This page separates reported trial results from statistical interpretation. Trial facts and numerical results on this page are restricted to the ClinicalTrials.gov record for NCT02441218. The registry provides the official trial record.
1. Trial at a Glance
SHIFT was a randomized, parallel-group, quadruple-masked phase 3 study comparing ivabradine with placebo in patients with chronic heart failure. The registered primary endpoint was a time-to-event composite of cardiovascular death, including death of unknown cause, or hospitalization for worsening heart failure.
| Feature | SHIFT |
|---|---|
| Trial name | SHIFT |
| Brief title | Effects of Ivabradine on Cardiovascular Events in Patients With Moderate to Severe Chronic Heart Failure and Left Ventricular Systolic Dysfunction. A Three-year International Multicentre Study |
| Phase | Phase 3 |
| Condition | Chronic Heart Failure |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 6505 |
| Interventions | Ivabradine; placebo |
| Trial status | Completed |
| Study start | 2006-09 |
| Primary completion | 2010-04 |
| Lead sponsor | Institut de Recherches Internationales Servier |
| Sponsor type | Other |
| ClinicalTrials.gov | NCT02441218 |
2. Clinical Question
The registered primary question can be expressed as a time-to-event comparison: among patients with moderate to severe chronic heart failure and left ventricular systolic dysfunction, how does ivabradine compare with placebo with respect to the time to the first occurrence of cardiovascular death, including death of unknown cause, or hospitalization for worsening heart failure?
Population
Patients with moderate to severe chronic heart failure and left ventricular systolic dysfunction.
Intervention
Ivabradine.
Comparator
Placebo.
Primary question
Whether ivabradine changes the hazard of the first event in the registered primary composite endpoint compared with placebo.
3. Trial Design
Ivabradine
- Drug intervention.
- Compared with placebo under the randomized parallel-group design.
- Included in the primary and secondary time-to-event comparisons.
Placebo
- Placebo intervention.
- Served as the randomized comparator.
- Included in the primary and secondary time-to-event comparisons.
What the design contributes statistically
Randomization establishes the treatment groups before follow-up outcomes occur, providing the basic framework for comparing event-time distributions between ivabradine and placebo. The parallel structure means the comparison is between two concurrently randomized groups rather than a crossover or factorial comparison. Quadruple masking is an additional design feature intended to keep treatment assignment concealed from four categories of trial participants or personnel; the ClinicalTrials.gov record identifies the masking level but do not specify those categories.
4. Endpoints
| Endpoint role | Registered endpoint | Time frame | Analysis type |
|---|---|---|---|
| Primary | Primary Composite Endpoint: First Event Among Cardiovascular Death (Including Death of Unknown Cause) or Hospitalization for Worsening Heart Failure. | All over the study (up to 42 months). | Time-to-event; adjusted Cox proportional-hazards model |
The registry defines the primary endpoint as the number of patients having experienced the Primary Composite Endpoint. Because the posted formal analysis is a Cox proportional-hazards model, the statistical comparison uses the timing of the first qualifying event and censoring information rather than treating the endpoint simply as a binary proportion.
Secondary endpoints with posted analyses
| Secondary endpoint | Time frame | Effect measure |
|---|---|---|
| Cardiovascular Death | From the date of randomization until the date of death, up to 42 months | Hazard ratio |
| Hospitalisation for Worsening Heart Failure | From the date of randomization to the date of first documented hospitalisation, up to 42 months | Hazard ratio |
| All-cause Mortality | From the date of randomisation to death, up to 42 months. | Hazard ratio |
| Death From Heart Failure | From the date of randomisation to death, up to 42 months. | Hazard ratio |
| Hospitalisation for Any Cause | From the date of randomisation to the date of first documented hospitalisation, up to 42 months | Hazard ratio |
| Hospitalisation for Cardiovascular Reason | From the date of randomisation to the first documented hospitalisation, up to 42 months | Hazard ratio |
| Unplanned Hospitalisation for Any Cause | From the date of randomisation to the first documented hospitalisation, up to 42 months | Hazard ratio |
| Unplanned Hospitalisation for CV Reason | From the date of randomisation to the first documented hospitalisation, up to 42 months. | Hazard ratio |
| Secondary Composite Endpoint | From the date of randomisation to the date of the first event, up to 42 months | Hazard ratio |
5. Statistical Methodology
Primary analysis: adjusted Cox proportional-hazards model
The registry reports a Cox proportional-hazards model for the primary composite endpoint. The analysis compared ivabradine with placebo using an adjusted model with beta-blocker intake at randomisation as a covariate.
In a Cox model, the exponentiated treatment coefficient is interpreted as a hazard ratio. The model estimates the relative instantaneous event rate between treatment groups while accounting for the covariate specified in the registry analysis text.
Hazard ratio as the effect measure
The effect measure reported for the primary endpoint was the hazard ratio (HR). An HR below 1 indicates a lower estimated instantaneous event rate in the ivabradine group relative to placebo under the fitted Cox model. An HR of 1 would represent equal modeled hazards, while an HR above 1 would indicate a higher modeled hazard in the treatment group.
Covariate adjustment
The primary analysis adjusted for beta-blocker intake at randomisation. Covariate adjustment can improve precision when the covariate is prognostically informative and is incorporated appropriately into the prespecified analysis. Importantly, adjustment does not turn an observational comparison into a randomized one; in SHIFT, the randomized allocation supplies the principal basis for the treatment comparison.
Wald testing
The registry states that the reported p-value was obtained using a Wald test. In the Cox model, the Wald statistic evaluates the estimated treatment coefficient relative to its estimated standard error. The p-value therefore tests a hypothesis about the model parameter; it is not a measure of how large or clinically important the hazard ratio is.
Two-sided 95% confidence intervals
The posted confidence intervals are identified as 95% and two-sided. A confidence interval communicates statistical precision around the estimated hazard ratio under the fitted model and sampling framework. It should not be interpreted as the range of effects that individual patients experience.
6. Results: Primary Composite Endpoint
The primary registered endpoint was the first event among cardiovascular death, including death of unknown cause, or hospitalization for worsening heart failure, assessed over the study period up to 42 months.
Primary composite endpoint
95% CI: 0.75–0.90 · P < 0.0001
Ivabradine vs placebo · adjusted Cox proportional-hazards model
An HR of 0.82 means that, under the adjusted Cox model, the estimated instantaneous rate of experiencing the first event in the primary composite was 82% of the corresponding rate in the placebo group. Expressed as a simple relative interpretation, this corresponds to a 18% lower estimated hazard for the composite endpoint.
The HR does not mean that 18% of patients avoided the event, that each individual patient had an 18% reduction in probability, or that the absolute reduction in the proportion of patients experiencing an event was 18 percentage points. Hazard is a time-dependent rate, and the HR summarizes a modeled relative comparison of those rates.
The 95% CI of 0.75–0.90 describes uncertainty around the estimated hazard ratio. It does not describe individual-patient variability or guarantee that the true treatment effect lies within this interval for every future population. The interval lies below 1, which is consistent with a lower estimated hazard for ivabradine under the model.
The p-value of <0.0001 is evidence against the null hypothesis represented by the model test; it does not quantify the magnitude of the treatment effect. A very small p-value can occur with a modest effect when the estimate is sufficiently precise, so the HR and its confidence interval remain essential.
Because the analysis uses a Cox proportional-hazards model, interpretation of one summary HR also depends on the proportional-hazards framework. The ClinicalTrials.gov record does not report a formal test or diagnostic for the proportional-hazards assumption, nor do they provide Kaplan-Meier curves or absolute event probabilities from which the time-varying behavior of the treatment effect could be assessed.
What the primary result establishes statistically
The posted analysis provides a quantified randomized comparison using an adjusted Cox model, a hazard ratio of 0.82, a two-sided 95% confidence interval of 0.75–0.90, and a Wald-test p-value of <0.0001. The result is consistent with the prespecified superiority hypothesis recorded for the analysis.
The registry data do not provide the underlying number of primary composite events by treatment arm, median event-free time, Kaplan-Meier estimates at particular time points, or a reconstruction of the survival curve. Those quantities should therefore not be inferred from the hazard ratio.
7. Secondary Endpoint Results
The registry contains nine posted secondary statistical analyses. Each uses an adjusted Cox proportional-hazards model with beta-blocker intake at randomisation as a covariate, with the hazard ratio as the effect measure and a Wald test for the p-value.
| Secondary endpoint | HR | 95% CI | P-value |
|---|---|---|---|
| Cardiovascular Death | 0.91 | 0.80–1.03 | 0.128 |
| Hospitalisation for Worsening Heart Failure | 0.74 | 0.66–0.83 | < 0.0001 |
| All-cause Mortality | 0.90 | 0.80–1.02 | 0.092 |
| Death From Heart Failure | 0.74 | 0.58–0.94 | 0.0140 |
| Hospitalisation for Any Cause | 0.89 | 0.82–0.96 | 0.0027 |
| Hospitalisation for Cardiovascular Reason | 0.85 | 0.78–0.92 | 0.0002 |
| Unplanned Hospitalisation for Any Cause | 0.88 | 0.81–0.95 | 0.0013 |
| Unplanned Hospitalisation for CV Reason | 0.84 | 0.77–0.92 | 0.0002 |
| Secondary Composite Endpoint | 0.82 | 0.74–0.89 | < 0.0001 |
Cardiovascular Death
Hazard ratio
95% CI: 0.80–1.03 · P = 0.128
The estimated hazard ratio was below 1, corresponding to a 9% lower estimated hazard under the Cox model. However, the 95% confidence interval extends from 0.80 to 1.03, crossing the value 1. The registry reports a two-sided Wald-test p-value of 0.128.
Hospitalisation for Worsening Heart Failure
Hazard ratio
95% CI: 0.66–0.83 · P < 0.0001
The HR of 0.74 corresponds to a 26% lower estimated hazard of first documented hospitalization for worsening heart failure under the adjusted Cox model. The confidence interval remains below 1, from 0.66 to 0.83, and the reported Wald-test p-value is <0.0001.
All-cause Mortality
Hazard ratio
95% CI: 0.80–1.02 · P = 0.092
The estimated hazard ratio corresponds to a 10% lower estimated hazard of death from any cause under the fitted model. The 95% confidence interval extends from 0.80 to 1.02, and the reported Wald-test p-value is 0.092.
Death From Heart Failure
Hazard ratio
95% CI: 0.58–0.94 · P = 0.0140
The HR of 0.74 corresponds to a 26% lower estimated hazard of death from heart failure under the adjusted Cox model. The confidence interval is 0.58–0.94 and remains below 1.
Hospitalisation for Any Cause
Hazard ratio
95% CI: 0.82–0.96 · P = 0.0027
The estimated HR corresponds to an 11% lower estimated hazard of first documented hospitalization for any cause. The 95% confidence interval is 0.82–0.96.
Hospitalisation for Cardiovascular Reason
Hazard ratio
95% CI: 0.78–0.92 · P = 0.0002
The HR of 0.85 corresponds to a 15% lower estimated hazard of the first documented hospitalization for a cardiovascular reason. The confidence interval ranges from 0.78 to 0.92.
Unplanned Hospitalisation for Any Cause
Hazard ratio
95% CI: 0.81–0.95 · P = 0.0013
The estimated HR corresponds to a 12% lower estimated hazard of first hospitalization for any cause when the hospitalization was classified as unplanned. The 95% confidence interval is 0.81–0.95.
Unplanned Hospitalisation for CV Reason
Hazard ratio
95% CI: 0.77–0.92 · P = 0.0002
The HR of 0.84 corresponds to a 16% lower estimated hazard of the first unplanned hospitalization for a cardiovascular reason. The 95% confidence interval is 0.77–0.92.
Secondary Composite Endpoint
Hazard ratio
95% CI: 0.74–0.89 · P < 0.0001
The estimated HR of 0.82 corresponds to an 18% lower estimated hazard of the first event in the secondary composite endpoint under the adjusted Cox model. The 95% confidence interval is 0.74–0.89.
8. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
Each posted efficacy endpoint is a time-to-event endpoint. The analysis therefore needs to account not only for whether an event occurred but also for when the event occurred and for patients whose event time is censored. A Cox model is designed for this setting and produces a hazard ratio as a relative measure of event rate between groups.
What does an HR of 0.82 mean?
An HR of 0.82 means that the modeled instantaneous event rate in the ivabradine group was estimated to be 82% of the corresponding rate in the placebo group, conditional on the fitted model. The simple relative interpretation is an 18% lower estimated hazard. It is not an 18-percentage-point reduction in event probability.
Why is the confidence interval important?
The point estimate alone gives only one estimate of the treatment effect. The 95% confidence interval communicates how precisely the effect was estimated. For the primary endpoint, the interval is 0.75–0.90, which is relatively narrow compared with the point estimate and remains below 1. The interval does not describe the range of outcomes for individual patients.
Why doesn't the p-value measure effect size?
The p-value addresses evidence against the null hypothesis represented by the statistical test. It depends on both the magnitude of the estimated effect and its uncertainty. Consequently, two studies can have similar hazard ratios but different p-values if their precision differs. For SHIFT, the effect estimate and 95% confidence interval should be considered together with the Wald-test p-value.
Why was beta-blocker intake included as a covariate?
The registry analysis text explicitly states that beta-blocker intake at randomisation was used as a covariate in the adjusted Cox model. Covariate adjustment can account statistically for variation associated with a prespecified baseline characteristic and can improve precision. The ClinicalTrials.gov record does not state the estimated regression coefficient for beta-blocker intake, so no independent effect of that covariate can be quantified from the registry analysis shown here.
What does a composite endpoint change?
The primary endpoint combines two event types: cardiovascular death, including death of unknown cause, and hospitalization for worsening heart failure. The statistical analysis concerns the first qualifying event. A composite can increase the number of observed events and therefore provide a broader outcome, but its interpretation depends on understanding which component events are contributing to the treatment effect. The ClinicalTrials.gov record provides separate analyses for cardiovascular death and hospitalization for worsening heart failure, which helps distinguish those components statistically.
Why is censoring important?
Time-to-event methods generally allow patients who have not experienced an event by the end of their observed follow-up to contribute information until their censoring time. The Cox model uses this timing information rather than requiring every patient to experience an event. The ClinicalTrials.gov record does not specify the censoring rules or missing-data procedures, so those details should not be inferred.
9. Primary Result in Statistical Context
The primary HR of 0.82 is a relative time-to-event measure. Under the adjusted Cox model, the estimated instantaneous rate of the first primary-composite event was lower in the ivabradine group than in the placebo group.
The 95% CI of 0.75–0.90 places the estimated treatment effect below the null value of 1 throughout the reported interval. This provides more information than the point estimate alone because it shows the uncertainty surrounding the estimated HR.
The two-sided Wald-test p-value was <0.0001. This indicates strong statistical evidence against the null hypothesis used for the model test. It should not be interpreted as the probability that the treatment has no effect, nor as the magnitude of the effect.
The ClinicalTrials.gov record does not provide absolute event rates, event counts for the primary composite by treatment group, median time to the primary event, or time-specific event-free percentages. Those quantities therefore cannot be derived reliably from the reported HR alone.
10. Safety Results
The ClinicalTrials.gov record provides serious adverse events by arm as affected patients divided by the corresponding number at risk.
| Safety measure | Ivabradine | Placebo |
|---|---|---|
| Serious adverse events | 1369/3232 | 1481/3260 |
Ivabradine
Serious adverse events affected 1369 of 3232 patients at risk in the ClinicalTrials.gov record.
Placebo
Serious adverse events affected 1481 of 3260 patients at risk in the ClinicalTrials.gov record.
The serious-adverse-event figures are descriptive safety data and are not presented in the registry analysis as a Cox-model treatment-effect estimate. They should therefore be kept conceptually separate from the efficacy hazard ratios.
11. What the Secondary Results Add
The secondary analyses help decompose the primary composite and broaden the statistical picture beyond its single combined endpoint. In particular, the registry reports separate hazard ratios for cardiovascular death, hospitalization for worsening heart failure, all-cause mortality, death from heart failure, several hospitalization definitions, and a secondary composite endpoint.
| Endpoint family | Reported HR range | Statistical feature |
|---|---|---|
| Mortality endpoints | 0.74–0.91 | Different mortality definitions produce different estimates and confidence intervals. |
| Hospitalisation endpoints | 0.74–0.89 | Several hospitalization definitions have HRs below 1. |
| Secondary composite | 0.82 | Estimate is numerically identical to the primary HR, but it is a distinct registered endpoint. |
The most important statistical distinction is that these endpoints are not interchangeable. Cardiovascular death, all-cause mortality, hospitalization for worsening heart failure, and hospitalization for any cause represent different event definitions. A hazard ratio for one cannot be substituted for another even when the numerical values are similar.
12. Confidence Intervals Across the Secondary Analyses
Confidence intervals provide a useful way to compare the precision of the posted estimates without turning the secondary endpoints into a ranking exercise.
| Endpoint | Estimate | 95% CI | Relationship to HR = 1 |
|---|---|---|---|
| Cardiovascular Death | 0.91 | 0.80–1.03 | Includes 1 |
| Hospitalisation for Worsening Heart Failure | 0.74 | 0.66–0.83 | Below 1 |
| All-cause Mortality | 0.90 | 0.80–1.02 | Includes 1 |
| Death From Heart Failure | 0.74 | 0.58–0.94 | Below 1 |
| Hospitalisation for Any Cause | 0.89 | 0.82–0.96 | Below 1 |
| Hospitalisation for Cardiovascular Reason | 0.85 | 0.78–0.92 | Below 1 |
| Unplanned Hospitalisation for Any Cause | 0.88 | 0.81–0.95 | Below 1 |
| Unplanned Hospitalisation for CV Reason | 0.84 | 0.77–0.92 | Below 1 |
| Secondary Composite Endpoint | 0.82 | 0.74–0.89 | Below 1 |
The two mortality analyses for cardiovascular death and all-cause mortality have confidence intervals that include 1, whereas the other posted secondary confidence intervals shown above remain below 1. This is a descriptive comparison of the reported intervals; it does not replace a prespecified multiplicity strategy or a formal comparison of treatment effects.
13. Why the Composite Endpoint Matters
The primary endpoint is explicitly defined around the first event among two clinically different outcomes. Statistically, this is important because the composite endpoint can be influenced by the frequency and treatment effect of each component.
Component 1
Cardiovascular death, including death of unknown cause.
Component 2
Hospitalization for worsening heart failure.
Event definition
The registered endpoint is the first event among these components.
Separate analyses
The registry separately reports cardiovascular death and hospitalization for worsening heart failure.
A composite endpoint can be statistically efficient when multiple clinically relevant events are combined, but the overall HR does not reveal the contribution of each component by itself. In SHIFT, the separate posted analyses are therefore important for interpreting what the primary composite represents.
14. Randomization and Masking
SHIFT was registered as randomized, with a parallel design and quadruple masking. These design characteristics address different sources of bias.
The statistical value of randomization is especially important for interpreting the Cox-model comparison. Covariate adjustment can account for the specified beta-blocker variable, but the primary causal framework remains the randomized comparison of ivabradine and placebo.
15. Analysis Population and Censoring Considerations
The ClinicalTrials.gov record does not specify an intention-to-treat definition, per-protocol definition, as-treated efficacy population, censoring algorithm, treatment discontinuation rules, or missing-data/imputation procedure. These details are important in a complete statistical analysis plan, but they should not be inferred from the presence of a Cox model alone.
For a time-to-event endpoint, censoring is particularly important because the Cox model uses patients' observed follow-up even when the event of interest is not observed. The validity of the resulting treatment comparison depends on appropriate event ascertainment and censoring assumptions. The ClinicalTrials.gov record does not provide enough detail to evaluate those assumptions empirically.
16. Multiplicity
SHIFT has one registered primary endpoint and multiple posted secondary endpoints. The primary endpoint has a formal superiority analysis with a two-sided 95% confidence interval and a Wald-test p-value. The secondary analyses also report p-values, but the ClinicalTrials.gov record does not state a multiplicity-adjustment procedure or hierarchical testing strategy.
| Analysis family | Number in the ClinicalTrials.gov record | Statistical interpretation |
|---|---|---|
| Primary endpoint | 1 | Formal adjusted Cox analysis reported. |
| Secondary endpoints | 9 | Formal adjusted Cox analyses reported. |
| Statistical analyses posted | 10 | All registry-reported analyses use Cox regression. |
Multiplicity matters because testing multiple hypotheses increases the opportunity for at least one apparently unusual result under a global null model. Without a documented multiplicity strategy in the ClinicalTrials.gov record, the individual secondary p-values should be described as the registry-reported Wald-test results rather than interpreted as though each had its own independently protected confirmatory error rate.
17. Interim Analysis and Other Design Features
The ClinicalTrials.gov record do not report an interim-analysis schedule, alpha-spending procedure, stopping boundary, non-inferiority margin, crossover design, factorial structure, Bayesian analysis, or missing-data/imputation method.
Interim analysis
Not reported in the ClinicalTrials.gov record.
Non-inferiority margin
Not applicable to the reported superiority hypothesis and not provided in the ClinicalTrials.gov record.
Crossover
No crossover design is reported; the registered design model is parallel.
Bayesian methods
No Bayesian analysis is reported in the statistical analyses posted on ClinicalTrials.gov.
This distinction is deliberate. A complete statistical discussion should identify both what the registry reports and what it does not report. Adding an unreported interim strategy, power calculation, or Bayesian procedure would create trial facts that are not supported by the ClinicalTrials.gov record.
18. Statistical Interpretation of the Secondary Findings
The posted hospitalization analyses all have HR estimates below 1: 0.74 for hospitalization for worsening heart failure, 0.89 for hospitalization for any cause, 0.85 for hospitalization for cardiovascular reason, 0.88 for unplanned hospitalization for any cause, and 0.84 for unplanned hospitalization for CV reason. Each corresponding 95% confidence interval is also below 1.
Cardiovascular death has HR 0.91 with 95% CI 0.80–1.03 and P = 0.128. All-cause mortality has HR 0.90 with 95% CI 0.80–1.02 and P = 0.092. Death from heart failure has HR 0.74 with 95% CI 0.58–0.94 and P = 0.0140. These endpoints should not be collapsed into a single mortality conclusion because their event definitions differ.
The secondary composite endpoint has HR 0.82 with 95% CI 0.74–0.89 and P < 0.0001. It is a separate registered endpoint and should not be assumed to have the same event definition as the primary composite merely because its HR is numerically identical.
19. Limitations
- Registry-level detail: the ClinicalTrials.gov record identifies the principal Cox model and its covariate but do not provide the full statistical analysis plan.
- No primary event counts by arm: the registry-reported primary analysis provides the HR, confidence interval, and p-value but not the number of patients experiencing the primary composite in each treatment group.
- No absolute event probabilities: absolute risk differences, cumulative incidence estimates, and time-specific event rates are not reported.
- No median event time: median time to the primary endpoint and secondary endpoints is not reported in the ClinicalTrials.gov record.
- No Kaplan-Meier estimates: the registry analysis identifies Cox regression but does not supply Kaplan-Meier curves or time-specific survival estimates.
- Proportional-hazards assumption: the Cox model relies on a proportional-hazards framework, but the ClinicalTrials.gov record does not provide diagnostics or a formal assessment of that assumption.
- Multiplicity: nine secondary endpoints have posted p-values, but the ClinicalTrials.gov record does not specify a multiplicity-adjustment strategy.
- Analysis populations: detailed efficacy and safety population definitions beyond the registry-reported safety denominators are not provided.
- Censoring and missing data: the ClinicalTrials.gov record does not specify the censoring rules or missing-data/imputation procedures.
- Component interpretation: the primary composite combines cardiovascular death, including death of unknown cause, with hospitalization for worsening heart failure. The overall HR therefore does not identify the contribution of each component without examining the component analyses.
- Safety reporting caveat: the registry states that the non-serious adverse-event section was completed by the sponsor to include only non-serious adverse events emergent on treatment during the study.
20. Why This Trial Matters Statistically
SHIFT is a useful teaching case because the registry provides a clean example of how a randomized clinical trial can combine a composite time-to-event endpoint, covariate-adjusted Cox regression, hazard ratios, confidence intervals, and Wald testing across a primary and multiple secondary outcomes.
| Concept | How it appears in SHIFT |
|---|---|
| Randomization | The trial is registered as randomized. |
| Parallel-group design | Ivabradine and placebo form two parallel intervention arms. |
| Quadruple masking | The registry identifies the study as quadruple masked. |
| Composite endpoint | The primary endpoint is the first event among cardiovascular death, including death of unknown cause, or hospitalization for worsening heart failure. |
| Time-to-event analysis | The primary and all nine posted secondary analyses are classified as time-to-event analyses. |
| Cox regression | The reported method is a Cox proportional-hazards model. |
| Covariate adjustment | Beta-blocker intake at randomisation is included as a covariate. |
| Hazard ratio | All posted statistical analyses use the hazard ratio as the effect measure. |
| Confidence interval | Each posted analysis supplies a 95% two-sided confidence interval. |
| Wald test | The registry identifies the p-value for each analysis as a Wald test. |
| Superiority testing | The primary and secondary analyses are recorded under a superiority hypothesis. |
| Multiplicity | One primary endpoint and nine secondary analyses are posted; the ClinicalTrials.gov record does not state a multiplicity strategy. |
21. Clinical Biostats Interpretation: Reading the Whole Statistical Story
The most informative way to read SHIFT is not to focus on a single p-value. The primary analysis gives four connected pieces of information: the treatment contrast, the effect measure, the uncertainty interval, and the statistical test.
1. Treatment contrast
Ivabradine was compared directly with placebo in a randomized parallel-group design.
2. Effect measure
The primary effect measure was a hazard ratio from an adjusted Cox proportional-hazards model.
3. Precision
The primary HR of 0.82 has a two-sided 95% CI of 0.75–0.90.
4. Statistical test
The primary Wald-test p-value was <0.0001.
This framework prevents a common statistical error: treating the p-value as the result itself. The p-value describes the evidence against a specified null hypothesis, whereas the HR describes the estimated relative event rate and the confidence interval describes uncertainty around that estimate.
A second important distinction is between relative and absolute treatment effects. The HR of 0.82 is relative and model-based. Without event counts or absolute cumulative-event estimates, it cannot be converted into an absolute risk reduction from the ClinicalTrials.gov record.
Finally, the primary endpoint is a composite. The secondary analyses show that the registry separately analyzed cardiovascular death and hospitalization for worsening heart failure, allowing the reader to examine components rather than assuming that the composite HR describes every component equally.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: NCT02441218 — SHIFT.
- Linked publication: PubMed PMID 20801500.
- Linked publication: PubMed PMID 20801495.
- Linked publication: PubMed PMID 32580655.
- Linked publication: PubMed PMID 28205056.
Continue through Clinical Biostats
Explore the statistical concepts behind randomized trials, survival analysis, hazard ratios, confidence intervals, and clinical-trial methodology.
25. Record Summary
SHIFT provides a useful example of randomized clinical-trial analysis centered on a composite time-to-event endpoint. The trial enrolled 6505 participants and used a randomized, parallel-group, quadruple-masked design comparing ivabradine with placebo. The primary analysis used an adjusted Cox proportional-hazards model with beta-blocker intake at randomisation as a covariate. The resulting primary hazard ratio was 0.82, with a two-sided 95% CI of 0.75–0.90 and a Wald-test p-value of <0.0001.
The secondary analyses extend the same statistical framework to mortality and hospitalization endpoints. Their HR estimates range from 0.74 to 0.91, with endpoint-specific confidence intervals and Wald-test p-values. The mortality endpoints should be interpreted separately from hospitalization endpoints, and all secondary p-values should be considered in the context of the absence of a documented multiplicity strategy in the ClinicalTrials.gov record.
The principal statistical lesson is that a hazard ratio is not an absolute risk reduction and a p-value is not an effect size. For SHIFT, the most complete interpretation combines the randomized treatment comparison, the registered composite endpoint, the adjusted Cox model, the hazard ratio, its 95% confidence interval, the Wald-test p-value, the component-specific secondary analyses, and the limitations of the information available in the registry record.