AS the full economic and social costs of the COVID-19 pandemic are realised over coming months, researchers and policy makers want to know the extent to which suicide rates may vary and, in turn, how they can respond effectively. A variety of predictive modelling analyses have begun in an attempt to forecast the likely scenario.

Modelling by the Brain and Mind Centre at the University of Sydney anticipates that increased unemployment, social disconnection and health services at capacity will result in an extra 750 to 1500 suicides per year over 5 years, with an increase rate of 25-50% per year. A separate study by the Epworth Centre for Innovation in Mental Health at Monash University estimates over a 10-year period an increase of 2761 deaths, or 276 extra deaths per year. Such estimates were based on a predicted 10-year recession, with an initial unemployment rate of 15%. This modelling suggests an increase in suicide deaths of 9% a year, an estimate much lower than that offered by the Brain and Mind Institute.

Examining these two modelling studies alone highlights the wide range of results, and potential risks of relying on modelling studies to design suicide prevention interventions. Modelling studies are based on assumptions that are made using data from previous research. For instance, suicide has been associated with periods of unemployment, social isolation, increases in domestic violence, and where support or treatment is not available. In contrast, suicides may be reduced by external factors such as employment benefits to remediate extreme hardship, social connectedness derived from the shared hardships, similar to that sometimes seen in wartime, or by the availability and adequacy of health care services.

Beyond potential risk factors, we currently have little evidence from previous pandemics and epidemics as to their effects on suicide rates. COVID-19 also differs from previous pandemics and epidemics not only in terms of its direct impact on illness and loss of life worldwide, but also due to dramatic physical distancing measures and nationwide lockdowns. It is therefore difficult to ascertain the direct causal links between COVID-19 and suicide. Additionally, as the situation changes rapidly, with new restrictions imposed and/or eased, it is difficult to even begin to map a coherent picture.

To better understand the likely impact of COVID-19, we were curious to see if suicide modelling had been undertaken outside of Australia, and what would be revealed in pre-published or published literature from the rest of the world. A literature review was conducted using Google Scholar, PubMed, MEDLINE and the World Health Organization’s COVID-19 databases. Search terms included “COVID-19” and “suicide”.  To estimate the percentage of deaths per year, we used estimates of suicide rates published from official sources or health authorities, when these were not provided within the research article.

We found five studies, covering the US and Canada, and two studies estimating global effects (see Table 1). These reported a wide variation in the estimated rates of suicide deaths — ranging from a 1% increase across the world and to 27.7% in Canada. Four out of five modelling studies used only one factor to predict suicide deaths (unemployment), whereas one modelling study assessed the impact of both unemployment and social isolation/loneliness.

Table 1 shows that majority of modelling studies have only examined the effect of unemployment on suicide rates. The assumed effect of COVID-19 on unemployment rates varies between studies (5-24%), resulting in wide variation in estimated excess suicide deaths (ranging from less than 1% to nearly 28%). One major limitation of the modelling is that, aside from unemployment, numerous other factors associated with COVID-19 that may either increase or decrease suicide rates have been excluded. These include relative levels of spending on social welfare, availability and access to mental health services, and other protective measures such as adequate housing and food security. Another limitation is that the modelling does not take account of regional variation in risk factors. This may explain the discrepancy between the findings in Table 1, and the rates of 25-50% estimated by the Brain and Mind Centre, which drew on data from Western Sydney and the North Coast, where extreme unemployment levels might eventuate. The discrepancy could also be due equally to the assumptions of their decision analytic approach, the use of multiple variables, or the untested assumptions of the model.

Author/ Country Country Variables introduced in the prediction Estimate of effect of variable (ie, unemployment number or rate) Number of extra deaths from suicide predicted (range) Percentage increase in deaths over previous estimates (range)
McIntyre and Lee Canada Unemployment rates based on assumptions about lockdown times 7.5-16.6 unemployment rate for 2020

7.2-14.9 unemployment rate for 2021

418-2114 5.5-27.7%
McIntyre and Lee US Unemployment rates based on assumptions about lockdown times 5.8-24

unemployment rate for 2020

9.3-18 unemployment rate for 2021

3235-8164 3.3-8.4%
Bhatia US Unemployment 33.5 million extra unemployed in 2019 9107 19%
Weems, Carrion US






Unemployment, social isolation and loneliness 6 million additional unemployed

16.6 million individuals impacted by loneliness

3819 (UE)

5617 (L)

Total deaths: 9437

7.9% (UE)

11.6% (L)

Overall: 19.5%

World Unemployment social isolation/


10 million additional unemployed

130 million impacted by loneliness

5136 (UE)

35 142 (L)

Total deaths: 40279%

0.6% (UE)

4.4% (L)

Overall: 5%

Kawohl and Nordt World Unemployment 5.3-24.7 million jobs lost: range of unemployment of 5.1-5.7 million. 2135-9570 0.3-1.2%

Table 1: Estimated impacts of COVID-19 on suicide rates from published/pre-published research papers. If not stated in paper, figures were estimated based on published suicide rates from the country/region modelled: suicide rate of 48 344 in 2018 (US source); suicide rates for the world of 793 823 in 2017 (WHO, 2017). UE = estimated deaths from unemployment. L = estimated deaths from loneliness.

So, what can be drawn from this review of COVID-19 related suicide rise?

First, if modelling of suicide rates such as those receiving recent media attention is intended to directly inform service planning, we recommend that these models be published after methods to reduce confounding are applied, and that rigorous, independent reviews are commissioned to test their assumptions. Over time, an evidence base could be developed on the accuracy (and hence usefulness) of predictive models in this context. Given their current limitations, we think it would be misguided to rely on modelling studies to design suicide prevention interventions.

Second, we propose that it is important to focus on data which emerge from studies of suicide rates and risk factors. Empirical data are a more useful resource than modelling studies in informing response planning as they are free of the assumptions which inform modelling studies. Further, as the COVID-19 pandemic is a unique event unlike any other in recent history, collecting the right data can provide insight into factors specific to COVID-19 that increase or decrease the rate of suicide.

The problem is we do not yet have timely, full or unbiased data on risk factors, suicide risk, ideation and attempts, hospitalisations, deaths, or uptake of services or service usage. This leaves governments and service providers in the dark. We urge governments to immediately invest in new representative data collections and progress the national suicide prevention and state suicide register systems. Systems that quickly pull various data sources together are needed not just in the context of global or national disasters like COVID-19 but to act to reduce rates of suicide more broadly.

Dr Samantha Tang is a Research Officer at the Black Dog Institute.

Dr Mark Deady is a Research Fellow at the Black Dog Institute.

Dunkan Yip is a Policy Advisor at the Black Dog Institute, having previously advised federal and state government in areas such as social policy, climate and energy, infrastructure, and economic reform.

Professor Helen Christensen is the Director and Chief Scientist at the Black Dog Institute and Professor of Mental Health at UNSW Sydney.



The statements or opinions expressed in this article reflect the views of the authors and do not represent the official policy of the AMA, the MJA or InSight+ unless so stated.


Should we be aiming for COVID-19 suppression or eradication?
  • Eradication (52%, 74 Votes)
  • Suppression (33%, 47 Votes)
  • Neither ... let it run its course (15%, 21 Votes)

Total Voters: 142

Loading ... Loading ...

4 thoughts on “COVID-19 and suicide: variation and response

  1. Anonymous says:

    Agree with Phillips point of data collection, however the first thing that needs addressing with that regard is how they report these incidents or admissions because to my knowledge if a patient has a physical injury or a more predominant medical concern, they are reported as such and the data collected is recorded for only 1 issue and not necessarily the true reason for presentation at hospital or why assistance was required by the first responders therefore we need to improve the way we capture the full detail in the first instance to gain true insight.

  2. A/Professor Jo-An Atkinson, Professor Ian Hickie, Dr Adam Skinner, Dr Yun (Christine) Song, A/Professor Kenny Lawson, Dr Ante Prodan & Dr Frank Iorfino at the Brain and Mind Centre, University of Sydney says:

    This article represents a grossly inadequate critique of modelling studies exploring the likely impacts of COVID-19 on deaths by suicide. It simply fails to provide adequate comparison of the type, scope, structure, time horizon, quality, consistency, and validity of studies available (in both the academic and wider modelling literature) to justify the conclusions. Much rigorous modelling is done by reputable agencies that produce real-time reports for government, and other substantive agencies and sources. Most is not directed simply at producing academic reports. While we support the call for greater transparency in modelling (see transparency does not in itself represent a reliable indicator of the utility or robustness of predictive modelling.

    In failing to understand the difference between traditional regression models (that articulate statistical associations between a risk factor and an outcome) and complex systems modelling and simulation (that articulates a complex causal mechanism underpinning the interrelationships between risk factors / dynamic drivers of mental health outcomes) the authors have compared apples with oranges to suggest the general unreliability of predictive modelling.

    There are obvious inconsistencies in the authors’ argument where, on the one hand they recognize that the extent to which unemployment impacts suicide outcomes depends on ‘external factors such as employment benefits to remediate extreme hardship, social connectedness derived from the shared hardships, similar to that sometimes seen in wartime, or by the availability and adequacy of health care services,’ many or all of which vary between studies; yet on the other hand they use variation in results of modelling studies across countries and across regions within countries to claim that the ‘wide range of results’ are a ‘limitation’ of modelling studies that pose ‘potential risks of relying on modelling studies to design suicide prevention interventions.’

    The following statements are particularly bemusing:

    – ‘Empirical data are a more useful resource than modelling studies in informing response planning as they are
    free of the assumptions which inform modelling studies;’ and
    – ‘…collecting the right data can provide insight into factors specific to COVID-19 that increase or decrease the
    rate of suicide.’

    Are the authors suggesting:

    A. That we should wait for the crisis to unfold, collect data on how many suicides occurred, then more reliably predict what happened in the past? Imagine if we had engaged in this approach for COVID-19 virus transmission! Thankfully, despite imperfect data and very little specific knowledge of novel coronavirus, multiple reputable Institutes rapidly deployed their advanced systems modelling expertise to provide our national cabinet with critical tools for managing the uncertainty of disease dynamics, testing alternative assumptions, integrating new data and evidence as it came to hand, and weighing response options in the midst of an evolving crisis; an approach that prevented tens of thousands of unnecessary COVID-19 deaths and an approach being undertaken by the Brain and Mind Centre of the University of Sydney to inform national mental health and suicide prevention responses.

    B. Or are authors suggesting that we take previous empirical studies that are apparently ‘free from assumptions’ (which of course they are not – a fundamental teaching of elementary epidemiology and biostatistics) and make assumptions about what might work to prevent suicide for the current crisis (even though the current situation is, by their own assessment, a ‘unique event unlike any other in recent history’ and despite stating in equally spectacular contradiction that modelling studies are limited because they ‘are based on assumptions that are made using data from previous research’). Without rigorous modelling and simulation of possible future trajectories, on what basis do the authors propose decisions should be made to respond effectively to the crisis? What combination of initiatives should we invest in? What impacts should we expect? What targeting, timing, scale, frequency, and intensity of investments are needed to respond effectively and sustainably to the mental health crisis? Should we simply guess at the answers to these questions? Or should we bring together best evidence, data and expert and local knowledge in a way that helps us ‘map a coherent picture,’ quantify it, validate its performance, and simulate forward, test alternative assumptions, understand their implications and weigh up the trade-offs in a robust and disciplined way as we make important strategic decisions that will impact people’s lives in fundamental ways.

    It is understandable that there are still major misunderstandings in mental health regarding predictive modelling. To address these difficulties, we refer to an earlier plain language summary by national and international leaders in the field of modelling and simulation ( We encourage others who are genuinely interested in how we best plan our response to the current COVID19-induced mental health crisis to read that summary.

  3. Philip Morris says:

    By now we should have some inkling of whether the suicide rate is going to increase. Suicide attempts are a good guide to what can be expected in terms of completed suicides. So ambulance records, emergency department records, police records and records from phone crisis services of attempted suicide should be reviewed to see if the rate of attempts has changed during Covid-19. In the meantime, suicide prevention interventions should be instituted with the main thrust being to identify and prevent isolation of at-risk individuals.

  4. Andrew Renaut says:

    Everyone can absolutely expect suicide to increase substantially. You don’t need a study to show that! And that’s not even taking into account the morbidity associated with the deterioration in mental health of thousands of people to the point where they simply can’t function with day to day activities. And then of course we have all of the premature deaths with the 2 major killers in this country – cancer and heart disease – because of missed diagnoses and delayed treatments.

    It simply beggars belief that the advice given by so called “doctors” to politicians is so spectacularly wayward. The lot of them should hold their heads in shame. They are an absolute disgrace to the profession and humanity. And for the rest of my “colleagues” to do nothing more than sit on their hands is equally lamentable. Utterly shocking!!

Leave a Reply

Your email address will not be published. Required fields are marked *