We are open honest and trustworthy and look at every issue from all perspectives
Reducing hospital readmissions in the NHS
Reducing hospital readmissions in the NHS – or why the best prediction algorithm is as good as the human that can use it!
A hospital readmission according to the NHS is one that occurs in less than 30 days after a first-time hospitalization and for the same reason. For example, a patient is rushed to A&E with a lung infection and 10 days after being discharged, she ends up back in A&E with the same or similar diagnosis.
This phenomenon has taken colossal size in recent years, having multiple implications in terms of bed pressure, quality of care and of course cost. To tackle the issue, the UK department of health - in a very questionable 2011 act - decided to penalize readmissions by withholding payment of the second hospitalization from local trusts. The rationale being that it is a poor first time care that results in the patient needing hospitalization again in such a short time. Penalties since, have been reported in the billions of pounds, resulting in various NHS trusts merging, hospital managers losing their jobs and millions cut from trust budgets. The academic community reacted by an array of publications trying to explain and eventually solve the problem primarily by trying to establish mechanisms that could profile or even predict the frequent attendees, in an attempt to prevent them from coming back through the doors of the A&E. In the plethora of methods invented to predict which patients may be coming back, the one that performed best in terms of prediction accuracy is the Nuffield Trust PARR30 model.
At EXUS we chose to tackle the problem ourselves. At the time (mid-2015) we had all the right ingredients to come up with a promising approach. We were already working with Croydon NHS Trust R&D Director @John Chang, Professor of Pharmacy and clinical testing expert @Reem Kayyalli and the team had just been completed by a seasoned data scientist @AnujSharma.
The idea was simple, we first built a prediction algorithm based on the data that was made available to us from Croydon hospital. We would then build a software on top that could help the hospital avoidance team manage the patients post-discharge. In this way we would know who to prioritise for a follow up (to minimize already scarce medical personnel resources) and we would also have a complete post-discharge picture of the high-risk patients. The nurses would then use a predefined set of questions to assess each case over the phone and decide on the best possible intervention from a selection of primary and community care services available in the region.
The first version of OPTIMAL was released in March 2017 to two experienced Croydon nurses that embraced the idea. Our prediction algorithm performed equally well with the PARR30 model for a fraction of the input parameters, allowing the algorithm to become operational in the hospital setting. 12 months, two releases and one additional hospital site later we had enough data to evaluate the approach. A little under 1500 patients provided consent and went through the OPTIMAL system. Randomized control trial allowed only for half to be followed up. The other half was used for evaluation.
Results showed a 2% reduction in readmitted patients in the intervention group. Although a seemingly small percentage, if extrapolated to the 30,000 annual admissions for Croydon this is translated to 600 avoided readmissions, 3,000 bed days or £1m in costs saved.
Most importantly we had a 99% positive feedback from patients on the follow-up phone call. Almost all of the patients that participated in the survey were happy that someone took the time to see how they were after they left the hospital. We cannot evaluate how this affected the result or if it contributed to the reduction in readmitted patients but it made the entire effort worthwhile and is proof that the science alone is not enough.