Guidance

Psychotropic drugs and people with learning disabilities or autism: discussion

Published 22 March 2019

This study was designed to establish a simple method for monitoring the impact of the STOMP programme on prescribing of psychotropic medications to people with learning disabilities, autism or both.

The aim was to do this in a way which did not need expensive additional data collection and which could provide retrospective data to clarify any prior contextual trends.

It used data from a well-established and continuing research database consisting of anonymised records from a sample of practices covering (initially) about 5% of the population of the UK ^{[footnote 1]}. Autistic people and people with learning disabilities were identified from diagnoses recorded in practice notes. The primary measures were quarterly figures for the prevalence of prescribing of drugs belonging to 6 groups of interest. A number of additional measures were also studied including:

• whether patients had records of diagnoses constituting recognised indications for the drugs they were being prescribed
• rates of starting and stopping prescribing
• the extent of prescribing in doses above recommended maxima
• the extent to which multiple drugs from 1 group or drugs from multiple groups were used

It looked for evidence of changes in the trends in these measures following the start of the STOMP programme in June 2016.

1. Complexities in data processing

The data available to us came from routine note-keeping systems, not a research data collection environment. It would have been recorded by thousands of different clinical staff, all, to some extent with their own idiosyncratic ways of writing case-notes, in the course of live clinical practice.

Data were processed and made available to us by the company running the THIN data service. This inevitably introduced further filters as a result of essential processing to monitor and maintain data quality and to ensure secure anonymisation.

Whilst the system for collecting and processing data did not change during the period studied, the subjects of the data were in constant flux. Practices joined, and (more commonly) left THIN, new patients registered with, and left participating practices, and those that stayed, aged. So the group in each age band changed incrementally each year.

An important aspect of studies of this type is the selection of codes to be used for identifying subjects and clinical observations. Our approach is described in the methods section of this report. The code lists used are in the annex.

The most important deficiencies we identified in the data were the low levels of identification of learning disabilities in children and young people, and of autism in adults aged over 30. Both look seriously deficient, probably for different reasons.

In the case of children with learning disabilities, this is probably a feature of the fact that learning disabilities are commonly identified either by school special educational need co-ordinators, or, particularly if it arises from some syndromic cause, by paediatric services. In either case GPs may be reticent to record a learning disability in case-notes until they consider that this has been fully established. In the case of autism, this is complicated by the fact that the rates of identification in some age bands look higher than expected.

Diagnosed rates of autism for the start of the study period at ages 5 to 9, 10 to 14 and 15 to 19 were 1.6%, 1.6% and 0.9% respectively. At the end of the study period, corresponding rates were 1.7%, 2.3% and 2.2%. These are substantially higher than the rates quoted in the national adult psychiatric morbidity survey ^{[footnote 2]} though not statistically significantly so mainly because the psychiatric morbidity survey is based on a relatively small sample (for these purposes) and thus has wide confidence intervals.

The survey estimates from the 2007 and 2014 samples combine give age specific rates at 16 to 34, 35 to 54, 55 to 74 and 75 and older as 1.6% (0.8% to 3.3%), 0.1% (0.0% to 0.4%), 1.0% (0.4% to 2.1%) and 0.4% (0.1% to 1.7%). At older ages, diagnoses of autism almost disappear in our GP data.

The age specific data in the psychiatric morbidity survey shows a similarly sharp fall to the group now aged 35 to 54, but higher rates in older people which we do not see. From these gaps we concluded that data derived from GP notes are unlikely to be a reliable guide to the patterns of prescribing either for children with learning disabilities or for autistic adults over the age of 30.

There is also reason to be concerned about the rate of GP records of psychosis. The learning disability health and care dataset ^{[footnote 3]}, using a very similar code-set, reported that GPs identify a prevalence of severe mental illness of 8.8% in people with learning disabilities aged 18 and older.

The prevalence of psychoses identified in adults with learning disabilities was 9.1% in our study. This is more than double the upper estimate of the best recent study of the prevalence of psychosis in people with learning disabilities.

Cooper and her colleagues ^{[footnote 4]}, used the best available research methods to study patterns of mental illness in the population of people with learning disabilities known to GPs in Greater Glasgow. They found a population based rate for psychosis of 2.6% to 4.4% depending on the diagnostic criteria chosen. This suggests that GP diagnosis records of psychosis are over-inclusive. This could have led us to overestimate the proportion of people taking antipsychotics for whom a recognised indication was present.

Three aspects of the processing were potentially problematic. The difficulties described (in the previous 2 chapters) in calculating intended daily dosage and duration for about 35% of prescriptions meant that some of these data had to be estimated. The results obtained by this approach looked reasonable, but clearly could not be taken as precise measures. Initially we had intended to study the frequency of dose increases or decreases as well as starts and stops. We concluded this was not reliably possible from this data source.

Secondly, the rapid decline in numbers of practices providing data to the THIN system reduced the statistical precision of our findings in the second half of the study period. We understand this reflects practices switching from Vision practice software to competitor brands.

The inevitable consequence for the study findings was that with declining number of subjects, confidence intervals for the measures widened and statistical power fell. By the end of 2017, numbers had reached the lower end of what our preliminary power calculations suggested was necessary to identify levels of change in clinical practice that would be considered relevant for the programme.

Thus, whilst the findings reported here can be reasonably confidently asserted, the THIN data source itself looks unlikely to be usable for monitoring the programme going forwards. Fortunately, alternatives are available. Lastly, we had assumed that our lack of information about subjects’ months and dates of birth would not pose any difficulty. In the event smaller movements seen in some of our measures, tended to be obscured by the resulting saw-tooth patterns on quarterly rate charts.

2. Primary findings

With these provisos, the data appeared to give a reasonably plausible view of the state of, and trends in psychotropic prescribing for the groups studied.

There were very high prevalences of prescribing of antipsychotic, antidepressant and antiepileptic drugs to people with learning disabilities. These were similar to those seen in the preliminary PHE study ^{[footnote 5]}. Rates tended to rise with age. The rate of prescribing antidepressants increased throughout the period. Antipsychotic and antidepressant drugs were commonly given together. Prescribing of multiple drugs from the same class was much less common.

Fewer than 40% of adults with learning disabilities prescribed antipsychotics had a diagnosis of a psychotic or major affective disorder. Around 75% of people prescribed antidepressants had a record of depression or anxiety. Antipsychotics were rarely prescribed for autistic children under 10 (0.5%), but 2.5% of autistic young people aged 10 to17 and 5% of those aged 18 to 24 were prescribed these drugs. Antidepressant prescribing to autistic children was also very rare under age 10, but seen in rising numbers of autistic people aged 10 to17 (from 2% to 4%) and 18 to 24 (12% to 13%).

In the post-STOMP period, there were favourable changes in trend for overall prescribing prevalence for antipsychotics, antidepressants and anxiolytics for adults with learning disabilities. The only drug group for which the trend in prescribing rates for autistic children and young people changed was hypnotics for which the initial increasing trend accelerated.

Among the more detailed measures of prescribing quality we studied, for adults with learning disabilities there were improvements in the post-STOMP period in the trend in prescribing of antidepressants without indications, prescribing of more than 1 antidepressant and prescribing of multiple drug groups. Rather surprisingly, the trend towards reduction in prescribing of antipsychotics for this group seemed to be largely confined to people with (rather than without) a diagnosis of psychosis.

The only one of these wider measures to show any change in trend for autistic children and young people was within-group polypharmacy for antipsychotics, where again the initially increasing trend became steeper.

3.6% of the adults with learning disabilities who were being prescribed antipsychotics were being prescribed a combined dose for this drug group exceeding recommended limits, 1.8% were being prescribed a dose of a single drug higher than the recommended maximum. There was no apparent trend in high-dose prescribing.

2.1 Comparison with other studies

Two studies have examined similar issues in the current or recent English context, for people with learning disabilities. The earlier Public Health England study, using Clinical Practice Research Dataset data (CPRD-Gold), reported similar treatment prevalence figures for most drugs groups ^{[footnote 5]}. Exceptions were antidepressants (earlier study 14.1%, present study 18.3% - this was consistent with the pre-STOMP trend we found) and anxiolytics (earlier study 4.4%, present study 9.6% - in this case we found no pre-STOMP trend).

Some figures in this study for within-group polypharmacy differed from those in the earlier study. The earlier study found that 22.5% of adults prescribed an antipsychotic were prescribed more than 1; we found a falling trend in this with our quarterly figures falling from 12% to 8%. The present study also found lower rates of within-group polypharmacy for antidepressants (6.3% vs 10.8%), and antiepileptics (38.8% vs 43.3%).

Sheehan and his colleagues at University College London ^{[footnote 6]} reported the incidence of new treatments with antipsychotics in people with learning disabilities over a fifteen year period from 1999 to 2013. Their findings were not directly comparable with ours as they measured rates of first ever treatments with antipsychotics for individuals whereas we measured new treatment episodes, taking 56 days non-exposure as the marker of the end of a treatment period.

Predictably the rate we found (251 new treatments per 10,000 currently unexposed people per year) was numerically higher than the rate they found (132 first ever treatments per 10,000 never-exposed people per year). Their 1 directly comparable observation was of the proportion of people taking antipsychotics for whom there was no record of a clinical condition recognised as an indication for drugs of this class. They found this figure to be 71%, slightly higher than the 62.9% we found across the whole period we studied. Our study looked at practice about a decade later than theirs. However, assuming that both studies accurately reflected practice at the time, it is not possible to say whether this reflects a change in prescribing or in diagnostic recording.

Our study showed that 14% of autistic children and young adults, were prescribed drugs from at least 1 psychotropic drug group studied. PHE’s previous study recorded a similar prescribing rate (6.9% of people aged 0 to 18 and 17.8% of people aged 18 to 24 were prescribed at least 1 psychotropic drug studied).

Murray et al ^{[footnote 7]} used the same dataset as ours to study prescribing in people aged 0 to 24 with autism spectrum disorder (ASD) between 1992 and 2008. They found that psychotropic drugs were prescribed to 29% of patients which was substantially higher than the 14% reported in our study. However, the drug groups in Murray et al’s study were wider than ours, notably also including beta blockers.

2.2 Impact of STOMP

The first thing that should be stated clearly is that we cannot say whether any of the changes in prescribing trends between the pre and post-STOMP periods occurred as a result of the STOMP programme or merely at the same time.

To explore this, we would require more detailed analysis than is possible with this type of sample data. It would require a much larger dataset in which it was possible to identify local areas and to try and correlate the timing of changes with the timing of local STOMP activity. Our principal focus was to look for changes in prescribing trends occurring in the later part of the period studied which at least possibly could have been results of the STOMP programme.

The measures we were able to implement that would show evidence of resulting changes are the quarterly prescribing prevalence for each drug, the prevalence for people without a recognised indication, prescribing of multiple drugs and the rate of starting and stopping.

One feature of all these measures is that they do not show any change for an individual until their prescriptions for a drug actually stop. This could realistically take a year from the initial health check where the issue is first raised. This makes the trend measures relatively insensitive. We had hoped to be able to use dose reductions as measure likely to be sensitive to earlier changes, but the problems with dose calculation made this unworkable.

We were asked to estimate the national impact of the trend changes we found, assuming the practice recorded in our data set was representative of the whole of England. Our approach to doing this is described in methods chapter. It should be noted that the numbers quoted are very rough estimates and not the actual numbers. In this case we estimated the differences in numbers by the end of 2017, 18 months after the start of the programme, and those derived from the natural continuation of the trends up to June 2016.

Table 1 shows these estimates for adults with learning disabilities. We have predicted these estimates only for those measures that showed a change in trend, following the launch of STOMP or where there were enough data points post-STOMP to draw any plausible estimates.

Table 1: Rough estimates of the differences in numbers by the end of 2017 and those derived from the natural continuation of the trend up to June 2016 (projected), for adults with learning disabilities for various measures, in England.

Measure	Estimated numbers by the end of 2017 compared to those projected
Overall prevalence
Antipsychotics	~500 fewer
Antidepressants	~300 fewer
Anxiolytics	~2,000 fewer
Polypharmacy
More than 1 antidepressant	~400 fewer
More than 1 drug group	~2,500 fewer

For autistic children and young people, fewer changes were seen. Those that were seen, ran counter to the aims of STOMP. This could be because a greater proportion of the prescribing for children and young people is done by paediatricians or child psychiatrists and thus, not visible in general practice data. Alternatively, as children were a later focus for the STOMP programme, it may simply be too early to see the impact in this area.

3. Interpreting the findings

The study did not find that there has been a substantial decrease in the prevalence of antipsychotic and antidepressant treatments prescribed to people with learning disabilities, autism or both. This is not surprising, mainly because it is probably too soon to see a substantial impact on this final end point.

The STOMP programme provides information and good practice guidance to a range of people involved in the care of people with learning disabilities.

Targets include:

• GPs and primary care colleagues
• informal carers
• mental health service staff
• care staff working in adult social care
• people with learning disabilities themselves

The aim is that anyone with learning disabilities, autism or both who is currently prescribed psychotropic drugs should have their need for these reviewed.

A likely possibility is that this review would first be suggested in the context of an annual health check. If an individual was found to be having a psychotropic for which primary care staff could not see a current justification, it is likely that a psychiatric assessment would be arranged to advise whether the medication should be withdrawn, and if so over what time period. Discussions with care staff would be likely to follow, then a decision to reduce the dose and stop, if possible over several months.

Prescribing decisions affect many care staff and decisions to withdraw drugs taken without involvement of the care staff involved with an individual are likely to raise anxieties which may well ultimately frustrate the process.

The ‘Call to Action’ methodology was successfully applied to a similar drug related issue ^{[footnote 8]} to reduce the use of antipsychotic drugs to manage problem behaviours in dementia. However, during this programme, there were a number of financial incentives provided to GPs, primary care trusts and mental health trusts.

As yet the STOMP programme has not used specific financial incentives. This should be redundant for adults with learning disabilities because review of current medication is supposed to form an integral part of annual health checks for which specific payments are available ^{[footnote 9]}.

The Royal College of General Practitioners (RCGP) is a signatory of the STOMP pledge and has agreed specific advice in support of its goals. It may be, however, that GPs need more support to undertake a psychotropic drug review and to embark on a drug reduction programme.

A recent study in Netherlands ^{[footnote 10]} investigating the preparedness of physicians in undertaking drug reviews and reduction programmes, reported that almost 60% of the physicians felt that they needed further education and support in doing so. One option is to train a cohort of clinical pharmacists or nurses specifically to assist GPs in undertaking such an endeavour. Moreover, prescribing for patients with learning disabilities, autism or both is often initiated in specialist care. Therefore, GPs might be reluctant to embark on changing the medicines without significant support from specialists.

Finally, it is clear that reduction of psychotropic drugs is more likely to succeed if there is clarity about the nature of the behaviours that caused the prescribing in the first place and alternative approaches in place to manage the behaviour should it re-emerge ^{[footnote 11]}. The most widely accepted approach is Positive Behavioural Support (PBS).

The widely acclaimed STOMP clinic in Sunderland uses the combined resources of a specialist clinical pharmacist working together with a positive behavioural support team ^{[footnote 11]}. Many organisations are committed to training their staff in such approaches but GPs may be reluctant to reduce psychotropic drugs without such supports in place.

Further research about attitudes of GPs to their ability to withdraw psychotropic medication especially antipsychotics without input from specialists is recommended. Similarly, evidence of availability of PBS training to staff would also be interesting.

Thus, programmes like this, take time to have any measurable impact, given the ground work that obviously needs to be done and the setting factors that need to be addressed before professionals, people with learning disabilities, autism or both and their families and carers are likely to be confident in starting medication reduction or discontinuation.

This study shows that changes to prescribing have started to steer in the right direction, at least for adults with learning disabilities. With the proviso about the need to switch to a data source which does not have a rapidly declining participant base, the monitoring mechanism developed looks likely to provide a reasonably good picture of national developments.

4. Strengths and limitations

The strength of this study is in establishing a comprehensive methodology that addresses many of the concerns and complexities in analysing THIN data. The study was successful in setting up a mechanism for monitoring trends in the prescribing of psychotropic drugs. It was obviously not able to report on people with learning disabilities or autistic people who do not have this diagnosis recorded in GP case notes.

The code lists were developed with advice and opinion from experts in the field and were also compared and matched to another similar study carried out using CPRD GOLD data. These lists are made available in the annex for others seeking to study this, or related topics.

The data source is a nationally representative sample of patients in the country. However, numbers are too small for regional or more local analysis. In any case, the area of residence of subjects is not available to researchers to ensure anonymity. Sample numbers were too small to report on autistic people aged over 24 or children with learning disabilities. These are major omissions. The data on which we are reporting does not cover patients who are temporarily registered with a GP practice. We have assumed that these would be proportionately negligible.

The data source cannot tell us about patient’s compliance in taking the drugs prescribed. Whilst it potentially provides information about some side effects of the drugs studied, such as weight gain, diabetes, and long term neurological disorders, we did not have time to explore these.

For comparing prescribing trends before and after the launch of the STOMP programme we used ordinary least squares regression to identify trend slopes in the 2 periods. The statistical confidence of these analyses was limited by the small number of data points in the post-STOMP period and the degree of scatter around best-fit trend lines. A fuller statistical exploration would benefit from a time series to allow for the possible presence of interdependencies in prevalence estimates at different time points.

5. Recommendations for future work

This was the first study in setting up a monitoring system to evaluate the impact of the STOMP programme. The study demonstrated that a database of GP transactions such as THIN can be effective in monitoring trends and patterns of psychotropic prescribing for people with learning disabilities, autism or both.

However, the numbers of children with learning disabilities and autistic adults identified in our dataset were too small for realistic trend analyses. The fall in numbers in practices participating in the THIN database also raises doubt about the viability of this specific source for future work. Other research datasets focussing on other practice information systems are available. Future work should focus on one of these. This could get around the problem with dwindling subject numbers but it is unlikely to improve the problem with failure to identify children with learning disabilities or middle aged or older autistic adults.

The programme code developed for the work is likely to require some modification for use with another database; the code lists developed should be transferrable, although the switch of NHS primary care data coding systems to SNOMED from April 2018 will require some translation.

6. Conclusion

This study was successful in setting up a mechanism to monitor the impact of the NHS England STOMP programme and analysing prescribing trends. The statistical models and analytic strategies set up in this study are likely to give a good indication of national trends in prescribing rates and patterns. In the longer term this type of work is likely to be most useful in identifying detail about prescribing patterns. However, for the NHS it is likely also to be important to have a more comprehensive national set of measures using data drawn from all general practices. This would inevitably be more limited in scope but could be aggregated to provide an authoritative measure of the national trend whilst also providing reliable and compatible information about local progress.

The conclusions from the current model were limited due to the lower number of data points in the post-STOMP period. However, in theory, if the number of data points increase and the current sample size meets the sample size calculation, the current model should be able to monitor future trends. Even with the fewer data points, the current study showed that the changes were beginning to happen in the desired direction for adults with learning disabilities, following the launch of STOMP.

7. Acknowledgements

Expert advisory group

Dr Ashok Roy, Carol Roberts, Dr Duncan Edwards, Dr Dominic Slowie, Dr Heather McAlister, Dr Ian Davidson, Kevin Elliott, Dr Kirsten Lamb, Dr Mark Lovell, Peter Pratt, Dr Rory Sheehan, Vivien Cooper.

Quality assurance of data and analyses

Rachel Roche, Public Health England

Statistical advice

Mark Dancox, Public Health England

8. Citation

Cite as: ‘Mehta H. and Glover G. Psychotropic drugs and people with learning disabilities or autism, 2019. Public Health England’

9. References

1. Quintile IMS. THIN Data Guide for Researchers. 1701:1- 105 (2017) ↩

2. NHS Digital. Adult Psychiatric Morbidity Survey: Survey of Mental Health and Wellbeing, England, 2014. Official statistics, National statistics, Survey (2016). Accessed at https://digital.nhs.uk/data-and-information/publications/statistical/adult-psychiatric-morbidity-survey/adult-psychiatric-morbidity-survey-survey-of-mental-health-and-wellbeing-england-2014 [Accessed 27 Nov. 2018] ↩

3. NHS Digital. Health and Care of People with Learning Disabilities: Experimental Statistics: 2016 to 2017 (2017). Available at https://digital.nhs.uk/data-and-information/publications/statistical/health-and-care-of-people-with-learning-disabilities/health-and-care-of-people-with-learning-disabilities-experimental-statistics-2016-to-2017 [Accessed 27 Nov. 2018] ↩

4. Cooper and others. Psychosis and adults with intellectual disabilities. Prevalence, incidence, and related factors. Social Psychiatry and Psychiatric Epidemiology (2007) 42(7) pp530 to 536. Available at https://link.springer.com/article/10.1007%2Fs00127-007-0197-9 [Accessed 27 Nov 2018.] ↩

5. Glover G., Williams R., Branford, D., Avery, R., Chauhan, U., Hoghton, M. and Bernard, S. Prescribing of psychotropic drugs to people with learning disabilities and/or autism by general practitioners in England. Public Health England. (2015). Available at http://webarchive.nationalarchives.gov.uk/20160704152031 [Accessed 27 Nov 2018.] ↩ ↩²

6. Sheehan R, Hassiotis A, Walters K, and others. Mental illness, challenging behaviour, and psychotropic drug prescribing in people with intellectual disability: UK population based cohort study. BMJ. 351: h4326 (2015). Available at www.bmj.com/content/351/bmj.h4326 [Accessed 27 Nov. 2018] ↩

7. Murray and others. Pharmacological treatments prescribed to people with autism spectrum disorder (ASD) in primary health care. Psychopharmacology 231(6):1011 to 1021 (2014). ↩

8. Health and Social Care Information Centre (HSCIC). “National Dementia & Antipsychotic prescribing audit 2012” (2012). Available at: www.rcpsych.ac.uk/pdf/nati-deme-anti-pres-audi-summ-rep.pdf [Accessed 27 Nov 2018.] ↩

9. NHS England. Annual health checks (2018). Available at www.england.nhs.uk/learning-disabilities/improving-health/annual-health-checks/ [Accessed 27 Nov 2018.] ↩

10. Gerda de Kuijper Annette A. J. van der Putten. Knowledge and expectations of direct support professionals towards effects of psychotropic drug use in people with intellectual disabilities. J App Res Intellect Disabil Volume30, IssueS1 (2017). Available at www.ncbi.nlm.nih.gov/pubmed/28467003 [Accessed 27 Nov. 2018] ↩

11. David Branford, Anne Webster, Carl Shaw, David Gerrard, Nigget Saleem, Stopping over-medication of people with an intellectual disability, autism or both (STOMP) in England part 2 – the story so far, Advances in Mental Health and Intellectual Disabilities (2018). At https://doi.org/10.1108/AMHID-02-2018-0005 ↩ ↩²