An industry expert says analysing previous trends in how, when and where emergency care is needed could change the way the healthcare industry uses resources to deliver better outcomes for patients
With the constant strain on emergency medical services across the world making it increasingly difficult to deliver effective patient care at the point of need, Matthew Cooke, the chief clinical officer at French management consulting corporation Capgemini, writes on his belief that predictive analytics could vastly improve how the healthcare industry deploys its resources.
It surprises many people that the workload of emergency care is predictable.
We can’t say when an individual will suffer a heart attack, but we can make accurate predictions about how many emergencies will occur and when.
Ambulance services have used this information for many years to position resources in the areas of highest workload.
Some are very obvious — the accident hotspots on Friday and Saturday nights in areas with most pubs and clubs, whereas medical illness is more common during the daytime.
Likewise, minor injuries presenting to A&E decrease during heavy rain, asthma attacks increase after thunderstorms.
There are large numbers of studies showing how presentations at A&E vary by time of day, day of the week and even seasonal.
Every A&E in England knows there are two peaks of arrivals, one at lunchtime and another early evening; they also know that Monday is the busiest day of the week.
It is therefore surprising that the staffing levels often don’t reflect these variations.
Most emergency departments will also tell you that a surge of patients arriving is one of the most common causes of delays for patients to be seen and treated.
How can predictive analytics ease the burden on healthcare services?
Predictive analytics is widely used in many other sectors to accurately understand the workload and therefore plan the resource allocation.
Supermarkets understand when customers will arrive and what they are most likely to buy during a particular visit — allowing them to manage staffing and stock levels accordingly.
The workload of an emergency department can vary in several ways, but most important are the numbers arriving and the case mix.
The timing of the arrival of less severe injuries, such as sprains and cuts, is highly predictable.
However, the timing of the arrival of a seriously injured person is less predictable and may require 40 times the resource.
I recently spoke with a group of NHS analysts who confirmed they do use predictive analytics.
However, they meant using a six-week rolling average to understand daily numbers of A&E attendances.
The danger of using averages to calculate staffing means you will be understaffed half of the time.
Using daily figures will not help you cope with a surge of activity over a single hour.
Modern mathematical techniques have the capability to understand and predict A&E numbers by the hour and the type of case.
At present it is not known how accurate the predictions could be, but I am certain it could be more precise than current systems.
Predicting accident surges and workload changes
We know that a diverse range of factors influence attendances at A&E, including the weather, pollution, illness outbreaks, local events, religious festivals, local health scares.
What would happen if we added all the known factors in to one mathematical model that understood the interaction of these hundred different factors?
Could we start to predict and understand when and how surges will occur?
If surges are predictable then preventative action can be taken, especially related to staffing levels.
This would result in fewer queues and therefore fewer delays in treatment and better outcomes.
It may also mean we could adjust the skill mix by hour of the day, to have the correct specialty support available for the emergency department at the time the person presents.
As the model is increasingly used then the accuracy of the machine learning would increase too.
Currently this type of analysis is mostly used for strategic planning, but what if the mathematical model was live and could predict the workload over the next few hours?
Would the system have the ability to respond appropriately?
Is our workforce flexible enough to respond to such short-term changes in demand?
Using limited healthcare resources more effectively
In a resource-restricted healthcare system there will always be a need to focus on those who will benefit most and avoid exposing those with no advantage to risk of harm from healthcare interventions.
In England, women between 50 and 70 are invited for breast cancer screening as this is the population age group at most risk.
Some individuals outside this age group may have a higher risk but not be screened.
Others in this age group may be very low risk and be exposed to the risks of radiation.
Predictive analysis could enable more appropriate focus of a screening programme.
It would allow an understanding of how all the known risk factors relate not just to the risk of breast cancer but to those that are more amenable to treatment.
Predictive analysis using “big data” sets has the potential to improve healthcare resource utilisation, resulting in fewer waits and improved outcomes.
The detailed analysis is the first step, the system must then be able to respond to the predictions and become more agile in its use of resources.