The impact of the COVID-19 pandemic on self-harm and suicidal behaviour: a living systematic review

Introduction

The COVID-19 pandemic is causing widespread societal disruption and loss of life globally. By the end of June 2020 over 10 million people had been infected and over 500,000 had died (Worldometer, 2020). There are concerns about the impact of the pandemic on population mental health (Holmes et al., 2020). These stem from the impact of the virus itself on people infected, as well as front-line workers caring for them (Kisely et al., 2020), and on population mental health from the public health measures that have been implemented to minimise the spread of the virus – in particular physical distancing, leading to social isolation, disruption of businesses, services and education and threats to peoples’ livelihoods. Physical distancing measures have resulted in substantial rises in unemployment, falls in GDP and concerns that many nations will enter a prolonged period of deep economic recession.

There are concerns that suicide and self-harm rates may rise during and in the aftermath of the pandemic (Gunnell et al., 2020; Reger et al., 2020). Time-series modelling indicated that the 1918-20 Spanish Flu pandemic, which caused well over 20 million deaths worldwide, led to a modest rise in the national suicide rate in the USA (Johnson & Mueller, 2002; Wasserman, 1992). Likewise, there is evidence that suicide rates increased briefly amongst people aged over 65 years in Hong Kong during the 2003 SARS epidemic, predominantly amongst those with more severe physical illness and physical dependency (Cheung et al., 2008).

The current context is, however, very different from previous epidemics and pandemics. The 2003 SARS epidemic was restricted to relatively few countries. Furthermore, during the 100-year period since the 1918-20 influenza pandemic, global and national health systems have improved, international travel and the speed of communication of information (and disinformation) have increased, antibiotics are available to treat secondary infection, and national economies have become more inter-dependent. The availability of the internet and technological advancement has made it far easier for people to communicate and engage in home working and home schooling. However, there are marked social inequalities in relation to access to technology and ability to stay safe and continue to work, within and between countries. Public health policies and responses, and the degree of access to technology to facilitate online clinical assessments and treatments differ greatly between countries.

Key concerns in relation to suicide prevention during the pandemic include: uncertainty regarding how best to assess and support people with suicidal thoughts and behaviours, whilst maintaining physical distancing; people who have attempted suicide may not attend hospitals because they are worried about contracting COVID-19 or being a burden on the healthcare system at this time; diminished access to community-based support; exposure to traumatic experiences; and an economic recession may have an adverse impact on suicide rates (Chang et al., 2013; Stuckler et al., 2009). There have been increases in bereavement (with many being unusually complicated during the crisis), sales of alcohol (Finlay & Gilmore, 2020) and domestic violence (Mahase, 2020) – all risk factors for suicide (Turecki et al., 2019); the insensitive or irresponsible media reporting of suicide deaths associated with COVID-19 may be harmful; and in some countries access to highly lethal suicide methods such as firearms and pesticides may rise (Gunnell et al., 2020).

In the context of the COVID-19 pandemic there is likely to be a rapidly expanding research evidence base on its impact on suicide rates, and how best to mitigate such effects. It is therefore important that the best available knowledge is made rapidly available to policymakers, public health specialists and clinicians. To facilitate this, we are conducting a living systematic review focusing on suicide prevention in relation to COVID-19. Living systematic reviews are high-quality, up-to-date online summaries of research that are regularly updated, using efficient, often semi-automated, systems of production (Elliott et al., 2014). This paper reports the first set of findings from the review, based on relevant articles identified up to June 7^th 2020.

Aim

The overarching aim of the review is to identify and appraise any newly published evidence from around the world that assesses the impact of the COVID-19 pandemic on suicide deaths, suicidal behaviours, self-harm and suicidal thoughts, or that assesses the effectiveness of strategies to reduce the risk of suicide deaths, suicidal behaviours, self-harm and suicidal thoughts, resulting from the COVID-19 pandemic.

Methods

This living systematic review (Figure 1) follows published guidance for such reviews and for how expedited ‘living’ recommendations should be formed where relevant (Akl et al., 2017; Elliott et al., 2017). The review was prospectively registered (PROSPERO ID CRD42020183326; registered on 1 May 2020). An overview of our living review process is provided in Figure 1. A protocol (John et al., 2020a) was published in line with the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols guideline (Moher et al., 2015). Since publication of our protocol we have amended our methodology to: 1) search additionally the PsyArXiv and SocArXiv open access paper repositories; 2) include modelling studies within the scope of our review (e.g. to predict the likely impact of the pandemic on suicide rates); and 3) update our research questions to include adult self-neglect and parental neglect and fear of losing livelihood.

Figure 1. Workflow for updating the living systematic review review.

The process will be supported using automation technology and at three-monthly intervals the team will update the published version of the review. The latest and previous versions of this figure are available as extended data (John & Schmidt, 2020).

Eligibility criteria

Study participants may be adults or children of any ethnicities living in any country. Outcomes of interest are:

Studies must address one of the following research questions:

(i) What is the prevalence/incidence?

(ii) What is the comparative prevalence/incidence?

(iii) What are the effects of interventions?

(iv) What are the effects of other exposures?

Qualitative research

We include any qualitative research addressing perceptions or experiences around each outcome in relation to the COVID-19 pandemic (e.g. stigma of infection, isolation measures, complicated bereavement, media reporting, experience of delivering or receiving remote methods of self-harm/suicide risk assessment or provision of treatment; experience of seeking help for individuals in suicidal crisis); narratives provided for precipitating factors for each outcome.

No restrictions were placed on the types of study design eligible for inclusion, except for the exclusion of single-person case reports. Pre-prints were re-assessed at the time of publication and most current version included. There was no restriction on language of publication. We will draw on a combination of internet-based translation systems and network of colleagues to translate evidence in a language other than English.

Identification of eligible studies

We searched the following electronic databases: PubMed; Scopus; medRxiv, bioRxiv; the COVID-19 Open Research Dataset (CORD-19) by Semantic Scholar and the Allen Institute for AI, which includes relevant records from Microsoft Academic, Elsevier, arXiv and PMC; and the WHO COVID-19 database. A sample search strategy (for PubMed) appears in Box 1 from 1^st Jan 2020 to June 7^th 2020. We have developed a workflow that automates daily searches of these databases, and the code supporting this process can be found at https://github.com/mcguinlu/COVID_suicide_living. Searches are conducted daily via PubMed and Scopus application programme interface and the bioRxiv and medRxiv RSS feeds. Conversion scripts for the daily updated WHO and the weekly updated CORD-19 corpus are used to collect information from the remaining sources. The software includes a systematic search function based on regular expressions to search results retrieved from the WHO, CORD-19 and preprint repositories (search strategy available in extended data (John & Schmidt, 2020)). Our review is ongoing and we continue to investigate the use of other databases and to capture articles made available prior to peer review and assess eligibility and review internally. We therefore included PsyArXiv and SocArXiv repositories in our search strategy via their own open access platforms as we developed our automated system. For this version of the living review, Psy- and SocArXiv searches were carried out retrospectively on the 12^th of June, using a publication date filter for Jan 1^st 2020 – June 7^th 2020.

A two-stage screening process was undertaken to identify studies meeting the eligibility criteria. First, two authors (either CO or EE) assessed citations from the searches and identified potentially relevant titles and abstracts. Second, either DG, AJ or RW assessed the full texts of potentially eligible studies to identify studies to be included in the review. This process was managed via a custom-built online platform (Shiny web app, supported by a MongoDB database). The platform allowed for data extraction via a built-in form.

Data collection and assessment of risk of bias

One author (DG, AJ or RW) extracted data from each included study using a piloted data extraction form (see extended data (John & Schmidt, 2020)), and the extracted data were checked by one other author (AJ, or EE where AJ extracted data). Disagreements were resolved through discussion, and where this failed, by referral to a third reviewer (KH, NK or PM). Irrespective of study design, data source and outcome measure examined, the following basic data were extracted: citation; study aims and objectives; country/setting; characteristics of participants; methods; outcome measures (related to self-harm / suicidal behaviour and COVID-19); key findings; strengths and limitations; reviewer’s notes. For articles where causal inferences are made - i.e. randomised or non-randomised studies examining the effects of interventions or aetiological epidemiological studies of the effects of exposures – we used a suitable version of the ROBINS-I or ROBINS-E tool to assess risk of bias as appropriate based on the research question and study design (Morgan et al., 2017; Sterne et al., 2016).

Data synthesis

We synthesised studies according to themes based on research questions and study design, using tables and narrative. Results were synthesised separately for studies in the general population, in health and social care staff and other at-risk occupations, and in vulnerable populations (e.g. people of older age or those with underlying conditions that predispose them to becoming severely ill or dying after contracting COVID-19). Where multiple studies addressed the same research questions, we assessed whether meta-analysis is appropriate and would conduct it where suitable, following standard guidance available in the Cochrane Handbook (Deeks et al., 2019). The current document is the first iteration of our review. We have not considered it appropriate to combine any results identified so far in a meta-analysis.

Living review method

Details of the living review method, justification for its use and our transition plan are provided in our protocol (John et al., 2020a). We plan to maintain the review in living mode for at least 12 months, from publication of the protocol (25^th June 2020). We will undertake monthly screening and consider full updates at least every three months. We will extend the living mode at 6-monthly intervals if evidence is still being published regularly. We anticipate an end to the living phase of the review at most 24 months after initiation, at which point we plan to publish the cumulated evidence in the form of a standard systematic review. Any decision to update the review more or less frequently will depend on the likely impact of the new evidence on the conclusions of the review. Impactful evidence may be (i) evidence that affects policy and/ or (ii) substantial, high-quality research studies (e.g. a randomised trial or population-based observational cohort study). Since we have not as yet identified any new evidence that impacts on the conclusions of this review we next update will include studies up to the 7^th of October 2020 after four months.

Results

In total, 2070 citations were identified by 7 June 2020 from all electronic searches, after duplicates were removed (Figure 2). The cumulative numbers of articles over time that were identified by the search and included in the review are shown in Figure 3 and Figure 4.

Figure 2. PRISMA flow diagram.

The latest and previous versions of this figure are available as extended data (John & Schmidt, 2020).

Figure 3. Number of articles identified by database and respository over time.

The latest and previous versions of this figure are available as extended data (John & Schmidt, 2020).

Figure 4. Number of articles selected by database and respository over time.

The latest and previous versions of this figure are available as extended data (John & Schmidt, 2020).

Discussion

In total, 28 independent studies (29 articles) were included in this review covering a total of 33,345 studied individuals from around the world with a mix of low, middle and high income countries. Almost half of the articles were pre-prints published before peer review, or research letters that may not have been peer-reviewed. The majority of studies were case series or cross sectional surveys, almost all based on non-representative convenience samples. Only one study reported on the change in incidence of suicide or suicidal behaviour before versus after the onset of the pandemic (Pignon et al., 2020); this analysis was based on emergency psychiatric consultations for suicide attempt – and reported a decline, although levels of consultation could have been influenced by fears about using services or ideas of not burdening the health service rather than changes in incidence. A further study from China reported heightened levels of self-harm thoughts in pregnant women surveyed in the period after the onset of the pandemic, compared with levels reported amongst women surveyed at the same stage of pregnancy just before the pandemic (Wu et al., 2020b). No studies reported potentially harmful effects of lockdown/physical distancing measures in relation to our outcomes, although one study comparing the prevalence of suicidal thoughts and attempts in people living in USA states with varying timing and strigency of state-specific lockdowns found no evidence for such an ecological association (Bryan et al., 2020). Modelling studies that aimed to predict the impact of the pandemic on national or global suicide rates produced widely differing estimates of the likely impact and most focused on predictions based on previous studies of the impact of changes in unemployment levels on suicide. Three studies investgated service use patterns – one found a decline in ED visits for suicidal thoughts, one a decline in psychiatric emergency consultation for suicide attempt and the other reported an increase in contacts to a mental health digital platform but no changes in contacts for suicidal thoughts.

We identified eight case series reports of suicide attempts and suicide deaths, five based on news stories in India, Bangladesh and Pakistan. Given the relatively low quality of case series in the hierarchy of evidence, often reflecting small numbers and selection bias, but more importantly the lack of comparator data, drawing any reliable inferences from these studies is challenging. Furthermore, news reports report a non-representative sample of suicide deaths and often derive their information from bystanders and witnesses who are unlikely to know the full circumstances of the death (Khan et al., 2009). Nevertheless, these studies highlight circumstances surrounding apparently COVID-19-related suicides and flag the potential importance of factors such as economic difficulties, fear of the disease, and social isolation. Indeed in parts of the world without reliable suicide incidence data they may be the only source of information (Khan & Hyder, 2006).

The 12 cross-sectional studies investigated a range of issues. Findings indicated worries about COVID-19 and recent COVID-19 infection were associated with suicidal thoughts (Hao et al., 2020; Killgore et al., 2020a; Killgore et al., 2020b; Lee, 2020; Lee et al., 2020; Zhao et al., 2020) and, amongst pregnant women surveyed during the pandemic, thoughts of self-harm were higher than amongst those surveyed pre-pandemic. The one study comparing suicidal thoughts and behaviours amongst people living in areas with versus without physical distancing measures found no adverse association (Bryan et al., 2020). Surprisingly survey by Ammerman et al. (2020) from the USA indicated that social distancing was associated with reduced instances of suicidal thoughts early in the period of lockdown. Only one survey suggested it was nationally representative but lacked sampling details (Killgore et al., 2020a). Non-probability sampling lacks a sound theoretical basis for statistical inference (Neyman, 1934). Consequently, basic descriptive analyses and explorations of potential associations are appropriate but measures of uncertainty (i.e., confidence intervals around estimates of prevalence) are generally not valid. One study (Bryan et al., 2020) used panel quota sampling, but these sorts of adjustments for age, sex and ethnicity may miss other elements of bias and cannot account for groups not included at all, particularly if the response rate is unknown (Pierce et al., 2020). Four studies used convenience sampling which tend to attract volunteers who have access to the internet, are already engaged in research and have an interest in the topic. Hence responses may be unrepresentative of the general population, and associations observed among these healthy volunteers may not reflect associations that would be observed in others. Similarly, when assessing suicidal thoughts and behaviours, those in most distress or with co-existing mental illness, as well as older people, are less likely to participate in these sorts of surveys. There is no way to assess non-response bias in a convenience sample as might be possible in a probability-sampled survey (Pierce et al., 2020).

There was a large range in modelling estimates of the effect of the pandemic on suicide rates, varying between a 1% and a 145% rise. These differences between model estimates were partly due to differences in modelling assumptions, which are associated with considerable uncertainty. Given the methodological limitations, the uncertainty of assumptions about how the economies of individual countries will be affected, as well as international differences in financial supports given to businesses and people out of work, these predictive exercises can at best only provide a guide as to where action should be directed.

Data availability

Software availability

The development version of the software for automated searching is available from Github: https://github.com/mcguinlu/COVID_suicide_living.

Archived source code at time of publication: http://doi.org/10.5281/zenodo.3871366 (McGuinness & Schmidt, 2020)

License: MIT