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
SELECT was a randomized, parallel-group, quadruple-masked phase 3 trial evaluating semaglutide versus placebo in participants with overweight or obesity. The registry reports a single primary endpoint defined as the time from randomization to the first occurrence of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke.
| Feature | SELECT |
|---|---|
| Phase | Phase 3 |
| Therapeutic area | Endocrinology |
| Conditions | Overweight; Obesity |
| Design | Randomized, parallel-group |
| Masking | Quadruple |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 17604.0 |
| Arms | 2 |
| Interventions | Semaglutide; Placebo (semaglutide) |
| Primary endpoints | 1 registered primary endpoint |
| Primary endpoint type | Binary |
| Posted statistical analyses | 1 |
| Reported statistical method | Cox proportional-hazards model |
| Reported effect measure | Hazard ratio |
| Hypothesis type | Superiority |
| ClinicalTrials.gov | NCT03574597 |
2. Clinical Question
The central question represented by the registered primary endpoint is whether assignment to semaglutide, compared with placebo, changes the time to first occurrence of a composite cardiovascular outcome consisting of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke.
Population
Participants in SELECT with the registered conditions of overweight or obesity.
Intervention
Semaglutide.
Comparator
Placebo (semaglutide).
Primary question
Does randomized assignment to semaglutide change the time to the first occurrence of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke?
3. Trial Design
Semaglutide
- Semaglutide
- Randomized treatment assignment
Placebo
- Placebo (semaglutide)
- Randomized treatment assignment
4. Trial Timing and Registry Status
Trial start
The registry records 2018-10-24 as the trial start date.
Primary completion
The registry records 2023-06-21 as the primary completion date.
Registry status
The trial is recorded as COMPLETED, with results posted on ClinicalTrials.gov.
5. Primary Endpoint
| Endpoint | Definition / time frame | Analysis |
|---|---|---|
| Participants From Time of Randomization to First Occurrence of a Composite Outcome Measure Consisting of: Cardiovascular (CV) Death, Non-fatal Myocardial Infarction (MI), or Non-fatal Stroke | From randomisation (week 0) up to 240 weeks | Cox proportional-hazards model |
The registry defines the outcome as the number of participants with a first occurrence of the composite outcome consisting of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke. Cardiovascular death includes an undetermined cause of death presumed to be cardiovascular death. The outcome was evaluated using data from the in-trial observation period.
This is fundamentally a time-to-event question even though the registered primary endpoint is classified as binary. The posted statistical analysis explicitly treats it as a time-to-event endpoint and uses a Cox proportional-hazards model.
6. Analysis Population and Comparison
| Element | Registry-supported specification |
|---|---|
| Analysis population | FAS included all randomised participants. All participants were analyzed according to the treatment to which they were assigned at randomisation. |
| Groups compared | Semaglutide vs Placebo |
| Endpoint role | Primary |
| Endpoint classification for analysis | Time-to-event |
| Method | Regression, Cox / Cox proportional-hazards model |
| Effect measure | Hazard Ratio (HR) |
| Hypothesis type | Superiority |
The analysis population is important because treatment assignment is preserved in the primary comparison. Participants are analyzed according to the treatment to which they were assigned at randomisation rather than being reassigned according to treatment received.
7. Statistical Methodology
Time-to-event analysis
The primary outcome is not simply whether a participant experienced an event. The registry's analysis treats the outcome as time-to-event: the analysis considers the time from randomization until the first qualifying cardiovascular event, while participants without an event contribute information until the end of their in-trial observation period.
Kaplan-Meier estimation
Kaplan-Meier estimation is the standard descriptive framework for displaying the probability of remaining event-free over time in a time-to-event analysis. It accommodates right-censored observations, meaning that participants who have not experienced the event by the time their observation ends can still contribute information before censoring.
Here, di is the number of events at time ti and ni is the number of participants at risk immediately before that time.
The ClinicalTrials.gov record does not provide the underlying event-by-event and censoring data needed to construct an independent Kaplan-Meier curve. Accordingly, this page does not fabricate a curve from the summary hazard ratio and confidence interval.
Cox proportional-hazards model
The posted analysis used a Cox proportional-hazards model with treatment as a categorical fixed factor. This model estimates the relative hazard associated with treatment assignment while using the observed event times and censoring information.
An HR below 1 indicates a lower estimated instantaneous event rate in the semaglutide group relative to placebo under the fitted model.
Censoring
Participants without events of interest were censored at the end of their in-trial period. Censoring is therefore part of the primary analysis rather than a reason to exclude participants who had not experienced the composite outcome during their observed period.
8. Primary Result
Hazard ratio for the first composite cardiovascular outcome
95% CI: 0.72–0.89 · P < 0.0001
Semaglutide vs placebo · Superiority analysis
| Primary endpoint | Effect estimate | 95% confidence interval | P-value |
|---|---|---|---|
| Time from randomization to first occurrence of cardiovascular death, non-fatal MI, or non-fatal stroke | HR 0.80 | 0.72–0.89 | < 0.0001 |
A hazard ratio of 0.80 means that, under the fitted Cox model and over the analyzed in-trial follow-up, the estimated instantaneous rate of experiencing the first qualifying composite event in the semaglutide group was 80% of the corresponding estimated hazard in the placebo group. Expressed as a relative model-based comparison, this corresponds to an estimated 20% lower hazard.
The HR does not mean that 20% of participants avoided an event, that an individual's probability of an event was reduced by exactly 20%, or that the absolute difference in event probabilities was 20 percentage points. Hazard and probability are different quantities.
The 95% CI of 0.72–0.89 describes uncertainty around the estimated hazard ratio under the statistical model and sampling framework. It does not describe the range of effects that individual participants experienced.
The P-value of < 0.0001 addresses the strength of evidence against the relevant null hypothesis under the specified testing framework. It does not measure the size of the treatment effect. Effect size is described by the hazard ratio, while the confidence interval provides information about its precision.
Because this is a Cox analysis, interpretation also depends on the model's proportional-hazards framework. A single hazard ratio summarizes a relative hazard relationship over the analyzed period; it should not automatically be interpreted as a constant relative probability reduction at every time point.
9. Understanding the 95% Confidence Interval
The reported confidence interval extends from 0.72 to 0.89. The point estimate of 0.80 is therefore accompanied by a range that reflects statistical uncertainty around the estimated treatment effect.
Point estimate
HR 0.80 is the single estimated relative hazard from the reported Cox analysis.
Interval estimate
The 95% CI of 0.72–0.89 communicates uncertainty around the point estimate rather than individual-level variability.
What it excludes
The interval is not a prediction interval for the treatment effect in every future participant.
Why both matter
The HR describes the estimated relative effect, while the confidence interval describes the precision of that estimate.
10. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized arm as affected participants divided by participants at risk.
| Safety measure | Semaglutide | Placebo |
|---|---|---|
| Serious adverse events, affected / at risk | 2941/8803 | 3204/8801 |
These figures are presented as reported. They should not be substituted for the primary efficacy analysis because safety and efficacy answer different statistical questions and use different outcome definitions.
11. Statistical Methods Explained
Why use a Cox proportional-hazards model?
The primary endpoint concerns the time from randomization to the first occurrence of a cardiovascular event. A Cox model is designed for this type of censored time-to-event data and produces a hazard ratio as a relative treatment-effect measure.
What does an HR of 0.80 mean?
An HR of 0.80 means that the estimated instantaneous hazard in the semaglutide group was 0.80 times that in the placebo group under the fitted model. It can also be described as an estimated 20% lower hazard because 1 − 0.80 = 0.20.
Why isn't an HR of 0.80 the same as a 20% lower probability?
A hazard ratio compares instantaneous event rates, not cumulative probabilities directly. The cumulative probability of an event depends on the hazard over time. Consequently, an HR of 0.80 cannot by itself be converted into an absolute risk reduction without additional time-specific survival information.
What does the confidence interval tell us?
The 95% CI of 0.72–0.89 gives the uncertainty around the estimated hazard ratio under the analysis framework. A narrower interval generally indicates greater statistical precision than a wider interval, but the interval should not be interpreted as the range of individual treatment effects.
What does P < 0.0001 tell us?
The P-value quantifies evidence against the relevant null hypothesis under the specified statistical framework. It does not tell us that the treatment effect is large, clinically important, or equal to a particular percentage. Those questions require interpretation of the effect estimate, confidence interval, and absolute outcome measures.
Why does censoring matter?
Not every participant necessarily experiences the primary event during the period in which they are observed. The registry states that participants without events of interest were censored at the end of their in-trial period. Survival-analysis methods allow those participants to contribute information up to that point rather than treating them as if they had experienced the event.
Why does the analysis population matter?
The FAS included all randomized participants, and participants were analyzed according to their treatment assignment at randomisation. This preserves the randomized comparison for the primary efficacy analysis and avoids redefining treatment groups based on events that occurred after randomization.
12. What the Hazard Ratio Does — and Does Not — Mean
The reported HR of 0.80 indicates an estimated 20% lower instantaneous hazard of the composite endpoint for semaglutide relative to placebo under the fitted Cox model.
The HR does not tell us how many percentage points lower the cumulative probability of the composite outcome was. That requires absolute event probabilities at a specified time point.
The population-level HR should not be interpreted as saying that every participant receiving semaglutide experienced exactly the same 20% reduction in event risk.
The HR measures relative effect size. The P-value addresses evidence against the null hypothesis, and the confidence interval addresses uncertainty around the effect estimate. These are related but distinct statistical quantities.
13. Registered Endpoint vs Statistical Analysis
| Feature | Registered / reported specification |
|---|---|
| Primary endpoint count | 1 |
| Registered endpoint type | Binary |
| Analysis endpoint type | Time-to-event |
| Time frame | From randomisation (week 0) up to 240 weeks |
| Primary analysis method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Hypothesis type | Superiority |
| Analysis population | FAS; all randomized participants analyzed according to randomized treatment assignment |
| Censoring | Participants without events of interest were censored at the end of their in-trial period |
The distinction between the registered endpoint classification and the posted statistical-analysis classification is statistically important. The endpoint is described in the registry as binary because the outcome is ultimately whether a participant has a qualifying first event. But because the timing of that event is available for analysis, the posted analysis appropriately treats it as a time-to-event endpoint and uses a Cox model.
14. What the Supplied Registry Data Do Not Establish
The ClinicalTrials.gov record identifies one posted statistical analysis for the primary endpoint. They do not provide detailed information on several design and analysis topics that can be important in a full statistical analysis plan.
- Stratification: the ClinicalTrials.gov record does not specify randomization stratification factors or stratified Cox-model terms.
- Interim analysis: the ClinicalTrials.gov record does not specify an interim-analysis schedule or alpha-spending procedure.
- Multiplicity: the ClinicalTrials.gov record does not specify a multiplicity-adjustment strategy beyond identifying the primary hypothesis as superiority.
- Missing-data imputation: the ClinicalTrials.gov record does not specify an imputation procedure for missing covariates or outcomes.
- Non-inferiority: the primary hypothesis is superiority, and the ClinicalTrials.gov record does not specify a non-inferiority margin.
- Crossover: the ClinicalTrials.gov record does not describe a crossover analysis.
- Bayesian methods: no Bayesian method is identified in the ClinicalTrials.gov record.
15. Important Limitations and Interpretation Issues
- Composite endpoint: the primary outcome combines cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke. A composite summarizes these components together rather than reporting one homogeneous event.
- First-event analysis: the registered endpoint concerns the first occurrence of one of the qualifying components. Events occurring after the first qualifying event are not represented by the primary endpoint as defined.
- Hazard-ratio interpretation: the HR is a model-based relative measure and should not be treated as an absolute risk reduction or individual-level effect.
- Proportional-hazards framework: the Cox model provides a single hazard-ratio summary. Interpretation is most straightforward when the proportional-hazards framework is reasonable over the analyzed period.
- Censoring: participants without events were censored at the end of their in-trial period. Survival analyses depend on appropriate handling and interpretation of censored observations.
- Limited registry detail: the ClinicalTrials.gov record does not report detailed stratification, interim-analysis, multiplicity, or missing-data procedures.
- Safety versus efficacy: serious adverse-event counts and the primary efficacy endpoint describe different dimensions of the trial and should not be collapsed into a single statistical measure.
16. Why This Trial Matters Statistically
SELECT provides a useful teaching case for understanding how a large randomized clinical trial converts a clinically defined composite outcome into a formal survival-analysis question. The statistical story is not simply the reported P-value: it includes randomization, treatment assignment, event timing, censoring, the Cox model, the hazard ratio, and uncertainty around the estimate.
| Concept | How it appears in SELECT |
|---|---|
| Randomization | Participants were randomized to semaglutide or placebo. |
| Blinding | The trial used quadruple masking. |
| Time-to-event endpoint | The primary analysis evaluates time from randomization to the first qualifying cardiovascular event. |
| Composite endpoint | Cardiovascular death, non-fatal MI, or non-fatal stroke comprise the primary outcome. |
| Censoring | Participants without events were censored at the end of their in-trial period. |
| Cox model | The primary endpoint was analyzed using a Cox proportional-hazards model. |
| Hazard ratio | HR 0.80 summarizes the reported relative treatment effect. |
| Confidence interval | The 95% CI of 0.72–0.89 describes uncertainty around the HR estimate. |
| P-value | P < 0.0001 provides evidence against the relevant null hypothesis under the specified framework. |
| Analysis population | The FAS included all randomized participants and retained randomized treatment assignment. |
17. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The primary randomized comparison produced a hazard ratio of 0.80, with a 95% CI of 0.72–0.89 and P < 0.0001, using a Cox proportional-hazards model in the FAS.
Clinical interpretation
The primary endpoint is a composite of cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke. The statistical result describes the relative timing of first qualifying events between randomized groups; it does not by itself quantify an absolute reduction in event probability.
18. Reading the Primary Result Correctly
This visualization is a conceptual representation of the reported hazard ratio, not a Kaplan-Meier curve and not an estimate of cumulative event probability. The 0.80 value means the fitted relative hazard was 80% of the comparator hazard.
19. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
20. Related Statistical Calculators
21. Limitations of the Statistical Record
The ClinicalTrials.gov record provides a complete primary statistical result for the endpoint represented in the data, including the analysis population, comparison, model, hazard ratio, confidence interval, P-value, and censoring description. It does not provide the detailed underlying survival dataset required to independently reproduce the Cox-model estimate.
Similarly, the ClinicalTrials.gov record does not provide enough information to reconstruct a full Kaplan-Meier curve, calculate time-specific absolute risks, evaluate individual component effects separately, or reproduce additional statistical procedures that are not explicitly reported in the ClinicalTrials.gov record.
22. Sources
- ClinicalTrials.gov: SELECT — NCT03574597.
- PubMed: PMID 32916609.
- PubMed: PMID 33567185.
- PubMed: PMID 37952131.
- PubMed: PMID 42747817.
- PubMed: PMID 42610271.
Continue through the Clinical Biostats statistical pathway
Connect the primary endpoint in SELECT to deeper tutorials on survival analysis, hazard ratios, confidence intervals, randomization, and time-to-event methods.
23. Record Summary
SELECT is a randomized, parallel-group, quadruple-masked phase 3 trial with 17604.0 participants and two treatment arms. Its registered primary endpoint evaluates the first occurrence of a composite cardiovascular outcome consisting of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke from randomisation (week 0) up to 240 weeks. The posted analysis uses the FAS, retains randomized treatment assignment, and applies a Cox proportional-hazards model to the time-to-event endpoint.
The reported primary result is a hazard ratio of 0.80 for semaglutide versus placebo, with a 95% CI of 0.72–0.89 and P < 0.0001. Statistically, the most important distinction is between the relative hazard represented by the HR and the absolute probability of experiencing a cardiovascular event. The HR does not provide the latter by itself. The confidence interval quantifies uncertainty around the estimated relative hazard, while the P-value addresses evidence against the null hypothesis rather than effect magnitude.
The analysis also illustrates why trial design and statistical methodology need to be read together: randomization defines the treatment comparison, censoring allows participants without observed events to contribute information, and the Cox model converts the observed event-time information into a relative hazard estimate. The registry's serious adverse-event counts provide a separate safety dimension and should not be merged with the efficacy endpoint into a single statistical quantity.