Wearables and connected health futures

Future digital health visions include the remote monitoring of patients with wearable internet of things (IoT) devices that sense and communicate health and activity data, perform real-time assessments and enable timely AI-informed interventions. This revolution in healthcare delivery has the potential to improve patient health and wellbeing, support independent living and assist individuals who self-manage chronic conditions.

By using new sets of ‘big data’ from wearable devices, AI algorithms can glean new health insights, and automate the detection, prediction and monitoring of conditions. However, it will be important that AI-informed decisions are based on algorithms trained with data of appropriate relevance and quality. It will also be important to address the ‘human–AI interaction’ challenges of combining and reconciling expert clinical opinions and AI recommendations, particularly when AI algorithms are unable to explain their decisions.

Wearable sensing and persuasive health

Wrist-worn wearable trackers, such as popular consumer-grade Fitbit devices, have matured in recent years and are transitioning from saturating wellbeing markets into healthcare and other new markets such as private health insurance and ‘corporate wellness’. The devices typically include an accelerometer for step counting and an optical pulse sensor for estimating heart rate, as well as a temperature sensor and an electrodermal activity sensor for detecting changes in skin conductance caused by perspiration or stress responses. Data from these sensors, together with user supplied biometric information, can be used to infer and track things like calories burned, activity types, stress levels and sleep quality. Additional sensors can include gyroscopes and GPS sensors for orientation and location sensing, ECG sensors that can detect arrythmia, and additional optical sensors to sense blood oxygen saturation, all of which can be embedded into more functional smartwatch types of wearables that can support apps and the making of phone calls.

For you

Be part of something bigger, join BCS, The Chartered Institute for IT.

A significant aspect of modern wearable trackers and smartwatches is their ability to encourage and incentivise users toward setting and achieving activity goals and challenges that they can share with others. It is this persuasive ability that makes wearable technologies particularly attractive, not only as monitoring devices, but as health interventions in and of themselves.

More data, real-world data, but not better data

Traditionally healthcare assessments have been confined to measurements recorded in the hospital, clinic or GP surgery. Whilst clinically reliable, these recordings provide isolated assessments rather than longitudinal recordings, and they are acquired in unnatural contexts rather than in the individual’s real-world environment during the activities of their everyday living. In contrast, wearable sensors enable the acquisition of longitudinal and potentially continuous real-world data of variable quality that is generally inferior to clinically acquired data.

Wearable sensing challenges, inequity and bias

Accurate wearable sensing is challenging, particularly when sensors are restricted to the wrist. Of course, the wrist is a convenient and conventional location for a wearable. Women have worn timepiece bracelets since the 16th century, although it wasn’t until 20th century military campaigns required synchronised manoeuvres that men switched from pocket watches to the more accessible wristwatch. But the wrist, as an active joint and point of flexion in everyday activities and gestures, is not the ideal site for multimodal wearable sensing. This being the case, it is impressive that today’s popular consumer-grade wearables have evolved to sense as well as they do. But the accuracy of these devices varies with the physicality of the wearer, their physical characteristics and physical actions. While manufacturers are clear that consumer-grade wearables are not medical devices, there have been some user dissatisfactions with performance that have led to the filing of class actions, and there are growing concerns about the extent of bias and inequity embedded in sensing systems that are now used in health-related research and clinical trials and are transitioning into healthcare practice.

• Heart rate estimation: optical pulse sensing on the wrist relies on detecting minute changes in small amounts of reflected light due to passing arterial pulse waves. The performance of this sensing varies according to the optical properties of the wearer’s skin. The higher melanin content of darker skin blocks more light meaning that the heart rate estimation of darker skinned individuals may be less accurate. The effects of ageing on the optical properties of the skin can also deteriorate heart rate estimation and reduce the sensitivity of atrial fibrillation detection. Similarly, the body mass index (BMI) and skin temperature of individuals may also affect heart rate accuracy.
  The challenge of accurately sensing minute changes in reflected light is also exacerbated by movements of the wearer. When wearers are inactive it is generally easier to sense their pulse and estimate heart rate, but oftentimes it is the tracking of data during activity and not inactivity that is of interest.
• Step counting and sleep detection: steps can be incorrectly detected when wearers perform repetitive movements with their wrists, for example when chopping vegetables. However, wearing devices on non-dominant wrists, as recommended by manufacturers, can reduce this misdetection. Steps can also be missed, particularly the steps of slow walkers. The strong impact of a brisk stride is easier to detect than the gentle footfall of a slow or hesitant walker.
  Sleep can be incorrectly detected for individuals who are motionless but not asleep, and result in overestimates of sleep quality.
• Pulse oximetry: finger-worn pulse oximeters can fail to accurately estimate oxygen saturation for individuals with darker skin. Unfortunately, because devices do not calibrate according to skin colour, commonly used pulse oximeters can mislead because they overestimate, rather than underestimate, oxygen saturation. This failing was raised in 2021 by the, then, Health Secretary Sajid Javid and reported by the BBC and the NHS Race and Health Observatory.

Despite the challenges, there are many opportunities for further technological innovations to address bias and inequity and enhance patient care and individual wellbeing. For example, using the same sensors as conventional trackers, new wearables can support individuals with epilepsy by detecting epileptic seizure onset from spikes in electrodermal activity or variations in heart rate, and sense the rhythmic motor patterns or the lack of movement that can signify seizures. As well as recording seizure episodes, these devices can respond by messaging caregivers with alerts for timely assistance that can reduce injuries and, potentially, save lives.