Health of the nation

Geodemographic profiles use addresses or postcodes to classify people into groups which are homogeneous with respect to variables like income, housing tenure and life stage.  The original purpose of geodemographic profiles was to improve targeting for marketing purposes.  There is no point in sending marketing material for hearing aids to young families, for example, and geodemographic profiles help make more efficient use of marketing resources.

Geodemographic profiles have been conclusively shown to be predictive of mortality in the United Kingdom, both by Richards (2008) and Madrigal et al (2009).  This is because geodemographic profiles are about education level, wealth and income (amongst other variables) and these things are all strongly correlated with mortality.  But what if a profiler were specifically created with health outcomes in mind?  Would this be better than a geodemographic profile created for marketing?

We recently took the standard geodemographic profiler used by Madrigal et al (2009), namely Acorn from CACI Ltd.  The same provider also has a health-orientated profiler, called Health Acorn, and some summary statistics are given in Table 1.

Table 1.  Summary of Acorn and Health Acorn profilers from CACI Ltd.

Our first step is to find a baseline against which to measure models.  We start by fitting a model with Age, Gender and Time to establish its AIC as our reference point.  We use a survival model to make the best use of the available data.

We next try including either Acorn group or Health Acorn group, and see how the AIC improves in each case.  Any differences in improvement will mainly be due to the power of each profiler.  The results are shown in Table 2.

Table 2.  Explanatory power of Acorn and Health Acorn for model of pensioner mortality.  "Explanatory power" is the drop in AIC from moving from Age*Gender+Time model to Age*Gender+Time+Group.

Table 2 clearly shows that the geodemographic profiler, Acorn, is far more powerful at explaining mortality patterns than Health Acorn. This might be felt to be due to the larger number of groups, but the AIC automatically balances improvement in fit against extra parameters.  However, a more likely reason is that the Health Acorn groups are rather heterogeneous, thus weakening their power in explaining mortality variation.

Fortunately, we can improve on the weakness of Health Acorn group by optimising the underlying type codes into a bespoke, three-level lifestyle grouping.  We can also do the same with the Acorn type codes, and see how the explanatory power of each profiler changes.  At the same time we allow the impact of lifestyle (or healthstyle) to vary by age.  The revised comparison figures are shown in Table 3.

Table 3. Explanatory power of Acorn and Health Acorn for model of pensioner mortality. "Explanatory power" is the drop in AIC from moving from Age*Gender+Time model to Age*(Gender+Lifestyle)+Time.

Table 3 shows that this new approach has improved the apparent explanatory power of both profilers, with Health Acorn improving much more than Acorn.  However, Health Acorn is still not as good a predictor of mortality as Acorn is.  Since both profilers come from the same supplier, we can only conclude that the general geodemographic profiler, Acorn, is superior to the apparently health-orientated alternative, Health Acorn.

For those who are interested, I recently gave a presentation to the 2009 Life Convention of the results of a wider comparison of postcode profilers in the UK.