Mapping GP prescription data
GP prescription data is a valuable source of information
The analysis of GP prescription data enables healthcare providers and local authorities to understand the health needs of communities, plan support services, align resources effectively and strategise for future requirements. Analysis can also contribute to the planning and implementation of targeted awareness campaigns.
In this article, written by Allan Brimicombe (Head of Centre for Geo-Information Studies at the University of East London) & Pat Mungroo, GP prescription data has been used to map the prescribing of medication used to treat Schizophrenia and similar psychotic illnesses in England. The article also plots the rate of change over five years.
Allan Brimicombe has combined the data with our geodemographic classification, P² People & Places, to analyse possible causes, such as environment, and to identify patterns/relationships between communities and the prescribing of medicine.
P² People & Places is a comprehensive classification that we built using Census data and other lifestyle data. It provides detailed descriptions about the lifestyles, behaviours and attitudes of people living in the UK and is widely used by public and private sector organisations.
Please note that GP prescription data is open source and available to the public. It doesn't contain any personal data about individuals.
Allan Brimicombe has kindly permitted us to publish this article.
A Note on Maps of GP Prescribing for Schizophrenia and similar Psychotic Illnesses in England.
Allan J. Brimicombe & Pat Mungroo
Centre for Geo-Information Studies, University of East London
This note is written in response to international requests for more information following the BBC’s publication of the maps at: http://www.bbc.co.uk/news/health-39082595 dated 25th February 2017.
Two maps have been created to show the distribution of GP prescribing for Schizophrenia and similar psychotic illnesses in England on the one hand, and the rate of change over five years on the other.
The purpose of the maps is to explore geographical variation across England. This visualisation of psychosis prescribing not been achieved before, so seeing the geographical variation is in itself a potential eyeopener. In doing so, new questions can be asked both of the data, practice and policy that are focused on explaining and, where desirable, reducing these geographical variations.
The data for the maps comes from GP prescriptions data collected by the NHS in anonymised form. They are available from NHS Digital https://www.digital.nhs.uk/ . There are about 120 million records a year and to compare change between two different years requires analysing some 240 million GP prescription records .
We have extracted from the database of all GP prescriptions those prescriptions that are for various types of mental illness. These represent about 10% of all GP prescriptions at a cost of nearly £80m a month (all GP prescriptions cost about £710m a month; whilst the NHS recoup money through prescription charges we have no data on this in relation to the GP prescription data as not everyone has to pay a prescription charge). The GP prescriptions for mental illnesses have then been grouped by Local Authorities and Districts for further analysis. We have used the data for the twelve months from October 2015 to September 2016 (latest available data at time of download) and then compared it with the twelve months from October 2010 to September 2011, five years previous. Our maps focus on Schizophrenia and similar psychotic illnesses, but maps for other mental illnesses such as dementia can be produced. The data can be analysed down to individual GP practices and for individual medications.
The download, extraction and preparation of the data ready for analysis were carried out by our partners Terra Cognita Limited.
The causes of Schizophrenia and similar psychotic illnesses are not precisely known and may be due to a genetic predisposition, environmental factors such as stress, family dynamics and changes in lifestyle, the abuse of alcohol and drugs or a combination of these. In this respect, explaining the variations in the maps may not be straightforward.
The geographical pattern of GP prescribing for Schizophrenia and similar psychotic illnesses in Figure 1 is not uniformly spread. Geographical variation could arise if there were differences in:
- GP diagnosis and prescribing practices for psychosis. We consider that this would be more random across the country and would largely be aggregated out at Local Authority levels
- Genetic disposition, say by ethnicity. This would require reference to reliably comparable data at national level from around the world. The GP prescribing data contains no patient details and therefore no data on prescribed patient ethnicity is available.
- Environmental causes such as stressful life events and lifestyle issues such as drug and alcohol abuse. It would seem to us, from the pattern that environmental causes are exerting a greater influence on the outcome.
The top and lowest ten Local Authorities are listed in Table 1.
|The top 10 areas in England are:||The lowest 10 areas are:|
|North Kesteven (39)||East Dorset (9)|
|Coventry (35)||South Gloucestershire (10)|
|Rochdale (34)||Tewkesbury (10)|
|Cambridge (34)||York (10)|
|Hastings (33)||Epsom and Ewell (11)|
|Manchester (33)||South Northamptonshire (11)|
|Walsall (32)||Blaby (11)|
|Kingston upon Hull (32)||North Dorset (11)|
|Wigan (31)||Hambleton (11)|
|Liverpool (30)||Wokingham (11)|
The nature of environmental causes will differ from area to area obviating a one-size-fits-all solution, but can be analysed. The top 20% of areas as shown in the map encompasses both urban and rural areas, university cities and coastal towns. Looking for causes of a pattern requires systematic statistical analysis.
One way of getting an idea of the causes is to look at demographic and lifestyle classifications. We use the P² People & Places classification from Beacon-Dodsworth Ltd which splits the population into categories based on lifestyles, backgrounds and behaviour. An area the size of a Local Authority can expect to have a range of lifestyle groups. However, based on a regression analysis (Annex 1), the groups that have the strongest positive association with higher rates of GP prescribing for Schizophrenia and psychosis are:
Group N - Disadvantaged Families: Deprived young urban dwellers, lone parents with young children, and couples with children. Immigration is below average. Housing rented in dense, overcrowded areas. Long-term unemployment is common. In very poor health, smokers and consumers of junk food. Typically ex-industrial areas.
Group L Struggling Singles: Deprived young urban singles and lone parent families with pre-school children. Much recent immigration. Small households mostly privately rented. Poor or no qualifications across all ages means low employment or part-time work doing routine jobs. Found to be in poor health, sometimes work-related illnesses. Likely to have poor diet and smoke. Coastal and market towns, some provincial cities.
Group K Multicultural Centres: Culturally diverse young families with children, married and lone parents. Some students. Significant immigration both recent and 2nd generation. Housing rented and overcrowded. Generally poorly-qualified young unemployed of part-time manual workers in service sectors. Below average health, average diet although too many takeaways. Outer London, major cities and northern towns in England.
The group most associated with lower rates of GP prescribing for Schizophrenia and psychosis is:
Group C Middle England: Older married households in predominantly white areas, with low migration and low population density. Spacious rural detached dwellings, mostly owner-occupied. Well-qualified managers and professionals and other white-collar workers. Average health - eats well and spends on health and fitness; doesn't smoke. Lowland rural and suburban England.
Turning to Figure 2, it shows the annual percentage rate of change in GP prescriptions per thousand population for Schizophrenia and similar psychotic illnesses over the five years to September 2016. The pattern is very striking with clusters of above average increases in East Anglia and other rural areas. These changes do not have a strong association with lifestyle types and is therefore more likely to be due to differences in policies and practices in the way mental health services are commissioned across the country resulting in growth/decline in the numbers of patients living and being treated in the community.
The average percentage change in the rate of prescribing for England is an increase in 3.3 percent per year. The Local Authorities with the greatest increases and decreases are given in Table 2.
|The top 10 areas with greatest increases:||The top 10 areas with greatest decreases:|
|Isles of Scilly (+29)||Corby (-5)|
|West Somerset (+14)||Hastings (-4)|
|South Norfolk (+13)||Havant (-3)|
|Selby (+11)||Worthing (-2)|
|West Devon (+11)||Three Rivers (-2)|
|Gedling (+10)||Kettering (-2)|
|Broxtowe (+10)||Epsom and Ewell (-2)|
|North East Derbyshire (+10)||Runnymede (-2)|
|Mid Suffolk (+10)||York (-2)|
|Uttlesford (+9)||Wirral (-2)|
There are many possible causal variables we could look into over a range of geographical scales, and would likely take a lot of time and effort. A geodemographic classification (P² People & Places) has allowed us to quickly identify the characteristics of those neighbourhood types that nationally are positively associated with higher rates of prescriptions for Schizophrenia and similar psychotic illnesses. Of course the maps ask more questions than they answer, but we can start to ask the right questions.
The Centre for Geo-Information Studies, University of East London is a research centre specialising in extracting value from data. Clients include police forces, NHS trusts, local authorities, and private sector companies and organisations such as the International Olympic Committee. The Centre runs an MSc and Professional Doctorate in Data Science. http://www.uel.ac.uk/geo-information/
Professor Allan Brimicombe is Head of the Centre for Geo-Information Studies (CGIS). He works at the intersection of geography and geographical information systems, criminology, social statistics, IT and data science. Pat Mungroo is a Visiting Senior Fellow with CGIS. He has 30 years experience of working as a mental health practitioner. He does research into drug use and mental health problems.
Beacon-Dodsworth Ltd are a data and software consultancy that provide mapping and market analysis solutions to private and public sector organisations throughout the UK and Europe. P² People & Places is a comprehensive geodemographic classification that uses census and lifestyle data to categorise the UK population http://www.p2peopleandplaces.co.uk/
This note has not been peer reviewed. Ethical clearance was not required. We have no conflicts of interest; this piece of work is not externally funded.
P² People & Places classification at the higher level of the ‘tree’, by Lower Super Output Ares (LSOA), were aggregated by population count to Local Authority areas to give a matrix of Local Authorities and P² trees A-U (excluding G which is only present in Wales; U is unclassified). The population counts for each tree were transformed into a percentage of the Local Authority total population. The count of prescriptions for each Local Authority were transformed into per thousand total population. Thus all counts were transformed into proportions. One data problem is that there are a lot of zeros in the matrix (zero-augmented regressors) where a Local Authority has no population in a tree. Correlations with the rate of prescribing by tree are given in Box 1 using pairwise complete observations. Many are not statistically significant. We carried out multiple regression with case-wise removal of zeros iteratively to weed out non-statistically significant regressors to come up with the main trees positively associated with the rate of GP prescribing given in Box 2. Tree C is the only one that has a statistically significant negative association.
Box 1: Correlations of GP prescribing with each geodemographic types (excluding G and U)
|cor MH.dis2, use="pairwise.complete.obs", method="spearman"|
Box 2: Regression of trees N, K and L as predictors of the rate of GP prescribing
|lm(formula = MH01_1516 ~ N + K + L, data = MH.dis2, na.action = na.omit)|
|Estimate||Std. Error||t value||Pr (>|t|)|
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 4.647 on 124 degrees of freedom
(198 observations deleted due to missingness)
Multiple R-squared: 0.2464, Adjusted R-squared: 0.2282
F-statistic: 13.51 on 3 and 124 DF, p-value: 1.096e-07
Allan Brimicombe can be contacted at email@example.com
All you need to know about postcodes but were afraid to ask
The humble postcode has been around for years. We look at how postcodes are used and what led to their introduction.
How far is it to the beach
Using Beacon Dodsworth's scripting technology to showcase demographic and geographic trends.
What is GIS software?
GIS software is a problem-solving toolkit that will effectively enable you to manage large datasets for example from your customer database, understand locations and present information.
Where is the North
We've used the territory manager tool in Prospex GIS to split the UK into North, East, South and West with equal population counts.
A career in the GIS industry
Three months into my first professional software developer job with Beacon Dodsworth, and I love it.
Geodemographics and the University of East London
The University of East London explain how they have been using our P² People & Places geodemographic classification.
Mapping GP prescription data
An article by Allan Brimicombe (Head of Centre for Geo-Information Studies at the University of East London) & Pat Mungroo on using GP prescription data to understand health needs.
Lifestyle data and spending habits
Exploring consumer habits around baked goods using lifestyle data and geodemographic profiling unlocks a range of opportunities.
Household spend on fruit and vegetables
Using GIS mapping to visualise, analyse and present data in a geographical context.
Postcode to postcode drive time and distance
What happens if we want a postcode to postcode drive time lookup table?
What has the census ever done for us?
How Census 2011 can be used to help organisations with demographic analysis.
Customer data analysis solutions
Customer data analysis solutions provide actionable insight and a range of opportunities that can make a huge difference for businesses. Free data samples available
Your continued use of this site is taken as implied consent to receive cookies from us and our analytics partners.