A technology perspective on how we can use the Internet of Things to better serve patients in clinical trials
Joss Langford, CTO & Founder of Activinsights.

So, what exactly is the Internet of Things (IoT)? IoT products can be defined by having four characteristics:

• they are ubiquitous,
• they are connected,
• they can sense their environment,
• and have computing abilities.

IoT products and services can be used to simplify the remote collection of health data for patients, practitioners, and researchers. Here we will look at some of the ways in which we can change our thinking to better understand our patient’s needs and how we can alter our methods to ultimately produce better outcomes and reduce burden for the patient.

Changing our Thinking

IoT thinking is transformative – it helps us change our way of thinking to better understand our patients’ everyday environment, behaviours and needs. Data from IoT products can come from everywhere and anywhere. We must go beyond being patient-centric to being human-centric and look at the world from the viewpoint of the patient. For example, when clinical trials refer to ‘remote monitoring’, for a patient it is ‘monitoring at home’. Early in protocol development, using human-centric design will help us to understand the environment of the patient and view them as an individual first and as a patient second – patients want to live an everyday, normal life and not just be defined by their diseases.

Using Standardised Approaches

Standardised approaches help to reduce risk, cost and development times. Many of the most useful standards are not formal or technical. They can be as simple as using a normal watch strap on a wearable – something people are used to using in everyday life – to reduce cognitive burden and the need for instructions. These are known as de facto standards.

Formal technical standards allow interoperability between modules in a system, giving substitutability and preventing vendor lock-in. However, there will always be compromises to be made between what can be standardised and what must be bespoke.

Staying Connected

The strictest definition of an IoT product requires that it is discoverable and addressable on the Internet. In our studies, we probably don’t want products to be universally discoverable and we don’t need them to be permanently connected.

But how much ‘real-time’ connectivity is really needed?

Near real-time management and quality data can support study risk control, but only if they are actionable. We often see this when measuring non-wear – we are keen to understand how successful we are at deploying technologies but if we can’t do anything to change this once the study is in progress, then we may be adding a real-time measurement burden that we cannot action.

Real-time, continuous measurement is clearly a transformative aspect of home-based approaches and can really make a difference in our studies. While digital interventions may fundamentally need real-time data, in most measurement studies the perceived need for the timeliness of data must be balanced with the power and infrastructure requirements it brings.

Power & Infrastructure

Real-time is not free as connectivity needs both power and infrastructure. Power requirements will impact product size or the need for at-home charging. The size of a product may impact acceptability and having to remember to charge is another burden for the patient. Once the patient has the need to remove a wearable product, they are much more likely to forget to put it on again, reducing adherence.

When we insist on using infrastructure that we don’t need, we risk exacerbating technology bias. That might be because were excluding people through their digital literacy or their social economic status or geography. The levels of interconnectivity vary massively between rural and urban populations. We find that carefully defining the digital cadence of a study before specifying real-time requirements helps to resolve many of these questions.

Sensing

We need to recognise that IoT products in clinical studies author personal data automatically. This has profound impacts on how we think about managing that data. The data are behavioural and describes what the patient actually does rather than what they say they do, so we have an increased responsibility to explain exactly what they are sharing. These data are also transformative, but risks of privacy are obvious, and we need to consider trust as a potential patient burden.

Personal Data Principles & Rights

When it comes to personal data, there are regulations that vary by jurisdiction, but the principles and rights are reasonably consistent. One of the challenges is around the management of privacy. We must be very careful that when we use pre-scaled technologies, we clearly understand their pre-defined data flows and business models and ensure they’re compatible with our aims.

As we develop more digitally-enabled systems, individuals start to expect that if they can express consent or sign-up digitally, they should be able to exercise their rights digitally. We need to manage expectations and use this as an opportunity to create trust with our patients.

Sensor-Level Data Computing

Once the on-product software (known as firmware) has done the basic job of managing the product, it can then convert sensor-level ‘raw’ data into digital measures using algorithms.
These outputs may be meaningful health measures in their own right or they may become features in statistical or machine learning models.

Behavioural events can then be identified, characterised and their classification understood. This includes:

• Activities of daily living, for example, sleeping, sitting or walking
• Specific behaviours of interest, for example, scratching or tremor

At this level of algorithm development, sensor-level data are an essential resource for us to be able to create novel, verified algorithms that we can turn into digital health measures for validation.

Computing at the Edge

Once we understand the algorithm requirements, we can use a technique from the IoT space called ‘edge computing’. This means that we are going to move computing power from our servers to the edge of the cloud/network. We can do that once the algorithms are well-defined.

This allows us to compress data intelligently to improve connectivity. It makes pseudonymisation more effective and removes artefacts from the data that may disclose aspects of identity that we don’t fully understand. It minimises the data we collect and this, in turn, helps build patient trust.

Connected Computing

IoT thinking relies on a network of effective infrastructure which is not all within your direct control. Embracing interoperability and using technical standards allows us to simplify data integration from multiple sources (e.g. ePRO). This means that we reduce our dependence on using just one product or vendor and enhances human-centricity through data portability.

All Day, Everyday Data

Once we have extracted a sequence of events from sensor-level or on-board processed data, we can look at how these 24hr cycles start to build a picture of an individual. These include:

• Lifestyle patterns – activity levels, bout patterns, temporal patterns, behaviour types etc.
• Phenotype – chronotype, sleep duration etc.

When we start to see the world of the patient in the data, measurement over multiple days gives us even further insight, including:

• Types of days, response to environment / intervention & longer-term trends

We can then begin to understand how our interventions have impacted them or how their disease has changed over time.

This comprehensive view will often reveal insights on patient populations that can be used in stratification or enrichment in further disease prevention or clinical trials.

For more information on how Activinsights can help with your next study, contact us on info@activinsights.com or call +44(0)1480 862082.