Apple Heart Study: Assessment of Wristwatch-Based Photoplethysmography to Identify Cardiac Arrhythmias: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Apple Heart Study: Assessment of Wristwatch-Based Photoplethysmography to Identify Cardiac Arrhythmias (NCT03335800), a na na study. The primary outcome is Atrial Fibrillation (AF) of Greater Than 30 Seconds.

2. Clinical Question

The primary outcome focuses on a selected group: participants who received an irregular pulse notification and returned an ECG patch. It asks whether atrial fibrillation was detected during the subsequent monitoring period for this group. It does not, by itself, describe how often the app identified every case of atrial fibrillation among all participants, or how often a notification occurred in participants without atrial fibrillation.

That distinction matters when interpreting a device-identification study. A finding about notified participants who returned monitoring data addresses confirmation in that subset. A complete assessment of identification would also require information about participants who were not notified, those who did not return a patch, and the results of monitoring across the relevant study population. Those results and details are not reported in the registry record.

3. Trial Design

The Apple Heart Study was a completed, single-group study of the Apple Heart Study App. The record lists an enrollment of 419927 and identifies Apple Inc. as the sponsor. The study began on 2017-11-29, reached primary completion on 2019-02-21, and has a completed status.

The intervention is listed as a device, and the intervention name is Apple Heart Study App. The allocation field is NA, and the masking field is NONE. A single-group design does not establish a randomized treatment contrast. The record also explicitly identifies that there was no randomized comparison and that the arms are not a like-for-like control comparison.

The study’s stated purpose was to evaluate whether the app could use data collected on the Apple Watch to identify irregular heart rhythms, including rhythms from potentially serious heart conditions such as atrial fibrillation. The record says up to 500,000 people could participate; the listed enrollment was 419927. It does not report further details about recruitment, participant characteristics, or how the enrolled group compared with people who did not participate.

The primary outcome was defined for participants who returned an ECG patch. It concerns atrial fibrillation lasting more than 30 seconds detected during ambulatory ECG monitoring, for up to 8 days. Secondary outcome names include concordant atrial fibrillation with an app algorithm notification and self-reported contact with a health care provider. The record lists four outcome measures, but does not report the results for those outcomes. Further protocol details, including the study’s statistical plan and the reason for its design choices, are not reported in the registry record.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Overall Study

Analysis population for the primary outcome, as defined in the registry: Participant who returned ECG Patch

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

6. Primary Endpoint

The registry lists the primary outcome as Atrial Fibrillation (AF) of Greater Than 30 Seconds, measured over During ambulatory ECG monitoring (up to 8 days).

The primary endpoint is the proportion of notified participants who returned an ECG patch and had atrial fibrillation detected on the patch. The specified rhythm duration is greater than 30 seconds, and the monitoring period is ambulatory ECG monitoring for up to 8 days. The endpoint title refers to atrial fibrillation of greater than 30 seconds; its description clarifies that the measure is among participants who received an irregular pulse notification and returned the patch.

This endpoint is relevant to the study’s stated identification question because it compares an app notification with rhythm information collected by ambulatory ECG monitoring. In general, an endpoint based on a follow-up measurement can indicate how often a finding was confirmed among the people who completed that measurement. It is not a complete measure of overall identification performance unless the results for the broader study population, including people without a notification and people without returned monitoring data, are also available.

The registry record specifies a 97.5% confidence interval as the dispersion measure, but does not report the endpoint estimate, the interval limits, or the number of participants in the endpoint population. Thus the outcome definition is available, but its observed value and precision are not reported in the registry record. The record does not give further details about how ECG findings were assessed or how incomplete monitoring was handled.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

What does the endpoint proportion represent?

The primary outcome is a proportion: the fraction of a defined group who meet an outcome criterion. In this study, the defined group is participants who received an irregular pulse notification and returned an ECG patch. The criterion is atrial fibrillation lasting greater than 30 seconds detected during ambulatory ECG monitoring, which could continue for up to 8 days.

In general, a proportion is interpretable only alongside its numerator and denominator. The numerator is the number meeting the endpoint definition, and the denominator is the number assessed in the defined group. Neither the primary outcome estimate nor the number of participants contributing to it is reported in the registry record. The enrollment of 419927 should not be mistaken for the endpoint denominator: the outcome applies only to a specified subset, and its size is not reported.

What would a confidence interval add?

The registry record names a 97.5% confidence interval as the dispersion measure for the primary outcome. In general, a confidence interval communicates statistical uncertainty around an estimate, conditional on the data and method used. A narrower interval indicates greater precision under that method; a wider interval indicates more uncertainty. A confidence interval does not describe the spread of individual participant outcomes and does not guarantee that the true value lies in the interval.

The interval’s numerical limits are not reported in the registry record. The record also does not report the interval calculation method or explain why a 97.5% interval was specified. Without the estimate and interval limits, the endpoint’s precision cannot be assessed from the registry entry.

What can the design tell us?

The record classifies the study as single-group and lists no masking. Allocation is NA. These fields do not describe a randomized comparison between app use and a control condition. General statistical interpretation: a single-group result can describe what was observed in the measured group, but it cannot by itself separate an app-related effect from other explanations or estimate a randomized difference between groups.

The outcome is measured in notified participants who returned a patch. This selection is central to interpretation: results in that subset do not automatically describe all participants, people who were not notified, or people who did not return a patch. The numbers in those groups and the results for them are not reported in the registry record.

How should planning calculations be read?

Planning illustrations based on an observed effect are not observed power and are not evidence that the original study was designed around that effect. The stated planning assumptions are a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation, with the effect to detect defined as a stated fraction of the observed effect. The trial itself is single-group, so these planning assumptions do not describe a randomized comparison conducted in this study. The registry record does not report the original sample-size justification or its design assumptions.

In general, a planning calculation estimates the sample size needed under specified assumptions. Its result depends on those assumptions and does not show that a study had a particular probability of detecting an effect after the data were collected. Any planning illustration should therefore be distinguished from the observed study enrollment and from the unavailable primary endpoint result.

What analysis details are missing?

The registry record lists four outcome measures and zero analyses. It does not report the statistical test or estimation procedure used for the primary endpoint, any adjustment for participant characteristics, or the handling of incomplete or uninterpretable patch recordings. It also does not report analysis populations beyond defining the primary outcome population as participants who returned an ECG patch.

General statistical explanation: missing outcome data can affect both the number of participants included in an analysis and how well the analysed group represents the intended population. A returned patch is not necessarily the same as a complete, interpretable recording, but the record does not describe completeness or interpretability. No assumptions about such cases can be made from the registry entry.

Because the estimate, denominator, interval limits, and analysis details are absent, the registry record supports understanding the planned outcome definition and broad design, but not a numerical assessment of the primary result or its precision.

10. Exact Estimates for Each Arm

No two-arm comparison is shown for this trial, but each arm's result can still be summarized with exact interval estimates computed from the posted counts. The Clopper-Pearson interval is exact and conservative; the Wilson interval is a common closer-to-nominal alternative; the Bayesian credible interval uses a uniform Beta(1, 1) prior. For small groups the three can differ noticeably.

R code and output

arm    <- c("Apple Heart Study (AHS) App Users")
n      <- c(450)
events <- c(2)
for (i in seq_along(n)) {
  ci <- binom.test(events[i], n[i])$conf.int          # Clopper-Pearson exact interval
  cr <- qbeta(c(0.025, 0.975), 1 + events[i], 1 + n[i] - events[i])   # Bayesian, uniform Beta(1,1) prior
  wi <- (function(x, n) { p <- x / n; z <- qnorm(0.975); d <- 1 + z^2 / n
    c((p + z^2 / (2 * n) - z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d, (p + z^2 / (2 * n) + z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d) })(events[i], n[i])
  cat(sprintf("ARM|%s|%d|%d|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f\n", arm[i], n[i], events[i], 100 * events[i] / n[i],
              100 * ci[1], 100 * ci[2], 100 * wi[1], 100 * wi[2], 100 * cr[1], 100 * cr[2]))
}

ARM  Apple Heart Study (AHS) App Users  450  2  0.4  0.1  1.6  0.1  1.6  0.1  1.6

11. Multiplicity, Missing Data and Analysis Populations

The record lists four outcome measures and names two secondary outcomes: concordant atrial fibrillation with an app algorithm notification and self-reported contact with a health care provider. It identifies a primary endpoint concerning atrial fibrillation lasting greater than 30 seconds in notified participants who returned an ECG patch. Numerical results for the primary and secondary outcomes are not reported in the registry record.

When a study evaluates multiple outcomes, repeated statistical testing can increase the chance of a seemingly notable result occurring by chance. General statistical practice may address this issue by identifying a primary outcome in advance, adjusting for multiple comparisons, or interpreting secondary analyses as exploratory. The registry record does not report whether any multiplicity adjustment was planned or used, how the four outcome measures relate to the named outcomes, or whether the outcome hierarchy changed.

The record lists zero analyses. It does not report the statistical analysis plan, the timing of outcome specification, or any interim analyses. These details are not reported in the registry record, so no conclusion can be drawn about the approach to multiplicity.

The primary outcome population is described as participants who returned an ECG patch. The record does not report the number who returned a patch, the number with usable recordings, or how missing or incomplete data were handled. Other analysis populations are not reported in the registry record. In general, differences between the enrolled population and the analysed population can affect how broadly an estimate applies; without those counts and methods, that issue cannot be quantified for this study.

12. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 5% reporting threshold.

Most frequent serious adverse event terms

Serious events with the highest arm-level rate: atrial fibrilation (0.0%), chest pain (0.0%), syncope (0.0%), pancreatitis (0.0%), allergic reaction (0.0%).

The registry record lists one group, Apple Heart Study (AHS) App Users, with 419927 at risk and 62 serious adverse events. It also lists 2 deaths. These are safety figures recorded for the study group; the registry entry does not establish that the app caused any event or death.

The serious-event terms listed include Atrial Fibrilation, Chest Pain, Syncope, Pancreatitis, Allergic reaction, and Atrial Flutter. Each listed term has a displayed value of 0.0%. The record does not explain how those term percentages relate to the overall serious-event count of 62. It does not report further event-level detail, timing, or a causal assessment in the information.

General interpretation: a serious adverse event count is not the same as a count of participants with a particular event, and a count alone does not establish causation. The record’s safety group does not provide a randomized control comparison, so it cannot support a comparative estimate of safety against a control group. The listed deaths should likewise be read as recorded in the group, not as evidence that the intervention caused them.

13. What the Result Does, and Does Not, Mean

The registry record establishes that the Apple Heart Study evaluated an app intended to use Apple Watch data to identify irregular heart rhythms. It describes a completed, single-group study with an enrollment of 419927 and defines a primary outcome based on atrial fibrillation detected during ambulatory ECG monitoring among notified participants who returned a patch.

The primary outcome result itself is not reported in the registry record. There is no numerical estimate, denominator, or 97.5% confidence interval limit to interpret. The record therefore does not establish how frequently atrial fibrillation was detected in the specified group or how precise an estimate might be. Secondary outcome results are also not reported in the registry record.

Even if the primary estimate were available, this endpoint would answer a focused question about notified participants who returned a patch. It would not alone describe performance among all app users. A broader assessment would require results for participants who were not notified, participants who did not return a patch, and the relevant monitoring outcomes across those groups. Such results are not reported in the registry record.

The single-group design also has no randomized control comparison. This means the registry entry does not provide a randomized estimate of the difference between app use and a control condition. The design and endpoint definition should not be confused with evidence of a causal effect.

General statistical interpretation: an outcome definition, study enrollment, and completed status are not substitutes for a reported result. The enrollment indicates the listed study size, but not the number included in the primary analysis. The registry information supports describing what the study set out to measure and the group to which its primary outcome applies; it does not support a numerical conclusion about the outcome or a claim about overall identification performance.

14. Why This Trial Matters Statistically

This study raises a useful statistical distinction between evaluating a signal in a selected group and evaluating identification across a full population. The primary endpoint concerns notified participants who returned an ECG patch. That can address whether atrial fibrillation was detected in that subset, but it does not alone describe results among all participants or people who were not notified.

For device-identification questions, the denominator and the pathway into analysis are especially important. The primary outcome population depends on both receiving a notification and returning a patch. The registry record does not report how many participants entered that group or how many had usable monitoring results. Without these counts, readers cannot assess the extent to which the endpoint population represents the enrolled group.

The study also illustrates why a clearly defined outcome and a numerical result are different kinds of information. The endpoint duration, monitoring method, and target population are described, while the estimate and confidence interval are not reported in the registry record. A statistical interpretation requires both: the definition tells readers what was measured; the result and its precision tell readers what was observed and how uncertain that estimate is.

Finally, the single-group design matters. A randomized comparison is not reported, so the study record does not offer a like-for-like control contrast. Clear description of design, analysis population, missing data, and outcome precision is necessary to distinguish a focused measurement from broader conclusions about performance. Several of those analysis details and the primary result are not reported in the registry record.

15. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The largest interpretive limitation is that the primary endpoint estimate and its 97.5% confidence interval limits are not reported in the registry record. The number of participants contributing to the endpoint is also not reported. The outcome definition is available, but the posted record does not provide the numerical result needed to assess its frequency or precision.

The primary endpoint population is restricted to participants who received an irregular pulse notification and returned an ECG patch. The record does not report the number notified, the number returning a patch, or the number with complete and interpretable monitoring. It also does not report outcomes among participants without notifications. This prevents assessment of how representative the endpoint group was of all enrolled participants or how the result relates to identification across the full study group.

The study is single-group, with allocation listed as NA and masking listed as NONE. A randomized control comparison is not reported. The registry record also does not report participant characteristics, recruitment details, the statistical analysis plan, missing-data methods, or the rationale for the study’s sample size. These details should not be inferred from the enrollment figure or the completed status.

Safety interpretation is limited as well. The record lists 62 serious adverse events and 2 deaths, while the displayed serious-event term percentages are 0.0%. It does not explain the relationship between those entries or provide event-level causal assessments. The entries therefore do not establish that events were caused by the app.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Apple Heart Study: Assessment of Wristwatch-Based Photoplethysmography to Identify Cardiac Arrhythmias — NCT03335800.
• Registry record last updated 2020-03-30; status last verified 2020-03.
• Results first posted 2020-03-30.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

19. Record Summary

The Apple Heart Study was a completed single-group study sponsored by Apple Inc. It listed enrollment of 419927, an intervention named Apple Heart Study App, allocation as NA, and masking as NONE. The primary outcome was atrial fibrillation lasting greater than 30 seconds detected during ambulatory ECG monitoring for up to 8 days among notified participants who returned an ECG patch.

The registry record lists four outcome measures and zero analyses. It names secondary outcomes related to concordant atrial fibrillation with an app algorithm notification and self-reported contact with a health care provider. The primary endpoint estimate, confidence interval limits, analysis denominator, and secondary outcome results are not reported in the registry record.

The record lists 62 serious adverse events and 2 deaths among the app-user group. The study has no randomized, like-for-like control comparison. Accordingly, the registry entry describes the design and outcome definitions but does not provide the numerical primary result needed to assess its observed value or precision.