AI adoption within the National Health Service
In May 2023, the World Health Organization (WHO) released a statement calling for the safe use of artificial intelligence (AI) within all sectors of treatment. They called for the large language models (LLM) that are currently being used within society, e.g ChatGPT, to be monitored and rolled out with caution in large-scale healthcare use.
As I enter my fourth year of medical school, the use of technology is becoming more prevalent when it comes to the treatment of patients. Within my lifetime, UCL-affiliated hospitals have gone from widespread paper use for notes to computers, allowing for a centralised hospital record. While pagers remain in use, the preference now lies in the widespread adoption of mobile phones. The next technological step for healthcare provisions seems to be with the integration of AI.
So, how can we use it?
Shortages within the NHS
The current trends among healthcare staff are currently following a worrying trajectory. According to current predictions, NHS England predicts that the staff shortage will rise to 571,000 in 2036 from the 154,000 today. This is across the healthcare services in England alone. As such, wait times are increasing across all specialties. The number of people waiting to receive treatment in the UK now stands at over 700,000, with more waiting over 18 weeks to be seen by a consultant.
An argument for AI that some put forward is that it can be used as a way to reduce the burden of staffing shortages. Some hospitals in the United States did trial the use of AI during the earlier stages of the pandemic. While it did offload some work during the early stages of COVID-19, it also relied on the patient's decision making and accuracy of their descriptions..
While AI can follow guidelines, they are not good at providing a clear case-by-case plan for the patient. Patients can have multiple comorbidities and receive multiple pharmacological interventions. This can lead to ultimately harmful recommendations, which, without the supervision of a senior physician, could lead to potentially fatal outcomes. Even Google’s Chief Health Officer, Dr. Karen DeSalvo, has warned against the immediate use of AI and cautioned that limits need to be implemented to guarantee patient safety.
It should be noted that mistakes do happen when physicians give advice. It is common. But there is a reason why Multidisciplinary team meetings are present. Complicated cases can be discussed amongst several team members to provide a suitable treatment plan. Individuals seeking the use of AI will typically only review one model with no additional support. ChatGPT 3.5, for example, only works with data up to 2021. While two years may not seem like a long period of time, small changes can affect the outcome. If done incorrectly in the primary setting, patients can induce tissue damage.
How to use AI in the future
The optimal use case for AI will be as an assistive tool. Models can be trained to detect specific forms of a condition, e.g., detecting polyps that are suspected in patients with colon cancer. Studies carried out in 2022 found that missed cases of colorectal cancer were noticeably lower than standard physician detection rates. As such, the idea that AI is not replacing the department of radiology and instead acts as a second pair of eyes.
AI can serve as a useful tool that aids and elevates all healthcare staff. Doctors spend the majority of their time in clinics trying to document and write letters of referral or discharge to separate healthcare teams. The LLMs could be used to document findings, provide detailed outpatient plans, and help prioritise the list of jobs that need to be completed without the issue of models providing unverified information due to influence by third parties.
As a medical student, I believe that the next few years will be crucial to deciding whether we can have a safe implementation of AI within the NHS. Slow, cautious implementation within the roles of doctors, nurses, paramedics, and all healthcare professionals will be critical to ensuring the long-term effectiveness of this new piece of equipment within the proverbial toolkit.