For those who know me, I frequently say the number 42 to my peers in relation to why patients get admitted to hospital. We know the answer, that they are in hospital but what is the question, what brought them in in the first place or more importantly can we predict their chance of admission in the future? Over the next few years we are going to get a deluge of companies claiming their risk stratification tool is the best thing since sliced bread and that their algorithms have prevented x% of admissions in the last year and saved the NHS this much. This blog is about my modest practical experiences on this matter for those commissioning groups who are looking at this area. In fact I don’t even discuss about any particular risk tool. Ironically this blog also won’t tell you the answer but give a few simple concepts and I hope some of it’s content will help you.
I’m part of a motley crew of 5 other GP practices in the local area and we were part of a pilot which looked at integrated care. We also did a bit of telehealth which I can discuss at a later blog.
We meet weekly with our local nurses to discuss patients who have have attended A&E with certain co-morbidities and have been discharged from hospital in the previous week. The outcomes of our pilot were favourable although we have no current data and we’ve lost our psychologist and social care worker. With regard to this model of care I still think we have a long way to go.
I thoroughly recommend reading literature from the NuffieldTrust about this. I certainly learnt a lot from their publications. These 2 articles are especially good reading: Choosing a predictive risk model and Predictive risk and health care.
Regression to the Mean
Distilling this into admissions, regression to the mean relates to the fact that patients exhibit patterns of multiple admissions which become more frequent and ebb towards a baseline regardless of the intervention. So if we focus on a patient at the height of their multiple admissions, even if we do nothing they will stop going in.
Whatever you think about, think about catching the patient before they reach the end of the curve. Most risk tools will state this but bear in mind those who have a high risk of attendance might be those who have little rehabilitation potential ie if we focus all our energies on these patients, they will go in anyway. The reason for this is complex, but on the whole my belief is they have developed a learnt behaviour or attachment to the hospital to get them better. If we are to get a grip on the 20% reduction in the NHS budget, I believe we should be focusing on those who haven’t gone in that much and where we can kerb this learnt behaviour and catch them earlier in the curve. An example is a 90 year old patient who lives alone with COPD who hasn’t attended hospital in the last few years but who attends A+E with breathlessness. It should be all guns blazing on this type and risk tools need to identify these types too.
People and processes come first
No matter what risk tool you choose, it’s important to have the correct governance or structure within your organisation to operationalise everything.
Patients on a piece of paper churned out by the risk tool don’t keep them out of hospital: you need case load management, have a regular forum to discuss patients and actually see these patients with actions and resolutions and delegated responsibilities.
Organisational Development (OD) or what I phrase “The Vehicle” is probably as important as which patients to discuss or “The Content”. You can get a stalemate situation where GPs say “I need a risk tool to keep patients out of hospital” and the managers say “I’m only giving you a risk tool if you get your act together”. I think both are right and as with most things operationalising models use up resources. Commissioning groups need to look at this resource seriously and which pocket of funding the whole of risk profiling (including more than just the software) comes out of. The options for resources both financial and human are from GPs as a provider via their QP points vs a commissioning pool vs both.
Get as realtime to the data as possible
This point has 2 interpretations to it.
Obviously meet up as often as the data refreshes, so if your data refreshes monthly, interact with this data as close to when you get it as possible.
The other aspect is probably more important though but is not without it’s issues. It’s not rocket science, but I’m a firm believer that we need to get hold of patients as soon as they have their event, indeed in our group we get daily A&E activity data from our local hospital. In our forum, Monday mornings can be quite hairy as we wait for activity data from a local hospital on patients who have attended A+E and been admitted and discharged from the previous Monday to Sunday (the day before the meeting). It’s not quite real time (being daily) and we meet weekly but if I could I would ask the GPs in my group to meet daily. However we live in a more pragmatic world!
The concept is as soon as the patient appears on the radar and attends A+E or is discharged, we hit them as hard as possible with community based care to prevent them from getting used to the learnt behaviour of hospital care. We also focus on activity of A&E attendance rather than hospital admission as A&E attendance can be a predictor of subsequent admission. If anyone can find a better paper to reflect this, please let me know.
Here’s the problem though, most risk tools look at a merge between primary care data and hospital budgeted (SuS) data with the unfortunate side effect of being out of date by at least 3-4 weeks. At the moment we look at daily activity data from a local hospital which is more real time but is at the expense of accuracy. For me the benefits of immediacy outweights getting budgeted data.
In my opinion the main reason why risk tool companies look at budgeted (SuS) data is because
– arguably it’s more accurate than activity data.
– It’s in the hands of commissioners therefore easier to obtain.
– It’s more uniform across the UK so is easily implementable across the country with little change in their software.
– Conversely and more importantly it’s much harder to organise bespoke SLAs with local Hospitals to get realtime activity data.
Risk tools don’t focus on obtaining real time data due to the above and my hope in the future is that we will get a culmination of both types of feed (both budgeted and activity).
Subjective as well as objective
Not matter how good a risk tool which can be produced, we need to filter patients through subjective means or how we feel about their risk. Unfortunately this is a feeling rather than something more objective. As risk tools become more refined I hope this will improve as the quality of our data mining improves and as we get more quality data.
There are just too many variables and it is one of the reasons which separates us from computers. There have been many a time in our weekly meetings when I believe a discussed patient should be focused on. The data gives use a direction, but clinicians provide the focus.
I also apologise for the light hearted approach of using the force for such an important topic of patient selection but I’ve seen monthly lists churned out by risk tools being given directly to the community matrons (nurses working in the community whose role is to reduce hospital admissions) without any collaborative forum or filtering from any providers which is very scary! Data is a good slave but a poor master.
Choosing your patient
Why do they go into hospital?
So you have your patient which you are discussing in your forum having been risk profiled and have to decide what you want to do about them or if they should be focused on. We can try the force, but it probably won’t get far for us non Jedi Masters. I’ve enclose a diagram to try to be as systematic about the question to the answer of 42 above as possible. This is something which I go through when discussing patients which I hope will help.
MDT refers to our weekly Multidisciplinary Team Meeting and I’ve mentioned a little bit more on factors to determine whether a patient should have a telepod as this diagram is around the work I’ve done with telehealth.
In summary there are 3 main areas
– Is the reason for admission based on a culmination of one or more of a social, medical or mental cause and can we unravel this?
The idea is to fill this need with community based services to help prevention of admission for example if they have a psychological need, provide an outreach psychological service which offers face to face therapy for housebound patients.
– Does the patient want to change their habits?
This goes around learnt behaviours of multiple attenders who just want to go into hospital and will go in whatever we do. Examples of these are alcoholics, chaotic patients with diabetes or anxious patients with COPD. Our “Heart Sinks” which we know if we focus on we won’t stop them from going in. Unfortunately the system collapses if the patient isn’t eager to change and patients need to be active stakeholders in their care rather than a passive recipient.
– What is the profile of the patient?
As discussed regression to the mean is important here as I believe we should be focusing on patients who are just starting on the curve to prevent their learnt behaviour of reliance on the hospital. Activity or budget profile is important here and look at drilling down more around what type of patient they are: Frequent Flyers, Recent multiple attenders or One Offs with no potential to re attend. In my opinion we should be focusing on recent multiple attenders.
I hope commissioners of care find the above useful and it’s based on practical thoughts around a currently working model with a few pragmatic assumptions. There maybe a need to formalise this to make it more research based.
I also enclose a paper I wrote on this last year which gives more information around using integration and also on telepods.