"It is health that is the real wealth, and not pieces of gold and silver." - Mahatma Gandhi
I am trying to write an article on AI use in a few different parts of the market and society at large. The intent is to get a sense of the rate of change and build up habits of learning about how people use machines to improve their everyday lives.
I want to do a few different articles on different areas where data is used to try and keep me thinking about how technology changes the world and just broaden my horizions. So I started with a literature review on data in medicine within the U.K.
I felt that I found a place where data was incredibly important but while plenty of AI was in use; in contrast to LinkedIn and OpenAI where AI is discussed constantly as a revolutionary dynamic force what was discussed, celebrated and shown as having the biggest impact in the medical industry was often much more simpler data project undertaken to address specific situations.
Focus was on more classical tabular AI for classification processes or convulutional neural network aka computer vision to detect cancers.
I use this article to dig through what the NHS says about the use of data in medicine.
Showcasing improvements in the NHS
The NHS is quick to talk about its success with data. A quick google search brings up this.
source: https://www.england.nhs.uk/gp/national-general-practice-improvement-programme/
- Gresleydale Healthcare Centre, in Swadlincote, Derbyshire, worked together with its Patient Participation Group to implement changes in telephone and website systems that would help to address the ‘8am rush’. In this short video, practice staff and patients explain the improvements that have been made to make booking appointments easier and the impact this has had on patient and staff satisfaction.
- Abbey Medical Centre, in Kenilworth, Warwickshire, transformed the way patients accessed care by using its new digital telephony system to identify when telephone lines were busiest and ensure more reception staff were available at peak times. As a result, the number of abandoned calls at the practice fell. The practice also focused on improving access and care continuity. Ryan Smith, Non-Clinical Partner at the practice, said: “The General Practice Improvement Programme really helped us continue to improve our services so that more patients can get the care they need more easily. It’s particularly helped the patients who we previously saw most frequently, many as often as twice a week, by enabling us to focus on how we provide the best care for them, so they get the care they need with fewer appointments, which is better for them and means there are more appointments available for others who also need them.”
- Hillside Practice is a semi-rural practice close to Saltburn-by-the-Sea in North Yorkshire. It transformed patient experience with marked improvements in website use, patient safety measures and care navigation. The practice measured patterns of patient contacts to match capacity to demand. This helped the practice make full use of its multi-disciplinary team, e.g. social prescribing link workers, to support with non-clinical problems. Patient safety processes also improved as the practice enhanced its patient re-call process.
- Crabbs Cross Surgery, in Redditch, Worcestershire, made a range of improvements to improve patient access and experience. It invested in care navigation training for staff, put in place consistent processes to make use of the skills in its multi-disciplinary team and ensured that each patient saw the right professional at the right time. The practice improved its appointment allocations through more proactive care and tailored the length of appointments to ensure adequate time to respond to individual patients’ needs.
I think this challenged the view that I found even on the governments own website that the NHS is not good a digital transformation. I note the following attributes in all these advertised cases they are not top down implemented.
They are often much more low down that an national roll out of such and such technology. Transformation does not need to be big to be transformative.
There's a history here
There is a history with the NHS and digital transformation. It's not good and it's not pretty.
Guardian 10 billion wasted on systems that we did not get. Similar articles exist across newspapers.
source:
https://www.theguardian.com/society/2013/sep/18/nhs-records-system-10bn
NHS IT systems linked to death and harm. Ok in some cases it looks really bad; apparently neither rappers nor guns kill people IT transformation projects do.
source:
Some samples of studies talked about.
"
now have electronic patient records (‘EPR’) available, with a total of 189 trusts introducing the new paperless system, which is a government priority in England, investing £900m over the past 2 years to implement such systems. The latest deadline to implement EPR is set by the Department of Health and Social Care for 2026. EPR’s essentially allow everyone involved in a patient’s care to have access to their records and health information at the touch of a button; this includes hospitals, general practitioners and care homes.
Despite this, numerous IT failures have been linked to deaths and hundreds of instances of serious harm. A freedom of information request conducted by the BBC, revealed that around 200,000 medical letters had gone unsent due to widespread problems and glitches with NHS computer systems. In addition, further studies from the BBC in September 2023 revealed that more than 24,000 letters from Newcastle hospitals had not been sent from their EPR system and more than 400,000 letters had got lost in computer systems at hospitals in Nottingham."
That's a lawyers website so someone is getting sued over that. It might be me and you the taxpayer getting sued. It might be you and me who are the victims as well. That is a sorry state of affairs.
Relevant Research
Looking up the technology most in use I found the following where discussed. Convulutional neural networks and AI that I think is likely various forms of classification tools.
It seems that convulutional neural networks (CNNs) and robot vision used for classification have been a great success. These have seen 5-10% improvement of accuracy in diagnosing cancer from X rays or mammograms.
The same technology has been applied to identifying microorganisms using CNNs. Though the headline accuracy figures where not listed.
Database and normal websites is listed as allowing patients to receive antibiotics within 24 or 48 hours of testing when paired with AI that dispenses the prescription. The idea being the blood test is uploaded to the website an AI authorised in bulk the prescription a doctor checks it and it is then sent digitally to pharmacy and patient who can then go receive treatment without the need for a return visit to the doctor and further diagnosis.
sources:/bmcmededuc
"AI can be used to diagnose diseases, develop personalized treatment plans, and assist clinicians with decision-making. Rather than simply automating tasks, AI is about developing technologies that can enhance patient care across healthcare settings. However, challenges related to data privacy, bias, and the need for human expertise must be addressed for the responsible and effective implementation of AI in healthcare."
"AI is still in its early stages of being fully utilized for medical diagnosis. However, more data are emerging for the application of AI in diagnosing different diseases, such as cancer. A study was published in the UK where authors input a large dataset of mammograms into an AI system for breast cancer diagnosis. This study showed that utilizing an AI system to interpret mammograms had an absolute reduction in false positives and false neg- atives by 5.7% and 9.4%, respectively [11]. Another study was conducted in South Korea, where authors compared AI diagnoses of breast cancer versus radiologists. The AI- utilized diagnosis was more sensitive to diagnose breast cancer with mass compared to radiologists, 90% vs. 78%, respectively. Also, AI was better at detecting early breast cancer (91%) than radiologists 74% [12]."
The big picture is AI is used a lot for classification of diseases. It is obvious this works for X-rays and radiolagists where computer vision is very applicable and the data is data; but it is also applied everywhere below we will discuss claims of being able to data mine diagnostics out of notes. Practitioners become just another data ingestion methodology alongside everything else. I did note when talking about AI accuraacy it was 80 or 90% in many cases even when using case notes or data points you might think less traditional.
A frustration is I do not know how often human accuracy is. Also except when they die and the autopsy report finds out what they really died from it would seem any misdiagnosis is just missing data. If that's the case 80% or 90% accuracy cannot possibly be right and it's saying it's 10% less accurate than humans.
Until that is clarified I am glad I am treated by humans. But I am also glad an AI double checks there work. You will see later on this constraint is probably why AI transformation of the NHS is possibly limited.
Over The Horizion
Ok so let's see what technology could bring to medicine.
I found a lot of the stuff being said to being trialled had actually long history of use cases. It's confusing a lot of reports today will say genomics and data has a lot of promise. I can find reports from 2015 saying the same thing. I found one instance mentioning outright genetic modification in clinical use and others it was rarely used. If your a futurist like me its simultaneously a delight and a let down round every corner. I have really struggled to get a sense of what is happening now and what will happen soon.
Some trends I found in my research.
Smart hospitals. I really did want to know more but I couldn't download anything.
Cloud based data reporting solutions. Putting all the reports and AI tests in one place.
Genomic data use: combining statics, AI, genetic data in one place allows possible prediction of a variety of disease types and susceptibility to these diseases. Nothing is absolute and it feels like it comes out of a movie but genetic data is already being used. I put it in the over the horizion section because I've yet to see or hear it being offered on the NHS but a quick google search shows you can get it done it will cost you £840 at time of writing.
If this data was more available then we could use precision medicine where genetic data, lifestyle data could be analysed statistically using standard statistical ML techniques to tailor interventions down to individuals. I think a missing step is knowing over the lifetime of a patient what the value of having that data would be as it might be worth paying £840 for the service or it might not be.
The report report here said with genomic data it predicted the right cancer treatment 80% of the time with a sample size of 175. source: onlinelibrary.wiley.com
AI trained to pattern match is another trend. I felt critical reading it in one sense as I felt that potentially the technology could be as simple as a softmax classification model, random forest or the like. I feel like I could also build this but I probably lack the medical knowledge to build this and no one is crazy enough to trust me to build it.
source:
https://www.hdruk.ac.uk/news/new-ai-tool-may-offer-insights-into-patients-future-health/
There is something odd when articles are written about medical innovation and I'm thinking does it use Scikrit learn? I know that is flippant of me but at its base it's mostly just multiple class classification problem that random forests or softmax neural networks have been doing for a long time. There is no need to build a transformer architecture like used in chat GPT.
The NHS is using stuff that probably Kaggle could and possibly have and will do again a public competition on. The NHS probably sponsor those as well.
If your expecting one day to talk to a AI doctor who will send you to a AI surgeon then I didn't find that outright said no but I'm not sure that's what they are looking at.
Your consent is implied
I found that statement on a website it made me chuckle.
It sounds authoritarian but imagine your unconscious or dying. Use of your data matters but it makes sense that saving your life is a legal basis to make decisions.
"A health and care organisation may use your personal information in AI systems to provide you with individual care. AI can help a health and care professional reach a decision about your care, for example, diagnosing a condition you have or helping you choose a treatment option. In these cases, your consent to the use of your data is implied. Decisions will not be made by the AI system. Health and care professionals will always provide advice and allow you to make the final decision on the care and treatment you receive."
GDPR matters, your consent matters but how does that work balanced against saving your life?
I felt that was a important thing to share on how the NHS sees your data because I feel that as someone trained to think about GDPR and personal data it was a important thing about the NHS who will try and tell you everything up front but failing that they will first and foremost try to save your life.
I thought that was touching and puts things in perspective. Though if you thought this meant that there was some laxity here the definition down to the level of health records is made. This is to say that health data is somewhat of a special case.
Definition of health records found here.
purpose-of-the-gp-electronic-health-record/
The requirements for upfront data protection assesments. This probably puts a barrier to entry as a lot of checks and up front work is needed to check.
https://digital.nhs.uk/services/nhs-login/data-protection-impact-assessment
Central data silos
This is meant as a bit of a get to know the NHS and their data use. The data is managed more centrally by the CQRS.
https://www.england.nhs.uk/long-read/calculating-quality-reporting-service-cqrs/
The codes for describing what a given disease is also tightly controlled and defined here. It is deeply hierarchical data and the coding ie that an specific microbial infection is a child of general phnemonia more broadly of diseases of the lungs etc etc. it is something I have not thought about but specifity of the coding and the need for large vector databases to describe all the parents probably does not make SQL especially useful in this setting.
https://www.england.nhs.uk/long-read/clinical-coding-snomed-ct/
I think if you wanted to sell data products into this environment I would work to make sure was compliant with this use as looking up these codes in SQL is probably inefficient but also getting a vector database structured to do it and staying up to date would be a pain.
in fact talking about central data silos and above where I asked could I build my own diagnostic AI. The data is available tbrough the fingertips website here.
source:
10 year plans
Therw has been starting 2024 a development of a 10 year plan for the NHS. I read into it but found little detail.
"
To achieve this vision, we have many real challenges to overcome:
- legacy technology and commercial arrangements
- complex organisational and delivery structures
- a risk-averse culture
- limited resources to invest
- a critical need to build and maintain public trust[footnote 5],[footnote 6],[footnote 7]
At the heart of this vision are 4 guiding principles we should maintain to make this work:
- user need
- privacy and security
- interoperability and openness
- inclusion
And we need to draw on emerging thinking on designing technology safely, ethically and effectively for the values and interests of civil society.[footnote 8]
Ask what the user need is
Every service must be designed around user needs, whether the needs of the public, clinicians or other staff.
Services designed around users and their needs:
- are more likely to be used
- help more people get the right outcome for them – and so achieve their intent
- cost less to operate by reducing time and money spent on resolving problems
"
Though I suspect some foreword was written by someone else. It tben starts
- Put our tools in modern browsers
-
Internet first
-
Public cloud first
-
Build a data layer with registers and APIs
-
Adopt the best cyber security standards
- Separate the layers of our patient record stack: hosting, data and digital services
It then sets out the NHS priorities. The problem I have being that it lists a need to have best in infrastructure and digital services. Something I find confusing because surely if I'm buying the services I do not need the infrastructure and it is either a build first approach or a outsource first approach. I grant there is mixed options of partnering with others and hybrid ecosystems.
A annoyance I have here is when a company sits on the fence. We are a AI first company but cannot build our own. We are a infrastructure company who is cloud first and outsources everything. I don't mind what your strategy is but please do have one.
So the governments 10 year plan was we'd stop being patients and become users. Build services that worked to give us our intent of not dying. With both infrastructure and digital services. It sounds great if you've never actually done or been any of those things.
I do agree with the bulk of that but notice according to the government that at some point I and others became a user not a patient. It mentions achieving patients intent as if chouce in not dying was somehow a choice further I sort of expect the NHS to be risk adverse I suspect when taking risks kills people or affects peoples health you would be by inclination risk adverse.
I also found the government paper links back to another one. The paper behind it seems to be the earlier paper by the kings fund. Which I must admit I had not heard of before. It also seems more cogent I think it's worth a read to see how this is a much more thought out vision that sees both opportunities and risks.
Source:
https://assets.kingsfund.org.uk/f/256914/x/0158561602/nhs_70_what_will_new_technology_mean_2018.pdf
Key findings
• Technological advances offer significant opportunities to improve health care but
are not a silver bullet for the pressures facing the NHS. While there are really exciting developments in areas like genomics and precision medicine, we are a long way from being able to realise their full potential.
• Technology has the potential to deliver significant savings for the NHS but the service does not have a strong track record in implementing it at scale and needs to get better at assessing the benefits, feasibility and challenges of implementing new technology.
• Patients are embracing new technology and increasingly expect their care to be supported by it. For example, the majority of people say they would use video consultations to consult their GP about minor ailments and ongoing conditions.
• New technology could fundamentally change the way that NHS staff work – in some cases requiring entirely new roles to be created. The impact of these changes should not be underestimated.
• People generally have relatively little knowledge about how the NHS and commercial organisations use data for health research, which may be responsible for mistrust in some cases. Transparent public dialogue is needed about how data is currently used; what the opportunities are for the future; and how risks can be mitigated. While it is vital to balance the benefits of sharing data with concerns about security and confidentiality, these concerns should not be used as a barrier to progress.
This seems to be the earliest trend setter for the idea of much of trends in data within the NHS. It is both referenced under current and the previous government for their approach to data in the NHS and their 10 year plans.
The paper talks about genomic medicine much earlier than I had expected.
"These advances enabled the UK to become the first health system to introduce genomic medicine into mainstream health care in 2015. The 100,000 Genomes Project will sequence 100,000 genomes from around 70,000 people. Participants are NHS patients with rare diseases, plus their families, as well as patients with cancer. The project is already leading to patients receiving treatments that are likely to be most effective based on the genetic, lifestyle and environmental information of the individual in question – so-called ‘precision medicine’. That said, progress in general has been slower than expected due to the complexity of the science involved.3"
I feel that it put a downer on my hope for revolutionary genetic modification.
"Precision medicine (meaning use of genomics to make precise medical interventions) can be expensive. Treatments are usually developed for relatively small groups of people. The price-per-patient, then, has to be high to recoup the original development costs, which can make them appear poor value. Drug companies have tried developing tests to understand patient amenability for treatments alongside the treatments themselves – but the high failure rate in drug discovery makes this difficult.6"
Earlier it said similar about direct genetic modification. My desire to be a trans human space marine does not appear to be likely to be covered on the NHS anytime soon.
I am somewhat critical of this reports use. Where I have found referenced it is quoted as being in support of technological use in the NHS. I find this confusing as it feels to me a much more muted and conservative report.
Its investigation into remote consultations is far from a ringing endorsement.
"Where remote consultations reduce unnecessary referrals or outpatient appointments, they have the potential to save money. But, so far, evidence on cost is unclear. Where remote interactions subsequently require a face-to-face appointment, costs are likely to increase.19 There is also evidence that remote consultations can increase demand. After implementing remote care, leaders at Kaiser Permanente found that virtual visits via telephone and email increased from under 5 million in 2008 to over 10 million in 2013, while face-to-face visits remained largely the same.20"
Though it is this report that I found on the government 10 year plans seems much more balanced offering only that data if implemented well offers transformative benefits. I also note it's a 10 year plan not a 5 minute task of just implement AI and chat GPT which I feared at the outset I'd find someone somewhere seriously stating it's need to do the consultations.
"1. To reduce variation in clinical practice and become a learning system
Digital technology provides the opportunity for health care providers to embed best practice protocols. For example, if a patient attends hospital with a suspected stroke,
the electronic health record can prompt clinicians to undertake each task in the protocol. This approach can be taken a step further by automating protocols. At digitally advanced health care organisatons in the United States, the admitting clinician can trigger the
entire stroke workflow at the click of a button, organising laboratory tests, the CT scan, administration of intravenous medication, and so on.31 Aside from streamlining the system and supporting clinicians to make the right decision at the right time, in this approach data can be collected on how clinicians’ actions vary from protocol – and what impact that has on outcomes. In theory, if variation leads to better outcomes, the organisation can think about updating its protocols – although appropriate evidence standards would need to be met."
Where I understand the NHS is heading is to start with NHS practitioners to be promoted and monitored and while the above paragraph does not mention AI it is in the section for AI. It is also worth noting this was written prior to chat GPT.
Feasibility of changing the NHS was also interesting with numbers split in the report. It is not obvious that popular support exists for radical data led transformation of the NHS.
Where you see this impact
You can already see where this is impacting triage in that it is considered part of the core contract to have two way messaging systems and online triaging processes.
source: https://www.england.nhs.uk/long-read/digitally-enabled-triage/
When reading this I could not find examples. I infer this is the 111 service but I am unsure.
Conclusion
Intending to do a bit of reading on digital technology in health is wierd. I told a friend when I was writing this I don't think they can simulate a brain in the NHS but I might be surprised. I am right on both accounts.
There is a lot of stuff they do and have been doing that is very interesting and does not get talked about.
There is little mention of controversial use cases of AI like chat GPT diagnosing people or AI surgeons but to read between the lines reports like the kings fund report they where talking in 2018 of AI use in workflows.
The data policies probably mean that it's hard to create a product for them. Data protection must be certified up front and they have a lot of paperwork online of which I probably didnt read a fraction of a percentage.
I infer because of the disease coding methods they probably have to by necessity have bespoke software for everything it probably requires vector databases and is likely quite large.
There seems ah a confusing mix of directives from government (are we patients or users) and when they fail it's inferred to be complicated and messy.
The sort of technology you don't necessarily believe exists has been there quietly being tested and improved. I might joking say I want to be a trans human super soldier on the NHS but in 2018 someone was already thinking of its widespread use in the NHS which was ironically the same year the department for genomics celebrated 70 years of operations.
source: https://www.genomicseducation.hee.nhs.uk/blog/70-years-of-genetics-and-genomics-in-healthcare/
I also think I felt it was right it was not flashy. I felt that I wouldn't want people to take risks with my health or run data experiments on my hospital. I sort of do find a frustration with reports talking about say genomics only to find nothing really to show with any sudden massive changes being replaced with small iterations and improvement both understandable and explainable in the aftermath of past failed IT projects.
I think there is a lesson here in looking for AI in healthcare and data it is easy to expect and believe in revolutionary and sudden change. Doing the research what the NHS puts on its website as big succcesses are small iterative improvements constantly being undertaken at the lowest levels. Likewise their big centralised projects are considered abysmal failure. There more likely to talk about a dashboard or database use than a new AI revolution. I think there's something to that and it was worth digging through the articles.
Further I found it hard to articulate any great direction I found. There website on use of data in hospitals mentions envelopes specifically used for patient diagnosis. You definitely have a lot of paper when you still need websites to tell people what to do with it.
Source:
https://pcse.england.nhs.uk/help/medical-records/lloyd-george-envelopes-paper-records
Though even there of you look up businesses involved in digitising NHS records there are plenty. But I think it says something about the complexity of transformation in NHS as most moves are from technology to another, from database type A to database B. Here there are multiple mediums still in use, cannot be retired and are still in use as we sequence genomes and use AI in diagnosis.
It also is the point I go back to the original criticism about AI accuracy. Because we only really know we got the right or wrong diagnosis when someone dies and is autopsied. Given a lot of this data use focuses on clinicians maybe the NHS is right when they publish on their website they are using data and AI to reduce call times and not do a clinicians job. It puts true false positives and negatives into perspective there.
I like the idea it does not need to be gen AI I like that in most cases we are celebrating traditional data projects and tabula AI.
Though honestly I think even spending a couple of hours reading various articles I really have not read anything but a sampling of the complexity.
References
NHS website
https://transform.england.nhs.uk/information-governance/guidance/artificial-intelligence/
Report into NHS use of data
https://www.england.nhs.uk/digital-gp-good-practice-guidance/all-guidelines/
Add comment
Comments