In one sentence
Each morning, Symptom Radar compares last night with your own normal and tells you whether your body looks settled or under strain; Symptom Trend, a second layer on top of it, asks how fast things are changing.
What it reads
Ten readings, eight of them overnight, in three tiers. Main, at full weight: overnight resting heart rate, overnight respiratory rate, overnight temperature deviation and overnight blood oxygen. Secondary, at 0.7: deep sleep, REM sleep, sleep efficiency and time awake during the night. Supporting, at 0.4: daytime resting heart rate and daily average HRV. Readings that usually move together are counted once, so one change is never counted several times; that leaves five independent signs: your heart (overnight and daytime heart rate, HRV), your sleep, your breathing rate, your temperature and your blood oxygen.
What it compares them with
Your own baseline: the distribution of each reading across your last 30 nights, with sick days and the three days after them left out, so that a recent spell of being unwell never becomes your normal. Overnight heart rate and the four sleep readings count only nights from your usual sleep device, so a change of device does not look like a change in your body. Your readings are never compared with a population average; while your baseline is young, VT assumes at least a typical night-to-night spread, so a few calm nights do not make small changes look alarming.
What it takes into account first
Some everyday things move the readings without any strain behind them, and when you log them Symptom Radar gives the affected readings less weight: hard training, when your last seven days add up to more than a quarter above your weekly training goal (this needs a goal set in the app); alcohol, scaled by how much you drank; a late bedtime; a late meal, or a large late meal; high body discomfort the evening before; your period. "Late" and "large" are judged against your own habits, and VT needs about 20 logged examples from the last 90 days before it can tell what is unusual for you. When a sign you usually have is missing, Symptom Radar needs two signs to agree before it shows yellow, and the overall score cannot turn the day red, though low blood oxygen or one extreme reading still can.
What you see each morning
One of five states. Normal: everything sits within your usual range. Yellow: a few readings sit outside it; consider easing off training and giving rest priority. Red: several readings sit well outside your usual range and agree with one another; rest is strongly advised, and if you feel unwell, speak to a healthcare professional. Building baseline, in the first week, with a countdown of nights. Not enough data, when fewer than three readings are arriving after the first week, which waiting will not fix and checking your device's sync will. With the result come each reading's value against your baseline, the readings that drove the result named in plain words, a confidence (High when at least two of the main overnight readings were measured, Medium when one was, and Low when none were, while results are still provisional, or when a sign you usually have is missing; a night recorded on a different sleep device drops it one step), and, when you were in fact unwell, a place to say so, which VT uses to learn how its results relate to what actually happened to you.
What sits outside the baseline
Three rules can turn the day red on their own: blood oxygen below 94 percent; one reading far beyond anything in your recent history that nothing you logged explains; and breathing rate or temperature far outside your range with other signs agreeing. Fixed limits add a safety note to the result without changing its color: blood oxygen below 90 percent, overnight resting heart rate below 40 or above 100, breathing rate below 8 or above 25, temperature more than 1.5 degrees from the device's own baseline.
When it starts, and how it grows
Each reading needs 7 usable nights within the last 30 before it can be scored, and Symptom Radar starts once three readings reach that mark, usually on your seventh morning. Results are provisional until half your readings have 14 nights behind them, from about day 14; a provisional day cannot go red on its overall score, though it can for low blood oxygen, one extreme reading, or two independent signs both far outside range. Around day 30 each reading's normal is built from a full month of nights and is at its most reliable.
Symptom Trend
Symptom Radar asks whether last night was unusual; Symptom Trend asks how fast things are changing, so that a fast move is noticed before it is large. It reads two patterns: a jump, where a reading changes from one night to the next by more than it normally does for you, and a drift, where a reading creeps steadily in one direction over several days with no single night standing out. Its results are OK, Watch and Flag, and a Flag appears beside Symptom Radar's result; Symptom Trend never raises a day above yellow, and red stays with Symptom Radar alone. Drift reading starts once about two weeks of nights have been scored; jump reading needs 28 pairs of back-to-back scored nights for at least one reading, about four weeks after Symptom Radar's first result. Only unbroken pairs count: a missed night breaks the pair on either side of it, and a sick day spoils both pairs it belongs to, because those nights are not your normal. The illness episodes you record are used to check how far ahead of your own symptoms it tends to move.
What the evidence behind it says
The readings were chosen because they are the ones the literature has tested. In a Stanford cohort of about 5,300 smartwatch wearers, 32 developed COVID-19; 26 of them (81 percent) showed changes in heart rate, daily steps or sleep, and of the 25 with symptom dates, 22 changed before or at the day symptoms began, four of them at least nine days earlier (Mishra 2020). In 2,745 Fitbit wearers with a confirmed infection, respiration rate and heart rate were typically raised and HRV lowered around diagnosis (Natarajan 2020). Across 9 million morning readings in 28,175 people, heart rate ran about 6 percent higher and HRV about 10 percent lower on days people reported being unwell (Altini 2021). Those are population findings about readings; what VT does is compare your readings with your own, and say when several have moved together.
What it never does
It never sets a sick day from your readings; only you do that. It never names a condition. It never compares you with anyone else. And it never replaces the judgment of a person: if you feel unwell, see one, whatever the morning's result says.
How far to trust the result
As far as the readings it rests on, which the entries above describe one by one: measured for heart rate and oxygen, trend-grade for HRV, temperature and breathing, estimated for the sleep readings. Its confidence label says how many of the main readings it had; a result built on sleep readings alone is a weaker result, and it is labeled so.
Sources
- Mishra T, et al. Pre-symptomatic detection of COVID-19 from smartwatch data. Nat Biomed Eng 2020;4:1208–20. doi:10.1038/s41551-020-00640-6
- Natarajan A, Su HW, Heneghan C. Assessment of physiological signs associated with COVID-19 measured using wearable devices. NPJ Digit Med 2020;3:156. doi:10.1038/s41746-020-00363-7
- Altini M, Plews D. What is behind changes in resting heart rate and heart rate variability? Sensors 2021;21:7932. doi:10.3390/s21237932
- Vitality Tracker, How it works: Symptom Radar and Symptom Trend
- Vitality Tracker, What VT measures (catalogue of 15 October 2026).