In our role partnering sustainability and transformation partnerships and emerging integrated care systems, we are always looking for digital solutions to challenges involved in improving health and integrating care. Staging an innovation conference in 2018, we announced our partnership with artificial intelligence (AI) and predictive analytics experts PredictX aimed at using AI to answer challenges in care and health.
A good example of a real-world health and care challenge lies in Wolverhampton in the West Midlands. With a population of 262,500, Wolverhampton’s age distribution is similar to the rest of England but the healthy life expectancy (i.e. how long can expect to live unhampered by illness or injury) for males at birth is 58.2 years in comparison to 63.4 years in the rest of England. Similarly, the female healthy life expectancy is 58.7 years (England average 63.8 years).
With support of NHS Digital Demonstrator funding, City of Wolverhampton Council partnered with us to develop predictive analytics models to understand the care pathway of patients and whether they are accessing the right care package at the right time. It was hoped this would assist in targeting interventions to help people remain independent, in their own homes, for longer.
Action
Using health and social care data, PredictX and MLCSU created a model predicting hospital admissions from A&E. The data used included:
* Patient demographics
* Hospital data – including hospital location, department, arrival time and arrival mode
* Details regarding current social care packages
* Care pathways – including previous touchpoints
* Deprivation data.
After the data was explored and key features identified, machine learning models were trained to predict how many patients would be admitted.
What we found
Of the sample of 66,321 observed patients entering A&E, 3,615 of them were admitted. Our model accurately predicted 81% of these 3,615 patients would be admitted.
Overall, the hour of arrival and the length of time between arrival and departure most impacted on whether a patient would be hospitalised. Patients who arrived later and stayed at the A&E longer were more frequently admitted.
Long-term conditions also had a strong influence, particularly conditions like cancer and coronary heart disease (67.2% of all admitted patients (94.2% of those aged 65 and over) had at least one long-term condition).
Impact
Correctly predicting these factors gave us a basis for understanding the driving factors behind hospital admission. This provided the Wolverhampton team with a solid evidence base from which to effectively plan programmes and interventions which could reduce hospital admissions and the cost of care packages, while helping people remain independent for longer.
This can potentially lead to an opportunity for patients to receive the best care at the right moment – improving life expectancy in the borough as a whole.