Association between mirtazapine use and serious self-harm in people with depression: an active comparator cohort study using UK electronic health records

Background Studies report an increased risk of self-harm or suicide in people prescribed mirtazapine compared with other antidepressants. Objectives To compare the risk of serious self-harm in people prescribed mirtazapine versus other antidepressants as second-line treatments. Design and setting Cohort study using anonymised English primary care electronic health records, hospital admission data and mortality data with study window 1 January 2005 to 30 November 2018. Participants 24 516 people diagnosed with depression, aged 18–99 years, initially prescribed a selective serotonin reuptake inhibitor (SSRI) and then prescribed mirtazapine, a different SSRI, amitriptyline or venlafaxine. Main outcome measures Hospitalisation or death due to deliberate self-harm. Age–sex standardised rates were calculated and survival analyses were performed using inverse probability of treatment weighting to account for baseline covariates. Results Standardised rates of serious self-harm ranged from 3.8/1000 person-years (amitriptyline) to 14.1/1000 person-years (mirtazapine). After weighting, the risk of serious self-harm did not differ significantly between the mirtazapine group and the SSRI or venlafaxine groups (HRs (95% CI) 1.18 (0.84 to 1.65) and 0.85 (0.51 to 1.41) respectively). The risk was significantly higher in the mirtazapine than the amitriptyline group (3.04 (1.36 to 6.79)) but was attenuated after adjusting for dose. Conclusions There was no evidence for a difference in risk between mirtazapine and SSRIs or venlafaxine after accounting for baseline characteristics. The higher risk in the mirtazapine versus the amitriptyline group might reflect residual confounding if amitriptyline is avoided in people considered at risk of self-harm. Clinical implications Addressing baseline risk factors and careful monitoring might improve outcomes for people at risk of serious self-harm.


Antidepressant exposure
The start and stop dates of individual antidepressant prescriptions were defined using a published algorithm. 1 The start dates are the prescription dates unless they have been adjusted for any overlapping time. Prescription duration is not entered for most prescriptions and must be estimated from other information. The algorithm used includes 10 decision steps, which are shown below with the decisions used highlighted. qty = total quantity entered by GP for the prescribed product; ndd = derived numeric daily dose; numdays = number of treatment days; dose_duration = derived duration of prescription. All options that produce a missing value stay coded as missing unless otherwise stated. *For options 6d-e: If only one is stop available, use it; if two are available and equal, use that date; if two available and unequal (but within X days), use mean; if three are available and unequal, use mean of closest two if within X days. **Records with missing stop dates after step 7 are dropped.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)

Variable list and details
Covariates included demographic and lifestyle characteristics, comorbidities in the Charlson comorbidity index 2,3 and the QRisk score, 4 and prescriptions for other medicines.
Code lists Read code lists and drug lists are available on Zenodo (https://doi.org/10.5281/zenodo.4779024) and the ClinicalCodes repository (https://clinicalcodes.rss.mhs.man.ac.uk). New code lists were developed if a list was not available in the ClinicalCodes repository, 5 the CALIBER phenotyping resource, 6 or individual papers. To develop a new list, an initial list of terms was extracted based on key word searches and then screened by clinicians in the research team.
The depression code list was based on existing published lists 7,8 and the Quality and Outcomes Framework (QOF) business rules code release 9 .
Demographic and lifestyle characteristics Age in index year (based on year of birth), sex, ethnicity (using the England and Wales 2001 Census broad ethnicity categories -Asian or Asian British, Black or Black British, Mixed, Chinese or other ethnic group, White -most recent code prior to index date) 10 , socio-economic status (SES, quintile of Townsend Score) 11 , and practice region. Ethnicity was defined using Read-coded information in the primary care record, supplemented with available secondary care information if missing. Body mass index (BMI) was defined using recorded weight and height measurements. The most recent BMI, smoking status (never/ former/ current) 12 , and alcohol use status (non-/ former/ occasional/ moderate/ heavy drinker) 13 on or prior to index date were defined.
Prescription medicine Use of the following drugs were considered present if prescribed on or less than 6 months before index date: opioids, glucocorticoids, non-steroidal anti-inflammatory drugs (NSAIDs), other analgesics, statins, antipsychotics, anxiolytics, hypnotic agents.

Comorbidities
The following comorbidities were considered present if recorded in the primary care data on or before the index date: abdominal pain, inflammatory bowel disease, indigestion, liver disease, obesity, pancreatitis, peptic ulcer disease, renal disease, anaemia, atrial fibrillation, angina, cerebrovascular disease, congestive heart failure, diabetes mellitus, hypertension, myocardial infarction, 14 peripheral vascular disease, venous thromboembolism, alcohol misuse, anxiety, 15 contact with mental health services, eating disorder, insomnia, 16 intellectual disability, personality disorder, self-harm, substance misuse, appetite loss, living in a care home, hemiplegia, leg ulcer, palliative/end-of-life care, poor mobility, unexplained/unintentional weight loss, unplanned hospital admission, asthma, chronic obstructive pulmonary disease, dyspnoea, 17 sleep apnoea, acquired immunodeficiency syndrome (AIDS), cancer, recent cancer (past year), metastatic tumour, dementia, epilepsy, 18 fibromyalgia and generalised pain, Huntington's disease, 19 migraine, multiple sclerosis, neuropathic pain, Parkinson's disease, and rheumatological disease. An indicator of a hospital record for intentional self-harm was also defined.
Other variables Depression severity was defined as having a coded record of severe depression or depression with psychosis, scoring 15 or above on the Patient Health Questionnaire-9 (PHQ-9) scale, or scoring 16 or above on the Hospital Anxiety and Depression (HAD) scale. An indicator of having a record of completing a depression scale was defined. Other variables were: most recent dose of the first antidepressant on/prior to index date, current dose of first antidepressant at index date (including 0), whether or not the first antidepressant was still being prescribed at index date, the time between starting the first and second antidepressants, the first SSRI prescribed, and the calendar year of index date.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)

Propensity score model
Variables associated with the exposure and outcome and variables associated with the outcome only (in univariate or age-sex adjusted models) were used to estimate propensity scores. 20 Variables included in the propensity score model: age, sex, age-squared, body mass index, smoking status, alcohol intake, antipsychotics, anxiolytics, hypnotic agents, statins, substance misuse disorder, baseline selfharm (primary care records), peptic ulcer disease, pancreatitis, contact with mental health services, liver disease (mild), intellectual disability, insomnia, indigestion, hypertension, diabetes, cancer, anxiety, asthma, appetite loss, alcohol misuse.
Sensitivity analyses including all defined variables were also performed. Some variables were excluded as the models were not converging. Excluded variables were: AIDs, moderate liver disease, and Huntington's disease.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)  1 (0-1) 1 (0-1) 1 (0-1) 1 (0-1) 1 (0-1.5) KW chi2 (3)   Median dose over follow-up was calculated for each person, excluding times of non-use (dose=0). The summary data are the median (interquartile range) of these values according to study treatment group. Doses are converted to defined daily dose (DDD). SSRI selective serotonin-reuptake inhibitor.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance

Current antidepressant dose at index (DDD), median (IQR)
0 (0-1) 1 (0-1) 1 (0-1) 1 (0-1) SSRI selective serotonin reuptake inhibitor; IQR interquartile range; n number; BMI body mass index; DDD defined daily dose. For columns 1) and 2), only a subset of people had an index date, hence certain results are omitted for these columns. Characteristics defined with respect to index date if not otherwise specified. a Counts and percentages do not include missing values. b Severe depression: Read code for severe depression or depression with psychosis, scoring 15 or above on the Patient Health Questionnaire-9 (PHQ-9) scale, or scoring 16 or above on the Hospital Anxiety and Depression (HAD) scale. c Exclusion criteria: aged 100 years or more, record for schizophrenia, record for bipolar disorder, or hospital record for self-harm at index.