AI Adoption Case Study: using responsible AI to support ward-based pharmacy services

AI Adoption Case Study: using responsible AI to support ward-based pharmacy services

techUK's AI adoption collection of case studies showcases examples of how organisations are Putting AI into Action, either through the adoption of AI models within their organisations, or by developing AI tools that can be leveraged by others.

By shining a light on these use cases, techUK hopes to demonstrate examples of best practice from across sectors and from organisations of all sizes.

1. Challenge:

Ward-based NHS pharmacy teams need to balance their resources and manage their time efficiently to make sure they treat their patients effectively and safely. At Mid Yorkshire NHS Teaching Trust, there are around 1100 beds spread across 3 different hospital sites, highlighting the scale of the challenge of ensuring that high-risk patients are prioritised accordingly. The existing model for identifying high-priority patients is time intensive and requires a pharmacy team member to visit specific wards and manually identify patients with the highest risk.

Given that the pharmacy management process has been digitised, there was an opportunity to use the data to autonomously flag patients requiring immediate attention.

2. Solution

This led to the idea of developing a Prioritisation Tool which utilised Machine Learning to support pharmacists in achieving three key objectives:

• Identify high priority patients with the greatest clinical need
• Share the information in an easily accessible way, without undue burden
• Provide the reason why the patient is high priority for pharmacy services

In this case study, we uncover a proof of concept project undertaken by Hippo to design a Clinical Pharmacy Prioritisation Software Tool (CPPST) that leverages existing medicines management data and machine learning to support Mid Yorkshire NHS Teaching Trust frontline pharmacy services.

The Digital Medicines Lead, Jonathan Bevan, posed the question: if we utilised the digitised data from the current medicines management process, can we use it to predict which patients are considered to be high priority? The project sought to test this hypothesis.

The data required to feed into the development of the solution needed collecting and processing in a safe and secure manner, in line with agreed Information Governance principles so that patient privacy was respected. Following collection, key steps included data transfer (pseudonymised first by the NHS Trust), profiling the data to understand its availability and latency, and data analysis to understand the quality.

Using the historical data, various models were developed, trained and evaluated using both machine learning and generative AI. This was a far from linear process and relied on experimentation and iteration.

The first iteration was a statistical model that harnessed volumetrics to determine complexity of patient cases and high-priority cases. Building on this, further iterations drew upon research papers to incorporate the definition of 'complexity' providing further context to the medications data. Generative AI enabled further iterations to overlay this information with scoring and descriptions so that the pharmacist has an indication of why the patient is flagged as high priority.

Working in close collaboration with the pharmacy team, driven by the Digital Medicines Lead, and with patient safety at the heart of decision making, each model improved on the previous, building trust along the way. The aim was to design a tool that provided human readable outputs which could be easily understood and which integrated into working practices.

3. Barriers:

Alongside the physical challenge of developing a platform and harnessing data and Artificial Intelligence to achieve this, the project team also needed to navigate a level of nervousness from the pharmacy teams about trusting a tool that used machine learning and overcome barriers to make the technology more understandable and accessible. Building this trust would play a crucial role in ensuring that the project could succeed.

This journey began with a discovery phase aimed at understanding the needs of patients and how the pharmacy team currently worked. By conducting interviews, the pharmacy team shared the pain-points they encountered on a daily basis - including accessing multiple systems to identify which patients they need to prioritise seeing first in a busy clinical environment. There was considerable hope and openness to how a machine learning tool could support the team, but equally there was nervousness about putting trust in a tool they didn't fully understand.

A face-to-face workshop with representatives from the Mid Yorkshire Teaching NHS Trust, including clinical colleagues and colleagues from the Information Services Team was a great opportunity to build this understanding and trust. The team were invited to Hippo's Leeds offices to learn more about how machine learning works and some key considerations when using algorithms.

The Hippo team ran two interactive games, one using the family favourite 'Guess Who' (to help participants learn about algorithms and decision trees) and another that involved sorting coins to explain classification and labelling. Key to exercises were the learning points that a) the importance of accurate information when developing algorithms, b) questions need to be well defined and c) data attributes are important. The team discovered that by teaching AI/ML concepts in an interactive, fun way can improve the accessibility of the subject.

4. Impact:

The resulting Prioritisation Tool was able to identify complex and high-risk patient cases with high levels of accuracy (93-97%, <2.2% false positives), proving the hypothesis that the tool is both feasible and viable in a clinical setting as a decision support aid to complement the expert judgement of frontline pharmacists. The team also identified next steps to take the work forward, including further clinical validation and feedback loops, developing the quality and completeness of data, and enabling the use of real-time data.

"The proof of concept developed by Hippo has laid a significant foundation for creating an innovative and precise tool that supports our patient's safety and outcomes and staff wellbeing. Collaborating with Hippo was an amazing experience, marked by seamless integration as one team and a shared passion for the project. Their expertise and dedication have been invaluable, and I have personally learned so much from them".

Jonathan Bevan, Digital Medicines Lead at The Mid Yorkshire Hospitals NHS Trust

When looking at the people, processes and technology/data which is needed to be in place to succeed, this project sought to prove the theory that it is possible to predict high risk patients, help pharmacists do their job and ensure patients get the treatment they need.

A core element of this case study is to demonstrate that with a responsible approach, AI can be harnessed and used by all - but the process must start, evolve and finish with humans at the centre and patients at the heart.

Public link

Please use the above public link if you want to share this noodl on another website.