Outsider turns to his friendly neighbourhood Large Language Model for some novel suggestions on pharmacy negotiations

As the nights grow gradually longer, one feels autumn setting in. It feels all so familiar and yet different at the same time. Classically, in true community pharmacy style, we enter into our peak season in some form of existential crisis. Typically, this involves the old favourites of contract anxiety combined with a lack of preparation for seasonal vaccination programmes.

This year, Community Pharmacy England (CPE) has compounded these by reassuringly informing us that a sixth of the estate is under imminent threat of financial collapse. It’s not for this column to rigorously analyse the validity of this report, which may be extremely prescient or it may be wide of the mark. But Aesop’s fable of ‘The Boy Who Cried Wolf’ does spring to mind. 

Certainly, in reading the report, the figures look stark and concerning, and there is no doubt that costs have increased across the board. What will this mean for attendances at the national trade shows – will pharmacy owners be more or less likely to be suckered into buying a new shiny thing?

The most hyped shiny new thing that is likely to attract attention in community pharmacy, just as elsewhere, is the promise that artificial intelligence (AI) will enhance and fix everything. Search for any phrase on a major search engine and the first half of the page will be dominated by some AI assisted answer. Specifically, a Large Language Model (LLM) generative AI.

If you do attend a trade show, say in October, you are bound to stumble across people promising that the AI revolution will save you and your business. Something that might be necessary if CPE’s dire predictions for the future are accurate.

But what can a LLM actually do for community pharmacy?  To find out, I simply asked the AI. Its response:

“AI can assist UK community pharmacies by automating prescription processing, improving medication management through predictive analytics, and enhancing patient care with personalised health recommendations. It can streamline inventory management, optimise staffing, and reduce human errors. 

“Additionally, AI-driven chatbots and virtual assistants can provide 24/7 support, answering patient queries and offering health advice, thus freeing up pharmacists to focus on more complex tasks and improving overall efficiency.”

Sounds impressive – but how? To answer that you need to understand the way in which LLMs work. They’re essentially a probability engine: you feed them huge amounts of training data (the internet) and then they use probability to generate text that looks like normal human text. In principle, this is great, especially if your training data looks like the answer you want back.

Famously, in community pharmacy, we are useless at making records – see the history of advanced services for more information. This begs the question – where is the training data? Unless the LLM is fed a huge chunk of historic prescription data (which isn’t happening), it’s hard to see how a LLM can help there. Sure, ChatGPT can pull an answer from CPE’s website about how to endorse an SSP supply, but it can’t do it for you. Ask it more complicated questions and it soon starts to generalise and pull in less accurate information.

What then about patient interactions? I asked ChatGPT about lots of awkward scenarios, ranging from some recent health issues experienced by family members through to digging out some really old worked examples from textbooks. I was expecting it to be a car-crash, but it was pretty good. One of the strengths of LLMs is that the internet is big – that’s lots of training data.

It’s difficult to understand how it could be used in a true face-to-face consultation without being somewhat cumbersome. It’s not well suited to natural conversation.

The big downside is that you can’t use a LLM as a decision-making tool in a healthcare setting. Whatever answer you get from a LLM, you need to double check. Any LLM acting as a decision-making tool is likely to fall under the scope of NICE and that’s a regulatory hurdle that is unlikely to be overcome any time soon.

So, is there any place for AI in community pharmacy with the current generation of LLMs available? I gave ChatGPT one more go: How can CPE improve its performance? It replied: “CPE can improve its negotiating performance by leveraging comprehensive data on pharmacy contributions and patient outcomes to strengthen their case. It should enhance collaboration with stakeholders, streamline communication strategies, and advocate for evidence-based funding models. 

“Additionally, adopting a proactive approach to policy engagement and building strong relationships with government officials can help in securing more favourable contractual terms and ensuring sustainable support for community pharmacies.”

Perhaps there is a place for AI after all? And the financial crisis may be as real as the boy cries. If it is, I look forward to the major buying groups choosing to hold their annual conferences in Margate rather than Mauritius.

Outsider is a community pharmacy commentator