Urgent Alert

From 1 April 2022 this website will not be updated.

  For the latest North West London health and care information, visit www.nwlondonics.nhs.uk.

What approach should we take?

Over the course of co-design, the Population and Outcomes working group, which has included commissioners, clinicians, mental-health, social-care, public-health experts and lay partners, has agreed a proposal to group the population into ten groups. It is important to remember that this grouping represents the primary organising logic and is consistent across North West London.

As with any general framework, there will always be exceptions, no matter which primary organising characteristic we use. The point here is to create a broad framework, within which providers will be able to make specific decisions to tailor the offer of care to individuals. As such, it is important not to get distracted by the exceptions, and rather focus on the broad principles of what we are trying to achieve.

International examples of grouping can range from very simple, such as that from Press Ganey:

To more complex, such as Delaware:


Methodology that the working group used

In carrying out the grouping, we used three complementary methods. First, we gathered the judgement of multiple professionals and lay partners from across North West London. Then, we did an in-depth analysis of a fully integrated example data set gathered from Hammersmith and Fulham to test our hypotheses. Bringing together data from across acute, primary, community and social-care helped us to understand levels of service utilisation and cost for each group, which helps build a picture of population needs. Finally, we also looked at how populations had been grouped in other health systems both nationally and internationally. Using these three approaches, we reached consensus in the working group around how to group the population of North West London.

It is also worth bearing in mind that children were outside the scope of this current phase of the programme. This does not mean that there are no innovative models of care for children. Rather, this might be a focus of a future phase.

As part of our analysis of the data from Hammersmith and Fulham, we created a regression tree. A regression tree is a way of visualising how data clusters within a large data set. All of the data is put into a model, and each subsequent split in the data is the split that explains that largest amount of variance within that group. The results of the regression tree analysis are shown below.

The major implications from this regression tree are as follows. First, there is a split at over or under 75, which explains a significant amount of the variation in cost within the population. Second, there are six condition types represented: serious and enduring mental illness, dementia, learning disabilities, physical disabilities, cancer and long-term conditions. It is useful to know that these elements are significant in the data when we are thinking about grouping in North West London. A more detailed description of the regression tree methodology and results can be found in Supporting Material E: Review of Analyses During Co-design Phase that accompanies the toolkit.

Exhibit 4.1 

What is the grouping approach for North West London?

There are ten proposed groups that cut across health- and social-care, and represent the holistic needs of the individuals that fall into those groups (see Exhibit 4.2 below). As such, a model of care surrounding the serious and enduring mental illness group would address all care needs of the people in that group, whether they are mental, physical or social, and would address these needs across organisations. The idea is to address the needs of individuals, rather than the specific conditions or the specific type of care.

Exhibit 4.2 


Within the proposed grouping above, cancer is a special group, as it is the only group that people will move in and out of due to its bimodal nature. Because of this, the working group decided that if it was necessary to treat this group differently. People who have an active diagnosis of cancer will be given extra capitation money and more care coordination services, but will remain in their home group.

For the purposes of the data analysis, cancer was included in the hierarchy as the third category below physical disabilities and learning disabilities. They are included after LTCs in the exhibit above because of the perceived need for a more targeted whole model of care for all the groups after cancer, e.g. SEMI, organic brain disorders, learning disabilities and severe physical disabilities.

Exhibit 4.3


As mentioned previously, we need a hierarchy to ensure that groups are distinct and unambiguous. The hierarchy used was – (a) physical disability, (b) learning disability, (c) cancer, (d) advanced organic brain disorders, (e) serious and enduring mental illness, (f) long-term conditions and (g) mostly healthy.

The following exhibits show how the data from our integrated dataset was analysed based on the groupings described previously. Exhibits 4.4 and 4.5 show the total costs and average costs for the population groups. The size of the blue bubbles corresponds to the magnitude of the cost. As you can see from Exhibit 4.4, the mostly healthy and LTC groups by far have both the largest populations and the highest total costs. As you can see from Exhibit 4.5, learning disabilities and severe physical disabilities represent the highest average costs per person, yet have relatively small population sizes. More detail on this analysis can be found in Supporting Material E: Review of Analyses During Co-design Phase.

Exhibit 4.4 

The next exhibit shows the coefficient of variance for each of the population groups. The coefficient of variance is the standard deviation over the mean, and gives a standardised measure of the relative variation in the data across the groups. As can be seen in the exhibit, the 16-74 Mostly Healthy group has the highest coefficient of variance, which means that there is the most variation in costs for this group. This intuitively makes sense, as there are some mostly healthy people who will have no health- or social care costs in a year, and others who might be in a one-off accident and cost tens of thousands of pounds. The severe physical disability group has the lowest coefficient of variance of all the groups, which means that costs within this group are the most consistent.

Exhibit 4.6