Identifying people with a learning disability: an advanced search for general practice

Background People with learning disabilities (LD) have poor physical and mental health when compared with the general population. They are also likely to find it more difficult than others to describe their symptoms adequately. It is therefore harder for healthcare workers to identify the health needs of those with learning disabilities, with the danger of some problems being left unrecognised. Practice registers record only a proportion of those who are eligible, making it difficult to target improvements in their health care. Aim To test a Read Code search supporting the identification of people with a mild-to-moderate learning disability who are not currently on the learning disability register. Design and setting An observational study in primary care in West Yorkshire. Method Read Code searches were created to identify individuals with a learning disability not on the LD register; they were field tested and further refined before testing in general practice. Results Diagnostic codes identified small numbers of individuals who should have been on the LD register. Functional and service use codes often created large numbers of false-positive results. The specific descriptive codes ‘Learning difficulties’ and ‘Referral to learning disability team’ needed follow-up review, and then identified some individuals with LD who were not on the register. Conclusion The Read Code search supported practices to populate their registers and was quick to run and review, making it a viable choice to support register revalidation. However, it did not find large numbers of people eligible for the LD register who were previously unidentified by their practice, suggesting that additional complementary methods are required to support practices to validate their registers.


INTRODUCTION
Individuals with a learning disability (LD) are estimated to constitute 2% of the adult population. 1 Only around a quarter of affected adults are on the learning disability register at their local general practice. 1,2 A learning disability is defined as a: '... significantly reduced ability to understand new or complex information, to learn new skills (impaired intelligence), with a reduced ability to cope independently (impaired social functioning); which started before adulthood, with a lasting effect on development'. 3 A learning disability is often classified by IQ scores, with a score of 50-70 defined as mild, 35-50 moderate, 20-35 severe, and <20 as profound. 4 However, IQ is not a favoured way to define a learning disability because it does not represent the functional abilities people have. 4 Individuals on the registers tend to be those with more significant/severe intellectual impairment who are known to specialist services. 1 Less is known about the lives and care needs of adults with milder learning disabilities, 5 even though those with milder learning disabilities could still benefit from flagging on primary care systems, because they too experience high rates of physical ill health and have a reduced life expectancy. 6 There is evidence that people with a learning disability are less likely to receive appropriate care for chronic health conditions, including diabetes, 7-9 and tend to have poor selfmanagement abilities, 7 and worse outcomes than the general population. 10 Accurate identification and recording of milder learning disability is the first step in correcting this situation.
In the UK, the 2001 government white paper Valuing People aimed to reduce health inequalities experienced by this group. 3 Part of its plan was to introduce a goal that all general practices should hold a register of patients with a learning disability. In 2006, the Disability Rights Commission recommended the introduction of annual health checks for people with learning disabilities in England. 11 In 2008/2009 a directed enhanced service (DES) introduced health checks for people with a learning disability who were known to social services. 12 This DES has continued to feature in each contract, and in 2017/2018 there is an increased payment per health check.
Basing health check provision on LD registers that are populated solely on information about people known to social services is inadequate, as it cannot be assumed that the local authority will be aware of all patients with learning disabilities who may qualify for a health check, and those people with a milder disability may not be in receipt of social service support. 13 There is, therefore, a benefit in practices compiling their own registers for completeness. 12 Attempts have been made to improve the comprehensiveness of learning disability registers using computerised searches of clinical databases in primary care, but they have focused on diagnostic coding alone. 1,13 This paper describes the creation of a new search strategy that goes beyond diagnostic

Research Abstract Background
People with learning disabilities (LD) have poor physical and mental health when compared with the general population. They are also likely to find it more difficult than others to describe their symptoms adequately. It is therefore harder for healthcare workers to identify the health needs of those with learning disabilities, with the danger of some problems being left unrecognised. Practice registers record only a proportion of those who are eligible, making it difficult to target improvements in their health care.

Aim
To test a Read Code search supporting the identification of people with a mild-to-moderate learning disability who are not currently on the learning disability register.

Design and setting
An observational study in primary care in West Yorkshire.

Method
Read Code searches were created to identify individuals with a learning disability not on the LD register; they were field tested and further refined before testing in general practice.

Results
Diagnostic codes identified small numbers of individuals who should have been on the LD register. Functional and service use codes often created large numbers of false-positive results. The specific descriptive codes 'Learning difficulties' and 'Referral to learning disability team' needed follow-up review, and then identified some individuals with LD who were not on the register. coding to identify patients who have an LD but are not on the practice LD register. The search was originally developed as part of a case-finding exercise for a feasibility randomised controlled trial (RCT), OK Diabetes, 14 which aimed to identify and recruit adults with a mild-to-moderate learning disability and type 2 diabetes who lived in the community.

METHOD Developing search terms
Searches were devised using standard diagnostic codes (Read Codes) to help practices identify individuals who were potentially eligible but not on a learning disability register. They wanted to create a list of patients whose associated Read Codes indicated they might have a learning disability, or had accessed learning disability services, and who did not have a code that would put them on the LD register. The list of codes was needed as the basis for a search that could be run on any general practice health record system.
The research team held a series of meetings with system specialists and learning disability clinical experts. Developing a candidate list of codes required several meetings, based upon previous studies and clinical experience. 1,13,15,16 Complexity was added in that even diagnostic codes can have multiple entries in Read Codes. For example, Down's syndrome can be coded in five ways, two using the term Down's syndrome and three describing Trisomy 21, which may cause problems for audit and research. 15 The team decided to focus on four domain categories for the creation of the Read Code search -diagnostic codes, functional codes, service access codes, and descriptive codes.
Diagnosis codes identify conditions often associated with a learning disability but that do not automatically add a person to the LD register -for example, Asperger's syndrome, Autistic disorder, or Down's syndrome. Including these codes in a search would ensure the need for a clinician to confirm or discount a learning disability.
Examples of functional codes included 'difficulty making considered choices' (XaA3B) and 'problems with learning' (ZV400). Service use codes included 'attendance at special school' (XM1Zd) and 'referral to learning disability team' (XaJmc).
Descriptive codes were those that described the person's life and status but fell short of a diagnosis and did not describe everyday functioning. They included 'developmental delay' (X76B7) and 'learning difficulties' (13Z4E).

Field testing the searches
Working with the data quality team at the Commissioning Support Unit (CSU; this is an external organisation that supports clinical commissioning groups with commissioning evidence and research), the usefulness of certain diagnostic codes was tested by running anonymised searches of data held at CSU level for three cities in West Yorkshire, for people who were not on the LD register but had a diagnostic code indicating a learning disability.
CSU data was searched for one city in West Yorkshire using a wider range of Read Codes.
On the basis of the CSU level searches, the full search was modified, and two general practices then ran the complete search on their systems and made the authors aware of the number of hits each code created.
The search was revised, and two further practices reviewed the cases identified by the searches and fed back the results. When finalised, the search was published into SystmOne (the most prevalent clinical computer system in these areas) under clinical reporting (a full list of codes can be found in Appendix 1). For other systems, it was condensed into a zip file to be e-mailed to practices.
As a final check of the practical utility of the search, practices were invited to use the search to identify participants for referral to the OK Diabetes RCT, and recorded how often referrals resulted from the search.

Diagnostic codes at the CSU level
In one city in West Yorkshire (population 500 000), there were 26 people with codes for a diagnosis of Down's syndrome who

How this fits in
Learning disability (LD) registers are a key way to identify patients who could benefit from reasonable adjustments to primary care services and/or health checks. LD registers are usually based upon people known to social services, which can mean that those who have a milder disability are not identified on a register but can suffer poor health as a result of their LD. It is not time efficient to explore patient records at length to identify one or two more people with a learning disability. A quick method is required to validate the membership of the LD register. The search created by the authors in this study proved popular as it took only four clicks in the practice system.
were not on the LD register at their practice. Table 1 shows the results when other diagnosis codes were searched for people not on the LD register in three cities in West Yorkshire with a combined population of 1.4 million. Table 2 shows the number of patient records identified by codes that were considered in one city in West Yorkshire (population 500 000), after excluding those individuals on a practice learning disability register.

Other Read Codes at the CSU level
It was agreed the first two terms in the table returned too many results and would damage the specificity of the search. 'Learning difficulties' was considered too important a term to remove, despite its high return. 'Lives in care home' and 'Impaired cognition' were retained for testing purposes. 'Adult safeguarding concern' was removed due to concerns that clinicians would feel uncomfortable reviewing and referring this patient. Table 3 shows the results of the search for potential LD when run at Practice 1. The practice found two people who they felt did have a learning disability and were not on their register. Practice 1 has a larger than CCG average list size of approximately 10 000.

Read Codes -practice level
It resides in the fourth most deprived centile, with a high Quality and Outcomes Framework (QOF) achievement. 17 The search was conducted by a nurse who leads on learning disabilities in the practice. This practice had updated its LD register 3 years prior to this search, using a list provided by social services. Those individuals who were marked as 'don't know' were flagged with a GP to explore the next time the individual visited the practice.
In the search for potential LD when run at Practice 2, the practice found two people who did have a learning disability and were not on their register, and three who required further review. Practice 2 has a list size of approximately 4000, and it resides in the seventh most deprived centile with a high QOF achievement. 17 The search was conducted by a GP who leads on learning disabilities in the practice. The search found 14 people who required GP review to exclude -three people with depression, nine people with impaired cognition, one with Asperger's syndrome, and one attending a voluntary agency. This practice regularly updates its LD register.

Case finding for RCT
In practice, the simplicity of the search strategy proved popular as it took only four 'clicks' to run all of the searches. Members of the research team gave support on the use of the searches if required, but it was rarely asked for.
In all, 65% (n = 145) of general practices in the study catchment areas were involved in recruitment to the RCT. Of the 325 participants referred, the most successful methods of identification in primary care was cross-referencing learning and diabetes QOF registers (n = 116, 36%). The Read Code searches identified an additional 65 individuals (20%).

DISCUSSION Summary
A population-level search identified relatively small numbers of people who were not on a LD register who had LD diagnosis codes on their records. The findings raised disconcerting questions because even obvious diagnostic codes like Down's syndrome identified unregistered patients. This suggests the value of a practice-level search strategy where patients can be identified and added to the register.
From the two practices with which the search was developed, the implication is that the Read Code search will identify about two patients per practice who should have been on a learning disability register, at the expense of reviewing about 10 patients  who might have other needs for additional support but did not have a learning disability. For a practice with a list size of 5000 (4000 adults), there should be about 80 adults with a learning disability and, typically, about 20-30 will be on the practice LD register. The search, which is easy and quick to use, will add about 10% to the register. Most useful was the code 'learning difficulties', which identified the most patients with a potential learning disability, but is coded to mean a variety of thingsmild learning disability or severe dyslexia, for example 1 -and will therefore include some false positives. 'Referral to learning disability service' captured some individuals who were not on the register and therefore proved useful in the study. Diagnostic codes returned few results but would not create many false positives.
Read Code searches alone may be unlikely to identify those individuals with a milder learning disability who continue to experience inequalities in health care and outcomes. Further ways of identifying the hidden majority of adults with milder learning disabilities are required, both to support their inclusion in research and to deliver reasonably adjusted healthcare services to this significantly disadvantaged group.

Strengths and limitations
The search is quick to run when published to a primary care clinical system such as SystmOne and can result in the identification of new patients. Searches at city population level have shown there is a need for further register revalidation as people are being missed from registers.
The new search does not identify substantial numbers from the hidden majority of unregistered people with a milder learning disability. It is unclear why, but the answer is likely to be because this group do not have a diagnosis and may not self-identify as having a learning disability, and because GPs are reluctant to assign relevant codes even if the patient's difficulties may be recognised informally when direct contact occurs.
A further limitation of this study is that those clinicians who were willing to review the results were often already interested in learning disabilities, and may have already done significant work on validating their registers. The addition of extra codes not only increased the chances of finding people 'missed' by previous searches, but also increases the workload for clinicians who need to review the results.

Comparison with existing literature
This study developed the work of previous searches to validate LD registers, 1,12 including a greater number of codes to increase the sensitivity of the search. The decision to include functional codes responded to previous findings about the variability of LD coding and aimed to capture those individuals without a formal disease diagnosis. 14 It did not attempt to identify prevalence of LD in the general population. Instead, it focused on finding people not identified by the LD register who may benefit from the reasonable adjustment of services.

Implications for research and practice
More effective ways of identifying the hidden majority of adults with milder learning disabilities to support their greater inclusion in research need to be found. Current methods are time consuming and require local knowledge on a practice-bypractice basis. 16 The transfer to SNOMED CT (a structured clinical vocabulary for use in an electronic health record) may support further refinement of searches, but this is currently uncertain. Research is also required to explore how decisions are made about who is included on the LD register.
The DES's focus on people known to the local authority has meant that the register is often thought of as only for people with a more severe LD. 18 However, there are people who have a mild or moderate LD who would benefit from, and have a legal right to, reasonable adjustments who are currently not flagged in the system. 19 This new search is quick, and will identify people currently not offered such adjusted service provision. Numbers 5 and 6 in bold show that the practice found two people who they felt did have a learning disability and were not on their register. LD = learning disability.

Funding
OK Diabetes was funded by the National Institute for Health Research Health Technology Assessment Programme (project number 10/102/03).

Ethical approval
Ethical approval was granted for the study by the Yorkshire and Humber Research Ethics Committee (Reference: 12/ YH/0304).

Provenance
Freely submitted; externally peer reviewed.

Competing interests
The authors have declared no competing interests.