Charles Tallack, Head of the NHS Operational Research and Evaluation Unit, discusses how the New Care Models are using data to identify and protect our most vulnerable patients, reducing their chances of being admitted to hospital.
Our health and care services are working to deliver the most positive experience and outcomes for patients.
Instinctively, we know that some patients, particularly older people with long-term conditions, are much more likely to end up in hospital.
Calculating risks accurately for the wider community, and putting in place interventions which then prevent emergency hospitals admissions and other costly, undesirable consequences, is a little harder to achieve. And, as we know, prevention is better than cure!
Our team of analysts at NHS England have been looking closely at the approaches some of the new care model vanguards have taken to identify patients at risk in their communities.
The term we use for this is ‘risk stratification’. Yes, it sounds a bit ‘techy’ but it is widely recognised by health and care leaders and is being adopted by many Sustainability and Transformation Partnerships and accountable care systems.
Another phrase we use is ‘predictive modelling’ or ‘artificial intelligence’. These analytics are used in all sorts of fields, including financial markets and Google.
We looked at five new care model vanguards – Tower Hamlets, Sunderland, North East Hants and Farnham, Erewash and Morecambe Bay – that have introduced the ‘building blocks’ of an effective risk stratification process. We discovered that most are using predictive models and implementation is clinically-led.
Other models that depend on clinical judgement alone and threshold modelling – identifying individuals at historic risk rather than future risk – are less effective and can have misleading results.
For instance, threshold models typically pick out patients who have had a high number of admissions in the previous year. However, generally speaking, we would expect this patient’s number of admissions to decline the following year. Unusually high numbers are likely to be followed by lower numbers.
You can see this phenomenon in everyday life: for instance, a young sports star who adorns the cover of sports magazines in their breakthrough season, very often disappears from the spotlight the following year.
If we target these people we are very likely to miss those individuals who are actually at highest risk of being admitted the next year and so our predictive powers must improve.
Predictive models that use data from a range of sources have the highest accuracy. Morecambe Bay vanguard, for example, uses data from hospital patient activity and GP practices, data from patients with a number of medical conditions and demographic information about the local community.
The findings then appear on GP IT systems and doctors can then make a judgement on whether or not to refer a patient who is likely to have an emergency admission in the next 12 months onto integrated care community teams. Steps can then be taken to support that patient and keep them out of hospital unless it is clinically necessary.
During our research, we found that all vanguards were attempting to identify, and work with, the ‘highest risk’, usually the top 2%, of their population. However, we have to still bear in mind that not all admissions are preventable and we have to look at larger numbers of people to have a major impact on overall admission rates.
We also have to go further and think about ‘impactability’. Once patients at risk are identified, they can be offered personalised interventions – what works for one individual might not work for another. Priority may be given to patients with diseases that are particularly responsive to preventative care.
Many vanguards are introducing innovative approaches to doing this – multi-disciplinary teams that bring together health and care professionals for patient reviews, complex care teams that focus on patients with many complex conditions, health coaches to help patients learn self-management techniques, and social prescribing where patients are directed to activities in their community that will improve their mental wellbeing and fitness.
There are other important considerations that we have covered in our analysis, such as information governance which is critical for patient confidentially and data protection.
Our full findings on risk stratification are included in a Learning and Impact Study which is available on request through our evaluation community of practice at email@example.com.