Can the Apple Watch Detect Illness Before Symptoms Appear — What the Research Actually Shows

TL;DR

Research on populations of thousands confirms that wearable signals — elevated resting heart rate, lowered HRV, disrupted sleep — do shift in the days around illness onset. The science is real. But the honest answer to whether your Apple Watch can detect illness before you feel symptoms is: not reliably, not at the individual level, and not in any way Apple has claimed or validated. The studies that generated headlines were retrospective and population-scale; translating them to real-time individual alerts is a much harder problem. Apple's own Vitals app, introduced in watchOS 11, notifies you when multiple overnight metrics are unusual — but Apple explicitly states the feature is not intended for medical use.

---

What the research actually studied

The most frequently cited research on wearable illness detection comes from two groups: Michael Snyder's lab at Stanford and the Scripps Research Translational Institute.

In 2017, Snyder and colleagues published in PLOS Biology the results of tracking 60 participants wearing between one and seven biosensors continuously. The team collected nearly two billion measurements, including heart rate, skin temperature, blood oxygen, and activity. They showed that deviations from each individual's personal baseline could be detected and correlated with illness events — including Snyder's own Lyme disease infection, which showed as an abnormal heart rate and blood oxygen pattern on a flight before he developed any symptoms. The study's conclusion was appropriately framed as proof-of-concept: given a well-established personal baseline, wearable-detected outliers can correlate with health events. Eric Topol of the Scripps Research Institute, commenting on the work, called the prospect of detecting infections before they happen "very provocative."

The same group subsequently published research in Nature Biomedical Engineering applying the same logic to COVID-19: showing that smartwatch data could be associated with infection in advance of symptom onset (Snyder et al., Nat Biomed Eng, December 2020; DOI: 10.1038/s41551-020-00640-6).

The most rigorous quantitative data came from Scripps. Gadaleta, Radin, and colleagues published a study in npj Digital Medicine in December 2021 drawing on 38,911 participants from the DETECT study, of whom 1,118 tested positive for COVID-19 by PCR swab. Using a gradient boosting model trained on passively collected smartwatch and fitness-tracker data, they achieved an AUC of 0.83 in symptomatic individuals when all data was available. But when they restricted to passive sensor data before the test date only — the scenario that actually maps to pre-symptomatic detection — performance dropped to an AUC of 0.70 for all participants regardless of whether they had reported any symptoms.

An AUC of 0.70 means the model correctly ranked a COVID-positive person above a COVID-negative person about 70% of the time. That is genuinely above chance — but it also means a meaningful rate of both false positives and false negatives. For individual use in daily life, where the probability of having any given illness on any given morning is low, the practical signal-to-noise ratio is considerably worse than that headline figure suggests.

---

The signal is real — the problem is specificity

Why can't you rely on it? Because the physiological signals that shift during illness are not unique to illness.

Resting heart rate rises when your immune system mounts a response to infection — but it also rises after alcohol, after a hard training session, with poor sleep, or under sustained psychological stress. HRV drops in the days preceding illness — but HRV also drops the morning after wine, after a late flight, or during a difficult work week. Wrist skin temperature can elevate with fever — but as Apple notes explicitly, it also responds to alcohol, exercise close to bedtime, a warm bedroom, and a loose watch fit.

This is the specificity problem. The signal is real in aggregate — across thousands of people and many illness events, the pattern is statistically detectable. But for any one individual on any one morning, an elevated resting heart rate and a flagged wrist temperature are far more likely to reflect last night's dinner and an open window than the onset of influenza.

The Scripps paper measured this honestly. The stronger AUC numbers required self-reported symptoms to be included in the model — meaning the algorithm was partly detecting the fact that someone had already noticed they felt unwell. Strip out the symptoms and restrict to pre-test passive data only, and AUC falls to 0.70. That gap between 0.83 and 0.70 represents exactly how much informational load was being carried by the symptoms themselves.

---

What Apple Watch actually does with this

Apple introduced the Vitals app with watchOS 11 in September 2024. It monitors five overnight metrics: heart rate, respiratory rate, blood oxygen, wrist temperature, and sleep duration. After seven nights of sleep data, it establishes a personal typical range for each. If multiple metrics fall outside that range on the same night, you receive a notification the following morning along with context noting that "factors such as your medications, elevation, alcohol intake, or even illness can impact your metrics."

That framing is precise and defensible: illness is named as one of several possible explanations, not the conclusion. Apple then includes a footnote that deserves to be read in full: "Vitals app measurements are not intended for medical use. Please consult your healthcare provider prior to making any decisions related to your health."

This is not boilerplate padding. It reflects a genuine regulatory reality: Apple Watch has not been validated or cleared for illness detection. The Vitals app is a wellness awareness tool. It can prompt you to pay attention to how you feel. It cannot tell you what is causing the readings it flags.

---

Why individual prediction remains an unsolved problem

There is a meaningful gap between what population studies show and what individual prediction requires.

A study showing that, across thousands of people, resting heart rate tends to be elevated in the days before a confirmed COVID-19 test does not tell you that your elevated resting heart rate tomorrow morning means you are getting sick. It tells you that elevated resting heart rate is associated with illness on a population level — and also with many other things.

Closing that gap would require, at minimum: a model trained specifically on your physiology over a long enough period to distinguish your illness pattern from your alcohol pattern and your overtraining pattern; a disease prevalent enough in your environment to make a positive alert statistically meaningful; and prospective clinical validation — not retrospective analysis of data from people already known to have tested positive.

None of that exists yet for Apple Watch or any consumer wearable. The research community is actively working on it, and the direction is genuinely promising. But "promising in peer-reviewed research" and "validated for individual clinical use" are different categories.

---

What you can usefully do with the data

The fact that individual-level validation does not yet exist does not make the data useless — it changes how you should use it.

The most defensible role is as a prompt to pay attention. If your Vitals app flags multiple metrics as unusual and you also happen to feel slightly off, that convergence is meaningful context: rest, avoid strenuous exercise, and monitor how you feel over the next 24 hours. It is not a diagnosis and should not replace your own assessment of your symptoms.

What it should not do is replace clinical judgment in either direction — neither "my watch says I'm fine, so I won't see a doctor" nor "my watch flagged me, so I must be getting sick." Both readings misapply what the data can actually support.

The useful frame is personal baseline awareness over time. Across weeks and months, you learn what your overnight metrics look like when you are well-rested and unstressed. Deviations from that pattern — especially when multiple metrics move together and there is no obvious lifestyle explanation — are worth noting. Treating them as a wellness signal, not a medical alert, is both accurate to what the data represents and consistent with what Apple has validated.

---

Where Sam Health fits in

Sam surfaces your Apple Watch overnight metrics — HRV, resting heart rate, wrist temperature, sleep, and respiratory rate — in a single timeline, so you can see when multiple signals diverge from your personal baseline at the same time. That multi-metric view is exactly what the research suggests is more informative than any single metric in isolation. You can explore the full picture of what your Apple Watch sensors collect in our complete sensor breakdown for 2026.

Try Sam Health

Sources

• Track your overnight vitals with Apple Watch — Apple Support, published September 2025. Accessed 16 May 2026.
• Track your nightly wrist temperature changes with Apple Watch — Apple Support, published October 2025. Accessed 16 May 2026.
• Wearable sensors can tell when you are getting sick — Stanford Medicine News, 12 January 2017. Reporting on: Li X, Dunn J, Salins D et al., PLOS Biology, 2017. Accessed 16 May 2026.
• Pre-symptomatic detection of COVID-19 from smartwatch data — Snyder MP et al., Nature Biomedical Engineering, Vol. 4, pp. 1208–1220, December 2020. DOI: 10.1038/s41551-020-00640-6. Accessed 16 May 2026.
• Passive detection of COVID-19 with wearable sensors and explainable machine learning algorithms — Gadaleta M, Radin JM, Topol EJ, Quer G et al., npj Digital Medicine, 4(1):166, December 2021. DOI: 10.1038/s41746-021-00533-1. PMID: 34880366. Accessed 16 May 2026.