- Correspondence to: D M Rosewarne david.rosewarne{at}nhs.net
On Christmas morning, Santa Claus delivers presents to many good children worldwide. His operation has grown hugely since the illustrated poem Old Santeclaus with Much Delight introduced the modern Santa in 1821, when the world’s population was just one billion.1 Santa’s work seems to have been scaled up without customer dissatisfaction or bad press—so he must have a solid recruitment and retention plan for his elf workforce, making smart use of digital transformation.
The NHS, however, is struggling to follow suit. Its recent workforce plan will take time to implement.2 As performance continues to worsen,3 digital transformation is needed quickly so that NHS staff can be deployed more effectively.
Our everyday lives are seeing rapid advancements powered by artificial intelligence (AI),4 but NHS spending on information technology (IT) is relatively small—less than £1bn a year.5 Multimillion pound investment in health AI hogs the digital limelight, such as the £21m promised for AI in diagnostics,6 but less complex approaches are neglected.
Much more should be made of existing IT, with closer collaboration among IT departments and clinical staff. IT departments need capacity to provide a full and flexible service,5 and users need education to understand what can be automated to have the best impact. NHS systems tend to lack interoperability, but data from multiple databases can be harvested quickly by clinicians with help from IT colleagues.
Reducing delays
For example, outpatient clinics often request reports on imaging that were performed a week ago or more but haven’t yet been reviewed. One of the authors (DMR) tackled this problem by using a short computer program. It uses patient administration and radiology databases to provide a prospective list of outpatients who are due to attend in the next month whose imaging is still unreported. Prioritisation such as this aims to reduce delays in outpatient clinics. Formal evaluation of the process is currently in progress: the result of such interventions must be quantified.
If this sounds trivial, that’s because it is: the program was conceived and deployed in a few hours by DMR (the author who knows less about programming). Yet many hospitals face the same issue, and there are probably several full time staff equivalents across the NHS working on this niche problem. The digital transformation of such tasks is an example of process automation.
Demonstrable service improvements bring a quick return on investment—and on management support (if such changes aren’t supported, perhaps new management is needed). Legions of similar tasks could be done this way if clinicians had the training and time. A paradigm shift is required to generate and share such solutions, which individual NHS services can then adapt and use. NHS-wide repositories of programs already exist, such as https://nhs-pycom.net/resources.
Repetitive tasks
More recently, large language models (LLMs), especially the GPT (generative pre-trained transformer) family, have captured headlines. Their algorithms perform natural language processing to understand questions and generate responses. The quality of responses depends on how models are trained and how questions are asked. A valuable feature is that they can perform repetitive tasks cheaply.
We have used LLMs to categorise tens of thousands of radiology reports in a few days—for example, asking whether a report includes a diagnosis. These models can perform such tasks as accurately as humans but much more cheaply, costing about 0.1 pence per report. We estimate that it would cost at least 100 times that to do this by human hand, with worse error rates and higher burnout.
Other potential applications of LLMs include completing clinical audits for service improvement. One of our Christmas wishes is to reduce unnecessary imaging, such as the search for cancer in small lymph nodes in children’s necks or the search for brain tumours in patients with typical headaches. To achieve this we first need to show, using natural language processing of a very large number of reports, that the yield from such imaging is very low.
Like Santa’s elven crew, the digital elves of process automation and LLMs bear Christmas gifts for the NHS: potential efficiencies and cost savings, given sufficient management support, time, and training. Such bottom-up digital transformation would then be limited only by the imagination of NHS staff.
Footnotes
-
Competing interests: We have read and understood BMJ policy on declaration of interests and declare the following interests: RDM works for Health2Works, a digital heath company that stands to benefit from digital transformation in healthcare but has no income from any of the projects we have described. Our website (https://digitalelf.org) is a not-for-profit blog site.
-
Provenance and peer review: Not commissioned; not externally peer reviewed.
