Artificial intelligence (AI) has been slowly transforming the delivery of healthcare for the past few decades. Some of the first applications date back to the early 1970s, far preceding the recent wave of AI hype. The MYCIN project was one of those early examples. It was an expert system designed to identify bacteria causing severe blood infections and recommend the right dosage of antibiotics given a patient’s weight. This reduced the risk of medical error and increased the quality of treatment.
In the early 2000s, London’s Royal Free Hospital and the Department of Engineering Science at the University of Oxford developed Tallis - a platform that allows clinicians to add information, such as test results, to a patient profile. The software then recommends the most suitable course of action given the input data and current clinical guidelines. This tool has helped doctors to make decisions about how they should treat their breast cancer patients. Historically, AI has helped to improve certain areas of medical practice such as clinical decision-making, and the quality of treatment received by patients within a clinical setting.
What is different about what is happening now with AI?
First, the increase in the amount of data that are being collected in a machine-readable format and the increase in computing power has led to a multiplication of applications of AI in healthcare. Second, advances in machine learning techniques have increased the overall performance of these tools in specific areas of healthcare, such as pattern and imagine recognition.
This has led to several applications of AI that are positively impacting the way in which medical practitioners work. Healthcare professionals face a huge knowledge challenge as the pace and complexity of medical knowledge now 'exceeds the capacity of the human mind'.
In 2014 it was estimated that 2.5 million scientific articles are published every year in English-language journals. AI can help medical practitioners keep abreast with this. IBM’s Watson can process existing literature alongside patient data to aid diagnosis and then recommend treatment options. AI can also help healthcare practitioners better detect existing conditions.
Evidence shows that AI is enabling interpretation of mammograms 30 times faster than humans and with much greater accuracy, reducing the need for unnecessary procedures and misdiagnosis. AI can also be used to automate administrative tasks and allow healthcare practitioners to focus on more important tasks or tasks that require human skills such as empathy.
‘AI has a history of enabling change in healthcare’
The applications of AI in healthcare have broken out of the clinical setting and are directly reaching people at home. Patient-facing applications can empower individuals to better manage their long-term and chronic health conditions. Apps powered by AI algorithms have been developed to help people better deal with conditions such as diabetes.
These types of apps process blood sugar levels and send guidance and information to help individuals manage their disease. AI is helping change the nature of healthcare delivery as care no longer needs to be received within the boundaries of a clinical setting. AI can also encourage the transformation of the wider healthcare system, helping it to move from one that focuses on acute care to one that focuses more on prevention.
The promise of AI’s positive impact on the way healthcare is delivered is great, however several factors should be considered to make this a reality. Whilst ethics should remain one of the main focuses, the relevance and design of the applications of AI in healthcare should be strongly considered.
Applications need to be relevant and actually solve a pressing issue faced by healthcare practitioners or patients. It is a recurring comment at healthtech conferences or roundtables that there is, at times, a lack of clinical engagement in the conceptualisation of the problems faced by healthcare practitioners that could be solved by an application of AI. This results in medical staff being approached by well-intentioned engineers who think they have the solution to a healthcare problem that is poorly defined or not relevant.
Interfaces need to be designed to meet user needs. Designing an AI that sends too many notifications to a clinician might end up creating an unnecessary workload or might result in a clinician ignoring important notifications. The way in which information appears on an interface can have an impact in a clinician’s behaviour. Evidence has shown that the first and last item on a picking list have a higher likelihood of being selected. In addition, having a degree of transparency or explainability in the design of an AI that would show how an algorithm reached a certain decision and can help instill confidence.
AI has a history of enabling change in healthcare. Current advances should be welcome, but the hype must not cloud judgement around relevance and design.
Eleonora Harwich will take part in a debate on AI-enabled healthcare tomorrow (22 March) at the HIMSS UK Executive Leadership Summit in London. Find out more here.
Eleonora Harwich is Head of Digital and Tech Innovation at Reform, an independent Westminster-based think tank, where she started in 2015. Her work now focuses on how tech innovations can help public services deliver better outcomes for people. She has particular interest in the public-sector applications of artificial intelligence (AI). She has led and co-authored a paper on AI in the NHS. The project looked at how AI could help the NHS deliver its service transformation plans, as well as the challenges that need to be overcome to make this a reality. She is a member of the AI Programme advisory board for the Kent, Surrey, Sussex Academic Health Science Network and a member of the advisory board forBlockchain Live. She is also the London Hub Lead at One HealthTech a volunteer-led network that seeks to promote diversity in healthtech.
