Prior to the COVID-19 pandemic, antidepressant prescribing was increasing about 0.3% per month. We found that the pandemic did not significantly affect this trend; however, since March 2021 (ie, recovery period) there has been a notable decrease in antidepressant prescribing overall. When we analysed specific subgroups, we found a 15% increase in antidepressant initiation (new prescribing) in patients residing in care homes. We also found that prescriptions to those aged 0–19 and 20–29 years were lower than pre-COVID-19 trends. Prescribing to patients under 18 years is often shared with secondary care services.

A number of studies globally used ITSA to assess the impact of COVID-19 on mental health. Campitelli et al found a 1.43% increase in individuals dispensed antidepressants in nursing homes in Canada.19 In Israel, Frangou et al found COVID-19 was associated with an increase in antidepressant prescription fills.16 Wolfschlag et al found no change in antidepressant prescribing in one Swedish region although they note that Sweden did not experience a true ‘lockdown’, highlighting that the impact of COVID-19 restrictions may have been country specific.20

Using the UK’s Clinical Practice Research Datalink (CPRD) dataset, Mansfield et al found substantial reduction in depression, anxiety and self-harm primary care contacts in March 2020 that had not recovered by July 2020.21 We also saw antidepressant and new antidepressant prescribing were below expected trends in July 2020, but allowing more follow-up time, we found both had recovered by March 2021.

Also using CPRD, Taxiarchi et al found that prescribing for depression and anxiety decreased at the beginning of the restrictions period, yet they found this had recovered by the beginning of the recovery period.22 Carr et al performed an ITSA of mental illness and self-harm, including antidepressant prescribing in England,23 and found a 36.4% decrease in first antidepressant prescriptions in April 2020 that had recovered by September 2020. We saw a level shift of −29.8% for new prescriptions with restrictions, but we did not see new prescribing recover until later. Our ‘new prescribing’ outcome, however, includes those who had a previous prescription ≥2 years ago. This might contribute to our higher baseline rate of new prescriptions (3.3 vs 2.2 per 1000).

Macdonald et al used OpenSAFELY to assess antipsychotic prescribing to those in care homes, and with learning disability or autism. Comparing Q1 2019 to Q4 2021, they found decreased antipsychotic prescribing to those with learning disability or autism, and an increase in new prescriptions to those in care homes,24 in accordance with our findings for antidepressants.

Since 2019, NHS England has maintained indicators to monitor antidepressant prescribing to those with learning disability and we can roughly validate our numbers against theirs. Branford and Shankar25 found 10.3% annual prevalence (calculated as number of patients with a prescription in the last 6 months) of antidepressant prescribing in the general population and 20.7% in those with learning disability for NHS financial year 2020–2021. When we calculated prevalence in the last 6 months of NHS financial year 2021 with our data, we found prevalence rates of 12.2% and 22.8%, respectively. Small differences between analyses are normal and expected due to differences in underlying populations.

A key strength of this paper is its scale; using the OpenSAFELY platform we have been able to access raw, pseudonymised, single-event-level clinical events for >24 million patients in England, who are registered at NHS GP practices, that use TPP software. This allowed us to explore medication usage, diagnostic events and salient clinical and demographic information, including ethnicity, age and scores of deprivation.

There are, however, limitations to note. ITSA is a strong quasiexperimental study design that can help address confounding by using each population as its own control. But this uncontrolled ITSA cannot address competing risks. Greater awareness around inappropriate prescribing, for example, National Institute for Health and Care Excellence guidance that antidepressants should not be offered as first-line treatment for mild depression, is a competing factor that could have impacted prescribing at the same time as COVID-19. A future-controlled ITSA comparing those with severe versus non-severe depression could help contextualise this effect. Pragmatically, we assumed antidepressant prescriptions would be issued on a 4-week interval. If some patients are non-compliant or have longer prescribing intervals, their medications may not be issued every month. If the level of non-compliance changed over the course of our study or doctors changed their repeat prescribing practice, this could have impacted our results.

This research relies on accurate recording of diagnosis and clinical events within primary care records, but this is a limitation of all large Electronic Health Record database projects. Finally, this study describes rates of antidepressant prescribing, which may vary slightly from rates of dispensing or usage. However, within our sensitivity analysis, we have compared our findings with rates of antidepressant dispensing (OpenPrescribing data) and found the overall prescribing result to be similar.

In 2021, a national review by the chief pharmaceutical officer estimated that up to 10% of all medicines prescribed within the UK are no longer needed or not appropriate for continued use.26 This review highlighted a systemic problem of medicines issued without a documented indication. Our findings concur with this review, as we saw that up to one-third of patients issued an antidepressant do not have a diagnosis of depression or anxiety recorded within their notes, and this problem is even more prevalent in those with a diagnosis of learning disability. It is possible that GPs may have recorded other indications or recorded indications in free text.

In 2016, the STOMP initiative was launched whose4 aim was to reduce overprescribing of psychotropic medicines within the learning disability and autism populations. While it is positive to see that the pandemic did not trigger a large increase in antidepressant prescribing in those with learning disability or autism, we saw a significant increase in new antidepressant prescribing in patients who live in care homes, a particularly vulnerable group that includes those with learning disability. We need to continue to focus on these vulnerable populations to ensure that antidepressants are used appropriately.

We have shown we can use the OpenSAFELY platform as a tool to monitor the impact of directives at a comprehensive level within ‘at-risk’ patient populations across the UK. With appropriate permissions and where appropriate support can be obtained from relevant professional bodies, the OpenSAFELY platform is also technically capable of providing audit and feedback information about clinical practice, and changes in clinical practice, at single sites to support improvements in patient care.