According to research by PwC, AI could be contributing $15.7 trillion to the global economy by 2030 – more than the current output of China and India combined. The total revenue generated by AI-powered tools within healthcare specifically is expected to exceed $34 billion worldwide by 2025. At the same time, Gartner’s recent Hype Cycle reports have tracked the emergence and mainstream adoption of Knowledge Graphs as a transformational technology with impacts for both business and wider society.
Within the domain of healthcare, these technologies represent a ground-breaking opportunity to impact the bottom line for patients via the optimisation of evidence-based support in clinical decision making. It’s not too hard to see how efficacy data from Randomised Controlled Trials (RCTs) and observational studies could be combined with healthcare provider-patient demographics and internal total cost of intervention delivery data. The results could allow healthcare providers to calculate which treatments for the conditions experienced by the populations they serve could have the greatest benefits for the available funding.
Once the electronic health record becomes integrated into this process, the traditional division between research and treatment dissolves. Instead, the possibility of a new virtuous cycle of research driving treatment that drives research becomes a realistic possibility. This then unlocks the potential for truly personalised evidence-based clinical recommendations, which moves beyond the intrinsic limitations of aggregated clinical study results where their relevance depends on how well any particular patient happens to match one of the reported subgroups.
At the same time “with great power comes great responsibility” so the saying goes, and these technologies also, therefore, represent a risk in their potential for misapplication. Firstly, indicators should never be mistaken for targets: if a healthcare provider sees a deviation from forecasts in their treatment spend then it might be entirely justified e.g. due to changing population demographics or changing disease transmission characteristics. Separating such signal from noise and feeding it back into models is a critical component of adaptive response to maintain effectiveness.
Secondly, algorithms are hard to understand. That means circumstances, where they are applicable and not applicable, are also hard to understand. Context – as in the sum total of all situational data in the current environment – is always critically important in decision making, and in almost all circumstances clinicians will have more situational awareness than algorithms.
You might also like
Contextual concerns are also of primary importance in acute/novel emergencies and crisis management. A defining characteristic of such events is incomplete information, potentially both in terms of lack of visibility and lack of evidence: we may not understand the emerging crisis situation fully so we can’t know which historic evidence is relevant as a guide to action, and once we do understand the situation better we may still be constrained by a lack of historic data. A naive reliance on AI and predictive analytics in such situations are highly unlikely to result in positive outcomes. Instead, the focus should be on localised decision making and empowering clinical risk management practice at the point of care.
Technology empowering people
For that reason, the most powerful and effective application of these technologies is almost always concerned with decision support rather than decision making. Technology is at its most powerful when it empowers people with all currently available reliable, relevant evidence so they can make the most effective decisions.
During the Covid-19 pandemic, access to the latest clinical evidence has been a crucial factor in evolving healthcare policy and effective point-of-care decision making. The Cochrane Covid-19 Study Register is a precisely scoped, authoritative source of the latest clinical evidence on Covid-19 from around the world which does exactly that. It is used by policy advisers, guidelines developers and healthcare professionals to locate relevant, reliable clinical evidence about specific healthcare questions from the latest Randomised Controlled Trial (RCT), observational, modelling and other studies to support the fight against Covid-19.
As a long term technology partner of Cochrane, Data Language has designed and delivered multiple major releases of Cochrane’s Linked Data Platform knowledge graph and data services since its first prototype in 2014. This has resulted in an asset base of semantically interoperable, human- and machine-consumable structured micrographs at the heart of Cochrane’s clinical evidence pipelines, and a collection of loosely coupled, cohesive services which allow them to be accessed and recombined as required. In many ways, the pandemic represented a fundamental challenge to Cochrane’s working practices, as the need for all available evidence necessarily extended their focus beyond RCTs to a wide range of other study types and included studies that had not even yet reported their results. The flexibility of this platform architecture and their business workflows allowed them to deploy a rapid response team, adapt their practices, and for us to build and release their study register in a little over three weeks.
There is no better validation of these technologies correctly applied than their ability to enable such rapid adaptation in the face of such an unprecedented crisis.