Dr Rishi Das-Gupta—written evidence (LLM0037)

House of Lords Communications and Digital Select Committee inquiry: Large language models

Dr Rishi Das-Gupta

Dr Rishi Das-Gupta is Chief Executive of the Health Innovation Network, the Academic Health Science Network for south London. He has worked to improve healthcare, patient experience and outcomes for over 20 years in a variety of roles focused on shaping and implementing government policy, working with health and care providers and redesigning services. He is passionate about the use of technology to enable clinical and operational change and to empower patients to manage their own conditions. Rishi’s experience spans UK and US healthcare systems and he is a fellow of the Sciana programme collaborating with German and Swiss healthcare leaders. Prior to this role he has held executive roles at NHS trusts as CITO, Director of Innovation and as Director of Operations and worked as a medical doctor. Rishi also has experience working as a strategy consultant at Oliver Wyman and McKinsey and Company. In addition to his primary medical qualification (M.B.,B.S.). Rishi holds an M.A. in law (University of Cambridge) and an M.B.A. (London Business School).

Health Innovation Network South London

Hosted by Guy’s and St Thomas’ NHS Foundation Trust, we aim to speed up the best in health and care by connecting the NHS with industry to support the spread of all types of innovation from new technologies to ways of working and service improvements. We do this to improve services for patients, enable NHS efficiencies and support economic growth.

At a local level, we work to:

• Identify innovation and improvements to specific problems, whether that be within a healthcare, academic or business setting, creating an innovation pipeline.

• Empower innovators and those developing improvements to further their ideas and get them in front of the right people.

• Advance the uptake and spread of innovation and improvements by delivering programmes and initiatives within the NHS and social care.

Currently we are working with NHS England to investigate the possibilities that Ambient Voice Technology (AVT) has for freeing up clinicians time and improving patient-clinician interactions.

Questions

The Committee is seeking evidence on the following questions (there is no requirement to answer all questions in your submission):

Capabilities and trends

1. How will large language models develop over the next three years?

Broadly there are five areas I see LLMs contributing to improvements over the next three years.

a)       Efficiency/automation of current processes – particularly where these involve summarising and generating text (such as appointment or discharge letters).

b)      Improvement of data for population health/big data projects- through coding clinical notes (especially diagnoses) to allow effective use of disease registries and to standardise the quality of coding across health systems.

c)       Generation of personalised care experience and providing patients with real-time access to tailored information (including chatbots, health assistants and translation).

d)      Improvement of staff experience – though the use of tools such as ambient documentation (such as Ambient Voice Technologies (AVT)) to reduce time spent on administrative tasks.

e)       Training tools – development of improved training scenarios and more engaging online training based on the use of LLMs.

2)              What are the greatest opportunities and risks over the next three years?

LLMs offer an alternative way to interact with technology and to have conversational interfaces for clinicians and patients. For clinicians, automating documentation is a near-term opportunity that might be live in six months or less. For patients, rather than patient information being presented as a leaflet, we might offer it as a chatbot, virtual health coach and the language used might be tailored to the level of medical knowledge each patient.

However, the models will be constantly improving and software companies are unlikely to want to take the risk being associated with using the models in clinical practice. So, we are likely to see LLMs adoption first in fully-supervised processes such as documenting a note in a clinical interaction where the clinician was present throughout and signs-off the note, before their widespread use in summarising all notes in a way that doesn’t need checking.

This will be accelerated by suppliers improving both the validation of their models and the explainability (I recently saw a model which showed where each statement of a summary had been drawn from in the note).

a)    How should we think about risk in this context?

In the context of the NHS, the risk of not using LLMs (the ‘do nothing’ option) needs to be clearly laid out. Clinician burnout is a well-documented reality in the NHS,[1] [2] and the administrative burden of tasks such as note taking and ordering tests is one of the main contributors.[3]

AVT has the possibility to capture patient interactions in real time, creating a record of the key points which the clinician can check. If this was utilised across the 1.75m daily patient appointments this could save hundreds of clinical hours, dramatically changing the capacity within the NHS. This could be used to see more patients or decrease the amount of overtime being worked to improve work/life balance and help retain staff.

September 2023

3