The Healthcare industry is under tremendous strain: in addition to pre-pandemic pressures, the industry now faces both novel and exacerbated challenges, such as addressing a growing backlog of medical cases, continued testing at scale for new COVID-19 variants, and responding to conditions such as long COVID. Service delivery at scale and the response to these challenges has, in part, been enabled by innovations in Artificial Intelligence (AI) and data analytics, which have significantly transformed the day-to-day workings of many medical practitioners and the wider industry alike.
Over the past two years, Deloitte has brought together teams of clinicians, hospital managers, data scientists, digital designers, academics, and strategists to develop some of these innovative solutions.
However, has the Healthcare industry fully embraced AI and automation-based decision making? How has the increased focus on advanced analytics and AI filtered down to a people and talent level? How can we improve and scale automation and AI in Healthcare?
In this article, we interview Sunny Dosanjh, Deloitte’s Lead for the Healthcare AI and Analytics team, and his colleague Elena Turek to provide us with insights from the industry and discuss their perspectives on future strategies and technical solutions in national healthcare. Responses have been edited lightly for clarity.
Has the healthcare industry fully embraced AI and automation-based decision making? What are some of the biggest challenges the industry is currently facing?
AI in the Healthcare industry is at an advanced stage, and AI and advanced analytics are not new or unfamiliar to most organisations – you need only look at the successes of DeepMind (who develop AI research and mobile tools for the Healthcare industry) or Great Ormond Street Hospital’s Digital Research, Informatics and Virtual Environments (DRIVE) initiative, both of which have been around for many years now. More recently, the Department for Health and Social Care awarded £250 million in funding to the NHSX to establish an AI Lab, aimed at improving the health and lives of patients.
However, one of the biggest considerations with AI in the Healthcare space is the safety-critical nature of the industry, and the impact of getting AI ‘wrong’. Safety and ethics are paramount to us and every decision and piece of code needs to keep this in mind. Creating scalable AI in Healthcare, which can operate successfully and replicate results across the complex structure of the NHS, is not always easy. AI solutions often focus on using machine learning models, but these can be difficult to replicate in the absence of ‘perfect’ training datasets. The historical data available is not necessarily a clinically perfect dataset, as real-world medical practitioners make optimal decisions with not only clinical considerations in mind but also operational pressures with human variables. This can make replication of AI solutions at scale challenging.
Because of this, I think that there needs to be a shift from thinking about AI as just automation or machine learning, to all the various possible applications of AI and advanced analytics – with interpretability in mind. I think Deloitte’s RITA (Referral & Intelligent Triage Analytics) is a great example, as it uses Natural Language Processing, but is based on an algorithmic rule base which enables interpretability of the AI, aligns to the national NICE clinical standards, but still enables local customisation hospital to hospital.
So, how do you think we can further scale automation and AI in healthcare?
There are hundreds of automation and AI use cases in healthcare, thousands of start-ups, and billions of pounds being invested in this space. The challenge is not necessarily in identifying appropriate use cases, but in getting them to work at scale. I think there are two key things we need to do as an industry:
How has the interest in advanced analytics and AI filtered down to a people and talent level? How are these capabilities being developed, and what are some of the learnings for other industries?
As a community, medical practitioners are very collaborative – the aim for us all is to help save lives. Medics are also generally very data-led and are used to using evidence and data for decision-making and treatments. It’s very rare to come across a medic who disputes good quality data and analysis. However, data science and statistical capabilities within the industry can be limited, with a lot of the current focus being on things like data provisioning and reporting. For example, we are recently seeing a shift towards investment into Population Health Management and a creation of linked data records. To really see the value from this we are going to need readily available skills to transform, analyse and interpret the data.
One of the reasons for this is talent development. We need to professionalise the data science and analytical capabilities within the NHS market, specifically considerations on hiring, capability development/progression and roles within an NHS environment. Whilst that may sound relatively straightforward, this will need radical steps from Chief Data Officers entering the Boardroom through to clarity on role requirements across a range of disciplines such as data scientists, data engineers, cloud engineers and data translators.
Thanks Sunny, that’s interesting. If the right advanced data and AI capabilities don’t always exist in a single organisation, can you give us a view of what data sharing in the industry looks like and if there is more collaboration on this? We have seen a growing impetus to increase and improve cross-departmental data sharing in government following publication of the National Data Strategy, is healthcare a good microcosm or is there still some way to go on this?
I would say that overall data sharing in healthcare is still siloed, although increasingly we are seeing good examples of data sharing, especially over the last year when we look at COVID-19 reporting. The technical infrastructure for data sharing often already exists, but we could be better at communicating the value of data sharing and analysis to the management community to enable more evidence-led decision making and investment. I think there needs to be a real impetus for improving communication around the value of data, enhancing transparency of how it is being used, and making analysis intuitive for the management community to interpret.
In addition, I think we need to start with what problems need to be solved at what levels and work back from there. For example, you will always need operational analytical capability within a hospital to support real-time decision making for clinical and operational teams, whereas ICSs may require linked datasets across Primary Care and Secondary Care to really understand their population to make more informed strategic decisions and these could be 1-2 months outdated.