Subscribe to DSC Newsletter

Machine Learning in Hospitals: Easing Wait Times in the ER

Like many emergency rooms in the United Kingdom, the A&E department at Salford Royal NHS Foundation Trust, Greater Manchester, faces high congestion. This results in treatment delays and access issues. The Data Science team at the Northern Care Alliance (NCA) National Health Service (NHS) Group of hospitals is implementing support mechanisms to ease wait times, using machine learning and regression to better predict peak demand times and improve the flow of patients from intake to discharge. 


I recently spoke with Karim Webb, Data Science Manager and Robyn Hamilton, Data Scientist at the Northern Care Alliance NHS Group, regarding some exciting developments with the use of data science to ease wait times, support clinicians and provide better patient experiences at the hospital. Karim believes that his team is on the leading edge of this discipline within the UK healthcare economy. However, one of the biggest challenges the team faces is that, “… data science is still a relatively new discipline…so finding stakeholder engagement to drive it forward is a challenge.” In fact, while most NHS trusts have Business Intelligence teams, they have yet to engage fully with Data Science. Karim hopes that by highlighting the accomplishments his team has made in a relatively short time period, data science will become more prevalent in the healthcare arena.

The Issues

As there wasn’t any prior data science work in their area, the team had to start from scratch. They faced big, NHS-wide challenges, such as:

  • Limited number of hospital beds. Beds were at capacity. In addition to a queue of patients in the A&E department needing to be admitted to the hospital, people in those beds that needed to go home couldn’t, mainly because of under-funded support systems.
  • Lack of funding. When the NCA DS team started their work: funds were being cut.  The cuts to social care meant that patients who could potentially go home with social care support (e.g. bandage changes) weren’t able to, which filled beds longer than it potentially needed to.
  • A backlog of patients. The team focused on the A&E, because it is the entrance point for a large proportion of their patients. Backups start in the A&E; four-hour treatment targets from registration to being treated weren’t being met, and even if the targets were met, the patients still had nowhere to go because of the lack of beds. 
    Identifying areas that could improve the flow of patients was the team’s top priority.

Out with the old…

Previously, the hospital had used averages to predict when congestion would occur. Historically the NHS only used NHS data sets. The team is in the process of hooking into data from local government sources or councils to get a wider view, supplement existing data, and gain a better understanding of the population demographics. Understanding how demographics are changing is vital to the teams’ work; As one of the largest cities in the UK, Manchester and Salford is experiencing a change in demographics and it is important for hospitals to understand the population they serve, with large numbers of people flowing in and out. The theory is that using a variety of data sets bolsters the quality of the predictions, analysis and insight, ultimately leading to improved patient experiences. Karim notes that the depth of data they have access to is “really exciting” and the supplementation of clinical data with non-clinical data sets is a pioneering type of development 

The Big Picture


The team is incorporating machine learning algorithms and predictive modelling into reporting frameworks to provide insights into:

  • Aiding clinical decision making, to provide hospital-level insights for Salford and building first of type readmission and text mining workstreams. Although the team isn’t there to replace clinicians, they want to demonstrate how data science can be used as a tool to aid in the diagnosis and management of patients.
  • Driving organizational strategy, building on existing Accident and Emergency (A&E) workstreams to create deeper insights. 
  • Allocating resources more effectively; showing how and where existing resources could be better used.
  • Pioneering the use of non-health based datasets to support insights on local populations. 

The Results


The team has already achieved some very interesting results in a relatively short period. For example, the readmissions model is live in Urgent Care, and is being scaled to the Emergency Assessment Unit and other NCA hospital sites.

Robyn explained:

“If we breach 11.10% readmission after 30 days, we could be subjected to intervention by government bodies. For the first time we are not being charged because of the readmission rate.”

While there have been several initiatives that have helped drive down this number, the readmissions model has been an integral part of the process.

The team also worked on creating risk scores for every patient, so they can be “smarter” about a patient’s risk, including who should be sent home and what provisions discharged patients might need, in order the to optimize the chances of them not being readmitted. In other words, the goal was to personalize the discharge method and design a healthcare package a little better, with high risk patients getting the support they need. As far as implementing those changes,. “We provide the clinicians with insight and the risk score,” Karim said. “…the challenge is engaging with clinicians and senior managers to implement change based on the insights provided.”

The Future of Health Care?

 

In the immediate future, data science can be used to improve the hospital experience, from identifying bottlenecks and confirming or disproving diagnosis & treatment narratives to improving patient experiences. Karim stated, “We can help [clinicians] understand their patients in ways that weren’t previously possible.”

Although data science isn’t there to replace consultants (specialists) or state with certainty that an event will or won’t happen, Karim notes that it can support decision making in a way that hasn’t been possible before. 

Views: 1090

Comment

You need to be a member of Data Science Central to add comments!

Join Data Science Central

Comment by Norberto J. Sanchez on August 11, 2019 at 9:03am

This article is an interesting read, but it lacks substance. A description of the algorithms used and the new metrics would have been very useful. 

Videos

  • Add Videos
  • View All

© 2019   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service