Arrhythmias & electrophysiology

Accounting for inconclusive results and repeated testing: A framework for evaluating wearable electrocardiogram diagnostic performance with application to Apple Watch and artificial intelligence-enhanced interpretation

Heart RhythmPublished Diagnostic accuracy study

Authors' conclusion

Naive reporting inflates performance, whiereas pragmatic reporting deflates it. AI-enhanced interpretation materially reduces inconclusive results and improves ITD accuracy, providing a more usable and reliable pathway for real-world AF detection.

Study context

Population
Participants with Apple Watch electrocardiogram (ECG) recordings for atrial fibrillation (AF) detection at a teaching hospital in Irelandafter exclusions
Participants
296 participants
Comparison
Apple Watch vs AI
Outcome
Wearable ECG diagnostic performance (sensitivity/specificity) in AF detection under naive (exclude inconclusive), pragmatic (count as incorrect) and ITD (permit 3 attempts) frameworks
Methodological detailsprospective, observational, single teaching hospital, Ireland
  • Prospective
  • Observational
  • Single teaching hospital, Ireland

Performance

As reported in the abstract

Sensitivity/specificity for AF detection
Test or modelNaive sensitivity/specificityPragmatic sensitivity/specificityIntention-to-diagnose (ITD) sensitivity/specificity
Apple Watch96.1%/97.9%78.1%/81.0%92.2%/91.0%
AINot reported in the abstract98.4%/96.1%98.4%/96.6%

Naive sensitivity/specificity

Apple Watch96.1%/97.9%
AINot reported in the abstract

Pragmatic sensitivity/specificity

Apple Watch78.1%/81.0%
AI98.4%/96.1%

Intention-to-diagnose (ITD) sensitivity/specificity

Apple Watch92.2%/91.0%
AI98.4%/96.6%

Further findings

Kappa

OutcomeApple WatchAIReduction in inconclusive outputs
Inconclusive outputs92%
RepeatabilitySubstantial for Apple Watch, near-perfect for AI0.770.96

Original publication

Accounting for inconclusive results and repeated testing: A framework for evaluating wearable electrocardiogram diagnostic performance with application to Apple Watch and artificial intelligence-enhanced interpretation. Heart Rhythm. Published 7 November 2025. doi:10.1016/j.hrthm.2025.11.003

Summarised from the published abstract; every number is checked against it

Get them every week

New studies on this topic in My Week and in your Monday e-mail.

We e-mail you a sign-in link and a code. No password, no spam.