Using AI and Machine Learning to aid Prescription Processing

Posted by: helenoleary - Posted on:

Head shot image of Chris Suter

Hi, I’m Chris Suter and I’m the Head of Digital Platforms and Innovation for all digital, insight and technology solutions here at the NHS Business Services Authority (NHSBSA).

I joined in February 2017 and since then; I have been engulfed in the world where artificial intelligence (AI) and Machine Learning (ML) is moving from a science fiction concept to the forefront of patient and clinicians day to day working life within the healthcare space.


As an Arm’s Length Body of the Department of Health and Social Care, The NHSBSA provides a range of services to NHS organisations, NHS contractors, patients and the public. These include Dental Services, Prescription Services and NHS Jobs. The NHSBSA supports the processing of prescriptions created predominantly in primary care and we are responsible for processing on average 45 million prescriptions per month, of these 18 million are paper and the rest are received via the Electronic Prescription Service (EPS). The average cost per prescription for the NHS is £20, which results in around £9.4 billion pounds paid every year, which represents approximately 10% of the overall annual NHS costs.

I wanted to look further into identifying areas where we see the most inefficiencies, which in turn have an effect on services, lead to frustration and have a cost impact, and see how we could improve this.

For the last 16 weeks we have carried out a piece of work to see if we can introduce Machine Learning into the prescription processing function to remove years’ worth of hot fixing and workarounds which include manual checks and validations.

Historically…

Prescription processing is carried out every month by the NHSBSA, and includes on average 11,000+ pharmacies boxing up tons of paper prescriptions and sending them to the NHSBSA to scan and process. As you will appreciate not all prescriptions look the same, some are printed, some are electronic and not to mention the handwritten ones which can be often hard to make out!

When introducing any system you can’t predict the future and this is usually the main driver for continued change. Unfortunately, this can result in future compromises and complexities, and the NHSBSA is no exception to this challenge. The current process and systems restricts our ability to make fast paced changes and offer solutions for better access to more detailed data.

What we set out to achieve using Machine Learning...

With any innovation concept, the goal is to take the current system, ways of working, and see if new technologies offer improvements.  These can be operating efficiencies, cost savings or removing complexity.  
To achieve this we have been working closely with our cloud and technology partners Microsoft and Amazon Web Services, working together to understand the issues and concerns within the current process and what technologies where available to help. We started with three main goals:

Reading accuracy 
As close to 100% prescriptions data automatically extracted from printed paper prescriptions. This would include patient and drug data. Determine what was possible when faced with handwritten prescriptions.

Processing Improvements
Determine what efficiencies were possible in the end to end process.

Future Use Cases
Opportunities for our partners to add value from the data we collect (i.e. anomaly detection, medical insight, drug usage and statistical analysis).

Security is Job Zero…

As it has been drilled into me for a number of years, we are all responsible for the security of the systems we create.  The introduction of AI and Machine Learning should also follow the same rules and boundaries with no exceptions.

This piece of work needed to consider the security and controls of sensitive Patient Identifiable Information (PII). Therefore we carried out a number of assurances against the best practice guides technologies including the:

• “Code of Conduct for the use of AI” within the NHS from the Department of Health and Social Care
• “Calidicott Principles” from the Department of Health and Social Care
• “Cloud Security Principles” from the National Cyber Security Centre

Working as One…

From the get go, we wanted to embed the NHSBSA teams from Development, Platform and Data with our partners’ technology teams from Microsoft and Amazon Web Services. This was to enable cross training, knowledge sharing and understanding of the Machine Learning methodology. This is critical to ensure we were self-sufficient to house future developments, upgrades and technology adoption.

Over the 16 weeks of working with our partners, this been one of the most surprising outcomes. Interaction between the teams from the NHSBSA and our partners has been a huge success, with energetic two way conversations taking place and knowledge flowing both ways, which has helped generate further insights into the process and achieved a better outcome.

The Outcome…

From a standing start, with just an idea of using new technologies, we have been able to build a number of micro-services. These are able to take a prescription image and accurately extract data, carry out validation and create insights. The microservices include:

Image re-processing
Dealing with toner marks, human tick marks, misalignment, handwritten amendments and print alignments.

Machine Learning models to extract data
Using image extraction models, text extraction models, medical and drug models.

Image processing and storing at scale 
Using cloud technologies to provide compute and storage using serverless technologies and software as a service package.

Validation of the data captured
Utilising the current data sources the NHSBSA has, including repeat prescription data, NHS drug catalogue, Prescriber register and Pharmacy register.

Data Analytics
Creation of a data analytic module to enable analysis. This could be for current medication trends, future drug forecasting for payment and supply, and creating health dashboards.

Reduction of Processing Cost
Reducing the number of manual resources required to extract data, this enables the NHSBSA to invest and transfer these people to more value added activities.

Unlocking Potential…

One of the key outcomes of this activity is the future potential of the solution we have created. Not only will it solve current problems, but it also has been designed to be easily transferable to other paper-based scenarios not only in the NHSBSA but potentially within the entire NHS family.  

Internally, the NHSBSA now has a scanning facility, a secure cloud platform and the skills to programme and train machine learning models. This has the potential to benefit other areas of the business that heavily rely on paper forms such as Pensions, Student Bursary and Citizen Services. 

What Next…

So as the 16 weeks proof of concept is now up it doesn’t stop there, the NHSBSA along with Microsoft and Amazon Web Services, are planning to implement this service at scale over the next six months. In addition to this, we will be further validating the findings and seek to expand and take full of advantage of this solution into other form types.


7+

Leave a Reply

Your email address will not be published. Required fields are marked *