Longitudinal changes in wearable-captured physiology before sudden cardiac death
A single case report: in a 76-year-old man with advanced structural heart disease who died minutes after high-intensity indoor cycling, a retrospective change-point analysis of his Apple Watch and Oura data found a coordinated, sustained step-change from his own baseline most clustering 4-6 months before death, alongside a fatigue complaint; it does not show that wearables predict cardiac events
Consumer wearables record cardiovascular metrics continually, yet their use for detecting physiological changes that precede cardiac events remains limited.
This case report describes a 76-year-old man with structural heart disease who experienced sudden cardiac death minutes after high-intensity indoor cycling, and looks back over his Apple Watch, Oura Ring and Peloton records for change-points before the event. Full text read (owner's PDF, 2026-10-03); supplemental tables and methods were not available.
Participants
n = 1: a 76-year-old man with structural heart disease.
Full text, checked 2026-10-03: Past medical history — palpitations and one syncope in 2018; cardiac MRI in 2018 showed moderate LV dilation with mildly reduced systolic function (EF approximately 49%), mild-to-moderate mitral and aortic regurgitation and transmural late gadolinium enhancement consistent with prior myocardial infarction; coronary calcium score 1,350; runs of supraventricular tachycardia; ablation of AV nodal reentrant tachycardia in 2018; worsening mitral regurgitation on echocardiography in 2022; hernia repair in April 2024 followed by 6 weeks of strict rest.
Protocol
Retrospective review of clinical records alongside continually recorded data from a Peloton Bike, an Oura Ring Gen 3 and an Apple Watch Series 5 and 7 (full text, checked 2026-10-03: Investigations; Figure 3).
Full text, checked 2026-10-03: Investigations — of 39 candidate wearable-derived signals, 35 were eligible; analysis focused on an approximately 18-month window from October 15, 2023; change-point detection by Pettitt's test with a confirmatory bootstrap cumulative sum, Benjamini-Hochberg correction; 10 core signals carried into the detailed analysis, each compared 30 days before vs 30 days after its change-point (Cohen's d); a second change-point tested by binary segmentation.
Full text, checked 2026-10-03: Workout intensity and heart rate kinetics — all 530 recorded Peloton sessions (2018-2025) examined; heart-rate kinetics from 145 workouts with adequate heart-rate data compared with the American Heart Association recommended exercise ceiling for his age.
No autopsy was performed.
Key Results
The event (abstract): sudden cardiac death minutes after high-intensity indoor cycling. Full text, checked 2026-10-03: History of presentation — a 30-minute high-intensity session, his 530th recorded Peloton workout; three minutes later the Apple Watch detected a fall; the defibrillator analysis showed ventricular fibrillation; no return of spontaneous circulation after 39 minutes of advanced cardiac life support.
Abstract: a change-point analysis showed a coordinated step-change, mostly bunched 4-6 months ahead of the sudden cardiac death: heart rate variability fell by half, resting and nocturnal heart rate went up, and deep sleep and the 6-minute walk distance went down.
Abstract: these shifts came at the same time as a complaint of fatigue and a murmur, which led to echocardiography: the left ventricle was severely dilated with ejection fraction preserved (approximately 60%), and there was severe mitral, moderate aortic and trace tricuspid regurgitation, in keeping with advanced structural heart disease.
Full text, checked 2026-10-03: Investigations; Figure 4 — a significant change-point in 30 of 35 signals, all surviving Benjamini-Hochberg correction.They clustered in two windows: mid-April through July 2024 (gait, sleep-duration and activity signals, consistent with the hernia repair and convalescence) and September 23 to December 4, 2024 (the "autumn 2024 step-change": nocturnal respiratory rate, then HRV and resting and nocturnal heart rate between October 3 and October 25, then sleep architecture, blood oxygenation and 6-minute walk distance).
Full text, checked 2026-10-03: Investigations (30 days before vs after each change-point) — Apple Watch daytime SDNN 179 → 91 ms (d = −2.2); Oura nocturnal RMSSD 180 → 63 ms (d = −2.8); resting heart rate 44 → 49 beats/min (d = +2.2); average nocturnal heart rate 50 → 52.5 beats/min (d = +1.1); deep sleep 39 → 18 min (d = −1.4); 6-minute walk distance 465 → 415 m (d = −2.4).Nocturnal respiratory rate fell and blood oxygen saturation rose, directions without a straightforward decompensation reading. HRV, respiratory rate and deep sleep stayed suppressed to the end of the record and nocturnal heart rate rose until death.
Full text, checked 2026-10-03: Workout intensity and heart rate kinetics; Figure 6 — in the final 79 days he completed 69 indoor sessions, more than his total for all of 2024 (59), including a 24-day streak with few rest days; 13 of the 69 ranked above the 75th percentile of all his workouts by average power, and the final workout ranked in the top 10%.Across 145 workouts the median heart-rate trajectory converged on, and in longer workouts surpassed, the AHA recommended ceiling for his age (128 beats/min); the 13 hardest 2025 workouts frequently exceeded 140 beats/min, and the final workout approached 150 beats/min, above the age-predicted maximum of 144 beats/min (220 − 76).
Authors' own limit (full text, checked 2026-10-03: Conclusions): the case does not show that wearables predict sudden cardiac death, since a retrospective n-of-1 cannot; that would need prospective personal-baseline change detection with prespecified alarm rules validated against adjudicated outcomes.
Practical Application
**"Can my watch or ring warn me before a heart problem?" In one 76-year-old man with advanced structural heart disease who died suddenly minutes after high-intensity indoor cycling, a change-point analysis done in hindsight found a coordinated, sustained step-change in his wearable data with most of the changes bunched 4-6 months before death — his heart rate variability fell by half, resting and nocturnal heart rate climbed, and deep sleep and the 6-minute walk distance dropped — at the time of a fatigue complaint and before the echocardiography findings.The signal was the sustained shift across several measures away from his own baseline, not any single value; a single case analysed knowing the outcome cannot show that such shifts predict cardiac events.**
Inference, not tested here: in an older athlete or one with known heart disease, a sustained shift over weeks in several nightly signals away from their own baseline, together with unexplained fatigue, is a reason for a medical check rather than a lighter week; this case is one illustration of that pattern, not evidence that the check would have changed the outcome.
Never read a single low HRV value as a cardiac warning: the authors themselves note there is no reference range for a single HRV value (full text, checked 2026-10-03: Discussion), HRV moves in either direction with ordinary training adaptation in elite endurance athletes (REF-361) and falls after even small amounts of alcohol (REF-1011).
Treat "deep sleep" on its own as a weak signal: consumer wearables stage sleep poorly against polysomnography (REF-364); here it was one of several measures moving together.
Intensity in known heart disease: in this man intra-workout heart rate repeatedly went above the age-adjusted ceiling recommended for him while training volume and intensity escalated in his final months; the authors suggest wearable heart-rate data could inform individual intensity targets, which they did not test.Not from this paper; the base's own practice: for an athlete with known heart disease, intensity limits come from their cardiologist, and the assistant does not set them.
Limitations
n = 1, and the analysis was done in hindsight with the outcome known; the paper calls it a retrospective n-of-1 that cannot show prediction.
Multiplicity: change-points were found in 30 of 35 signals; the 10 core signals were chosen after looking, and an earlier cluster (April-July 2024) had an ordinary explanation (hernia repair and convalescence).
No data on how often such step-changes occur in people who do not have an event — no false-positive rate, no sensitivity, no specificity.
No autopsy; the mechanism of death is not established, and the case cannot separate the structural disease, the escalating training and the final high-intensity session as causes.
Population: 76 years old with prior infarct scar, LV dilation and mitral/aortic regurgitation; this does not transfer to healthy or younger athletes.
Supplemental methods and tables (signal definitions, medication history, per-signal results) were not available for this entry.
Inference, not tested here: the baseline HRV values (daytime SDNN 179 ms, nocturnal RMSSD 180 ms) are far above what is usual for a 76-year-old in sinus rhythm; in a man with documented supraventricular runs, ectopic or irregular beats may inflate wearable HRV, so the physiological meaning of the HRV fall is uncertain. The paper does not discuss this.
Inference, not tested here: medication changes in 2024 (tamsulosin started in May, stopped on October 31) fall near the step-change window and are not analysed as a confounder.
How this entry was checked and changed
Why this level
the rubric row for a single case study is C+; graded one step down because the analysis was done in hindsight knowing the outcome, change-points were found in 30 of 35 signals tested, and there is no comparison with people who had no event (no false-positive rate)