 2.6%; P = 0.8151 |
| Proximal DVT | RD -0.8%; 95% CI -2.8% to 1.3%; P = 0.4715 | RD 0.1%; 95% CI -2.1% to 2.3%; P = 0.9325 |
| Total DVT | RD -1.4%; 95% CI -7.3% to 4.5%; P = 0.6463 | RD 2.7%; 95% CI -3.2% to 8.6%; P = 0.3740 |
| Symptomatic DVT | Fisher exact; P = 0.0385 | Fisher exact; P = 0.1414 |
| Pulmonary embolism | Fisher exact; P = 1.0000 | Fisher exact; P = 1.0000 |
| Death | Fisher exact; P = 1.0000 | Fisher exact; P = 1.0000 |
| Bleeding events | Fisher exact; P = 0.8209 | Fisher exact; P = 1.0000 |
This table is best read as a map of the analyses rather than as a single composite statistical conclusion. The endpoints use different analysis populations, and some provide risk differences while others provide only Fisher exact p-values in the ClinicalTrials.gov record.
13. Statistical Methods Explained
Why was a risk difference used for the primary endpoint?
The primary endpoint is binary, so each participant either did or did not experience the registered total VTE or all-cause mortality endpoint during the treatment period. A risk difference expresses the absolute separation between the two event probabilities. This is particularly natural for a non-inferiority framework because the prespecified margin itself is expressed as an absolute percentage-point difference.
Why is the 9.2% margin more important than the p-value for non-inferiority?
Non-inferiority asks whether the treatment is sufficiently close to the comparator, not simply whether the observed difference differs from zero. The registry specifies a 9.2% absolute margin and states that non-inferiority can only be shown with the confidence interval. Thus, the upper confidence limit is compared with 9.2% rather than treating a conventional p-value against zero as the decisive criterion.
What does a risk difference of -1.3% mean?
A risk difference of -1.3% means the estimated absolute event probability was 1.3 percentage points lower in the dabigatran 220mg group than in the enoxaparin group. It does not mean a 1.3% relative reduction, and it does not describe the outcome of every individual participant.
Why can a confidence interval include zero and still support non-inferiority?
Zero represents identical event risks, but identical risks are not the only possibility compatible with non-inferiority. For the 220mg comparison, the interval extends from -7.3% to 4.6%. Although zero is included, the entire upper end remains below the 9.2% margin. The same logic applies to the 150mg comparison, whose upper confidence limit is 8.7%.
Why was Fisher exact testing used for some secondary endpoints?
Fisher exact testing is an exact method for comparing binary outcomes between two groups. In the registry-reported RE-MODEL analyses it was used for symptomatic DVT, pulmonary embolism, death, and bleeding events. The registry provides p-values for these analyses but does not provide corresponding effect estimates for the ClinicalTrials.gov record.
Why is the analysis population important?
The primary endpoint uses a registry-defined Full Analysis Set, while other endpoints use specialized populations such as the Full Analysis Set - pDVT, Full Analysis Set - tDVT, Full Analysis Set - op, or the Treated set. These definitions determine which participants contribute to each analysis. A result from one population should not be transferred to another population without justification.
What does the reported p-value tell us?
A p-value quantifies how unusual the observed data would be under the statistical null hypothesis used for that particular test and analysis. It does not measure effect size, clinical importance, or the probability that the null hypothesis is true. This distinction is especially important in RE-MODEL because the primary analyses are non-inferiority analyses whereas many secondary analyses are superiority tests.
14. Confidence Intervals and Precision
The primary confidence intervals provide more information than the point estimates alone. They describe the statistical uncertainty around the estimated risk differences and make the non-inferiority decision interpretable in relation to the 9.2% margin.
| Comparison | Point estimate | 95% CI width | Upper confidence limit |
|---|---|---|---|
| Dabigatran 220mg vs enoxaparin | -1.3% | 11.9 percentage points | 4.6% |
| Dabigatran 150mg vs enoxaparin | 2.8% | 11.8 percentage points | 8.7% |
The confidence intervals are fairly broad relative to the point estimates. For the 220mg comparison, the interval includes both a potentially meaningful reduction and an increase in event risk, while for the 150mg comparison it likewise extends from a negative risk difference to a positive one. The non-inferiority conclusion is therefore driven not by a narrow interval around zero, but by the fact that the upper limits remain below the prespecified margin.
A 95% confidence interval describes uncertainty about the population-level effect estimate under the statistical framework used for the analysis. It does not mean that 95% of individual participants experienced effects inside that interval, nor does it provide a range of individual treatment responses.
15. Non-Inferiority vs Superiority
| Feature | Primary analyses | Many secondary analyses |
|---|---|---|
| Hypothesis type | Non-inferiority or equivalence | Superiority |
| Main effect measure | Risk difference | Risk difference for several endpoints; p-value only in some posted analyses |
| Key reference | 9.2% non-inferiority margin | Null hypothesis for superiority comparison |
| Critical interpretation | Upper 95% CI below 9.2% | Magnitude, confidence interval, and p-value considered under the stated superiority framework |
This distinction prevents a common statistical error: applying a superiority interpretation to a non-inferiority result. A primary risk difference of 2.8% with a p-value of 0.3553 does not by itself answer whether dabigatran is non-inferior. The relevant question is whether the confidence interval remains within the prespecified non-inferiority boundary.
16. Multiplicity and Multiple Comparisons
The trial has two posted primary comparisons involving the same primary endpoint: dabigatran 220mg versus enoxaparin and dabigatran 150mg versus enoxaparin. The ClinicalTrials.gov record identifies the hypothesis type and non-inferiority margin for each comparison, but they do not provide a multiplicity-adjustment procedure or alpha-allocation scheme for these two primary comparisons.
The same caution applies to the numerous secondary endpoints. A collection of p-values across many outcomes should not automatically be interpreted as a collection of independent confirmatory tests. The ClinicalTrials.gov record does not provide an adjustment strategy for the secondary endpoint family.
17. Randomization and Blinding
Randomization
The registry classifies allocation as randomized. Randomization creates the intended basis for comparing treatment groups while reducing systematic differences in treatment assignment.
Double masking
The registry classifies masking as double. Blinding can reduce the influence of treatment knowledge on participant behavior, clinical management, outcome assessment, and other aspects of trial conduct.
Parallel design
The registry classifies the design model as parallel. The ClinicalTrials.gov record does not describe a crossover design or a factorial structure.
Prevention objective
The primary purpose is registered as prevention, with thrombosis-related binary outcomes assessed during the treatment period.
Randomization and double masking are design features rather than statistical tests. Their value is that they support a fair comparison between treatment groups, after which the prespecified statistical analysis translates the observed binary outcomes into estimates and uncertainty measures.
18. What the Primary Risk Differences Do — and Do Not — Mean
The risk difference of -1.3% is an absolute treatment-group difference in the probability of the primary endpoint. It does not mean that dabigatran reduced every participant's risk by 1.3%, nor does it represent a relative risk reduction.
The risk difference of 2.8% is an estimated absolute increase in event risk relative to enoxaparin for the primary endpoint. It does not mean that every participant had a 2.8 percentage-point increase in risk.
The 95% CI of -7.3% to 4.6% for 220mg and -3.1% to 8.7% for 150mg quantify uncertainty around the corresponding risk differences. They are also the key quantities for evaluating the 9.2% non-inferiority margin.
The primary p-values, 0.6648 and 0.3553, are outputs of the reported statistical tests. Neither p-value tells the reader how large the treatment difference is. The effect size comes from the risk difference, while its uncertainty is described by the confidence interval.
19. Limitations and Interpretation Issues
- Incomplete event-count detail: the registry statistical-analysis data provide effect estimates for the primary and several secondary endpoints, but do not provide the underlying event counts for those comparisons. Event counts should therefore not be reconstructed from the risk differences.
- Different analysis populations: primary, thromboembolic, symptomatic-event, death, and bleeding analyses use different registry-defined populations. Results should be interpreted in their specified population.
- Non-inferiority margin: the 9.2% margin is central to the primary inference. The ClinicalTrials.gov record does not provide the clinical or historical justification used to establish that margin.
- Multiplicity: two primary comparisons and multiple secondary endpoints are reported, but the ClinicalTrials.gov record does not specify a multiplicity-adjustment strategy. P-values should therefore be interpreted according to their individual posted analyses rather than automatically treated as a fully adjusted family of tests.
- Stratification metadata: stratified analysis is listed among the analysis concepts, while the explicit analysis notes state that the risk-difference calculation was without stratification. The latter is used here to describe the numerical method rather than inventing a stratified calculation.
- Binary endpoint structure: the treatment-period primary endpoint does not call for time-to-event methods in the posted analysis. A risk difference describes the proportion experiencing the endpoint during the specified period.
- Secondary p-values without effect estimates: several Fisher exact analyses have p-values but no corresponding risk-difference estimate in the ClinicalTrials.gov record. Their effect magnitude cannot be inferred from the p-value alone.
- Safety denominator: bleeding-event analyses use the Treated set, while serious adverse-event counts are presented as affected/at-risk counts by arm. These should not be substituted for efficacy denominators.
- Registry data boundary: ClinicalTrials.gov does not provide, in the ClinicalTrials.gov record, additional baseline characteristics, subgroup results, crossover information, interim-analysis details, Bayesian methods, or missing-data/imputation procedures. Those features are therefore not inferred here.
20. Why This Trial Matters Statistically
RE-MODEL is a useful teaching case because it shows how the statistical question changes when a randomized trial is designed around non-inferiority rather than superiority. The same binary outcome can generate very different interpretations depending on the null hypothesis and prespecified margin.
| Concept | How it appears in RE-MODEL |
|---|---|
| Randomization | Randomized allocation in a three-arm parallel phase 3 prevention trial |
| Blinding | Double masking |
| Binary endpoint | Total VTE and all-cause mortality during the treatment period |
| Risk difference | Primary effect measure expressed as a percentage |
| Non-inferiority | Primary comparisons use a 9.2% absolute risk-difference margin |
| Confidence interval | Primary non-inferiority determination relies on the upper 95% CI |
| Wald / z-test | Normal approximation used for primary and several secondary risk-difference analyses |
| Fisher exact test | Used for symptomatic DVT, PE, death, and bleeding-event comparisons |
| Analysis populations | Full Analysis Set and endpoint-specific populations are used across analyses |
| Multiplicity | Two primary comparisons and multiple secondary analyses are reported; the ClinicalTrials.gov record do not specify adjustment |
The most important lesson is that a statistical result cannot be interpreted independently of its design. A risk difference of 2.8% means something different when the prespecified question is "is the treatment within a 9.2% non-inferiority margin?" than it would in a trial whose sole objective was to demonstrate superiority against zero difference.
21. A Practical Non-Inferiority Reading of RE-MODEL
The effect is absolute and is expressed in percentage points.
The registry states that the absolute risk difference has to be below 9.2% for the non-inferiority criterion.
Both upper confidence limits remain below the prespecified margin.
The confidence interval may include zero while still remaining entirely within the acceptable non-inferiority boundary.
This sequence is more informative than looking at the p-value first. It starts with the clinical effect measure, identifies the prespecified tolerance boundary, and then asks whether the uncertainty interval crosses that boundary.
22. Trial Timeline
RE-MODEL begins
The registry lists the trial start as 2004-11. The study is a phase 3 randomized, double-blind, parallel prevention trial.
Primary completion
The registry lists primary completion as 2006-05.
Statistical results available
ClinicalTrials.gov reports 12 outcome measures and 16 statistical analyses, including two formal analyses of the primary endpoint.
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
Explore calculators that support the binary-outcome and clinical-trial concepts used in this analysis:
25. Sources
- ClinicalTrials.gov: RE-MODEL, NCT00168805.
- Linked publication: PubMed record for PMID 22995531: PubMed 22995531.
- Linked publication: PubMed record for PMID 22709460: PubMed 22709460.
Continue through the Clinical Biostats statistical pathway
Move from this trial's design and results to deeper tutorials on non-inferiority, confidence intervals, risk differences, categorical tests, and clinical-trial methodology.
26. Record Summary
RE-MODEL provides a clear example of how non-inferiority changes the interpretation of a randomized clinical trial. The primary endpoint was binary, the effect measure was an absolute risk difference, and the registry specified a 9.2% non-inferiority margin. For dabigatran 220mg versus enoxaparin, the reported risk difference was -1.3% with a 95% CI of -7.3% to 4.6%; for dabigatran 150mg versus enoxaparin, the reported risk difference was 2.8% with a 95% CI of -3.1% to 8.7%. In both cases, the upper confidence limit remained below the registry-specified margin.
The secondary analyses demonstrate the importance of keeping effect measures, hypothesis types, and analysis populations distinct. Several endpoints use Wald / z-test analyses of risk differences, while symptomatic DVT, pulmonary embolism, death, and bleeding events use Fisher exact testing. The ClinicalTrials.gov record also distinguish the Full Analysis Set from endpoint-specific populations and the Treated set.