Meet the team
General Paediatrics Ambulatory (out of hospital) care
Bradford Teaching Hospitals NHS Foundation Trust
- England - Yorkshire and Humber
- Dr Danyal Akarca PhD Candidate Neuroscience & AI (MRC Cognition and Brain Sciences Unit Cambridge University)
- Ruaridh Arwyn Mon-Williams Mechanical Engineering Final Year Student (Edinburgh University)
- Sam Relins Data Science Student (Leeds Institute for Data Analytics, University of Leeds)
- Prof Hissam Tawfik Professor of Computer Science (Artificial Intelligence) (Leeds Beckett University)
What is the positive change that has emerged through new collaborations or partnerships during Covid-19 that your project is going to embed?
In Bradford we run a children’s hospital at home service. It has reduced hospital referrals from primary care and reduced admissions for children seen in hospital too. The service was developed with families, primary care and our local CCG. It won a HSJ award in 2018 and was CQC rated as ‘outstanding’ this year. It has also been endorsed by the RCPCH as a service model post CoVID. https://www.qicentral.org.uk/news/covid19-spotlight-paediatric-ambulatory-care
We have developed a bespoke training programme for our nurses who deliver the service. The model has improved workforce integration and joy at work.
The model has been codified to aid adaptation and replication and to share freely with other systems in the NHS. It is described as an ‘aspirational’ service in our Integrated Care System.
This you tube video shows the impact on one family https://www.youtube.com/watch?v=WNHFFKG12Nc&feature=youtu.be . There have been no adverse events and the service has received excellent family feedback.
What does your project aim to achieve?
Currently around 17% of children who are accepted into our hospital at home service (according to our preset clinical parameters) still end up in hospital.
At present our decisions for entry into the service are based on traditional factors so
1. decisions for referral are made on clinical assessment criteria alone
2. decisions are based on guidelines based on evidence and local decisions but not necessarily on the individual characteristics of the child e.g. demographic background
3. the rigid measures do not account for our local population and cannot adapt readily
4. less apparent contributors to the likelihood of successful referral including the important wider determinants of heath do not contribute to the decision for a referral
We want to answer this question for every child: ‘Given this child’s presentation of illness, their non-clinical risk factors and background information – what care is appropriate? 1) the hospital at home service or 2) assessment and admission to hospital?’
How will the project be delivered?
The project aims to develop a proof-of-concept (POC) machine learning tool which learns from prior and ongoing patient contact to allow clinical teams to make the best decision for their patient (i.e to help them consider the best place of care). We want to focus on our asthma/ viral wheeze pathway initially in this POC.
Phase 1 : This will bring together specialists with clinical expertise ( including paediatricians, primary care and public health doctors) and technical expertise (data analysts and data scientists) to decide on the clinical/non clinical criteria to be used.
Phase 2 : once clinical/non clinical criteria have been selected we will extract the data from patients looked after by the team over the last 30 months (which include over 500 patient datasets) to study various machine learning models to see which provide(s) the best predictive, clinical and technical fit.
Phase 3: A clinical user and system requirement specification will be produced and tested
How is your project going to share learning?
The problem of inappropriate use of urgent secondary care services in Bradford is not unique. We therefore developed our Ambulatory Care Experience (ACE) hospital at home service to be easily adapted and used by other acute systems. ACE has a simple design and we ‘packaged’ it for spread. The service, model, training and pathways can be used ‘off the shelf’.
Although the problems we face are similar the patients we serve and the places they live are quite different. Incorporating clinical and non clinical factors in decision making using a machine learning tool may better allow teams to provide the right care, at the right time, in the right place, for their local population. It will allow clinicians to learn from the patients they see.
It is hoped this POC will ultimately result in a much more effective and quality service tailored to children in different locations across the country.
How you can contribute
- Has anyone in Q had experience of using/ developing machine learning in urgent care triage?
- What are the potential ethical and technical issues that may arise?
- Is there anyone in Q who could provide external challenge to development that we could call on if successful?
- Is there any one who would like to develop a similar service in their area?
- Does the Q community have any thoughts about the acceptability of using machine learning techniques to triage patients?
- Could this technique be used to triage children in hospital -in A&E and on the wards?
- Could this technique provide a way of predicting problems before they arise?
- If successful could this be used for adult triage?
|4 Apr 2021||Clinical user and system requirement specification (clinical and technical teams)|
|3 May 2021||A data-integration protocol for POC AI implementation|
|5 Jul 2021||Model validation report for clinical settings|
|4 Oct 2021||POC AI models produced and compared|