The opportunities technological advancements present to our current healthcare system
Writer and Artist: Sophie Maho Chan
Editor: Ebani Dhawan
In the age of Alexa and driverless cars, what can technological advancements offer to medicine? In his best-selling book, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, Dr Eric Topol, celebrated cardiologist as well as founder and director of the Scripps Research Translational Institute, shares his vision of how artificial intelligence (AI) can transform all aspects of healthcare. This includes everything from booking appointments to diagnosing diseases, making treatment decisions to conducting surgeries. Gone are the days of wasting hours to see your GP only to leave the hospital empty-handed. By harnessing Big Data such as whole-genome sequences and personal medical histories—argue experts like Dr Topol and UK’s Health Secretary Matt Hancock—AI can not only help us eliminate clinical inefficiencies but also personalize care and revitalise patient-doctor relationships.
Of course, we are nowhere near achieving this grand vision. Very few AI algorithms have undergone clinical trials; even fewer have actually been put in practice. Most algorithms today also have narrow scopes of application. Nonetheless, studies are cropping up suggesting that AI can outcompete professionals in interpreting medical scans, diagnosing cancers and predicting suicide risks. Correspondingly, healthcare has emerged as a major playing field for tech giants like Google, Amazon, Apple and Microsoft. This is particularly evident in London, where five NHS trusts—including the University College London Hospitals (UCLH) NHS Foundation Trust—have partnerships with the Google-owned AI company DeepMind. This has formed the foundation of the Streams app used today by Royal Free London and Imperial College Healthcare to predict acute kidney failures.
There is much to unpack in understanding the role and potential of AI in healthcare, which is why the topic will be explored in a three-part series. This introductory article will be followed by an overview of the diverse applications of AI in healthcare, looking at some of the most exciting (and at times eerie) frontiers of innovations. The series will then be concluded with a discussion on the implications of using AI, emphasizing what this means for patient-doctor relationships and data privacy.
But first, we must familiarise ourselves with the basics: what is AI and why bother implementing it in healthcare in the first place?
A crash course in the history of AI and how it works
Conceptually, AI dates back to at least 80 years ago, but it was only at a conference at Dartmouth College in 1956 hosted by John McCarthy and Marvin Minsky where the term ‘artificial intelligence’ was coined, referring to intelligent machines that have the ability to achieve goals like humans. In years to follow, AI took the computer science world by storm, especially with contributions from Frank Rosenblatt. The New York Times reported in 1957 that “The Navy revealed the embryo of an electronic computer that it expects will be able to walk, talk, see, write, reproduce itself and be conscious of its existence”. Within two years, ‘machine learning’, the idea that computers can learn without being explicitly programmed with rigid rules, emerged. This continues to form the basis of many developments today.
After a period of inactivity and fundings cuts known as the ‘AI winter’, the 1980s welcomed a new era of innovation, this time focusing on commercial algorithms and ‘deep neural networks’ (DNN). DNN refers to a type of machine learning algorithm that can train itself by processing multilayered networks of data. To understand what this means in the context of medicine, consider a DNN machine trained on interpreting chest X-rays. First, the machine is fed with a dataset of chest X-rays labelled with diagnoses from expert radiologists. Once sufficiently trained, the machine is ready to be exposed to an unlabelled chest X-ray image. This will go through multiple layers of artificial neurons within the machine, with each layer responding to different features (like edges and shapes) in the image. As the X-ray image reaches higher layers, increasingly complex features are processed and eventually the machine’s interpretation is generated from the top layer. The larger the training dataset and the greater the number of layers, the better the output.
From driverless cars to facial recognition, when we speak of AI today we mainly refer to DNNs. These algorithms have immense pattern recognition abilities, making them suitable for diagnosing cancer from medical images, recognising mental health conditions from voice recordings and identifying genetic conditions from facial characteristics.
Our current medical climate: The ‘bad’ and the ‘ugly’
Many shudder at the thought of a ‘robot’ dressed in a lab coat mechanically conducting check-ups. But the truth is that healthcare today is already dehumanised. In the UK, the average GP appointment lasts for 9 minutes and 22 seconds. The number drops to 48 seconds in Bangladesh. Clearly, doctors do not receive enough time to get to know a patients’ story, understand their context or deliver personalized care; this is particularly problematic for patients with underlying long-term conditions that make up 50% of all GP appointments in the UK. Furthermore, according to surveys, the top descriptors patients associate with their doctors are “hurried”, “busy” and “rude”. This has been largely credited to the inordinate amount of time doctors spend typing notes during appointments instead of making eye contact with patients. Yet alarming data suggests that up to 80% of doctors’ notes are copy-and-pasted.
Faced with explosive growth in demand for healthcare, doctors’ workloads are soaring and quality of care is being increasingly sacrificed for quantity of appointments. One-third of GPs in the UK report dissatisfaction in the amount of time they get to interact with patients. Numbers of patients per GP is rising. Depression, burnout and suicide are becoming increasingly common among healthcare workers. As many as 20% of doctors in the UK suffer depression and each year 300 to 400 physicians commit suicide in the US.
As for patients, superficial contacts with doctors can be linked to misdiagnosis. In the UK, misdiagnoses cost the NHS £1 billion over six years (the same cost it would take to train 4,453 doctors). Cancer is responsible for particularly alarming statistics. Four out of 10 cancer patients are misdiagnosed at least once. One in five patients in a UK-based survey waited for at least half a year to get a correct cancer diagnosis.
Misdiagnoses lead to mistaken and/or excessive treatments. It is speculated that up to a third of operations performed in the US may be unnecessary. This not only exposes patients to unnecessary treatment risks but also generates stress, increases financial burden and wastes resources.
How can AI help?
Clearly, AI can reduce doctors’ workloads. They can also work 24/7 with consistent accuracy and attention; no holidays, complaints or burnout. In any given time, algorithms can also ‘see’ more patients and process more information than even top medical experts at a significantly lower cost.
AI also presents novel powers. Of the 10,000 human diseases that exist, a human doctor can only recall a handful. Furthermore, it takes decades of experience to develop diagnostic accuracy and decision-making abilities. Frankly, this can never measure up to an algorithm that can (in theory) draw from data shared by thousands of doctors worldwide as well as keep up-to-date with the latest biomedical research. Algorithms will also have no problem combining knowledge across disparate medical fields. The ultimate goal is that by creating a platform for AI to integrate existing information with an individual patient’s Big Data—including personal medical histories, genome sequences, gut microbiota, family histories and even real-time analytics—we can achieve a new era of personalised medicine.
Sounds idealistic? Perhaps. Many hurdles must be overcome if we want to achieve anything close to this. We must also ensure that AI does not replace doctors, but instead enhance their work and give them the time to do what they do best: care.
But while AI-driven hospitals may seem distant, as you will see in the next article, ground-breaking developments are taking place at this very moment.