Last month, six industry partners, and 70 students, postdocs and senior academics took part in a Data Study Group at The Alan Turing Institute.
The aim of the session was to develop data science solutions to challenges for health and wellbeing.
So what did we learn over the five days and what were some of the outcomes?
Alleviating pressures on A&E
Dr Nik Haliasos, neurosurgeon and Dr Darryl Wood, A&E consultant from Queen’s hospital worked with researchers to develop a tool which could enable A&E nurses to safely prioritise and highlight patients with time critical conditions such as stroke, sepsis and heart attack.
Dr Haliasos said: “Queens A&E is currently seeing 240,000 attendees every year, leading to more than 50,000 admissions. This trend has seen a 20% increase in attendances over the last 5 years. During winter, these pressures become even more acute. This rhythm is very difficult to sustain in the long term, and with the available resource and infrastructure there is no way we can continue delivering the excellent and safe service we strive to.
”What we need is an early warning prediction algorithm that can flag up the severity of presenting cases and which can help with diagnosis and chances of admission.”
He added: “Nurses will still do the usual observations during the triage examination, and the tool will provide advice based on the information she/he has fed into the computer. The advice generated by the predictive algorithm would then be evaluated by the healthcare staff to aid them in reaching the optimal decision for each specific patient.”
The group of researchers working on this challenge during the Data Study Group developed an algorithm that will generate diagnostic options based on clinical variables and the free text nurses have inputted into the deep learning network. After an appropriate pilot trial within clinical grounds the algorithm would be embedded into the hospital’s IT systems. Dr Wood who will introduce the tool to his A&E colleagues, is optimistic that preliminary testing in a controlled environment could be implemented within 3 months.
Better prediction of emergency hospital admissions
Samuel Oduro, Senior Information Analyst at NHS Scotland Information Services challenged researchers in the Data Study Group to develop a methodology that would more accurately predict the risk of a patient being admitted to hospital as an emergency in the next 12 months.
After five days the multidisciplinary group of eight came up with a model which improved risk calculation scores based on historic patient data.
Explaining the significance of the model developed by the researchers, Sam said: “The new model that the group has developed is a preliminary solution but already performs better than existing models. The new tool provides the enhanced accuracy we need and will help to boost our relationship with GPs.”
To shed light on how the existing model falls short, he added: “The existing SPARRA model, which is purely statistical, mainly predicts the most obvious cases known to clinicians but cases that are more complex and unknown to GPs can sometimes be scored lower. As a result, some GPs have less confidence in SPARRA and in some cases have been reluctant to use it.
“It’s important that GPs engage with the system. With this information and, in conjunction with their own data and professional judgement, GPs can review whether a patient is at risk of being admitted and put in place a preventative treatment plan that can help to avert admission.”
Enabling the development of cures for TB, malaria and cancer
Martin Jones, Deputy Head of Microscopy and Prototyping at the Francis Crick Institute arrived at the Data Study Group with the goal of developing a set of algorithms that could automatically recognise and segment features of cells in electron microscopy images. A breakthrough in this area could lead to scientific advances that could enable the development of cures for TB, malaria and cancer.
Setting the context for the challenge and how he hoped the Data Study Group could help, he said: “The problem is that the automated microscope currently in use can produce 1000 images overnight. This means we actually have the data to get the results, but the challenge is extracting meaning – since the data is being generated at a rate that is outstripping our capacity to analyse it. The process is very slow and we hope data scientists can help us to extract it more quickly, which will lead to faster advances.”
Martin is optimistic that machine learning and deep learning techniques applied during the Data Study Group will eventually provide flexible solutions that can be extended and used by other research teams at the Crick in the long-term.
What were the highlights of the Data Study Group experience?
Dr Haliasos said: “Working in a very collaborative setting with people who have the expertise. They have demonstrated great will and it has been very engaging for them. This I think, has been one of the most important features, together with the available computational resources. This enabled us to make significant steps forward in a very short period of time.”
Sam Oduro said: “Data science is definitely the solution – my expectations have been surpassed.”
He added: “Working with a group of researchers thinking so deeply about the challenge was definitely the highlight of the collaboration. The way they critiqued the existing problem and came up with a solution was phenomenal.”
Martin Jones said: “I’m very pleased with the outcomes achieved this week. The researchers came up with some cool ideas and I have made some important connections. I found the best people to tackle the problem. There is nowhere else that you will find such concentration of expertise all in one place and this has made the experience very worthwhile.”
The six partners that took part in the Data Study Group on health and well-being were:
Queen’s Hospital A&E
NHS Scotland Information Services Division
The Francis Crick Institute, with support from Wellcome Trust
Care Quality Commission
Cochrane with EPPI-Centre, UCL with support from Microsoft
The Centre for Cancer Prevention, QMUL with support of Cancer Research UK
Find out more about the challenges and work of Turing Data Study Groups.
The next Data Study Group will take place on 11-15 December 2017.