ChatGPT and the future of healthcare

7 July 2023

AI in healthcare: providing optimal outcomes for patients and institutions alike

RMB recently hosted the third healthcare innovation fourm focused on ChatGPT and the future of healthcare. RMB’s Healthcare Lead Banker Jefferson Murdoch welcomed the keynote speaker, Andreas Cambitsis, and the panel.

Says Murdoch: “To meet our shared goal of improving access to affordable, quality healthcare for all, it is our responsibility to catalyse all the advantages that technology – and artificial intelligence (AI) in particular – affords us. In this forum, we explored how ChatGPT can drastically change the future for a healthier South Africa.”

Keynote speaker, Andreas Cambitsis, is a former Endeavor board member and co-founder of technology and analytics company BSC (Business Science Corporation). With a Masters from Cambridge University spanning aerospace engineering and operations, he is a strong advocate for using AI and natural language processing to drive innovation and improve business outcomes.

The panel consisted of Dr Nonceba Koranteng, Senior Radiologist at Envisionit Deep AI; Joost Pielage, Chief Technology Officer at Quro Medical; and Danny Saksenberg, Chief AI Officer and Co-Founder of Emerge.

Demystifying ChatGPT and LLMs

ChatGPT is an AI large language model (LLM), meaning it is trained on language or text. To interact with it, you ‘ask’ it to generate text based on a specific ‘prompt’ – your human input. An LLM is a digital neural network inspired by the workings of the human brain. As such, it learns in much the same way as our brains do, identifying patterns and forming connections between concepts and ideas, then applying them when prompted to complete a task.

However, it is essential to be aware that the quality of the answers provided by LLMs is directly related to the quality of the prompts inputted. ChatGPT is trained using almost the entire public internet – about 1.5 trillion words – and its job is to produce what you have asked for. It deals better with concepts than with facts, so it will confidently answer your questions even if it does not know the answers by extrapolating and making things up. The technical term for this is ‘hallucination’.

Both anecdotal and empirical evidence suggest that AI complements humans extremely well – rather than replacing them, as many fear it will. In the healthcare sphere especially, it can augment the work of professionals, enabling them to make the best use of their time and expertise to provide optimal outcomes for patients and institutions alike.

AI in healthcare themes

Some of the key take-outs from the discussion, included:

Personal productivity

ChatGPT is a lot like having free access to a very intelligent intern. Like any clever but inexperienced individual, it will make mistakes – so it’s up to you to identify the best ways to communicate exactly what you want from it. This is where the work comes in: you will need to spend time experimenting with the model to determine what works and what does not. Communication is via prompts, and already a field is being developed around this called ‘prompt engineering’ – one can study this via free or paid online courses or learn through trial and error.

There’s no shortcut, really. To be applicable to your professional specialty, you need to know how to use ChatGPT effectively and input prompts that will deliver useful information, but a good YouTube channel or blog can offer you some ways to get started.

Stay abreast of new developments – ChatGPT 4 already has outstripped the capabilities and efficiencies of its predecessor, ChatGPT 3.5, which is currently free to use and became available in March 2022.

It’s important to note that because of ChatGPT’s tendency to ‘hallucinate’ when it does not know something, it cannot be left to run unsupervised – the human element remains indispensable. As with a talented intern, an experienced eye cast over its work will ensure that the model is performing at its best. Remember: AI won't necessarily take your job… but the people who use AI very likely will.

Personal productivity in healthcare: a case study

Dr Nonceba Koranteng, a senior radiologist consultant at Envisionit DPI, a company with both a South African and UK base, gives an African perspective of how AI can be utilised to augment radiologists’ performance in their day-to-day work. Given the shortage of radiologists in South Africa, extreme backlogs form very quickly.

 Dr Koranteng is helping to design AI systems to help make radiologists more efficient and enable them to diagnose more quickly. The model uses pattern recognition to cut through the noise and triage the most urgent cases, augmenting (rather than replacing) radiologists’ daily tasks and ensuring resources are directed where they are most needed.

Process improvement

Using APIs and open-source AI such as ChatGPT and other AI-based tools can speed up internal processes as well, enhancing productivity and using robotic process automation (RPA) to automate repetitive tasks in functions from finance to logistics and beyond.

Organisations with greater IT capability will be able to harness AI to help with designing new products and solutions across the healthcare spectrum.

Products and services

Within healthcare, AI-related tools have incalculable applicability across the entire industry, in areas including clinical decision support, personalised medicine, patient engagement and education, disease predication and surveillance, telemedicine and more.

Clinical decision support

LLMs are being trained on specialist medical knowledge with great success: among other things they are extremely good at diagnostics. This opens up many use cases complementing physicians’ expertise. Because they have access to a vast quantity of data, LLMs are able to flag up possibilities humans might miss. Providing reminders like this (based on the information used to prompt the model) can help to catch those cases that would otherwise fall through the cracks.

Personalised medicine

Treatment protocols today are based on broad averages, not personalised to suit an individual person’s genetic makeup. At a fundamental, biological level, patients vary widely – so a treatment protocol that works for one person won’t necessarily work for another. LLMs like ChatGPT open up the possibility of creating treatment protocols customised on a genetic level.

Patient engagement and education

Some studies already show that most patients prefer interacting with an AI physician than an actual doctor. This is not to understate the importance of human physicians – only to say that not all doctors communicate well with patients. Owing to time constraints, they usually produce highly technical discharge summaries that may be difficult for a patient to understand. Using that input, however, ChatGPT can provide a layperson’s – even a child’s – version of that information that is informative and beneficial to patients. Already Endeavor has an app that does just that.

Disease prediction and surveillance

It’s big news when medicine saves the life of someone who’s really sick; less so when it prevents them getting sick in the first place. This makes this the least ‘sexy’ use case of AI – and yet the one that may have the furthest-reaching impact.

AI beats humans at pattern recognition hands-down, and has done so for years. Given the right data set, machines can far exceed the ability of humans to recognise the early signs of everything from cancer to heart disease – AI is seeing patterns that we can’t. The possibilities for early intervention and prevention of noncommunicable diseases is immense.

Telemedicine and AI: a case study

 Quro Medical is pioneering a treatment model in which acutely ill patients are treated at home, instead of in a general ward, using remote patient monitoring capabilities. The patients’ vital signs are monitored in real time 24/7 and streamed into the cloud.

On a general ward the level of monitoring is fairly low – a nurse might run basic tests every four to six hours and another few hours may pass before a physician interprets those results and discovers that their patient has deteriorated overnight. As populations grow older, the risk of hospital-borne infections increases, too.

 Quro analyses a patient’s vital signs in real time, enabling immediate intervention. By leveraging AI and regular software engineering to combine multiple parameters, they can flag those patients most in need of care. Models can analyse ECG data, for example, at a very granular level, and detect cardiac events without labour-intensive activity that requires humans to analyse data manually. When the AI identifies an irregularity it is immediately transmitted to Quro’s dashboard for a clinical team to investigate further.

These are just a few of ChatGPT’s applications in healthcare – there are many others, including medical imaging analysis, mental health support, research and drug discovery, etc. It is also very useful in the academic context, helping the medical staff of the future to learn more efficiently and to specialise on a more granular level.

ChatGPT isn’t perfect, but its usefulness in lightening the load of heavily burdened and overworked clinical staff is undeniable. In our resource-constrained context, any help is valuable help – and the South African medical community needs all the help it can get.