Optimising capacity and cutting costs with our AI model
Hywel Dda University Health Board
We partnered with Hywel Dda to support their frontline care staff with AI-powered decision-making.
reduction in avoidable delays to discharge.
million in annual efficiency savings, based on an average cost of £400 per bed day. Not a cash savings as no capacity is being closed.
Background
As patients arrive at A&E, Hywel Dda faced some difficulties in proactively identifying individuals who may benefit from extra support, and those requiring advanced planning to ensure they are discharged on time. This resulted in extended stays and greater resource usage, ultimately leading to a more costly, intensive care package and poorer patient outcomes. These challenges were due to data quality and the limited insights they could extract from it; compounded by outdated, clunky legacy systems.
Solution
We implemented our AI model to lift value from the data, to drive operational decision-making. We trained the model on data captured in a user-centric way, to ensure a clear understanding of both the reasons for data collection and the value it provides to individual clinicians. We also lifted data from some of their core systems to provide a much richer understanding of the individual patient for improved decision-making.
AWS services were integral to delivering a scalable, secure, and efficient solution tailored to their needs. Amazon Simple Storage Service (S3) provided reliable and cost-effective storage for large volumes of data, enabling seamless data retrieval and archiving, which was critical for the client’s data-intensive workflows. To implement advanced machine learning capabilities, Amazon SageMaker allowed the team to build, train, and deploy models quickly, significantly accelerating the development of predictive analytics for the client’s core applications. Amazon Elastic Compute Cloud (EC2) offered flexible compute power to support both the client’s routine and peak workloads, ensuring high performance during critical operations. For containerised workloads, Amazon Elastic Kubernetes Service (EKS) simplified the deployment and management of Kubernetes clusters, allowing the team to orchestrate scalable and reliable microservices effortlessly. Meanwhile, Amazon Relational Database Service (RDS) ensured the client’s database infrastructure was highly available, secure, and easy to manage, removing the burden of manual administrative tasks like backups and patching.
Impact
The solution has provided tangible results quickly. Hywel Dda can use the AI model to understand what happens when they pause, fast forward and rewind the system to better understand their choices and the impact of those choices in real time. In the 12 months following rollout, Hywel Dda recorded significant improvements in operational performance across its acute sites. These improvements should be considered alongside other confounding factors which may also impact timely discharges, including the RAAC major incident in August 2023. Resulting in 32% reduction in avoidable delays to discharge (amounting to approx 85 beds freed up and £12.5 million in annual efficiency savings*, to be reinvested in UEC flow and elective care).
*Based on an average cost of £400 per bed day. Not a cash savings as no capacity is being closed.
“Faculty have been incredibly passionate and dedicated. They’ve thrown themselves into understanding our challenges, and helped us get under the skin of those challenges. They have that depth of knowledge of what AI is and what data science can do to unlock value from our data.”