AI and data-driven insights have the potential to overhaul our healthcare systems. But barriers lie in the way.
A data tsunami is building up in every aspect of our lives – and the health space is definitely no exception. From electronic health records (EHRs) and physician’s notes, to the vertiginous rise in genomics data due to plummeting costs of DNA sequencing, and the continuous data influx from a rapidly expanding range of wearable devices… The generation of health data is rapidly outstripping our ability to store and analyse them.
A healthy dose of innovation
Healthcare represents a particularly sensitive sector, and progress in the field promptly makes the news – as seen with the development of the DeepMind Streams app that alerts clinicians to acute kidney injury. It also represents a particularly lucrative market, demonstrated by huge investments from tech giants such as Google and Apple on one side of the spectrum, and by the booming number of startups focusing on developing data-driven insights to improve patient treatment.
Even in the public sector, the creation earlier this year of NHSx, the special unit meant to bring digital, data and technology to the NHS, shows that the time is ripe for harnessing the value of health data.
Yet, the healthcare industry encounters significant headwinds in leveraging data analytics and developing novel AI-based solutions. And of all the visionary ideas turned into scientifically sound algorithms, only a very small number actually makes it to real world application.
Let’s look more in detail at some of the most significant challenges a business venturing into this market faces – from handling the data in the first place, to bringing a product up to clinical uptake.
Into the data jungle
Healthcare is definitely a data-rich sector, so scarcity of information is not a problem – and the NHS database is particularly valuable with respect to other countries, since it has comprehensive records that go back decades.
However, access to health data is often very difficult from a regulatory point of view, and there are extreme differences in terms of quality and accessibility. Typically, health data is messy, disperse and often siloed in a multitude of medical imaging archival systems, pathology systems, EHRs, electronic prescribing tools and insurance databases.
While things are moving in the right direction, i.e., with the development of unified data formats such as Fast Healthcare Interoperability Resources, there is no easy and quick fix. No fancy algorithm can be developed without proper data collection and cleaning – and in many cases, this phase can take months.
A practical solution – the partnership
Until companies keep reinventing the wheel and developing their own internal tools for data cleaning with huge costs in terms of time and money, progress will be slow. This scenario is not dissimilar from the problems faced by the drug development sector, where the economics of the sector is being pushed to breaking point by the time and cost involved in the registration of a new drug.
The same, brilliant solution that was created to tackle research bottlenecks could be applied: a series of pre-competitive collaborations of public-private character, both with the government and the academia. Public-Private Partnerships (PPPs) such as Europe’s Innovative Medicines Initiative have already proven their potential in drug development; similar collaborative models in the health data space could greatly accelerate standardisation and help create libraries of open source tools for data curation.
Engagement by design: a dialogue with health providers
Unlike other sectors, technologists dealing with health data face hard barriers connecting with the end user of the products they develop. Not only are they remote from the patients’ perspective, but they are also detached from the clinicians’ practice and unable to communicate effectively with them due to highly specialised lingos and experiences on both sides.
Establishing a continuous dialogue and direct collaboration with health practitioners, though, is essential to break the so-called AI chasm in this sector: turning genial ideas into impactful products with real uptake. There is an endemic resistance within the healthcare sector to reform and change by information technology, and doctors consider tech more often a distraction than a benefit in direct clinical service. User-centric design and considerations of clinical applicability and patient outcomes are essential to overcome this barrier.
Enter the startup
The involvement of clinicians from the very early stages is one of the main strengths of healthcare startups, which are widely accepted as a major force in healthcare innovation. The first cohort supported by KQ Labs – the Francis Crick Institute’s accelerator programme for data-driven biomedical startups – for example, showed how initial teams are usually formed by clinicians and data technologists working side by side.
But there are examples of successful partnerships between larger tech companies and clinicians – take for example the collaboration between DeepMind and London’s Moorfields Eye Hospital on developing an algorithm for recommending referral decisions based on the analysis of highly detailed eye scans.
Crucially, the system developed has a good level of explainability, that is, it provides information that helps explain to eye care professionals how it came to a decision – thus increasing the level of trust for doctors.
Another notable point in this collaboration is that results were published in a highly reputable peer-reviewed journal. The healthcare sector is traditionally and rightfully cautious and usually accepts peer-reviewed research as a minimal standard of validation for novel products and technologies. Ignoring this culture and avoiding peer-reviewed assessment – performing what has been called ‘stealth research’ – does not foster trust and is one important factor delaying the clinical uptake of novel tech tools.
Earning public and patient trust
Industry access to health data is of course also a particularly sensitive matter for those the data come from – patients. Earning the trust of patients and of the general public is a crucial element, but a not trivial one.
In general, the involvement of private companies is looked at with more suspicion than access to data by academic institutions, and in the best case seen as a necessary evil. DeepMind may have got it right in their collaboration with Moorfields, but their previous partnership with London’s Royal Free made the news for breaching patients’ rights and led to an ICO ruling that the Royal Free had failed to comply with the Data Protection Act when it provided patients’ data to the company.
The absorption of DeepMind Health within Google Health has raised big concerns, and a similar clamour is now accompanying the recent acquisition of Fitbit by Google; in both cases, the public questions what use Google will make of their health and fitness data.
A step in the right direction
Transparency and stronger engagement with society are the only way to avoid public backlash against the use of AI in health, and to remove some of the barriers restricting access to health data. The creation of the likes of Partnership on AI, which establishes a common ground between leading companies, organisations and people affected in different ways by artificial intelligence, represents a step in this direction.
Collaborations with initiatives such as Understanding Patient Data, which aims to support conversations with the public, patients and healthcare professionals about how health and care data is used, could also help improve the public’s opinion.
Sign up to our free weekly newsletter for the latest from our expert guest contributors.