Ian Bolland caught up with Dr David Plans, CEO and founder of BioBeats. They discuss the work BioBeats does within the NHS, and the health service’s aim to become a world leader in machine learning (ML) and AI within the next five years.
Tell us about the work BioBeats has done with the NHS and other healthcare services and systems?
Our founding team took part in NHS projects within Stoke-on-Trent writing smartphone applications for chronic disease reporting. When we started building the current BioBeats platform, we worked within systems such as the Chelsea and Westminster hospital to experiment with mental health tracking in patient journeys such as cardiac surgery. Within the context of healthcare provision within insurance bodies, we have worked within mental health delivery services with one of the largest global insurance companies.
How does the NHS get to a point where it becomes a “world leader” in AI and machine learning?
The NHS has a unique opportunity to become a world leader in machine learning efforts for preventative disorder and disease detection, as it has a longer history than most socialist healthcare systems, and therefore (although records can be chaotic) a better data history. However, whilst the body of data the NHS generates is vast and rich, a lot of work has to be carried out in order to make that data useful for machine learning: it needs to be shaped into the right format, meaningful features must be identified and included, missing data removed as well as sensitive information, etc. The work of pre-processing and cleaning data is a significant and resource-intensive part of any machine learning work. As such, it means that in order for the NHS to take advantage of the opportunities it has, it will have to employ a substantial workforce just to do the manual dirty work of cleaning, transforming and make the data ‘useful’. And this is only the first necessary step: partnerships such as Deepmind will need to broaden, as it is unlikely that the NHS will find a broad enough pool of AI workers within itself. Provided internal government resource, through funding initiatives such as Innovate UK but also broader industrial-academic funding bodies such as NIHR, EPSRC, etc. is made available specifically for machine learning projects that bring more AI startups in direct collaboration with the NHS, we should be able to collaboratively achieve preventative targets in oncology, cardiology, and mental health.
Is the ambition realistic?
Given enough time and resources, this ambition is not unrealistic. The key question is not necessarily whether the ambition is in and of itself unrealistic (no ML project is unrealistic given rich enough data of the right scale), but how long it will take for all the right bodies, funding, partnerships to coalesce. The NHS is not one organisation: it is a vast, complex, hybrid system, and its complexities mean it isn’t always possible for ‘it’ to move quickly. It isn’t realistic to assume the NHS will foster ML innovation in a cohesive, integrated way. More realistically, partnerships will have to be formed within its constituent bodies at a granular level.
What changes do they need to undertake to get to this stage?
There are around 250 hospital, mental health and ambulance trusts in England and over 160 Clinical Commissioning Groups (CCGs), who buy services from providers in often disparate ways that vary across trusts. Hospital and mental health trusts are largely dependent on CCGs buying services from them, such as surgery, outpatient visits, etc. and understandably, there’s some challenges in integrating all this. Projects such as the 'Five year forward view' and the 'better care fund' have been created to enable better integration of services, but it’s important that these take into account how to bring industrial and academic knowledge and technologies into the commissioning stream.
What kind of effect does this target for the NHS have on the healthtech industry?
It means that healthtech industrial partners need to be increasingly aware of how commissioning services works as a process. Building machine learning services begins with understanding the challenges and processes behind a service and its needs. For example, in mental health, the current disconnect between people presenting symptoms at their GP and commissioned mental health services could be bridged by services that use sensor data and user interaction to detect mental health decline, involving digital therapeutics much earlier on. But in turn, digital therapeutic services need to understand and build connections to referral systems within the mental health trusts.
