Welcome back to our podcast HTN Let’s Talk, sponsored by Spirit Health!
For this episode, we interviewed Dr Janak Gunatilleke, director for healthcare data and analytics at KPMG UK. We discussed how AI can solve clinical problems, what is preventing AI solutions from being adopted, and his thoughts on to tackling the ongoing challenges in this area.
To start, Janak spoke about his current role and career in health tech.
Working in healthcare for more than 17 years, Janak started his working life as a doctor. Over the last 14 years since leaving clinical practice, he has worked in a variety of areas within healthcare, including working for large and SME management consulting firms, freelancing as a consultant, operational roles and for a health tech startup in London.
“Last year I completed a Masters degree in data science and published a book exploring the potential of AI in healthcare, challenges to successful implementation and how to do things better,” Janak said. “I’ve also been on the evaluation panel for the NHS Transformation Directorate AI Awards. Now I’m a director at KPMG where I lead our healthcare data and analytics work in the UK.”
Janak explained how KPMG focus on three areas within data and analytics. The first area is about “helping clients think about how to make the best use of their data, how to develop an associated data strategy, and how to realise the benefits and improve citizen outcomes,” he stated.
He noted that there have been a lot of national policies published in the last year, such as Data Saves Lives and the Goldacre report, along with a number of national initiatives such as the federated data platform and secure data environments. Janak emphasised the importance of regions and organisations having an understanding around how this all fits together, and what it means for them at a local level. “We work with ICSs and help them understand this better, how data can ultimately help them deliver their overarching strategies,” he explained. “This includes coming up with prioritised road maps of key initiatives that they need to implement to get them where they need to be.”
The second key area for KPMG focuses on implementation and follows on from the data strategy. “A lot of the NHS still mainly relies on descriptive analytics, where the focus is on analysing events that have occurred in the past,” he said. “Due to challenges in collating, cleansing and managing data, some of these insights can be weeks old.” In his experience, he added, whilst there are a few organisations leading the way, “for many the focus should be on getting some of the fundamentals right – the data governance or infrastructure.”
Janak highlighted the importance of developing workforce capability, to equip them with the skills, confidence and the ability to adopt the different mindset that is needed.
“We work with organisations through our unique analytics learning programme, to help develop the technical and non-technical skills of data analysts,” he explained. “We aim to better equip them for an evolving data and technology landscape in the NHS, and for them to effectively enable particular initiatives.”
The third area of focus for KPMG looks at identifying common problems faced by clients and developing a repeatable solution and effective product to deploy quickly to help. “For example; we developed a great product that helps hospitals with their strategic workforce planning, which helps them understand their demand better, the impact of their workforce retirement rate and where the gaps are.”
With regards to AI, Janak commented that many people tend to be interested, “but the reality is that it’s quite hard to implement AI solutions at scale in the NHS at the moment. We’re doing a lot of work to help NHS organisations with the foundations that they need to move along the analytics maturity curve, and to begin to harness the full potential of advanced technologies such as AI.”
Janak explained that he usually comes across three main challenges.
The first challenge is when there are “a number of initiatives and projects in progress, but there isn’t really an overarching vision or a structured plan. There is limited activity because either it’s not clear for the client where to start, or where the leadership is not bought into how data and analytics can add value.”
The second challenge comes about when the solution is technology led. Janak said, “It’s like getting it the wrong way round – having a cool piece of tech and trying to shoehorn it in by looking for a problem to solve. We really need to start with the problem.”
The third challenge is where the focus is solely on technology, with a lack of attention paid to other crucial factors such as stakeholder buy-in, integration into existing workflows, and training and support for end users.
“We work to help overcome these challenges, resulting in projects that deliver sustainable value,” Janak said. “For example; data strategies define a clear vision which the stakeholders can buy into, and which includes a prioritised road map. That sequences initiatives across the transformation timeline.”
He noted the importance of taking a user-centred approach, where the team “spends time upfront, talking to you and understanding the current situation and the challenges. As well as gaining stakeholder buying, this really clearly helps articulate and quantify some of the problems.”
Finally, he noted how they take multidisciplinary approach. “For example; if it’s a data strategy project, we don’t just have data analysts and technologists in the team. We bring in colleagues who have expertise in other areas, for example health operations or customer engagement. This really helps us to take a more holistic approach in terms of how to solve the problem and helps us address some of those key non-technical elements.”
How AI can solve clinical problems
Janak highlighted a controversial statement from Vinod Khosla, co-founder of Sun Microsystems. In 2017 at the Health Innovation Summit in San Francisco, Kholsa suggested “that machines will replace 80 percent of doctors in healthcare, and the future will be driven by entrepreneurs, not medical professionals.”
In terms of where we are now, Janak pointed out that AI is nowhere near replacing 80 percent of doctors. “If you look at real life results of AI-based solutions in healthcare, the results haven’t really been very impressive,” he commented.
The lack of adoption is reflected in academic publications where “the majority of studies are retrospective, so looking back instead of looking forward,” he said. He noted that small scale reviews and evaluation reports have found that actual AI adoption tends to be limited towards specific departments or use cases.
“Worryingly, there were a couple of reviews done on 647 AI tools that have developed during COVID and helped with COVID management,” he said. “The reviews found that actually none of them were really fit for clinical use, and that only two warranted any further evaluation. This brings us back to a question of where AI can help. Is there real potential?”
When identifying the potential of applications of AI to add value in healthcare delivery, Janak explained that there are two dimensions to consider. The first dimension involves looking at it from the perspective of the non-clinical area and frontline delivery of care. “In between those areas, there’s space for increasing productivity – it has some elements of back office work, but some elements of delivering care,” Janak said.
The second dimension is around the four stages of a patient journey and delivery of care: planning, prevention, delivery of care and ongoing management of care. If you put those two dimensions together, Janak said, “You get a nice grid which helps us look at identifying use cases for AI with more ease. Population health, for example, sits on that grid between back office work and prevention work. There’s an intersection there about enhancing care delivery, around things like supporting triage of patients, thinking about risk and so on. With ICSs and soon-to-be-established intelligence functions within the ICS, I’m excited at the opportunities within that particular space where AI could have a massive impact.”
Current challenges of AI solutions
Looking at the current challenges around AI solutions and what is preventing them from being adopted, Janak said: “I like to first look outside of healthcare at industries and companies that are using AI well. For example; household names like Netflix, Uber or Amazon. Whether it’s personalising recommendations or making sure that their operations are more efficient, they’re actually using AI to deliver value to themselves and to their users.”
In considering why it might be easier for them, Janak noted: “In most of these cases, data is readily available to these companies and in large quantities that helps them develop and refine effective AI models. So most of these companies have more than 100 million users, which gives them a readily accessible pool of data.”
He explained that these companies have “invested huge amount of money in developing sophisticated methods” to make the collection and use of this data “much more efficient within their companies.”
In addition, he added, these companies are solving real life problems and doing it in a seamless manner; for example, Netflix personalising a user’s viewing habits. Viewers don’t notice that they are using AI, he pointed out; it just happens and makes your life easier, or adds value and insight to the task you are trying to achieve.
They also have more room for error, Janak noted. “Maybe a recommendation provided by Netflix wasn’t perfect, the user might not follow the recommendation, but there’s no real damage done.”
Considering the challenges to AI implementation in healthcare, Janak highlighted three main categories: people, systems and technology.
On people, Janak said, users need to be able to trust and have confidence in the output of solutions. “Individuals involved need to have the knowledge to make informed buying and use decisions,” he said. “Also, frontline staff need to have headspace and protected time to engage with the development and implementation of these solutions.”
With regards to systems, challenges include integration into established clinical pathways. “Some of these clinical pathways have been there for 10, 20, 30 years, and the solutions that are deployed need to take those into account. You need to think about the potential disruption those solutions might cost to those established processes.”
From a technology perspective, “access to data within healthcare is not easy. We have to consider things like bias, and solutions need to work as intended across different organisations and different populations which can vary quite a bit.”
To overcome some of these challenges and develop a better methodology to design, deploy and operate AI solutions in healthcare, Janak made a number of suggestions.
“Firstly – really focus on solving a real problem. Always start with the problem, rather than with the solution or with the technology.”
Secondly, he encouraged the use of a roadmap that considers the lifestyle of AI solution development and the different stakeholders that need to contribute and work together. “The road map has five stages,” Janak said. “It starts with identifying the problem; designing it; developing a solution that is safe, effective and scalable; implementing the solution; and then continuous improvement and monitoring.”
Janak’s third suggestion for overcoming AI challenges centred around enablers. “Enablers are a combination of elements that need to be achieved during a solution, and some that just need to be there during the solution lifecycle. They can be at solution level, an organisation level, regional or national level. No matter how good you are or how hard you try, it’s difficult to succeed if some of these enablers are not there.”
The future of AI in healthcare
Looking at the future of AI in healthcare, Janak said: “I think it’s going to be a gradual process.” He noted that there tends to be a lot of interest in activity in areas such as radiology, where there has “always been a lot of digitised data and a more digitally savvy workforce.”
He added a belief that there is potential for AI to make a difference in the back office area. “There’s less risk to manage, which means you can experiment a little bit more than you can when it’s to do with delivering care. I think there’s a lot more that we can do, and much more quickly, in this area. If the right solutions are developed and implemented in the right way, it could have a massively positive impact on the workforce – we can get rid of some of the more mundane tasks and help with the more complex, to really help the workforce operate at the top of their license. This could be really beneficial in the current environment where there’s so many vacancies.”
He added: “On the flip side, the workforce does generally need to have the knowledge, skills and the confidence to choose and work with some of these solutions, and to be able to use them safely through understanding some of the risks and limitations.” He re-emphasised the importance of giving staff time and space to meaningfully engage with AI, adding: “I think there needs to be something done about how to create the right roles, the resourcing and the funding to make this happen.”
Overtime, Janak said that he believes we will start to see solutions succeed in wider clinical areas and on the frontline. “But I think there needs to be a number of things that needs to happen with the right investments and the right people working together,” he said.
We would like to thank Janak for sharing his time and thoughts with us. Janak’s book, ‘Artificial intelligence in healthcare: unlocking its potential’ is available here.