In this final part of Ivan Koychev's blog series, he considers the risks and rewards of the rapid development of digital technology for dementia research and healthcare.
With the incidence of dementia set to triple by 2050, we need to consider all new approaches in the care of dementia patients. Dementia is the most widespread form of neurodegenerative disorder and it is associated with an immense societal and personal cost. In the UK, as in other countries, the financial pressures on the health care system has never be higher and is projected to increase steadily.
The largest cost in dementia care happens when individuals need to move into care and lose their independence, when the risks associated with independent living are too great. As a clinician, it is often difficult to make this very significant decision in someone's life, as often we have to do it on the basis of indirect information. Novel technologies are now beginning to change this and are being used to monitor a patient's function and to support them to remain independent.
Notable examples are smart homes with a variety of reminder systems (e.g. to take medication, to attend appointments) as well as detectors that alert carers of unattended cookers, overflowing bathtubs and wandering risk (front doors being opened in the middle of the night). In current clinical practice repeated risky incidents point to needing to take patients into care. Through the use of technology the aspiration is that we can control these risks, empower the patients, and ultimately support them to remain in their own homes for longer.
Is there a cloud in the silver lining of wearable tech?
Machine learning is the tool of choice for many organisations looking to find patterns in large quantities of complex data – whether it is credit for loan applications, finding errors in legal contracts or dementia research. Machine learning runs self-learning programmes based on diverse data and for this reason it is susceptible to garbage-in garbage-out syndrome. In a now infamous case in 2016, biased data led Microsoft's AI chatbot to develop extreme right wing tweets in less than a day on Twitter, because it was programmed to learn by talking to real people on Twitter (Qz.com).
We need the tightest scrutiny when using digital tech to detect and forecast the progression of cognitive decline. Challenge from outside an organisation is vital and bringing together mixed skilled groups helps minimise bias. At a datathon last year, DPUK, DFP and the Alan Turning Institute brought together data scientists, imaging specialist, and dementia researchers to assess the value of machine learning for the pilot study (Institute 2018). Datathons inspire new ways of thinking and, although it is early days still, indications are that machine learning may be a tool for making brain health predictions based on brain imaging.
The dementia research community, busy collecting a range of rich and diverse real-world data, has yet to establish tested models and algorithms that reflect the processes at play in the lives of those at risk of dementia. And since most digital devices in dementia research are only in the initial stages of development, we have limited data on acceptability in patients. However, it is clear from trials that the use of digital technology for monitoring and supporting those affected by dementia offers real hope.
Digital data can give a semblance of control over one's health and its management, but it also gives rise to real concerns about data security and personal autonomy. There are well documented harbingers of these risks – in 2018 Facebook, the giant of data collection and algorithmic analysis, leaked details of more than 50 million users (nytimes.com). There are also concerns for health insurance or job prospects – but can we really foresee all implications? There is no doubt in my mind that the rapid development of digital technologies will represent a high gain, high-risk avenue for healthcare and dementia research in particular.
For those interested in the complex issues arising from cyber security, I recommend Marc Goodman's accessible book on the topic (Goodman 2015).