Uncertainty on home-recorded readings

How a product tells a person that a reading it took at home (an ECG, an overnight oxygen level, a cycle prediction) may not be reliable, cannot be classified, or should be read as a trend rather than a fact. The pattern covers the wording, the visual treatment, and what the product asks the person to do next.

A number on a phone looks certain. A home ECG from the wrist, a blood-oxygen average from a ring, a fertile-window prediction from logged dates: each is an estimate with known failure modes, and the person reading it usually has no way to see the error bar.

Products that hide uncertainty invite two harms at once, false alarm and false reassurance. Products that show it honestly make the reading useful, because the person knows when to repeat it, when to ignore it, and when to take it to a clinician.

Consumer devices in the library sit on both sides of the regulatory line, and their wording tracks it. Withings' ScanWatch (the subject of an FDA clearance the company announced in 2021) and its Body Scan scale carry an explicit “Inconclusive” ECG result, meaning the recording cannot be classified, with the reasons and the advice to talk to a doctor. Oura's blood-oxygen feature states that the ring is not a medical device and asks the person to focus on the long-term trend rather than any single night. Flo says outright that ovulation cannot be predicted with full accuracy and that its predictions must not be used as contraception.

For clinicians, a home reading arrives without the context a clinic measurement has (posture, time, device fit, what the person was doing) and a product helps by attaching that context, and its own confidence, to the reading before it is shared. The FDA's human factors guidance is the relevant frame: use error on a home device is a design outcome, and an unclassifiable reading is a normal case to design for, not an edge case.

Uncertainty is often shown as a colour, amber for “inconclusive”, grey for “poor recording”, and colour alone is not a message (WCAG 2.2 Use of Color, 1.4.1): the state needs a word, and the word needs to be what a screen reader announces. Trend charts need a text summary of what they show. The instruction that follows an uncertain reading: “keep still and record again”, “talk to your doctor”, should be a labelled action, not buried in help text.

The plain-language bar is high. The NHS service manual reports that more than 6 in 10 adults struggle with health content that includes numbers and statistics, which is exactly what a confidence range is.

Four products in the library document how they handle uncertainty, and they use three different devices to do it. Withings names the uncertain state as a result in its own right: “Inconclusive”, with a list of causes, and “Poor Recording” with the suggestion to record again, which treats the failure to classify as information rather than an error. Oura reframes the unit of measurement: a single night's oxygen average is expected to vary, so the product asks for attention to the trend. Flo changes the output itself: when its algorithm cannot predict a cycle length with confidence, it widens the fertile window to around 14 days rather than narrowing it falsely.

Each is a different answer to the same question (what do we show when we do not know) and the library will grow examples of each as products describe them. What no page in this set describes is how an uncertain reading reaches a clinician, if it does; that hand-off is the subject of the escalation pattern.