Antidepressant drug prescription and incidence of COVID-19 in mental health outpatients: a retrospective cohort study

We conducted an observational, retrospective, matched cohort study of individuals admitted to the 4 inpatient care units (Bethlem Royal Hospital, Lambeth Hospital, Lewisham Hospital, and Maudsley Hospital) affiliated with the South London and Maudsley (SLaM) NHS Foundation Trust during the 1st wave of the COVID-19 pandemic of 2020. SLaM provides near-monopoly comprehensive mental health services to a geographic catchment of 1.3 million residents in four boroughs of south London. SLaM has used electronic health records across all its services since 2006 and its Clinical Record Interactive Search (CRIS) platform was set up in 2008 to provide researcher access to de-identified data from these records within a robust governance infrastructure. CRIS has been subsequently developed through a range of data linkages and natural language processing (NLP) algorithms [25], and the platform has provided data for over 250 peer-reviewed publications to date. Using CRIS, we extracted data on admissions of patients aged 18 years or older to SLaM inpatient facilities between 1 April and 31 December 2020. PCR or antigen tests for COVID-19 were routinely performed at admission and during the inpatient stay over that period. The criterion for inclusion in this study was the conclusive positive or negative COVID-19 test result(s) during inpatient stay in a SLaM inpatient unit. Characteristics of the study cohort are presented in Table 1.

For exposure, we used a natural language processing (NLP) algorithm to identify the medications mentioned in the patient’s record in a 6-month window before or after referral, which provides a validated proxy measure for drug receipt [26]. The list of medications can be found in Table 2. We established use of the following medication classes: atypical (not mirtazapine), monoamine oxidase inhibitors (MAOI), mirtazapine, serotonin and norepinephrine reuptake inhibitors (SNRI), selective serotonin reuptake inhibitors (SSRI), tricyclic antidepressants (TCA), and any antidepressants mentioned above within 31, 62, or 90 days preceding the index hospital admission.

The primary outcome measure was the result of the laboratory test for COVID-19 (antigen or PCR). Any incidence of a positive COVID-19 test result during the inpatient stay was categorised as ‘positive’. For categorisation as ‘negative,’ all COVID test results during inpatient stay were required to be negative.

We ascertained age at the time of admission, gender, and ethnicity (dichotomised to white and non-white) as recorded at the time of hospital admission. We identified the diagnosis given closest to hospital admission in structured fields. According to the International Statistical Classification of Diseases and Related Health Problems (WHO ICD-10) [27] criteria, we established the following diagnosis groups:

Mental and physical health problems as well as functional difficulties were scored using the Health of the Nation Outcome Scales (HoNOS) [28]. This is a routinely used measure in British mental health services and the most recent scores were extracted at the time of the index admission. Each subscale is rated on a scale ranging from 0 (no problem) to 4 (severe or very severe problem); to simplify interpretation, we dichotomised the scores to ‘minor or no problems’ (scores 0 or 1) and ‘mild to severe problems’ (scores 2 to 4).

Initially, χ² tests and Z-score statistics were used to analyse COVID-19 test results for each covariate. Logistic regression models were then assembled to quantify odds ratios (ORs) for the associations between antidepressant medication receipt and incidence of COVID-19 positive test result, applying the above sub-categorisation of antidepressants and timing as secondary analyses. 95% confidence intervals (CI) and P-values for ORs were calculated, and P-values < 0.05 were considered statistically significant. Initially unadjusted logistic regression analyses were carried out, followed by adjustments for sociodemographic factors (gender, age, ethnicity) and then further adjustments for those significant HoNoS subscales and primary mental health diagnosis. Finally, primary analyses (only patients receiving any antidepressant medication) were stratified by primary mental health diagnosis measured using ICD-10 diagnosis at the time of index hospital admission. All statistical analyses were conducted with STATA version15 [29].

The study has a number of important limitations, largely due to its retrospective nature and focus on a cohort of mental health patients. It cannot demonstrate a causal relationship between medications and COVID-19 test results; indeed, the protective effect is consistent with either decreased infection rate or increased recovery rate from COVID-19.

The time period of the study was limited to the first wave of the COVID-19 pandemic, which presented novel adverse effects on mental health [40]; also, during this period, COVID-19 was associated with the original strain of SARS-CoV-2 rather than its subsequently documented variants. Owing to the numbers of participants and prevalence of multiple drug prescriptions, it was not possible to investigate the association between COVID-19 and individual drugs; compared to SSRIs, the number of participants receiving other drug classes was low (Table 1), and it was also not possible to assess older-generation drugs, including fluvoxamine. The study did not account for dosage regimen; also, it was not possible to determine medication adherence from the available data.

Finally, although the association of interest appeared to be present across a range of diagnostic groups, the possibility cannot be ruled out that antidepressant use may have been a marker of personal or behavioural factors conferring protection, e.g. compliance with societal restrictions and/or personal protection measures in place at the time. To address the above limitations and to further corroborate the findings of this study, randomised prospective clinical studies for selected ADs in the general population will be of essence.