Automating the clinical coding process in an NHS Trust — ep. 2
In my last post I gave an introductory piece about my professional background, described from my perspective the issues a manual coding process faces and touched briefly on the potential of an automated coding service. The response from colleagues both within my Trust and across the sector has been fantastic and if you’re still here reading then I’d like to personally thank you — as I said in my previous post this is my first foray into committing my thoughts to ‘paper’ in such a way and it’s proving challenging in all the right ways. I do wonder though whether my ‘personality’ comes across at all through this medium…and well if it does just accept my condolences in advance.
In this post I’d like to further expand upon the potential of an automated solution as I perceive it whilst referencing work already carried out in this field, vital work that I feel we in the NHS can build upon. I’d also like to devote some of my writing time to discuss the vital human component at the heart of this potential development in the form of my, your, our clinical coding workforce.
You may recall I previously included a simplified process map of the typical manual clinical coding process, RAG rated to highlight pain points in terms of inefficiency;
Consider each stage of this process and there’s multiple intervention points where automation would be likely to have a definitive impact. Indeed, since taking on the Head of Coding role at Kettering General Trust I’ve had many chats with people frankly far more intelligent than I around automation and the number of potential intervention/use cases for it across the healthcare sector is staggering.
But let’s stay focussed on the coding process as it’s something I’m slightly more knowledgeable about. In Figure 2 I’ve taken the same simplified process and given some examples to explain the ‘art of the possible’ — I’ve included annotations describing stages of the process but in essence in this scenario we have a functioning EPR with SNOMED CT capability (commonplace) and an AI solution in the form of a sequence-to-sequence NLP;
I’ve conservatively reflected the ‘amber pain’ of the coder audit and optimisation in order to reflect the relative inefficiency experienced whilst training against the initial output of the AI solution, this will decrease as training of the solution continues and prediction accuracy of free text increases.
The technology to automate this process already exists and is well proven…medical coding in the United States has been automated for some time and continues to develop further. Consider this recent NVIDIA blog piece which describes a healthcare tech start-up in San Francisco and it all sounds…rather familiar;
After every doctor’s appointment or procedure, a clinician’s summary of the interaction is converted into these codes. When done by humans, the turnaround time for medical chart coding — within a healthcare organization or at a private firm — is often two days or more. Natural language processing AI, accelerated by GPUs, can shrink that time to minutes or seconds.
San Francisco-based Fathom is developing deep learning tools to automate the painstaking medical coding process while increasing accuracy. The startup’s tools can help address the shortage of trained clinical coders, improve the speed and precision of billing, and allow human coders to focus on complex cases and follow-up queries.
The door is open.
We need only walk through it.
As I said at the outset of this post I wanted to expand upon the future of the human in this automated process. This is regularly glossed over or outright ignored for various reasons and I feel it’s at the heart of this topic.
Speaking frankly — some of my fellow coders have worries about automation…this is all based on my time spent within the profession. These worries are primarily centred around job security concerns but may also be underwritten by a more generalised lack of digital literacy (a topic expertly handled by colleagues of mine including Anna Awoliyi, Dione Rogers and Kerry Rodger to name only a few).
Coders…in my mind the automation of the coding process is the next step in a process of natural evolution…consider it in the broader history of coding itself;
How do we standardise and measure our performance? Code the documentation.
How do we validate and provide assurance on that output? Train your coders and audit against national standards.
How do we improve documentation for this purpose? Engage with your clinicians.
How do we improve this process further? Automate with coder validation.
Consider the role of a conventional coding auditor/trainer, throw in some SNOMED CT knowledge, add in some basic healthcare informatics capability and you’re not far away from the coder in the automated process. Automation does not exist to replace your role and nor can it;
Moseley first began to use AI-enhanced coding a couple of years ago, she was suspicious of it because she thought it might put her out of a job. Now, however, she believes that will never happen and human coders will always be necessary. She notes that medical coding is so complex and there are so many variables, and so many specific circumstances. Due to this complexity, she believes that coding will never be fully automated.
She has effectively become a code supervisor and auditor — checking the codes the system assigns and verifying if system recommendations are appropriate for the specific case. In her view, all coders will eventually transition to roles of auditor and supervisor of the AI-enabled coding system. The AI system simply makes coders too productive to not use it.
Hopefully this helps lays out the foundation to my thoughts around coder involvement with AI, that human aspect that is vitally important to success. I won’t rehash excellent work done elsewhere in terms of change management and effective leadership — I suppose I’d just reiterate to those keen on driving this innovation across the NHS that success will depend (in addition to other things) upon our honesty and openness as digital leaders for our respective workforces.
In my next post I’ll likely go into a bit more detail about our exploratory forays into this area locally within Kettering General Hospital…Though who knows, all things are mutable and we shall see where my intervening week takes me (exciting!). I’d also like to encourage you to reach out to me if you’re interested in this — a collaborative approach is the watchword here so get in touch via my Twitter. Hopefully you’ll still be here to read episode 3 and until then take care.
