This paper is useful for two reasons.

First, it looks at commercial wearable data at a very large scale rather than a small laboratory prototype. The study used data from more than 8,000 participants and over 1.3 million hours of Oura Ring recordings, which makes it a useful counterpoint to some of the smaller wearable studies I have been reading.

Second, the authors focus directly on a problem that is becoming central to my project: physiological data only becomes meaningful when it is interpreted relative to the individual, the context and the quality of the signal.

The paper is about detecting adverse physiological events such as COVID-19-related changes, fever and shortness of breath. It is not about autism, sensory overload or airport travel. What transfers to my project is the methodology: baselines, confounders, sensor noise and uncertainty all matter.

What the paper actually studied

Phipps et al. examined whether data from commercial Oura smart rings could support early detection of adverse physiological events using machine-learning algorithms.

The paper describes four major challenges:

1. Heterogeneity between and within people — physiological values differ substantially across individuals and can also vary within the same individual.
2. Confounders — factors such as physical exertion, caffeine, alcohol and other conditions can alter physiological signals without representing the target event.
3. Sensor-level noise — wearable data is imperfect and noisy.
4. Imprecise labels — self-reported events do not always align neatly with the underlying physiological change.

The authors therefore explore participant-specific baseline correction and label-correction techniques rather than relying on one universal threshold.

A very large dataset does not remove the interpretation problem

The scale of the dataset is striking: more than 8,000 participants and over 1.3 million hours of wearable data from Oura rings.

But the size of the dataset does not make the underlying physiology simple.

The paper explicitly argues that physiological heterogeneity makes general population-level algorithms less precise. A resting heart rate that is normal for one person may be unusual for another. The same applies to measures derived from inter-beat intervals, HRV and peripheral skin temperature.

This is very relevant to my project because it strengthens the case against designing one threshold such as:

“If heart rate exceeds X, the traveller needs support.”

That would ignore the person's normal baseline, physical activity, environmental context and ordinary day-to-day variation.

Key lesson for my project

A universal physiological threshold is not defensible. The system needs to understand change relative to the individual, not simply whether a value is “high” or “low”.

Baseline correction is more important than absolute values

One of the main contributions of the paper is the use of participant-specific baselines.

The authors show that correcting physiological values relative to a person's own baseline can improve detection performance. For the COVID-19 analysis, the ROC AUC increased from 0.725 without correction to 0.777 when baseline and label correction were applied to both training and test data.

I do not need those exact numbers for my final design, but the principle is important.

For my airport system, I am increasingly thinking that the useful question is not:

“Is this heart rate high?”

but:

“Is this meaningfully different from what is normal for this person in a comparable situation?”

Even that does not tell me why the change happened, but it gives the sensing a more defensible starting point.

Different physiological events were not equally detectable

Another part I found useful is that performance differed substantially depending on the event being detected.

The reported AUCs were very high for self-reported fever, reaching 0.994 in one corrected condition, but much weaker for self-reported shortness of breath, where the highest reported AUC was 0.635.

This is important because it shows that having a wearable and a large dataset does not mean every physiological or subjective state becomes equally inferable.

Some events have a stronger physiological signature than others.

For my project, that is a useful warning against assuming that because wearables can detect some illness-related changes, they can therefore detect something much less clearly defined such as sensory overload or a need for support.

What the Oura Ring was actually measuring

The Oura Ring Gen2 data used in this research included measures derived from the ring's optical and temperature sensing, including:

• heart rate;
• inter-beat interval information;
• HRV measures such as RMSSD;
• peripheral skin temperature;
• movement/activity information.

This surprised me because the ring can collect a substantial amount of physiological information in a very small, discreet form.

From a design perspective, that immediately makes the ring form factor interesting for my project. It is less visually medical than many chest straps or research wristbands, and it could potentially feel more like an ordinary personal object.

However, the sensing capability and the physical form are two separate questions.

Ring or wristband?

This paper has made me take the ring option much more seriously.

Why a ring is attractive

• discreet and familiar form
• compact
• continuous physiological sensing
• heart-rate / IBI / HRV-related data
• skin-temperature information
• activity / movement information

What I still need to question

• whether finger-worn sensing is comfortable for autistic users
• whether it works reliably while carrying luggage or using the hands
• battery / charging and continuous use
• whether the sensor set matches the final system architecture
• whether a wrist device gives me signals that a ring cannot

There is also one important limitation for my current sensing shortlist.

I have been considering EDA + PPG / heart rate + accelerometry. The Oura Ring data in this paper gives me strong examples of PPG-derived cardiac measures, skin temperature and movement, but it does not provide the EDA channel I have been considering.

That means the design decision may not simply be “ring versus wristband aesthetically”. It may depend on which signals I can actually justify needing.

If EDA remains important, a wrist-based device may still be more suitable. If later research suggests that heart-rate / pulse-derived measures plus movement and context are sufficient for my prototype, a ring becomes much more plausible.

Confounders are not edge cases

The paper also reinforces that physiological changes can be caused by ordinary behaviour.

The authors discuss confounders such as:

• physical exertion;
• caffeine;
• alcohol;
• differences in resting physiology;
• sensor noise.

For an airport system, the equivalent list becomes even longer:

• walking;
• rushing;
• carrying luggage;
• stairs;
• temperature;
• fatigue;
• caffeine;
• excitement;
• frustration;
• illness;
• time pressure.

This means the system cannot treat every physiological change as a meaningful accessibility event.

Movement and contextual information are therefore not optional extras. They are part of how the system avoids overinterpreting the body.

What this means for my airport system

This paper strengthens the architecture I am currently moving towards:

personal baseline + physiological change + movement/context → cautious interpretation → transparent check-in → traveller decides

I am becoming less interested in absolute physiological thresholds and more interested in change from personal baseline.

For example:

physiology changes relative to baseline + high movement

→ physical activity may explain the change.

physiology changes relative to baseline + low movement + increasingly crowded/noisy context

→ perhaps the system has more reason to offer a check-in.

The key word is still perhaps.

The wearable can help identify that something has changed, but the traveller still decides whether that change means anything and whether support is useful.

Design implication

Biosensing should help the system notice change from a personal baseline, not classify the traveller against a universal definition of “normal”.

What this paper does not establish

This study does not show:

• that Oura Ring data can detect sensory overload;
• that illness-detection algorithms transfer to autistic airport support;
• that HRV or skin temperature identifies when a traveller needs help;
• that one baseline method will work for every person;
• that a ring is automatically more acceptable than a wrist wearable;
• or that physiological detection should automatically trigger an intervention.

The study's strength for my project is the way it treats baseline, confounders, noise and uncertainty, not the specific illness-detection model.

What changed in my thinking

This paper has strengthened two parts of my design direction.

The first is methodological:

Physiological sensing only makes sense relative to the individual.

I now think a personal baseline should probably be part of the system architecture rather than an optional technical detail.

The second is physical.

I had been thinking mainly about wrist-based sensing because devices such as the Empatica wearables appear repeatedly in the literature. I did not realise how much physiological information a ring could collect.

I am now genuinely between ring and wrist-based sensing as possible physical directions.

However, I do not want to choose the form first and then justify the signals afterwards. The next step is to decide which physiological channels are actually necessary, then compare which form can collect them with the least burden.

Source

Phipps, J. et al. (2024). Early adverse physiological event detection using commercial wearables: challenges and opportunities. npj Digital Medicine, 7, 136.

Key points from my reading: Oura Ring Gen2; more than 8,000 participants and 1.3 million hours of wearable data; participant-specific baseline correction; confounding from behaviour and physiological variation; sensor noise; label uncertainty; heart rate, inter-beat interval / HRV-related measures, peripheral skin temperature and activity; large differences in detectability between physiological events.