Guidance

Psychotropic drugs and people with learning disabilities or autism: methods

Published 22 March 2019

1. Study design

This study was a descriptive quantitative one. It combined a repeated measures prevalence design with a review of trends. It studied prescribing prevalence and changes at quarterly points from January 2010 to December 2017.

The study was limited to England and hence, we only focussed on patients registered in active English practices during this period. The data source, as mentioned in the introduction section, was THIN which was accessed from the Public Health England (PHE) SQL server 2017 ^{[footnote 1]}.

In addition to this, we had a group of experts including clinicians, pharmacists, researchers and self-advocates who helped with interpreting this data and advised on caveats and nuances involved in prescribing of psychotropic drugs for people with learning disabilities, autism or both.

2. Study population

The study population comprised all patients registered with participating practices in the study period, where data met the following criteria in the THIN dataset:

• Medflag = R (Acceptable record)
• Patflag = A or C (Acceptable records)
• Regstat = 01, 02, 05 or 99 (patients having registration status of applied (01), permanent (02), transferred out (05) or died (99)

Our patients of interest were people with a diagnosis of learning disabilities, autism or both, or of a condition reliably predicting it (for example, Down’s syndrome), using Read codes version 2. Patients were included in the study if they had a qualifying diagnosis based on the Read codes list (table 3 in the code list) and had at least 1 day of registration between 1 January 2010 and 31 December 2017.

The patients of interest were divided into 2 categories:

• people with learning disabilities (with or without autism)
• people with autism but without learning disabilities

3. Visibility window

THIN data are taken from live records of patient care. The patient group they cover changes continuously as people register with or leave participating practices and as practices join or leave the scheme. We established a ‘visibility period’ for each patient. This started when they registered with a participating practice, or when their practice joined the data collection with data quality good enough for inclusion (whichever was the later). It ended when the patient left the practice or died, or when the most recent data were collected from it (whichever was the earlier).

4. Code lists

The following code lists were developed to calculate various measures. These lists were reviewed by the expert reference group, and wherever appropriate and feasible, their suggestions were incorporated. Wherever possible, we also compared our code lists with a similar on-going piece of work using CPRD GOLD data ^{[footnote 2]} and included all the codes that were relevant to our study.

4.1 Diagnosis codes list for identifying patients with learning disabilities, autism or both

People with learning disabilities, autism or both were identified using code lists developed for the QOF learning disabilities register, the experimental QOF autism register NICE Business rules document for identifying autism ^{[footnote 3]} and lists of appropriate additional diagnoses developed by Gloverand others [^4] and Sheehan and others ^{[footnote 5]}.

4.2 Drug list

Former BNF drug groups included in the study:

• 4.1.1: Hypnotics
• 4.1.2: Anxiolytics
• 4.2.1: Antipsychotic drugs
• 4.2.2: Antipsychotic depot injections
• 4.2.3: Antimanics
• 4.3: All antidepressants
• 4.4: CNS stimulants
• 4.8.1: Control of epilepsy

For analysing prevalence of prescribing rates, drug group 4.8.1 was divided into 2 subgroups: those with additional antimanic properties (carbamazepine, lamotrigine and all forms of valproate) and others. For combination drugs, eg Amitriptyline hydrochloride with Chlordiazepoxide, we replaced the prescription record with 2 records, one for each of the components.

4.3 Indication read codes list

Indications for various drug groups could be derived from either the Summary of Product Characteristics (SPC) ^{[footnote 6]} or the BNF ^{[footnote 7]}. We decided to use the BNF, as it is more commonly used by clinicians as an appropriate guide to prescribing. We did not look at the BNF for children as identifying whether individuals had recognised indications was substantially more complex and less likely to be fully recorded in general practitioners’ case notes. Hence, this list is only for the patient group- adults with learning disabilities.

A list of BNF mental health indications was constructed for each of the drug groups. It is important to note that this list was created for drug groups and not individual drugs. Hence, we included only those indications that could be considered relevant for the entire drug group. This resulted in 26 broad categories of indications (such as psychosis, anxiety, depression). Challenging behaviour was one of the categories. Read codes used to identify challenging behaviour were derived from a previous study ^{[footnote 5]}.

Table 1 shows the list of indication categories, for each of the main drug groups. Where available we used code lists from quality and outcomes framework business rules. We also searched for clinical terms including keywords relating to the indication. In some cases, where a term could be included in more than one indication, for example ‘other and unspecified manic-depressive psychoses’, it was included in both depression and bipolar disorder.

Table 1: Relevant mental health indications, by drug groups

Drug group	Indications
Antipsychotics	psychosis, schizophrenia, bipolar disorder and mania
Antidepressants	depression and anxiety
Anxiolytics	anxiety
Antiepileptics with antimanic properties	epilepsy and bipolar disorder
All antiepileptics	epilepsy

4.4 Maximum dose list for antipsychotics

We constructed a list of the BNF recommended maximum daily dosages for various antipsychotic drugs. In instances, where dosages varied across different indications for a drug, we considered the highest of these as the maximum recommended dose. Where the maximum dosage was quoted on a weekly or monthly basis, we calculated the daily maximum dosage by dividing the monthly (weekly*4) dosage by 20 ^{[footnote 8]}. This list was only for maximum dosages for adults. We did not look at dosages for children as these depend on child’s weight (especially in cases of stimulants and anti-epileptic drugs) and there was no plausible way to calculate this using the data source we had.

5. Statistical analysis

STOMP was launched in June 2016. Thus, we took quarter 2 of 2016 as the start of the STOMP programme. The aim of the study was to establish whether there has been any change in the trend in prescribing of psychotropic medication following the launch of STOMP.

We did this by fitting trend lines to the quarterly prevalence figures for prescribing levels for each drug group. We used ordinary least squares regression and compared the trends up to (pre-STOMP), and after (post-STOMP) the end of June 2016. We also performed sensitivity tests on the quarter chosen as the launch of STOMP, by choosing one quarter before or after 2016, quarter 2, to see whether this made any difference to the regression slopes. This made no substantial difference to the regression slopes. We compared the confidence intervals around the data points and calculated the standard error for the difference in proportions, to ascertain whether the rates at the start and the end of the study period were any different.

Separate models were fitted to the pre-STOMP and post-STOMP prevalence data, using quarter as an explanatory term. Intercept and slope terms were estimated for each of the fitted models and t-statistics used to judge whether or not the slope terms for each of the models differed significantly from zero using the 5% level of significance. 95% confidence intervals were also calculated for the estimated model parameters.

Statistical analysis was done in R using the linear models (lm) package. We have shown charts only where there was a significant change in trend between the 2 slopes. If not, we have simply quoted the modal parameters in text.

Wherever appropriate, we compared the current prescribing rate with the extrapolated rate of the pre-STOMP trend line. This allowed us to estimate the difference between the current level of prescribing and the likely level if the pre-STOMP trend had continued.

In PHE, THIN datasets are stored in Microsoft SQL server 2017 databases. We did the initial patient and event selection and a lot of the initial processing in Transact SQL. We did subsequent statistical analysis and produced charts and tables in R ^{[footnote 9]}.

All the rates/proportions calculated were based on average number of patients in the reported time-period, unless otherwise stated. The fairly large number of patient days actually accounted for a comparatively small number of patients. We considered this overstated the precision of the estimates since prescribing for individuals on consecutive days is not independent. Hence, in calculating the proportions and confidence intervals, we took the number of patients as our measure of sample size. Whenever we refer to the term ‘patients’ in this study, it is to be read as the average number of patients. Unless otherwise stated, confidence intervals were calculated using the Wilson method for standard errors of proportions using the Epitools package in R ^{[footnote 10]}. PheCharts package in R was used to draw the charts.

5.1 Estimating sample size

We estimated the sample size required for people with learning disabilities based on the QOF learning disabilities register and the prevalence of prescribing recorded from the previous study. We could not estimate the sample size for autistic people, as there is no national register to indicate the number of autistic people in England. The calculations were based on determining how big a group would need to be in order to detect differences between rates, before and after the launch of STOMP.

Estimates of likely numbers of cases are presented in Table 5 in the annex. Calculations were undertaken for estimated sample sizes of 5,000 to 9,000, assuming we require 80% power to identify a difference with 95% confidence. From a starting prevalence of 17%, the study would be able, with this power, to identify a change of 15% (up to 19.5% or down to 14.5%) assuming there were 6,000 or more patients. With 8,000 or more patients a change of 10% in prescribing prevalence (up to 18.7% or down to 15.3%) would be identified. These calculations were done using the pwr package in R ^{[footnote 11]}.

5.2 Calculating duration and daily dosages in THIN

The key elements required for analysing the prescribed amounts of drugs were the proposed daily dose and the date and duration of each prescription. This information was not always available in full. GPs write 2 types of information on prescriptions. Some is recorded in information systems in a coded form easily available for analysis. However, GPs often also write free text instructions. Some instructions are simple and can be easily coded (for example ‘Take 2 tablets 3 times a day’). Others are more detailed.

This is made more complex in prescriptions for people with learning disabilities because of the extensive use of liquid formulations of drugs. This expands the range of ways in which doses can be described. The THIN database provides coded translations for the more common instructions. These accounted for roughly 65% of prescriptions for our study. However, for the remaining 35%, including a disproportionate number of prescriptions of depot antipsychotic drugs, translations can only be obtained through special applications and at substantial additional cost.

Where the available information supported it, we calculated daily doses from the prescription instructions and the duration of the prescription from the quantity prescribed and the daily dose.

Where the required information was not all available but the patient had repeated prescriptions for the same drug, we assumed that the duration was the time between prescriptions and estimated the daily dose from this and the total quantity prescribed. We were unable to use data from single prescriptions where detailed prescribing data were not available. These made up 1.05% of prescriptions of the drugs of interest for study patients.

5.3 Translating findings to estimated national numbers

Wherever appropriate, we calculated a rough estimate of our key findings for the number of adults with learning disabilities in England. This was done by taking an average of the QOF learning disabilities register number for the last 3 years (2014 to 2015, 2015 to 2016 and 2016 to 2017) ^{[footnote 12] [footnote 13] [footnote 14]}. The definition for this QOF measure had changed from those aged 18 and over, to all ages in 2014 to 2015 ^{[footnote 12]} and hence, we could only calculate the average number based on the above 3 years. This is the recorded number of people with learning disabilities.

The diagnoses codes used to identify people with learning disabilities were far more in our study than the ones used in the QOF register and hence, we could not directly estimate the numbers nationally. Hence, we developed a correction factor which was the proportion of all the learning disability codes used in our study to that of the QOF register codes. We applied this correction factor to the identified number of adults with learning disabilities in our study. Prescribing rates were then applied on this number to estimate the numbers nationally.

It should be noted that this was a very rough estimate and by no means should be regarded as a definitive number. This was only calculated to look at the scales of numbers nationally. There is no such similar register for autistic people and so we could not estimate these numbers, based on our study.

5.4 Calculating measures

Following an initial description of the patient groups involved, the analysis was divided into 5 broad measures.

5.4.1 Prescribing rates

For each drug group we calculated the proportion of patient days for which there was a prescription for a drug included in the group. We further calculated this proportion based on the presence or absence of indications for the drug group.

From an initial study of the data and consideration of the way drugs can legitimately be used both to treat current illness episodes and to prevent relapse, we concluded that it was not feasible to define discrete periods during which patients were considered to have conditions for which particular drugs are indicated. Instead, drugs were considered indicated for a patient if at any time they had a record of a condition for which the drug was recorded as indicated, based on our indications list.

On this basis, for each drug group, we classified exposed patients, as exposed with an indication or exposed without an indication. Wherever appropriate we calculated the percentages of people without an indication, who had challenging behaviour, but no record of an indication.

5.4.2 Patterns of prescribing

Following the prevalence of prescription, we analysed various possible patterns of prescribing psychotropic drugs.

Within-group polypharmacy

This measure calculated the percentage of exposure by the number of drugs prescribed from the same drug group, on a given day. In case of antipsychotics, depot form of the same drug, was considered as one drug eg: Flupentixol and Flupentixol decanoate, were counted as 1 drug and not 2. We also calculated the average number of drugs prescribed on a given day.

Between-group polypharmacy

In addition to identifying number of drugs prescribed within a drug group, we also calculated the number of drug groups prescribed on a given day. Moreover, we calculated the proportions of exposed patients in each of the drug groups, who were exposed to each of the other psychotropic drug groups studied.

Inceptions and terminations of episodes

This measure calculated the rate of new episodes of prescriptions (inceptions) of drug groups or terminated episodes of prescriptions (terminations) of drug groups. For both these counts, we used a handover interval of 56 days. So, a prescription for a drug counts as inception for the drug group if there was no prescription of drug in the same group which would have reached (date and duration) later than 56 days before it. For terminations, it was counted as a termination if there was no prescription of drug in the same group for 56 days following the current date.

We included the following 2 criteria for person years at risk of inception:

• an inception should be more than 90 days after the patient’s visibility start date. This was to avoid false counting inceptions due to patients joining a new practice
• patients should not be prescribed the drug group of interest in at least 56 days preceding the inception date

For terminations, patients should be prescribed the drug group of interest and have no succeeding prescriptions for at least 56 days from the termination date.

With the above mentioned criteria, it was possible for a patient to have more than one inception or termination for a drug group, as long as they met the criteria mentioned. Both the rates of inceptions and terminations were calculated as rate per 10,000 person years and we used the Byar confidence intervals for rates to calculate the confidence around these rates.

High- dose antipsychotics

The last measure in the pattern of prescribing was calculating the proportion of adults who were prescribed antipsychotics in excess of the BNF (BNF for adults) recommended limits.

We analysed this measure in the following 2 ways:

• prevalence of antipsychotic prescription where at least one antipsychotic drug was prescribed in excess of the BNF recommended maximum dose for that drug
• prevalence of antipsychotic prescription where the combined total antipsychotic dose exceeded the BNF maximum - using the Royal College of Psychiatry ^{[footnote 15]} recommended approach of expressing the dose for each drug as a proportion of its recommended maximum and taking the total of these figures

6. Presentation of results

For each measure, we calculated the overall rates across the entire study period (1 January 2010 to 31 December 2017), followed by trends and finally the statistical trend tests. For all the trends analysis, data were analysed by quarters (except inceptions and terminations, which were calculated annually) and confidence intervals were calculated for each data point.

The trends were represented by the following patient and age groups:

• learning disabilities (with or without autism): 18 to 24, 25 to 44, 45 to 64 and 65 and over
• autism: 0 to 09, 10 to 17 and 18 to 24

Study quarter mentioned in the charts or text starts with the year followed by the quarter. For example, quarter 2 of 2016, is referred to as 2016- 2. Wherever appropriate, we quoted the confidence intervals in brackets for the rates or proportions calculated. Trend charts with trend lines, had 3 vertical lines going across them.

These lines referred to the following 3 events:

1. 2011: 2: The Panorama programme ‘Undercover Care’ about Winterbourne View
2. 2012: 4: Transforming Care was published
3. 2016: 2: Launch of STOMP

Following the trend charts, the results of the statistical trends tests were mentioned.

7. Citation

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

8. References

1. SQL Server Management Studio 2017. (2018). Seattle: Microsoft. ↩

2. Quality improvement reports for GPs. CPRD: UK data driving real-world evidence (2018). Available at: www.cprd.com/generalpractitioner/QualityImprovementProject.asp [Accessed 27 Nov. 2018] ↩

3. Primary Care Domain Specification Development Service. Business Rules for QoF Pilot 11 (2016/17): Autistic Spectrum. 0.1 Version. NHS Digital; (2016) ↩

4. Sheehan R, Hassiotis A, Walters K, et al. 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] ↩ ↩²

5. Mhra.gov.uk. (2018). Medicines Information: SPC & PILs. [online] Available at: www.mhra.gov.uk/spc-pil/ [Accessed 8 Oct. 2018] ↩

6. Joint Formulary Committee. British National Formulary. 73 ed. London: BMJ Group and Pharmaceutical Press (2017) ↩

7. Cohen, H. Casebook in clinical pharmacokinetics and drug dosing. New York: McGraw-Hill. (2015) ↩

8. R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. (2017) Available at: www.R-project.org/ [Accessed 27 Nov. 2018] ↩

9. Tomas J. Aragon. Epitools: Epidemiology Tools. R package version 0.5-9. (2017). Available at: https://CRAN.R-project.org/package=epitools [Accessed 27 Nov. 2018] ↩

10. Stephane Champely. pwr: Basic Functions for Power Analysis. R package version 1.2-2 (2018). Available at: https://CRAN.R-project.org/package=pwr [Accessed 27 Nov. 2018] ↩

11. . NHS Digital. Quality and Outcomes Framework - Prevalence, Achievements and Exceptions Report England, 2014-15 (2015). Available at https://digital.nhs.uk/data-and-information/publications/statistical/quality-and-outcomes-framework-achievement-prevalence-and-exceptions-data/quality-and-outcomes-framework-qof-2014-15 [Accessed 27 Nov. 2018] ↩ ↩²

12. . NHS Digital. Quality and Outcomes Framework 2015- 16: Recorded disease prevalence, achievements and exceptions Regional level (2016). Accessed at https://digital.nhs.uk/data-and-information/publications/statistical/quality-and-outcomes-framework-achievement-prevalence-and-exceptions-data/quality-and-outcomes-framework-qof-2015-16 [Accessed 27 Nov. 2018] ↩

13. NHS Digital. Quality and Outcomes Framework 2016- 17: Recorded disease prevalence, achievements and exceptions Regional level (2017). Accessed at http://digital.nhs.uk/pubs/qof1617 [Accessed 27 Nov. 2018] ↩

14. Royal College of Psychiatrists. Psychotropic drug prescribing for people with intellectual disability, mental health problems and/or behaviours that challenge: practice guidelines. Faculty of Psychiatry of Intellectual Disability. FR/ID/09. (2016). Available at www.rcpsych.ac.uk/pdf/FR_ID_09_for_website.pdf [Accessed 27 Nov. 2018] ↩