Back

20/02/2019

Using AI and ML to predict staffing and patient demand in A&E departments

Latest figures from January show that A&E waits across the country have reached their worst level since the 4-hour waiting time target was introduced. With only 84.4% of patients being treated or admitted within the 4-hour target, making it drastically lower than the acceptable 95% threshold. These failings are estimated to have directly impacted over 330,000 patients, with increased waiting times, poor visibility and difficulty finding beds all contributing to the end result of inadequate patient care.

According to HSJ “the deterioration came as demand rose to record levels, there were 564,000 emergency admissions in the month – 7.2 per cent higher than the same month last year and the highest number on record.”

The figures come as a surprise to many, as this winter saw relatively lower occurrence of the flu and adverse weather when compared to previous years, yet A&E wait figures were worse. Several hospitals have reported that there was an increase in older, sicker patients requiring admission, which was far more than they had forecasted and planned for, thus having a significant impact on bed availability, resources and overall patient flow and outcomes.

When viewing this problem, it seems that many hospitals are struggling with forecasted vs actual patient flow not corresponding. The setback of using tools that depend solely on forecasting constraints in the absence of a responsive solution results in corrections for patient access and flow being slowed down or completely put out of kilter. With this awareness, we have further developed our Emergency Department (ED) Insights and Analytics solution as means of providing hospitals with a dynamic solution which provides actual live ML driven predictive insights rather than snapshot forecasted wait times overview. Our solution uses a number of advanced data science and machine learning techniques to predict demand and patient flow for A&E departments. These predictions are then fed through to the live analytics module to measure the predicted value against live data. This gives hospitals real insights into whether they are running at the expected capacity relating to workforce and other resources. This is then used to drive the alerting mechanisms for clinicians and managers of ED/A&E departments or other departments.

The solution considers ambulances, urgent care, rostering data and more whilst providing deep insights into patients with major conditions, children, comorbidities and patients with minor injuries. All targeted at ensuring all conditions can be considered when planning resource allocation. The solution is programmed to enable teams to really align clinical and operations workforce to demand within the department.

Predicting A&E attendances and emergency admissions is now an integral part of delivering unscheduled care and patient flows in healthcare. The algorithm provides a risk assessment for re-attendances, deterioration and more. The module can easily connect with other systems in the hospitals such as allocate, ESR, rostering spreadsheets, job planning processes, ambulance attendances, 111 datasets and GP systems to provide a rounded view of demand and capacity.

The impact our clients have reported show an 85% stepped reduction in triage waiting times, with other clients performing at 98% on average against the 4-hour national A&E target, since using our A&E module to identify their service challenges. This has allowed them to make informed changes based upon real data. The integration of the module also saw our clients on average report a saving of £97,000 service efficiencies per annum.

In order to combat poor visibility and planning, hospitals should consider integrating similar solutions as a means of improving the overall A&E 4-hour patient access standard. D&D is here to support organisations wishing to move to a more dynamic solution which provides richer insights and better planning.

.

.

.

.