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
STROKE-AF was a randomized, parallel-design diagnostic trial evaluating the rate of atrial fibrillation through 12 months in subjects with a recent ischemic stroke of presumed known origin. Participants were assigned to Reveal LINQ™ insertable cardiac monitoring or a control arm.
| Feature | STROKE-AF |
|---|---|
| Trial name | STROKE-AF |
| ClinicalTrials.gov identifier | NCT02700945 |
| Status | Completed |
| Therapeutic area | Cardiovascular |
| Condition | Stroke, Acute |
| Design | Randomized, parallel |
| Masking | None |
| Primary purpose | Diagnostic |
| Enrollment | 496 |
| Intervention | Reveal LINQ™ Insertable Cardiac Monitor |
| Lead sponsor | Medtronic Cardiac Rhythm and Heart Failure |
| Sponsor type | Industry |
| Trial start | 2016-03 |
| Primary completion | 2020-08-03 |
2. Clinical Question
The primary statistical question was whether the rate of atrial fibrillation through 12 months differed between subjects with a recent ischemic stroke of presumed known origin assigned to Reveal LINQ™ insertable cardiac monitoring and those assigned to the control arm.
Population
Subjects with a recent ischemic stroke of presumed known origin. The registry condition is listed as acute stroke.
Intervention
Reveal LINQ™ Insertable Cardiac Monitor, a device used for continuous cardiac monitoring.
Comparator
Control Arm.
Primary question
What is the rate of atrial fibrillation through 12 months, with AF defined using the prespecified duration and adjudication criteria?
3. Trial Design
Insertable cardiac monitoring
- Reveal LINQ™ Insertable Cardiac Monitor
- Randomized study arm
- AF detection evaluated through 12 months
Control arm
- Control Arm as identified in the registry
- Randomized study arm
- AF detection evaluated through 12 months
The ClinicalTrials.gov record does not specify a randomization ratio, stratification factors, or a factorial structure. Those design features therefore are not assumed here.
4. Endpoints
| Endpoint | Registry definition | Time frame | Statistical framework |
|---|---|---|---|
| The Rate of AF Through 12 Months in Subjects With a Recent Ischemic Stroke of Presumed Known Origin | AF will be defined as an AF event lasting more than 30 seconds. The first AF episode detected and adjudicated by the endpoint adjudication committee will be used for this analysis. | 12 months | Time-to-event; log-rank test; hazard ratio |
The endpoint is expressed in the registry as a rate through 12 months and is analyzed as a time-to-event endpoint. That distinction matters statistically: participants can contribute information for different lengths of time before experiencing AF or being censored.
5. Statistical Methodology
Kaplan-Meier estimation
The registry analysis specifies a time-to-event framework and states that the 12-month Kaplan-Meier estimate will be reported for each arm. Kaplan-Meier estimation is appropriate when participants can have different observed follow-up times and some observations are censored before the event occurs.
Here, di represents events at time ti, while ni represents participants at risk immediately before that time. For this trial, the corresponding event is the first detected and adjudicated AF episode meeting the registry definition.
Log-rank test
The reported formal comparison used a log-rank test. The log-rank test compares the observed pattern of time-to-event occurrence between randomized groups over the analysis period rather than comparing only the proportions observed at a single time point.
Hazard ratio
The reported effect measure was a hazard ratio. A hazard ratio compares the estimated instantaneous event rates between groups within a time-to-event framework. For this analysis, the treatment effect is expressed as the hazard of the first detected and adjudicated AF event in the Reveal LINQ monitoring group relative to the control group.
The registry analysis notes that a hazard ratio greater than 1 indicates superiority of the continuous monitoring arm for the AF-detection endpoint. The hazard ratio is a relative time-to-event measure; it is not itself a percentage of participants with AF.
Two-sided hypothesis test
The registry-reported analysis notes define the null hypothesis as equal hazards through 12 months and the alternative as unequal hazards through 12 months. The reported confidence interval is two-sided, and the reported p-value is <0.001.
HA: hC(t) ≠ hT(t) for t ≤ 12 months
The registry describes hT(t) and hC(t) as the hazard functions of first detected and adjudicated AF for subjects with and without the Reveal LINQ diagnostics for AF, respectively.
6. Primary Result: Atrial Fibrillation Through 12 Months
The posted statistical analysis compares Reveal LINQ™ Insertable Cardiac Monitor with the Control Arm for the rate of AF through 12 months. AF was defined as an event lasting more than 30 seconds, with the first qualifying episode detected and adjudicated by the endpoint adjudication committee used for the analysis.
Hazard ratio for first detected and adjudicated AF
95% CI: 2.6–21.3 · P < 0.001
Two-sided confidence interval; log-rank test; superiority hypothesis.
The estimated hazard ratio of 7.4 means that, within the reported time-to-event analysis, the estimated instantaneous rate of the first qualifying AF event was higher in the Reveal LINQ monitoring group than in the control group. In the context of a diagnostic monitoring trial, this is consistent with more AF being detected when continuous insertable cardiac monitoring is used.
The estimate does not mean that 7.4 times as many participants developed AF, and it does not mean that 740% of participants experienced AF. A hazard ratio is a relative time-to-event measure, not an absolute event probability.
The 95% confidence interval of 2.6–21.3 describes statistical uncertainty around the estimated hazard ratio. It is wide relative to the point estimate, so the precise magnitude of the relative difference is less certain than the direction of the reported association. The interval remains above 1, which is consistent with the reported two-sided P < 0.001.
The p-value addresses evidence against the specified null hypothesis; it does not measure the size or clinical importance of the effect. The effect size is described by the hazard ratio and its confidence interval, while absolute AF rates and Kaplan-Meier estimates provide additional information about event probability over time.
Because this is a time-to-event analysis, interpretation also depends on censoring and on the assumptions underlying the hazard-based analysis. The ClinicalTrials.gov record does not report a formal assessment of the proportional-hazards assumption, so no such assessment is asserted here.
7. Understanding the 7.4 Hazard Ratio
The magnitude of the reported estimate is easiest to understand by separating three different quantities: the hazard ratio, the probability of an event by a specified time, and the statistical evidence against the null hypothesis.
Relative event rate
The hazard ratio of 7.4 is a relative time-to-event measure comparing the instantaneous rates of first qualifying AF between randomized groups.
Absolute event probability
A Kaplan-Meier estimate answers a different question: the estimated probability of remaining free of the event through a particular time point.
Statistical evidence
The P-value <0.001 describes the evidence against the specified equal-hazard null hypothesis; it is not a measure of how large the treatment effect is.
Precision
The 95% CI of 2.6–21.3 shows that the point estimate should not be treated as an exact measurement of the underlying hazard ratio.
These distinctions are especially important in a diagnostic trial. An intervention that increases detection of a previously unrecognized condition can produce a higher observed event rate because more events are identified. The statistical analysis quantifies the difference in observed time-to-detection; it does not by itself establish that the intervention changes the underlying biological incidence of AF.
8. Serious Adverse Events
The ClinicalTrials.gov record reports serious adverse events by randomized arm as affected participants divided by participants at risk.
| Arm | Serious adverse events | Affected / at risk |
|---|---|---|
| Reveal LINQ™ Insertable Cardiac Monitor | Serious adverse events | 58/242 |
| Control Arm | Serious adverse events | 65/250 |
These figures are reported as affected participants over participants at risk. They are presented descriptively here; the ClinicalTrials.gov record does not provide a formal statistical comparison of serious adverse events.
9. Statistical Methods Explained
Why was a time-to-event method used?
The primary endpoint concerns the first occurrence of AF during a defined follow-up period. Time-to-event methods preserve information about when the event occurred and can accommodate participants whose observation ends before the event occurs.
What does a hazard ratio of 7.4 mean?
A hazard ratio of 7.4 indicates a substantially higher estimated instantaneous rate of the first qualifying AF event in the Reveal LINQ arm relative to the control arm under the reported analysis. It does not mean that 7.4 times as many people necessarily had AF by 12 months.
Why is the confidence interval so important?
The point estimate alone does not show how precisely the treatment effect has been estimated. The 95% CI of 2.6–21.3 gives a range of values compatible with the statistical estimation framework. Its width indicates substantial uncertainty about the exact magnitude of the hazard ratio even though the interval is entirely above 1.
What does the log-rank test contribute?
The log-rank test provides a formal comparison of the time-to-event experience between the two randomized groups. It evaluates the pattern of observed event occurrence over follow-up rather than simply comparing one final proportion.
Why isn't the P-value an effect-size measure?
The P-value <0.001 indicates strong evidence against the stated equal-hazard null hypothesis under the statistical model and testing framework. It does not quantify how large the difference is. The hazard ratio describes relative magnitude, while the confidence interval describes uncertainty around that estimate.
Does a hazard ratio automatically prove proportional hazards?
No. A hazard ratio is a useful summary of relative event rates, but interpretation of a single hazard ratio is most straightforward when the relative hazards are reasonably stable over time. The ClinicalTrials.gov record does not report a formal proportional-hazards assessment, so this page does not claim that the assumption was verified.
What does the AF endpoint actually count?
The registry definition is specific: AF is an event lasting more than 30 seconds, and the first AF episode detected and adjudicated by the endpoint adjudication committee is used for the analysis. Thus the primary analysis is based on time to the first qualifying adjudicated episode rather than the total number of AF episodes.
10. Analysis Population and Event Definition
The ClinicalTrials.gov record identifies the randomized enrollment as 496 participants and identify the primary comparison as Reveal LINQ™ Insertable Cardiac Monitor versus Control Arm. The formal primary analysis is a superiority comparison using a log-rank test and hazard ratio.
| Analysis element | Reported information |
|---|---|
| Enrollment | 496 |
| Randomization | Yes |
| Arms | 2 |
| Primary endpoint | Rate of AF through 12 months |
| Event definition | AF event lasting more than 30 seconds |
| Event used | First AF episode detected and adjudicated by the endpoint adjudication committee |
| Analysis type | Time-to-event |
| Comparison method | Log-rank test |
| Effect measure | Hazard ratio |
| Hypothesis type | Superiority |
| Confidence interval | 95%, two-sided |
The ClinicalTrials.gov record does not report a separate analysis population definition, censoring rules, missing-data imputation procedure, stratification variables, or Bayesian analysis. Those elements are therefore not added to this analysis.
11. Missing Data, Censoring, and Follow-Up
Because the primary endpoint is a time-to-event endpoint, participants who do not experience the defined AF event during their observed follow-up can contribute information up to the point at which their observation ends. This is the statistical role of censoring in Kaplan-Meier analysis.
This distinction is important because survival estimates depend not only on the observed events but also on how follow-up is defined and how incomplete observation is handled. A statistical analysis should reproduce those rules from the protocol or statistical analysis plan when they are available.
12. Randomization and What It Contributes
Randomization is a central feature of the trial design because it creates the comparison between the Reveal LINQ monitoring arm and the control arm without allowing investigators to choose treatment assignment for individual participants.
Randomization does not guarantee identical groups at baseline, nor does it eliminate every source of uncertainty. Its principal statistical role is to support a causal comparison by making treatment assignment independent of measured and unmeasured baseline characteristics in expectation.
In STROKE-AF, the randomized comparison is paired with objective endpoint adjudication. However, the registry explicitly notes that assignment to ICM versus control was not blinded. That distinction matters when considering potential sources of bias, even though the study endpoint was adjudicated using objective criteria.
13. Limitations
- Subjective attribution of stroke mechanism: the registry notes that attribution of stroke mechanism is subjective and may have led to enrollment of a population at higher risk of underlying embolism.
- Real-world relevance of attribution: the registry also states that this attribution reflects real-world practice and that the sample is representative of patients in whom ICM decisions need to be made. This is the investigators' stated interpretation of the population and is presented here as an attributed claim rather than an independent conclusion.
- Unblinded assignment: assignment to ICM versus control was not blinded. The registry notes, however, that the study endpoint was adjudicated using objective criteria.
- Precision of the hazard ratio: the 95% CI of 2.6–21.3 is considerably wider than the point estimate alone might suggest, indicating uncertainty about the exact magnitude of the relative effect.
- Diagnostic interpretation: a higher detected AF rate with continuous monitoring reflects increased detection of qualifying AF events. The reported analysis by itself does not establish a corresponding reduction in stroke or another clinical outcome.
- Unreported analytical details: the ClinicalTrials.gov record does not specify detailed censoring rules, missing-data procedures, stratification factors, or formal proportional-hazards diagnostics.
- Serious adverse events: the ClinicalTrials.gov record is descriptive by arm and does not include a formal statistical comparison of serious adverse events.
14. Why This Trial Matters Statistically
STROKE-AF is a useful statistical teaching case because the primary endpoint combines a clinically specific event definition with a time-to-event analysis and a diagnostic monitoring intervention. The trial illustrates how detection intensity can affect an observed event endpoint and why the interpretation of a hazard ratio requires attention to the endpoint definition.
| Concept | How it appears in STROKE-AF |
|---|---|
| Randomization | Participants were randomized between Reveal LINQ™ monitoring and a control arm. |
| Parallel design | The registry identifies a parallel two-arm design. |
| Time-to-event endpoint | Time to the first qualifying AF episode is analyzed through 12 months. |
| Kaplan-Meier estimation | The registry analysis specifies a 12-month Kaplan-Meier estimate for each arm. |
| Log-rank test | The formal comparison uses a log-rank test. |
| Hazard ratio | The treatment effect is reported as HR 7.4 with a 95% CI of 2.6–21.3. |
| Confidence interval | The two-sided 95% CI quantifies uncertainty around the hazard-ratio estimate. |
| P-value | The reported two-sided P-value is <0.001. |
| Endpoint adjudication | The first qualifying AF episode is detected and adjudicated by an endpoint adjudication committee. |
| Unblinded assignment | The registry identifies the study as having no masking. |
| Diagnostic trial | The primary purpose is classified as diagnostic. |
15. Interpreting the Trial Without Overreading the Result
The randomized comparison produced a hazard ratio of 7.4 for the first qualifying AF event, with a two-sided 95% CI of 2.6–21.3 and a reported P < 0.001 using a log-rank test.
The result provides evidence of a difference in the time-to-detection of qualifying AF between the monitoring and control groups. The confidence interval indicates that the exact magnitude of that difference is substantially uncertain despite the strength of the statistical evidence against the equal-hazard null hypothesis.
The reported primary endpoint does not by itself quantify reduction in recurrent stroke, mortality, or another clinical outcome. It also does not establish that the biological incidence of AF was changed by monitoring. The direct statistical result concerns detection of the defined AF endpoint through 12 months.
The hazard ratio is relative. For clinical interpretation, an analyst would ordinarily examine the Kaplan-Meier estimates and absolute event probabilities at clinically relevant time points alongside the hazard ratio and confidence interval. The ClinicalTrials.gov record does not provide the arm-specific 12-month Kaplan-Meier percentages, so they are not added here.
16. ClinicalTrials.gov Registry Caveats
The registry's registry-reported limitations provide two important cautions for interpretation.
Stroke mechanism attribution
The attribution of stroke mechanism is described as subjective and may have resulted in enrollment of a population at higher risk of underlying embolism.
Blinding
Assignment to ICM versus control was not blinded, while the study endpoint was adjudicated using objective criteria.
The registry also states that the attribution reflects real-world practice and that the sample is representative of patients in whom ICM decisions need to be made. These statements are part of the registry's stated limitations and interpretation; they are not independently validated here.
17. Trial Timeline
Trial start
STROKE-AF began enrollment in March 2016.
Primary completion
The registry lists August 3, 2020 as the primary completion date.
Registry status
The ClinicalTrials.gov record classifies the trial as completed.
18. Secondary Endpoint Results
The ClinicalTrials.gov record indicates that 2 outcome measures were posted and that 1 statistical analysis was posted. The formal statistical analysis in the ClinicalTrials.gov record is the primary endpoint analysis described above.
19. Multiplicity, Interim Analysis, and Bayesian Methods
The ClinicalTrials.gov record does not report a multiplicity procedure, interim-analysis plan, alpha-spending method, or Bayesian statistical method. These design topics are therefore not attributed to STROKE-AF on this page.
| Design topic | Information in the ClinicalTrials.gov record |
|---|---|
| Non-inferiority margin | Not reported in the ClinicalTrials.gov record; the hypothesis type is superiority. |
| Crossover | Not reported in the ClinicalTrials.gov record. |
| Factorial design | Not reported; the design model is parallel. |
| Multiplicity adjustment | Not reported. |
| Interim analysis | Not reported. |
| Missing-data imputation | Not reported. |
| Stratification | Not reported. |
| Bayesian methods | Not reported. |
This absence of information is itself relevant when reconstructing a statistical analysis. A trial-results page should not convert a common method for this endpoint into a claim that the specific trial used that method unless the registry or other permitted source data support it.
20. What the Confidence Interval Tells Us
95% confidence interval for the hazard ratio
Point estimate: 7.4 · Two-sided 95% CI
The interval is important because a single estimate of 7.4 can create an impression of greater precision than the underlying statistical evidence supports. The lower confidence limit is 2.6, while the upper limit is 21.3. Thus the data are compatible with a range of substantially different magnitudes of relative hazard.
The interval also stays above the null value of 1.0. That feature is consistent with the reported two-sided P < 0.001 for the equal-hazard hypothesis. The two statistics answer related but different questions: the confidence interval describes uncertainty around the effect estimate, while the p-value summarizes evidence against the null hypothesis.
21. Endpoint Adjudication and Measurement
The primary endpoint is unusually explicit about how an event enters the analysis. AF must last more than 30 seconds, and the first AF episode detected and adjudicated by the endpoint adjudication committee is used.
This definition establishes the measurement rule before considering the statistical comparison. It is important because the statistical model can only analyze the event as defined by the study.
The registry's limitation regarding lack of blinding is also relevant here. Because assignment was not blinded, the possibility of differences in observation or care processes must be considered when interpreting a diagnostic endpoint. The registry simultaneously notes that the endpoint was adjudicated using objective criteria, which provides an important methodological safeguard.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: STROKE-AF, NCT02700945.
- PubMed: PMID 37902733.
- PubMed: PMID 36374508.
- PubMed: PMID 35369714.
- PubMed: PMID 34061145.
Continue through the Clinical Biostats methods library
Explore the statistical concepts that underlie randomized time-to-event analyses, including hazard ratios, Kaplan-Meier estimation, confidence intervals, and log-rank testing.
25. Record Summary
STROKE-AF provides a clear example of how a diagnostic intervention can be evaluated using a randomized time-to-event framework. The primary endpoint was the rate of AF through 12 months, with AF defined as an event lasting more than 30 seconds and the first detected and adjudicated episode used for analysis. The reported statistical method was a log-rank test with a hazard ratio as the effect measure.
The primary reported estimate was a hazard ratio of 7.4 with a 95% confidence interval of 2.6–21.3 and a two-sided P < 0.001. Statistically, this indicates a higher estimated hazard of first qualifying AF detection in the Reveal LINQ monitoring group than in the control group under the reported analysis. The result should be interpreted together with the endpoint definition, the time-to-event framework, the confidence interval, the unblinded design, and the registry's stated limitations concerning attribution of stroke mechanism.