Mapping unpaid carers using GP contract data
A guide for analysts
Unpaid carers play a crucial role in health, care, and wider society in the UK. There are at least 5.7 million according to ONS census data, and possibly many more. The value of the care they provide is estimated at £162bn per year, almost equal to NHS England’s funding in 2023.
The theme of Carers Week 2024 (10–16 June) is Putting Carers on the Map, highlighting how their unique challenges must be fully recognised and present on the political landscape in this crucial election year.
Equally important is for carers to be ‘on the map’ in a literal sense. They must be accounted for and known to the national and local institutions that can provide them with support. With 82% of unpaid carers concerned about their mental and physical wellbeing over the coming year according to a survey by Carers UK, it’s essential that as many as possible are on the map for services that can help ease their burden.
Too often, however, unpaid carers do not receive the support they need, either from the NHS or social care services. Despite widespread recognition of their importance and policy commitments to increase assistance, there is little information about how many are known to services that can support them. In our recent work, the Health Foundation’s Networked Data Lab (NDL) showed that unpaid carers are dramatically undercounted by local services. Analytical teams in five areas of England and Wales piloted patient-level linkage of GP and local authority electronic records, finding that data held by GPs and local authorities identified at most between 11% and 26% of the unpaid carers recorded in the 2021 census.
In spring 2022, a new indicator on unpaid carers registered to GP practices was added to NHS England’s GP Contract Services — England dataset. This provides one of the first publicly available, regularly updated sources on the recording of unpaid carers by primary care services. This dataset is published on a quarterly basis, typically with one month’s lag between collection for the final month of the quarter and publication. In close to real time, this offers opportunities to trace the degree to which unpaid carers are accounted for by their local services and to explore geographic variations in the recording of unpaid carers across England.
As a follow-up to our work on identifying unpaid carers, the NDL team explored the completeness, quality and coverage of this new data source. We transformed the GP core contract data to a local authority level to make them comparable to the measures of unpaid carers in the 2021 census. In the process, we uncovered vast differences in the degree to which unpaid carers are captured in GP data between local authorities.
In this blog, we will summarise what we learned about this new publicly available indicator and what we also learned from using it. We then assess to what extent GP contract data capture the unpaid carer population, and examine possible explanations for the disparities in unpaid carer recording between local authorities.
Unpaid carers in GP contract data
NHS Digital’s GP Contract Services — England dataset provides information on primary care programmes, including core contract components, enhanced services, vaccination and immunisation. Data on GP Contract Services are collected automatically through the General Practice Extraction Service (GPES), based on the records in GP practice clinical systems. It is not mandatory for practices to submit for this dataset.
The new indicator on numbers of unpaid carers registered to each GP practice was added to the GP contract data for 2022–23 Q1. Identifying unpaid carers from GP records is challenging, with a wide variety of SNOMED codes associated with unpaid care. Furthermore, some GP practices have registers of carers that are not linked to patient records. To simplify identification, guidance was circulated by NHS England in October 2022 listing two SNOMED codes that should ideally be used by GPs. In the latest business rules for the GP Core Contract data collection, a set of 30 SNOMED codes are listed.
In the initial quarters that this indicator was available, coverage was often incomplete, with entire Integrated Care Systems (ICSs) missing in some cases. For instance, all GP practices in the North Central London ICS were missing from the 2022–23 Q4 release. Although some GP practices are still missing, coverage has vastly improved, with 98.8% of practices in England (accounting for 99.5% of patients) included in the latest release.
To compare unpaid carers registered by GP practice to the number of unpaid carers by local authority reported in the 2021 census, the practice-level GP contract data was transformed into local authority-level data. For more information on this, see our technical appendix on GitHub.
Once transformed, we derived measures of unpaid carer coverage by expressing the number of unpaid carers recorded in a local authority in the GP contract data as a proportion of the number recorded in the same local authority in the census. For instance, if a local authority reported 2,000 carers in the GP data but 4,000 in the census, it received a coverage score of 50%.
How well are unpaid carers recorded in GP core contract data?
In total, the GP contract dataset captures approximately one-third of the number of unpaid carers recorded by the 2021 census, registering 1,580,101 unpaid carers in England (2.8% of the mid-2021 population), compared with 4,678,260 in the census (8.3% of the population).
In all local authorities, far fewer unpaid carers are reported in the GP contract data than the 2021 census but coverage varies widely. For example, Kensington and Chelsea, South Holland and Torbay report over 60% of the census figure in GP registers (66%, 66% and 62% respectively). In contrast, Basildon, Luton, Brentwood, Darlington and Bedford record less than 15% of the census figure.
What might explain the variation in coverage of unpaid carers in GP data?
Missing practices
In some local authorities, the fact that some large GP practices are missing entirely from the GP contract data may account for a portion of the difference between total unpaid carers registered in GP practices and those recorded in the census. For instance, three local GP practices are missing from Sheffield, accounting for 15,920 patients in the GP contract data and likely contributing to its low coverage of only 18.6% of unpaid carers in the census. However, this is an insufficient explanation by itself. Of the 10 local authorities with the lowest-recorded coverage, Sheffield is the only one missing any practices in the GP contract data. Missing practices may indeed affect individual local authorities’ measures of unpaid carers in the GP contract data, but explanations for most of the observed differences likely lie elsewhere.
Differences in reporting practices between Integrated Care Systems (ICSs)
Of the ten lowest-coverage local authorities, four are administered by the Mid and South Essex ICS (Basildon, Brentwood, Castle Point and Braintree), and two are covered by the Bedfordshire, Luton and Milton Keynes ICS (Luton and Bedford). However, when comparing local authorities grouped by the ICS that administers a majority of its Lower Super Output Areas (LSOAs), it appears that within many ICSs a wide range of coverage exists between administered local authorities. In most cases, local authorities are only serviced by a single Integrated Care System (ICS) — only 12 local authorities are administered by more than one ICS, with none being administered by more than two. While the local authorities in some ICSs cluster relatively closely (including those low-scoring authorities administered by the Mid and South Essex ICS, exhibiting a standard deviation of only 4 percentage points), in many other cases more variable rates of coverage are observed within the same ICSs. Overall, ICSs exhibit a mean standard deviation of 7 percentage points of coverage between the local authorities they contain.
Low data coverage where total recorded carer numbers are high
Local authorities with a very large number of carers recorded in the census tend to have low coverage in the GP contract data. All local authorities recording over 50,000 carers in the census (Birmingham, Leeds, Cornwall, and County Durham) have less than 32% coverage, and all with over 50% coverage recorded fewer than 27,000 carers. However, statistically, this is not a strong association (r = -0.08), as there is a very wide range of coverage in authorities recording low numbers of carers:
We also compared levels of coverage to:
· the proportion of carers in a local authority who were female (r = 0.03)
· % of households with at least one dimension of deprivation (r = -0.08)
· the proportion of carers who were over 65 years of age or under 25 years of age (r = 0.03 and r = -0.07, respectively)
· the proportion of carers who performed over 50 hours of care each week or who performed under 9 hours of care each week (r = -0.05 and r = 0.04, respectively)
There did not appear to be any associations between these metrics and coverage of census-reported unpaid carers in GP data, indicating that local factors outside of the available measures probably explain interlocal authority differences.
Limitations
While these comparisons can be used as a bellwether for improvements in recording of unpaid carers, the conclusions that can be drawn from them are limited. A greater number of unpaid carers appearing in the GP core contract data over time could simply indicate a greater total number of unpaid carers in the more recent period compared with when the denominator measure was taken, rather than a greater proportion being captured in GP data.
This is also a broader limitation of our analysis. Due to the time difference between the 2021 census and the GP contract data, the measures of coverage constructed implicitly assume that the number of unpaid carers in each local authority has not changed since 2021. Future analyses could refine this by projecting expected numbers of unpaid carers over time to provide more comparable, time-appropriate denominators.
Other nationally representative surveys, such as Understanding Society, the GP Patient Survey, and the Health Survey for England estimate much higher proportions of the population (16–19%) as unpaid carers. As such, our census-based denominator in this analysis may be an underestimate.
What more could be done to understand and improve identification of unpaid carers in GP data?
While our analysis compared the number of unpaid carers registered by GP practices with local authority-level measures from the 2021 census, numerous other comparisons may be made to alternative counts of unpaid carers. For instance, direct GP practice-level comparisons can be made with data from the GP Patient Survey, available through the Fingertips API.
Practice-level measures and comparisons may also highlight the differences we observed between local authorities, identifying practices or Primary Care Networks (PCNs) with high or low levels of carer recording. Uncovering the sources of difference between administrative areas could inform national and local policy regarding carer recording, laying the groundwork for improved identification and widened access to support.
Despite its limitations, the unpaid carers indicator in NHS England’s GP core contract data remains a promising new open source for assessing the degree to which carers are on the map. Carer identification matters: it is the first step towards the NHS and local authorities providing well-targeted support, and to fully recognising those who contribute so much to the UK’s health, economy, and wellbeing.
Special thanks to Caroline Fraser, Hannah Knight, and Charles Tallack for their contributions. We are part of the Networked Data Lab, situated in the Data Analytics team at the Health Foundation. Please do get in touch or check out our analytical resources on GitHub.
