Antidepressant Utilization and Suicide in Europe: An

Ecological Multi-National Study

Ricardo Gusmão1,2*, Sónia Quintão1, David McDaid3, Ella Arensman4, Chantal Van Audenhove5,
Claire Coffey4, Airi Värnik6, Peeter Värnik6, James Coyne7,8, Ulrich Hegerl9
1 CEDOC, Departamento de Saúde Mental, Faculdade de Ciências Médicas, Universidade Nova de Lisboa, Lisboa, Portugal, 2 Departamento de Psiquiatria e Saúde Mental,
Centro Hospitalar de Lisboa Ocidental, Lisboa, Portugal, 3 Personal Social Services Research Unit, LSE Health and Social Care, London School of Economics and Political
Science, London, United Kingdom, 4 National Suicide Research Foundation, Cork, Ireland, 5 Katholieke Universiteit Leuven, LUCAS, Leuven, Belgium, 6 Estonian-Swedish
Mental Health and Suicidology Institute (ERSI), Tallinn, Estonia, 7 Perelman School of Medicine of the University of Pennsylvania, Philadelphia, Pennsylvania, United States
of America, 8 Health Psychology Section, Department of Health Sciences, University Medical Center, University of Groningen, Groningen, The Netherlands, 9 Department
of Psychiatry, University of Leipzig, Leipzig, Germany

Abstract
Background: Research concerning the association between use of antidepressants and incidence of suicide has yielded
inconsistent results and is the subject of considerable controversy. The first aim is to describe trends in the use of
antidepressants and rates of suicide in Europe, adjusted for gross domestic product, alcohol consumption, unemployment,
and divorce. The second aim is to explore if any observed reduction in the rate of suicide in different European countries
preceded the trend for increased use of antidepressants.

Methods: Data were obtained for 29 European countries between 1980 and 2009. Pearson correlations were used to
explore the direction and magnitude of associations. Generalized linear mixed models and Poisson regression distribution
were used to clarify the effects of antidepressants on suicide rates, while an autoregressive adjusted model was used to test
the interaction between antidepressant utilization and suicide over two time periods: 1980–1994 and 1995–2009.

Findings: An inverse correlation was observed in all countries between recorded Standardised Death Rate (SDR) for suicide
and antidepressant Defined Daily Dosage (DDD), with the exception of Portugal. Variability was marked in the association
between suicide and alcohol, unemployment and divorce, with countries depicting either a positive or a negative
correlation with the SDR for suicide. Every unit increase in DDD of an antidepressant per 1000 people per day, adjusted for
these confounding factors, reduces the SDR by 0.088. The correlation between DDD and suicide related SDR was negative in
both time periods considered, albeit more pronounced between 1980 and 1994.

Conclusions: Suicide rates have tended to decrease more in European countries where there has been a greater increase in
the use of antidepressants. These findings underline the importance of the appropriate use of antidepressants as part of
routine care for people diagnosed with depression, therefore reducing the risk of suicide.

Citation: Gusmão R, Quintão S, McDaid D, Arensman E, Van Audenhove C, et al. (2013) Antidepressant Utilization and Suicide in Europe: An Ecological Multi-
National Study. PLoS ONE 8(6): e66455. doi:10.1371/journal.pone.0066455
Editor: Vinod K. Yaragudri, Nathan Kline Institute for Psychiatric Research and New York School of Medicine, United States of America
Received March 2, 2013; Accepted April 29, 2013; Published June 19, 2013
Copyright: ß 2013 Gusmão et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The OSPI project has received funding from the European Community’s Seventh Framework Programme, FP7/2007–2013 (http://ec.europa.eu/
research/fp7/index_en.cfm) under grant agreement nu 223138. The funders had no role in study design, data collection and analysis, decision to publish, or
preparation of the manuscript.
Competing Interests: Ricardo Gusmão, Sónia Quintão, David McDaid, Ella Arensman, Chantal Van Audenhove, Claire Coffey, Airi Varnik, Peeter Varnik and
James Coyne have declared that no competing interests exist. Ulrich Hegerl read the journal’s policy and has the following conflicts: Dr. Hegerl is an advisory
board member for Lilly and Lundbeck, a consultant for Nycomed and a speaker for Bristol-Myers Squibb. This does not alter the authors’ adherence to all the PLOS
ONE policies on sharing data and materials.
* E-mail: rgusmao@mac.com

Introduction with only the exceptions of Malta, Poland and Portugal where
increasing trends were present [1]. Despite the onset of the
Antidepressant use has continually increased in most European economic crisis, there is no strong evidence that national suicide
countries since the advent of selective serotonin reuptake inhibitors rates have increased but suicide remains a major public health
(SSRIs). Between 2000 and 2010 rates of use in Europe have problem, accounting for 60.000 deaths per year in the EU-27
continued to increase, with the highest DDD rates seen in Iceland, alone [1].
Denmark and Portugal [1]. Suicide rates vary greatly across the Suicide is strongly associated with poor mental health, especially
European Economic Area, but between 1980 and 2000 suicide mood disorders [3].
rates fell in all of the EU-15 countries plus Norway, with the Antidepressants are the most common treatment for mood
exceptions of Ireland and Spain [2]. From 1995 to 2010, the same disorders, but effective use of these medications requires admin-
decrease in suicide rate was observed across the EU-27 countries, istration to patients who have been properly diagnosed and then

PLOS ONE | www.plosone.org 1 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

adequately followed-up [4,5]. There is a consensus as to the was qualified by SSRI sales data having to be imputed prior to
importance of primary care doctors’ education programmes for 1990 due to a lack of sales data. Nonetheless, their finding suggests
improving the management of depression with antidepressants in that greater utilization of SSRI, particularly for adults, might be a
order to reduce the risk of suicide [6]. Furthermore, a number of cost effective strategy from a public health perspective, with one
multi-component suicide prevention programmes emphasise the suicide averted for every 300,000 pills sold. The same authors also
crucial importance of primary care education programmes to undertook further analysis with panel data covering the same time
facilitate optimal antidepressant prescribing [7]. period for 26 countries, including seven countries from central and
However, there are concerns about the efficacy and safety of Latin America, Japan, Israel, 13 EEA countries, the US, Canada,
antidepressants, with some authors suggesting that these medica- Australia and New Zealand. The conclusions were much the same,
tions are at best no better than placebo [8] and others that although a more powerful effect was shown with one suicide
antidepressants may actually increase the risk of suicidal behav- averted for every 200,000 pills sold. This analysis also noted that
iour, particularly in young people [9–11]. In contrast, still other there was no evidence in any change in patterns of psychotherapy
authors contend that there is a bias in these findings and that over the study period [32].
benefits are in fact greater than risk [12–16]. For instance, one Thus, a number of previous studies have used an ecological
meta-analysis of 27 RCT trials examined antidepressant prescrib- approach to look at some actions to help reduce the risk of suicide.
ing in children and adolescents to age 18 with a diagnosis of major Notwithstanding long held arguments on ‘ecological fallacy’ and
depressive disorder and showed that benefits appeared to far the danger of misinterpretation of findings of studies gathered
outweigh a small increased risk of suicidal behaviour [17]. using population level data [33,34], there are at least three reasons
The limited applicability of data from RCTs to public health for a greater use of this type of study design in respect of suicide
questions point to the importance of evidence from other types of research.
study design. For instance, analysis of US Veterans Affairs Medical First, escalating costs associated with increasing use of
System record data of more than 200.000 adults diagnosed with antidepressants in many countries suggests the need to examine
depression and followed up for at least six months, found long term effectiveness of antidepressants both in terms of a
statistically significant lower rates of suicide in those treated using reduced prevalence of mood disorders or reduced incidence of
any antidepressant [18]. Comparisons among such studies with suicide. The value of antidepressant treatment at a population-
very different approaches are difficult. Studies vary in basic health level has been challenged [35,36] and remains to be
terminology, definition of outcomes and time periods considered, demonstrated. Second, in order to demonstrate statistically in a
drugs and other interventions assessed, and statistical methods, controlled study that antidepressants produce a preventive effect in
leading to seemingly contradictory results. For instance, one respect of the profound but nonetheless relatively rare event of a
review of studies with naturalistic designs had equivocal findings completed suicide, we would need a sample size of 20.000 people
[19]. In contrast, a number of studies of the effects of warnings on randomly treated with either antidepressants or placebo [37]. This
the risk of suicide with use of antidepressants on subsequent usage may be difficult to achieve in practice given that suicidal risk tends
observed an increase in suicide rates in younger people [19].
to be an exclusion criterion in antidepressant trials, naturalistic or
Furthermore, differing approaches in these studies to controlling
experimental. Third, it would be unacceptable for ethical reasons
for potential confounds are challenging for integration and
to conduct a randomized controlled trial with suicide as an
interpretation. In some settings and contexts, economic develop-
outcome variable [38,39].
ment correlates with lower suicide rates [20,21], while alcohol
consumption [22–24], divorce both in men and women [25,26]
and unemployment [27,28] can correlate with higher suicide rates. Objectives
Nonetheless, there is potentially an important role for ecological Given the continued debate on whether evidence of substantial
studies, i.e., studies analysing data trends at a population rather increases in the rate of antidepressant prescription can be
than an individual level, to help to inform public health policy. translated into improved public health outcomes, and notably
This is advantageous where multiple areas or countries can be reduction in suicide, the present study aims to describe antide-
examined, in order to control better for region-specific factors that pressant utilization and suicide trends in European, largely EU,
may impact on suicide rates and use of mental health services. The Member States.
evidence from these studies is, however, also mixed. One review of Our first aim was to examine whether the growing use of
19 ecological studies found equivocal evidence for links between antidepressants had an effect on European suicide rates, exploring
suicide and antidepressants, with slightly greater reductions in the plausibility of competing explanations of associations with
suicide rates in the 1990s compared to the 1980s, especially when indicators such as adult per capita alcohol consumption, unem-
associated with higher initial suicide rates, being a man and older ployment and divorce rates, and GDP. Our second aim was to
age [29]. examine temporal relationships, i.e. whether any reduction in the
Wheeler et al [30] examined changes in country-specific suicide rate of suicide preceded any trend towards increased use of
trends in younger people following the introduction of regulatory antidepressants as revealed by shorter and longer time-series of
actions including the use of warnings on antidepressants in a simultaneous antidepressant utilization and suicide data.
number of countries in 2003 and 2004. They also found the
evidence to be equivocal with reductions in the rate of suicide Methods
observed in some countries and increases in others, albeit noting
weak evidence of an increase in suicide in young women. Sources of Data
Ludwig and Marcotte [31] pooled panel data concerning rates This ecological and naturalistic study analyses correlations
of suicide and the increased use of SSRI’s from the US, Canada, between datasets over a lengthy time period, covering 29
Australia and 24 European countries between 1980 and 2000 and European countries including all 27 European Union Member
estimated that overall an increase in sales of one pill per capita was States, with the exceptions of Malta and Cyprus, due to a lack of
found to be associated with a 2.5% decrease in suicide rates for the data on antidepressant utilisation in those countries. Data from
whole population. However, they acknowledged that this finding Croatia, Iceland, Norway and Switzerland were also included.

PLOS ONE | www.plosone.org 2 June 2013 | Volume 8 | Issue 6 | e66455

 Table 1. ATC N06 antidepressant utilization (DDD per 1000 inhabitants per day) and utilization growth, and registered suicide standard death rate (SDR suicide, all) and rates
variation, by year and country.

last 5-
period available years
country variables available years first year last year Diff mean growth %

1980– 1985– 1990– 1995– 2000–

last-first period per year 1985 1990 1995 2000 2005 2005–2009

Austria DDD/1000/day 1991–2009 19 9.1 56.54 47,44 49.78 521 27 115 39 27

PLOS ONE | www.plosone.org

SDR suicide, all 1980–2009 30 25.09 12.8 212,29 13.37 249 21.63 5 217 27 213 216 213
Belgium DDD/1000/day 1997–2009 13 29.3 67.76 38,46 61.43 131 10 43 22
SDR suicide, all 1980–20061 27 21.59 16.8 24,79 17.19 222 2.81 2 220 12
Bulgaria DDD/1000/day 2004–2009 6 5.89 7.61 1,72 6.03 29 5 74
SDR suicide, all 1980–2009 30 13.65 9.35 24,30 10.01 232 21.07 15 210 10 23 229 212
Croatia DDD/1000/day 2005–2009 5 16.45 22.9 6,45 19.56 39 8 39
SDR suicide, all 1985–2009 25 22.23 14.96 27,27 15.5 233 21.32 4 220 12 218 212
Czech Republic DDD/1000/day 1980–2009 30 2.7 35.88 33,18 30.93 1229 41 0 15 74 80 157 44
SDR suicide, all 1980–2009 30 22.23 12.4 29,83 12.42 244 21.47 28 26 216 28 26 210
Denmark DDD/1000/day 1980–2009 30 8.96 77.6 68,64 68.5 766 26 14 217 118 87 71 32
SDR suicide, all 1980–2009 30 31.99 9.9 222,09 10 269 22.30 216 217 229 223 217 23
Estonia DDD/1000/day 1999–2009 11 5.2 13.14 7,94 13.41 153 14 95 5

3
SDR suicide, all 1981–20092 29 36.74 18.25 218,49 17.32 250 21.72 213 49 236 228 23
Finland DDD/1000/day 1980–2009 30 3.45 64.21 60,76 57.82 1761 59 93 7 185 75 43 27
SDR suicide, all 1980–2009 30 25.24 18.26 26,98 18.18 228 20.93 25 21 210 218 218 4
France DDD/1000/day 1995–2009 15 28.9 49.26 20,36 49.47 70 5 37 24 1
SDR suicide, all 1980–2008 29 18.99 14.96 24,03 15.2 221 2.72 15 213 21 210 25
Germany DDD/1000/day 1986–2009 24 6.1 39.95 33,85 33.34 555 23 68 37 35 43
SDR suicide, all 1990–2009 20 15.47 9.51 –5,96 9.71 –39 –1.95 –10 –16 –11 –9
Greece DDD/1000/day 1998–2009 12 11.6 49.14 37,54 43.71 324 27 102 28
SDR suicide, all 1980–2009 30 3.2 3.02 –0,18 2.92 –6 –.20 23 –18 –1 –1 –2 –3
Hungary DDD/1000/day 1990–20096 20 3.77 25.45 21,68 24.19 575 29 60 124 68 13
SDR suicide, all 1980–2009 30 44.54 21.79 –22,75 21.94 –51 –1.70 –3 –12 –20 –4 –20 –6
Iceland DDD/1000/day 1989–2009 21 14.9 98.3 83,40 95.16 560 27 98 114 34 4
SDR suicide, all 1980–2009 30 12.33 11.49 –0,84 11.48 –7 –.23 22 7 –36 74 –36 –1
Ireland DDD/1000/day 1996–2009 14 17.96 55.51 37,55 49.91 209 15 41 22
SDR suicide, all 1980–2009 30 7.69 11.61 3,92 10.21 51 1.70 15 21 8 5 –11 8
Italy DDD/1000/day 1995–2009 15 8.88 36.39 27,51 33.3 310 21 87 75 25
SDR suicide, all 1980–20083 29 7.15 5.39 –1,76 5.24 –25 –.86 10 –13 2 –13
Latvia DDD/1000/day 2004–2009 6 4.15 6.14 1,99 5.59 48 8 42
SDR suicide, all 1980–2009 30 32.6 20.7 –11,90 20.23 –37 –1.23 –10 –11 57 –24 –27 –8

June 2013 | Volume 8 | Issue 6 | e66455

Antidepressant Utilization and Suicide in Europe

Lithuania DDD/1000/day 2002–2009 8 7.66 15.53 7,87 13.27 103 13 54

 Table 1. Cont.

last 5-
period available years
country variables available years first year last year Diff mean growth %

1980– 1985– 1990– 1995– 2000–

last-first period per year 1985 1990 1995 2000 2005 2005–2009
4
SDR suicide, all 1981–2009 29 35.16 31.47 –3,69 31.31 –10 –.34 –23 76 –2 –21 –15

PLOS ONE | www.plosone.org

Luxembourg DDD/1000/day 2003–2009 7 36.2 47.93 11,73 44.56 32 5 14
SDR suicide, all 1980–2009 30 12.88 10.8 –2,08 11.65 –16 –.53 11 17 –15 –5 –27 11
Netherlands DDD/1000/day 2001–2009 9 31.4 40.1 8,70 39.42 28 3 3
SDR suicide, all 1980–2009 30 10.56 8.52 –2,04 8.38 –19 –.63 6 17 –1 –4 2 –5
Norway DDD/1000/day 1980–2009 30 7.99 54.19 46,20 53.03 578 19 13 29 94 82 24 6
SDR suicide, all 1980–2009 30 12.67 11.45 –1,22 10.82 –10 –.33 12 6 –19 –3 –5 2
Poland DDD/1000/day 2004–2009 6 10.59 16.66 6,07 14.09 57 10 44
SDR suicide, all 1983–20095 27 12.91 15.78 2,87 14.37 22 .81 –2 6 3 0 5
Portugal DDD/1000/day 1995–2009 15 18.65 73.15 54,50 61.66 292 19 55 70 49
SDR suicide, all 1980–2009 30 8.08 7.85 –0,23 7.51 –3 –.10 24 –17 –12 –42 68 9
Romania DDD/1000/day 2004–2009 6 2.59 6.09 3,50 4.02 135 23 183
SDR suicide, all 1989–2009 21 11.48 11.17 –0,31 11.12 –3 –.14 34 –1 –8 –2

4
Slovakia DDD/1000/day 1996–2009 14 4.2 27.99 23,79 23.84 566 40 130 42
SDR suicide, all 1986–2009 24 17.31 10.3 –7,01 10.13 –40 –1.67 –14 –6 –11 –14
Slovenia DDD/1000/day 2006–2009 4 32.6 43.2 10,60 38.55 33 8
SDR suicide, all 1985–2009 25 33.64 18.71 –14,93 19.81 –44 –1.76 –17 –2 –1 –19 –15
Spain DDD/1000/day 1992–2009 18 10.2 59.06 48,86 53.86 479 27 76 69 24
SDR suicide, all 1980–2009 30 4.69 6.34 1,65 6.35 35 1.17 41 8 2 1 –9 –4
Sweden DDD/1000/day 1980–2009 30 6.47 72.64 66,17 70.09 1023 34 25 13 198 77 35 11
SDR suicide, all 1980–2009 30 19.04 12.33 –6,71 11.93 –35 –1.17 –10 –8 –11 –18 6 0
Switzerland DDD/1000/day 2004–2009 6 40.87 49.08 8,21 46.62 20 3 16
SDR suicide, all 1980–2009 30 24.94 12.5 –12,44 14.37 –50 –1.67 –5 –15 –8 –7 –13 –16
United DDD/1000/day 1991–2009 19 10.4 61.93 51,53 55.01 495 26 92 29 21
Kingdom SDR suicide, all 1980–2009 30 8.68 6.6 –2,08 6.47 –24 –.80 1 –11 –8 0 –10 3
Mean 15,28 13,69 43,91 40.33* 40,01 19,83
28,45 19,06 12,93 –6,16* 12,94 –0,81

For some countries, data is missing for some years:

1
Belgium, years 2000 until 2003;
2
Estonia, years 1983 and 1984;
3
Italy, years 2004 and 2005;
4
Lithuania, years 1983 and 1984; and 5Poland, years 1997 and 1998 (SDR suicide, all);
6
Hungary, years 1992 and 1994 (DDD/1000/day).
*mean weighted by number of years of data available.
doi:10.1371/journal.pone.0066455.t001

June 2013 | Volume 8 | Issue 6 | e66455

Antidepressant Utilization and Suicide in Europe
 Antidepressant Utilization and Suicide in Europe

Table 2. Pearson correlations between last 5-years means of DDD/1000/day and SDR suicide, million population and number of
suicide deaths within quartiles of countries by years of SDR suicide and DDD/1000/d simultaneous data.

all
countries countries by quartiles

19 or more years (P.75) 15 to 19 years (P50–75) 7 to 14 years (P25–50) 6 or less years (P,25)

Czech Republic, Estonia, France, Greece, Belgium, Bulgaria,

Denmark, Finland, Austria, Hungary, Ireland, Lithuania, Croatia, Latvia, Poland,
Germany, Iceland, Italy, Portugal, Spain, Luxembourg, Romania, Slovenia,
Countries Norway, Sweden UK Netherlands, Slovakia Switzerland

DDD/1000/day and –.41*** –.53*** –.46*** –.48*** .28

SDR Pearson correlation
SDR – last 5 years 12.94 12.08 10.15 14.21 15.33
available mean
DDD/1000/day – last 40.01 58.41 46.30 34.70 27.82
5 years available mean
Mid-year population 514.1 117.7 196.9 105.2 94.3
–2009
Suicide deaths – 60.903 14.563 16.508 14.857 14.975
last year available

***p#.001.
doi:10.1371/journal.pone.0066455.t002

Figure 1. Suicide and use of antidepressants, by country, more than 19 years of simultaneous data (P.75).
doi:10.1371/journal.pone.0066455.g001

PLOS ONE | www.plosone.org 5 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

Figure 2. Suicide and use of antidepressants, by country, 15 to 19 years of simultaneous data (P50–75).
doi:10.1371/journal.pone.0066455.g002

Figure 3. Suicide and use of antidepressants, by country, 7 to 14 years of simultaneous data (P25–50).
doi:10.1371/journal.pone.0066455.g003

PLOS ONE | www.plosone.org 6 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

Figure 4. Suicide and use of antidepressants, by country, 6 or less years of simultaneous data (P,25).
doi:10.1371/journal.pone.0066455.g004

Completed suicide data were obtained from the WHO Health (hypericum perforatum), because of a lack of data on consumption
for All European Mortality Database (WHO-MDB) [40]. This and/or consensus on average daily effective dose.
consisted of SDR for all cases of suicides (ICD10 codes X60–X84 In order to maximise time series data on antidepressant use in
and ICD9 codes E950–E959) for each available year for the period each country over the period from 1990 to 2009, three different
1980–2009. We assumed the suicide recording procedures DDD/1000/day data sources were used. Total wholesale figures
remained the same in the countries involved throughout the study were obtained from IMS Health for the period 2004–2009 (1995–
period [41]. Population data and national unemployment rates 2009 for Portugal and 1996–2009 for Ireland) and OECD
were obtained from the WHO European Region Health For All pharmacy sales data for the period 1990–2009 [46]. Data from
Database (HFA-DB) [42]. Unemployment comprised all working national statistical offices and published literature for 1990–2009
age individuals out of work, currently available for work, or were also used.
seeking work. GDP in US$ per capita was also obtained from the Country data from both the IMS and OECD were used
WHO HFA-DB [42]. Alcohol intake, defined as recorded adult wherever possible. For Bulgaria, Croatia, Ireland, Latvia, Poland,
(15+ years) per capita consumption of pure alcohol (APC) was Romania and Switzerland only IMS Health data were available.
obtained from the WHO Global Information System on Alcohol Units of antidepressants sold each year in the IMS database were
and Health (GISAH) [43]. The recorded crude divorce rate per converted into kilogrammes of active ingredient in order to
1000 population was obtained from OECD Social Indicators [44]. establish the total quantity of sold defined daily dose (DDDs),
The defined daily dosage (DDD) of a drug for adults is which were then divided by the country mid-year resident
determined by an independent scientific committee making use of population, in order to obtain global DDD/1000/day. Using this
the WHO Collaborating Centre for Drug Statistics Methodology procedure, we obtained units for total antidepressants, including
[45]. Data on DDD per thousand individuals per day (DDD/ tricyclic, atypical, SSRIs, Serotonin–norepinephrine reuptake
1000/day) for antidepressants were used in the analysis. This data inhibitors (SNRIs) and other antidepressants.
provides a rough estimate of use of these drugs and the proportion For Iceland, the Netherlands and Slovenia, OECD pharmacy
of the population receiving treatment with a particular antide- sales DDD/1000/day data was the only available source. In
pressant on a daily basis. Only antidepressants in class N06 of the addition to data from IMS and OECD, other DDD/1000/day
Anatomical Therapeutic Chemical Classification System (ATC) data were obtained from the published literature, namely in the
were included in the analysis [45]. Other ATC drugs classes were case of Austria [47], Hungary [48–50], and Italy [51] and directly
excluded, such as lithium, bupropion, combination with antipsy- from authors, as in the case of Denmark, Finland, Norway and
chotics and herbal remedies for depression such as St John’s Wort Sweden [38,52].

PLOS ONE | www.plosone.org 7 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

Table 3. Magnitude of correlations between suicide SDR and antidepressant utilization and potentially confounding variables.

Country DDD/1000/day GDP Alcohol Unemployment Divorce

Austria –.97 –.94 –.05 –.66 –.92

Germany –.96 –.77 .83 –.71 –.86
Hungary –.96 –.86 .90 .18 .46
Estonia –.96 –.78 –.88 .09 .72
Italy –.94 –.81 .69 .66 –.81
Sweden –.89 –.82 –.26 –.72 .03
Finland –.89 –.66 –.54 –.04 –.09
Lithuania –.88 –.67 –.33 .32
France –.87 –.86 .81 .02 –.54
UK –.86 –.90 –.82 .77 .43
Norway –.85 –.75 –.66 –.22 –.35
Denmark –.84 –.89 .30 .65 .37
Slovenia –.82 –.97 .27 –.22 –.48
Czech Republic –.80 –.85 –.75 –.84 –.43
Slovakia –.78 –.87 .47 –.26 –.79
Croatia –.75 –.87 .34 –.10
Spain –.73 .12 –.67 .42 .30
Bulgaria –.66 –.84 .01 .42
Ireland –.65 .38 .75 –.45 –.56
Romania –.65 –.40 –.12 .65
Belgium –.62 –.78 .80 .59 –.57
Switzerland –.59 –.93 .94 –.72 –.79
Latvia –.56 –.68 .07 –.36
Netherlands –.48 –.86 .82 .89 .29
Greece –.44 –.70 .57 –.54 –.52
Iceland –.12 –.32 –.27 –.40 .26
Poland –.08 .09 –.36 .56 –.31
Luxembourg –.01 –.63 –.46 –.60 –.27
Portugal .50 –.33 .46 .40 –.25

doi:10.1371/journal.pone.0066455.t003

Where there was overlap in available information covering the associations between suicide SDRs or DDD/1000/day respec-
same time period in any country, DDD/1000/day data were tively and GDP, alcohol consumption, unemployment and divorce
correlated to assess for consistency. After obtaining a very strong rates. We did not consider statistical significance because there are
positive correlational analysis (r = .98) from these different sources, correlated measures within each individual country and indepen-
averaged DDD/1000/day were used in the analysis. dent measures in different countries. Therefore, p-values can only
be obtained correctly through the use of a general linear mixed
Statistical Analyses model, which we discuss below. This requires taking longitudinal
The final analyses were performed using 870 observations from co-variation between measures into account.
29 countries covering varying timeframes ranging from a General linear mixed models (GLMM) combine the properties
maximum of 30 years (1980 to 2009) to just 4 years for of linear mixed models which incorporate random effects and
antidepressant utilization in Slovenia. There was at least 20 years generalized linear models which contain non-normal data. The
data for suicide rates in all countries, with 18 countries having data choice of a general linear mixed model (GLMM) allows for the
for all 30 years (Table 1). Rate of use for antidepressants and correlation of observations and analysis of incomplete longitudinal
completed suicides in the first and last years for which data are data. It is a statistical method for modeling outcome measures as a
available are presented, along with average annual trend data for function of fixed (population) effects, while simultaneously
each five year period covered. We did not use extrapolations based modeling individual subject parameters as random effects, and
on available trends in consumption to estimate likely consumption can accommodate time-dependent covariates as well as missing
of antidepressants for years where data were not available. observations [53].
We examined the strength of the association between SDRs for The GLMM is represented by
suicide and the use of antidepressants measured in DDD/1000/
day using Pearson’s correlation coefficient. We also used Pearson’s
Yi ~Xi bzZi di zei
correlation coefficient to examine the direction and magnitude of

PLOS ONE | www.plosone.org 8 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

Table 4. Magnitude of correlations between DDD/1000/day Table 5. Model estimates of fixed-effects with SDR suicide
and the other potentially confounding variables. rate as outcome.

Country GDP Alcohol Unemployment Divorce Regression

coefficient SE T p-value
Austria .80 –.91 .57 .81
Model 1
Belgium .92 –.79 –.65 .75
DDD/1000/day –.070 .022 –3.162 .002
Bulgaria .81 .26 –.17
Model 2
Croatia .94 .99 –.91
DDD/1000/day –.088 .026 –3.327 .001
Czech Republic .98 .80 .60 .37
GDP .018 .026 .707 .480
Denmark .91 –.44 –.86 –.03
Alcohol .129 .159 .809 .419
Estonia .92 .97 –.72 –.83
Unemployment –.015 .064 –.232 .816
Finland .89 .76 .21 .49
Divorce 1.273 .473 2.692 .007
France .71 –.60 –.84 .57
Germany .88 –.63 .69 .65 doi:10.1371/journal.pone.0066455.t005
Greece .94 –.26 –.83 .90
Hungary .92 –.47 –.18 .61 (AR) models of aggregate time-series data to adjust for serial
Iceland .79 .92 –.04 –.52 correlation in time series for each predictor (antidepressant use,
GDP, alcohol consumption, and divorce and unemployment
Ireland .92 .73 –.31 .77
rates).
Italy .88 –.91 –.97 .94
In order to assess the consistency of the GLMM results, we also
Latvia .97 .95 .29 performed a Poisson regression, an approach used in a number of
Lithuania .95 .60 –.45 previous studies [55–57]. In this case suicide SDR was the
Luxembourg .90 –.48 .85 .13 dependent variable, with DDD/1000/day as the predictor and a
Netherlands .94 –.91 .39 –.73 logarithm of base n of the number of years of available data per
country, with an analysis of effects of Type III tests. To do this we
Norway .89 .81 .20 .48
cleaned the original database of years simultaneously without
Poland .87 .78 –.85 .39
suicide SDR and DDD/1000/day, data were grouped by country,
Portugal .96 –.84 .64 .84 and the logarithm of base n time was created.
Romania .81 .66 –.27 Finally, to compare the effect of changes in the use of
Slovakia .97 .54 –.43 .95 antidepressants on suicide rates between two time periods,
Slovenia .80 –.95 –.78 1.00 1980–1994 and 1995–2009, an analysis was performed with an
AR adjusted model, using DDD/1000/day as the independent
Spain .92 –.87 –.78 .88
variable and testing the interaction of DDD/1000/day and time
Sweden .85 .52 .54 –.21
period. Significance was set at p#0.05 (two tailed). This
Switzerland .94 .46 –1.00 .20 demarcation of 1994, was chosen because it is the point where
UK .90 .82 –.92 –.78 SSRIs started to become available and so, it was expected to mark
an acceleration in the increase of DDD/1000/day over the
doi:10.1371/journal.pone.0066455.t004
subsequent 15 years.
Statistical analysis was done using SPSS software, version 17.0.
where Yi is an ni61 vector of ni observations on the i-th subject; b
is a p61 vector of known, fixed, population parameters; Xi is an Ethics Statement
ni6p known, constant design matrix for the i-th subject; and di is a These data are publicly accessible, with the exception of IMS
q61 vector of unknown, random individual parameters. The Health, and are aggregated at the population-level. Individual-
random parameters are subject-specific but the vector size is the level information, for instance on individual patients, was
same from subject to subject; Zi is an ni6q known, constant design unobtainable and therefore all data were analyzed anonymously
matrix for the i-th subject corresponding to the random effects di; without any privacy or confidentiality concerns. The Ethical
and ei is an ni61 vector of random errors terms. Commission of the Faculdade de Ciências Médicas of Universi-
The GLMM is unstructured in relation to time frame, since the dade Nova de Lisboa (medical institutional review board) where
series of available years of data vary from country to country. The the two first authors are affiliated considers no review is needed if
country is taken as the subject and a random effect, with DDD/ the data are anonymous and administrative.
1000/day, GDP, alcohol, unemployment and divorce as fixed
effects, year as a repeating variable and suicide SDR as the
Results
dependent variable. The year of data observation was not
considered a fixed effect because of its anticipated strong Trends in the Use of Antidepressants
explanatory power for variations in suicide observations, which On average there was 15 years of antidepressant utilisation data
would prevent analysis of the role of other variables. This is also available in the 29 countries. There were marked differences: those
the reason we did not use time series. countries with six or fewer years observations, with the exception
In line with previous work looking at temporal patterns in of Switzerland, were all countries that have joined the EU since
fluoxetine prescribing and suicide rates in the US [54], co-variance 2004 (Bulgaria, Latvia, Poland, Slovenia and Romania) or are in
analyses were performed according to first order auto-regressive the process of joining (Croatia).

PLOS ONE | www.plosone.org 9 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

Table 6. Poisson distribution regression with SDR suicide rate as outcome.

Regression
coefficient SE 95% Wald confidence interval p-value
Lower Upper

DDD/1000/day –.00018 .00004 –.00026 –.00010 ,.001

doi:10.1371/journal.pone.0066455.t006

Data on the use of antidepressants are presented in Tables 1 and and Romania but in all cases SDR rates were below the mean rate
2. Overall there has been an increase of 40.33 units DDD/1000/ for all 29 countries. The highest rates of reduction were seen in
day in the study period, equal to the weighted-average difference Denmark, Estonia, Germany, Hungary, Slovakia, Slovenia and
between the first and last years of DDD/1000/day in each Switzerland.
country, from an average of 13.69 to 43.91 in the first and last year
of availability respectively. Our data indicates continuous growth Correlation between Suicide and Utilization of
in the use of antidepressants over time, albeit varying across Antidepressants
countries, but with an average growth per annum of 19.83% in
As Table 2 indicates, countries were grouped in quartiles by the
DDD/1000/per day. The lowest rates of annual growth of just 3%
availability of annual data on DDD/1000/day and SDR for
were seen in the Netherlands and Switzerland followed by
suicides. A non-significant correlation was observable in countries
Bulgaria, France and Luxembourg (all 5%), with the highest
with 6 or less years of both DDD and SDR data (r = .28; NS), they
growth rate of 59% seen in Finland followed by the Czech
had the highest mean SDR suicide rate and the lowest DDD/
Republic (41%), Slovakia (40%) and Sweden (34%).
1000/year for the last five year period, but covered less than a fifth
The latest available five year data indicate that the use of
of the population under study and less than a quarter of suicides in
antidepressants varies markedly from just 4.02 DDD/1000 per
the last available year. For all other quartiles there is an inverse
day in Romania, 5.59 in Latvia and 6.03 in Bulgaria, to as much
statistically significant correlation, with an increasing use of
as 68.50 in Denmark, 70.09 in Sweden and 95.16 in Iceland.
There was an average DDD/1000/per day of 40.01 across all antidepressants and greater reductions in the suicide SDR.
countries. Figures 1–4 plot suicide SDR and the use of antidepressants for
the same country quartiles. In almost all countries, an increase of
DDD/1000/day seems to correspond with a decrease in suicide
Suicide Trends
SDR, although in countries where suicide rates are already low,
Suicide trends are presented in Tables 1 and 2. Over the study
antidepressants appear to have less impact. At first sight, the
period, SDR rates for suicide decreased by a weighted average of
notable exceptions are Iceland (Figure 1), Portugal (Figure 2), and
6.16, from an average SDR of 19.06 for the first year available to
Luxembourg (Figure 3) whereas for the countries in Figure 4 the
12.93 in the last year available, notwithstanding variations in the
paramount characteristic is the lack of antidepressant utilization
years of data available in countries. The mean decrease in the
SDR rate was 0.81%. There was little difference in the mean SDR data.
rate for the last five years of data at 12.94.
Marked differences remained in suicide SDR rates across Association between Suicide and Use of Antidepressants
Europe, yet there was also a high degree of consistency in those with GDP, Alcohol, Unemployment and Divorce
countries with the lowest and highest rates of suicide over the study We assessed the magnitude and direction of correlations in each
period. For the initial years of observation the highest rates were individual country in an exploratory analysis with Pearson’s
seen in Hungary (44.54), Estonia (36.74) and Lithuania (35.16), correlation coefficients. As Table 3 indicates, an inverse correla-
with the lowest rates seen in Greece (3.2), Spain (4.69) and Italy tion was observable in all countries between SDR for suicide and
(7.15). For the last year of observation the highest SDR rates were DDD/1000/day, with the exception of Portugal. There was also
seen in Lithuania (31.47), Hungary (21.79) and Latvia (20.7). an inverse correlation with GDP, with the exceptions of Ireland,
Similarly the lowest rates were seen in Greece (3.02), Italy (5.39) Poland and Spain. As Table 4 indicates there is also a consistent
and Spain (6.34). direction and magnitude of correlation between DDD/1000/day
Only Poland, Spain and Ireland had annual suicide rates higher and GDP in all 29 countries. No strong patterns are seen in either
in the last year compared with the initial year of observation. Tables 3 or 4 in respect of SDR and alcohol, unemployment or
There was also little change in Greece, Iceland, Norway, Portugal divorce.

Table 7. Model estimates with total suicide rate as outcome Effects of Use of Antidepressants, GDP, Alcohol
and the interaction with period. Consumption, Unemployment and Divorce on Suicide
The GLMM with DDD/1000/day as the only predictor of
suicide rates (model 1) and with all independent variables as
Regression predictors (model 2) are displayed in Table 5. Model 1 revealed a
coefficient SE T p-value significant effect of DDD/1000/per day (p = 0.002). In Model 2
DDD/1000/day –.479 .066 –7.232 ,.001 DDD/1000/day still presents a significant effect on SDR suicide
Period –6.991 .714 –9.791 ,.001 when adjusting for the other independent variables (p = 0.001). In
this model an increase of one unit in DDD/1000/day, adjusted to
DDD/1000/day6Period .413 .064 6.460 ,.001
the remnant independent variables, diminishes suicide SDR by
doi:10.1371/journal.pone.0066455.t007 0.088 units. Divorce also appears to have a significant effect on

PLOS ONE | www.plosone.org 10 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

SDR for suicide in this model (p = .007). An increase of one unit in Portugal is the only country where there is a positive correlation
the divorce rate, increases the SDR by 1.273. between DDD/1000/day and suicide SDR, considering the actual
In Table 6 we present a Poisson distribution regression model large utilization of antidepressants. This can possibly be explain-
where DDD/1000/day presents a significant, though only modest, able by the lack of precision of suicide register and over-estimation
effect on suicide SDR. The model has an adequate goodness of fit of undetermined violent deaths concealing suicides [41,59]. The
with a chi-square (27) = 1.118,83 inferior to the critical value for small populations of both Luxembourg and Iceland and therefore
a = .001 and the Omnibus Test, a likelihood chi-squared test, is the small numbers of suicides recorded, probably accounts for the
statistically significant. lack of any relationship with high antidepressant utilization. In
We do not present a Poisson model with the other independent other countries, such as Greece or Ireland, there was a sharp
variables because of a high level of missing data after cleaning the increase in DDD/1000/day and little change in suicide was
database for empty cells: the number of observations was 870 in apparent though a clear negative correlation was present. Perhaps
the GLMM but there are only 429 observations for suicide SDR in Greece, with an already very low suicide rate along the period a
and DDD/1000/year with Poisson regression. These would only ceiling effect is present whereas in Ireland suicide is more frequent
be 311 observations when considering all independent variables. in young and middle aged men, who typically present an
unfavourable help-seeking behaviour [60].
Effects of Antidepressant Utilization on Suicide Rates in Another key result was the demonstration of how the impact of
Different Time Periods antidepressant utilization on suicide changes as more annual
Table 7 presents the results of our analysis comparing the information becomes available on both DDD/1000/day and SDR
effect of DDD/1000/day on suicide rates for two different time suicide: a longer series of data means a stronger correlation
periods. This is a mixed model not structured within time, with between lower SDR for suicide and higher DDD/1000/day. This
countries with random effect, with DDD/1000/day and two suggest that those countries in this study with less than 6-years of
periods (1980–1994 and 1995–2009) as fixed effects, with an both types of data that did not show a significant inverse
interaction between DDD/1000/day and period, and year as a correlation between these indicators may well in future demon-
repeating factor. An effect modification between the two periods strate this finding as more data becomes available. Previous
is observable since the interaction is very significant (p,0.001). published studies with negative results should be analysed in view
In the period 1980–1994, the effect of DDD/1000/day on of this contingency.
suicide SDR was –0.479, and –0.066 (–0.479+0.413) in the A third important result was that two different regression
second time period. Thus there was a negative impact on suicide statistics confirmed that DDD/1000/day is an explanatory factor
rates in both time periods, though this was much more for suicide SDR, notwithstanding that the Poisson regression
pronounced for 1980–1994. meant reducing analysed observations from 870 to 426, therefore
reducing accuracy in relation to the GLMM. Though we cannot
assume a causal effect, when adjusting for other independent
Discussion
variables, adequate GLMM modelling makes DDD/1000/day an
Main Results explanatory factor for changes in the suicide SDR: a one unit
Our analysis indicated that for 15 years of data on average for increase of DDD/1000/day seems to diminish the suicide SDR by
the 29 countries included in our study, the use of antidepressants 0.088 units.
increased on average by 19.83% per year. By the end of this time A clear covariance is observable, at the country level, between
period for the whole study area there had been an average increase potentially explanatory suicide factors such as antidepressant
of 40.33 units of DDD/1000/day. Over a mean period of 28 utilisation, GDP, alcohol consumption, unemployment and
years, the overall SDR for suicide decreased at a rate of 0.81% per divorce, suggesting a differential impact on countries and implying
year, corresponding to a reduction of 6.16 in the SDR rate for GLMM and Poisson regressions were appropriate models for
suicide. There was a strong inverse correlation between these estimation.
trends. The fourth key result was the demonstration that DDD/1000/
How much of this increase of antidepressant use is needed to day had an effect on SDR for suicide both between 1980 and 1994
reduce the rate of suicide? A classical study from Sweden by and in the subsequent 15 years. With increasing DDD/1000/year,
Isacsson [38] covering the period from 1978 to 1996 found that suicide-related SDR was expected to decrease. However, this
suicide rates decreased consistent with a hypothesis that if the use decreasing trend decelerates over time: the analysis suggests that in
of antidepressant medication increased five-fold, suicide rates the first period, where in most cases high rates of suicide SDRs
would decrease by 25% assuming that depression treatment were seen, less of an increase in the use of antidepressants would
prevalence was approximately 1% and point prevalence of major be necessary to reduce suicide SDR whereas in the subsequent
depression was 5%. We have also already noted that in the large period, when suicide SDRs had become lower in most countries, a
multi-national studies by Ludwig and Marcotte covering the much higher rate of antidepressant utilization would be necessary
period from 1980 to 2000 [31,32], that an increase in SSRI sales to further reduce suicide SDR. Suicide-related SDR continued to
of one pill per capita is associated with a decline in suicide rates of decline in the second 15-year period, albeit at a much reduced
between 2.5% and 5% in different groups of countries around the rate. More importantly, nevertheless, it suggests that antidepres-
world. sant utilization DDD/1000/day increase had an important effect
The total population of the 29 European countries in our study in suicide SDR from the start, when suicide SDR started to lower
was 514.1 million in 2009. If 10 DDD/1000/day corresponds in Europe. This rebuffs most criticisms and scepticism on
approximately to 1% of population treatment point prevalence observable antidepressant effects on suicide decrease, usually
[58], our data suggests that there could have been an increase in stating that suicide had already started to decrease before
treatment for depression of 4% of this population correlating with antidepressant utilization exploded, in the nineties, therefore
a saving of 31.670 deaths by suicide in the last year covered, denying earlier generation antidepressant effects.
equivalent to 650 people treated for each life saved, per year.

PLOS ONE | www.plosone.org 11 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

Previous Studies far, which further complicates comparing results. Most studies
This study is similar to those of Ludwig and Marcotte [31,32] present differing correlations, linear regressions, Poisson regres-
since the observation unit is the country. Nevertheless, it presents sions and time series; we avoid this problem in our multi-country
several characteristics that might be seen as advantageous in analysis.
relation to the generalisability of results. First, it draws on a more
homogeneous set of countries, albeit with some substantial Ecological Design Considerations
differences in GDP, infrastructures and historical development of Because this is an ecological study, we emphasise that we cannot
countries that had formerly been part of the Soviet bloc prior to depict causal links and therefore these results must be interpreted
1990. Second, it covers more consecutive years of effects versus with great caution. Nevertheless, as argued earlier, we believe that
each 5 year calculation (average 15 years against 10 years, since in there is a case for this study design because there is a need to
Ludwig and Marcotte’s studies for 1980–1990 SSRI pills are validate the effectiveness and potential cost effectiveness of
extrapolated), the almost global extension of antidepressants (and antidepressants as an intervention for suicide prevention. As
not only of SSRIs), the use of DDD/1000/day, which is a measure suicide is a comparatively rare event, it would be very difficult to
of antidepressant utilization independent of national regulations, study in a controlled trial, not to mention any of the potential
costs and commercial specificities, allowing for comparisons, and ethical concerns that might arise in trial design [38,39]. Trials are
that can be related to utilization needs and treated prevalence. As also unlikely to be of a sufficiently long timeframe for adequate
previously explained, one DDD may be sufficient for one person- analysis of impacts on suicide, where data covering many years is
day of adequate treatment, and 10 DDDs per 1000 inhabitants per required. Thus ecological studies still have a place in the
day is therefore considered to represent approximately a 1% point evaluation of some interventions, namely public health interven-
treated prevalence [58] though we cannot assume users take tions. As in this case, conclusions drawn do not uncover causal
medication as prescribed. links but must be taken in view of knowledge available, face
This study has also many advantages over single national validity and plain common sense.
studies, though it does not substitute for them: it controls for the The debate on suicide-antidepressant trends at the public health
variability of factors that affect suicide rates at the country level level might be seen as disproportionate since the methodological
and it has more explanatory power on the role of antidepressants and interpretation problems that arise are present in all ecological
on suicide rates because of the number of observations involved. studies exploring any kind of correlations. For instance, there have
We also believe it purports more power than reviews and meta- been positive developments in mental health services policy,
analysis produced so far because of its consistency: close
delivery and provision in Europe in recent decades [61] but the
geographic and socio-political context in the last 10 years, use of
impact of mental health services provision on suicide rates assessed
available annual data series, inclusion of all families of antide-
through ecological studies has produced mixed results [47,48,62–
pressants and use of DDD/1000/day.
68]. Nevertheless, it would hardly be arguable not to implement
In fact, single studies and reviews [29] have used several
good practices and optimise mental health services when resources
definitions of antidepressant utilization, including costs, number of
are available.
packages or pills sold, number of prescriptions issued, defined daily
dosage (DDD) and defined daily doses per thousand individuals
per day (DDD/1000/day). These discrepancies hinder compara- Confounding Effects
bility and introduce probable sources of bias because drug costs, The association between alcohol consumption, unemployment,
pills dosages, quantities of pills per package and prescriptions, and divorce and suicide was inconsistent across the different
might oscillate longitudinally because of external, regulatory and countries in our analysis. How can an inverse correlation between
commercial reasons. Using DDD/1000/day represents a stable these variables and the suicide rate in some countries be
variable for the estimation of the exposure to drugs and the interpreted, when a positive correlation was expected?
proportion of the population that may receive treatment with a Considering alcohol consumption, even though patterns of
particular drug on a daily basis [45]. alcohol use differ from culture to culture, it is known that alcohol
Another source of bias in some previous studies has been abuse can contribute to an increased risk of suicidal behavior
associated with a major focus on SSRIs, and not on the analysis of [24,69]. Therefore, it would be expected that a reduction in
the use of the whole class of antidepressants, including SNRIs, suicide would be accompanied by a reduction in alcohol intake
atypicals and tryciclics. Patterns in the use of these drugs might [70]. Notwithstanding, in Hungary, between 1990 and 1998
vary considerably across countries. alcohol sales increased by 25% and suicide rates dropped by 20%
Moreover, these studies present substantial differences in the [48] and in a Hungarian suicide prevention project, the
periods of time that are analysed, both for antidepressant use and intervention region had a higher alcohol-related death rate both
recorded suicides, from as low as 2 years to as high as 30 years for before and after the program compared with the control region.
antidepressant use utilization series, and a similar but slightly Moreover, there was a decrease in alcohol-related deaths over time
better pattern for suicide time series. It is likely that these in both regions, and the expected improvement in the intervention
variations will have had an impact on results, as well as making region was not confirmed [71].
meaningful comparison difficult. Also, in a previous study performed in 1980–1982 in Portugal,
The extent to which data series are available across countries for the inverse association between suicide rates and alcohol cirrhosis
the same time periods can influence correlation results strikingly. deaths had a distinctive regional distribution: whereas in the north,
This is quite important to assess previous and future studies: results alcohol uptake and cirrhosis death was greater than in the south,
are far more reliable when longer yearly time series are present. suicide rates were much greater in the south [22], suggesting that
Our study presents an average of 15 years of both annual suicide alcohol addiction is on many occasions either a depression
and antidepressant utilization data, the largest figure to our equivalent or a self-medication strategy [24]. Overall, cultural
knowledge, albeit hampered by inconsistent time series data for differences are also important in explaining variation in the
antidepressant utilization across countries. There is also consider- associations across countries.
able variability of statistical procedures within studies published so

PLOS ONE | www.plosone.org 12 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

Explanations for Increased Prescribing of Antidepressants Eastern Europe countries where undetermined deaths are
Along with the perceptions of newer antidepressants of being considerable. Future analysis should consider ‘probable suicide’
safer and easier clinically to manage, the past 30 years has been i.e. the sum of registered suicides and undetermined violent deaths.
characterized by depression awareness campaigns and more Another limitation of our analysis is a lack of data on the use of
extensive medical training concerning depression and suicide psychological therapies, alongside or as an alternative to the use of
[48]. Training for primary care and other medical personnel antidepressants in treating depression and related disorders, and
concerning depression and suicide risk management has been a therefore potentially contributing to the prevention of suicides.
core component of many suicide prevention programmes in Many of these limitations can only addressed through improve-
Europe since the implementation of the Gotland study [7,72–75]. ments in epidemiological datasets, including more information
Moreover, there is increased awareness of the extent of the impact about treatment pathways; surveys might also be considered to
of poor mental health and the increased need for treatment and better obtain data on the use of antidepressants and other
support in Europe [76,77], varying from country to country [78], medications, as well as other psychosocial therapies among specific
that may also contribute to this increase in antidepressant population sub-groups.
prescribing. Finally, there has been some increased funding for Nonetheless, despite these limitations and our caution over the
mental health systems during the observation period. This may interpretation of our findings, the outcomes of the present study
have helped make antidepressants, along with other treatments for underline the need to better optimise the appropriate use of
depression more accessible [79]. antidepressants as part of routine care, given that many people
who may benefit from their use do not receive them, while
Limitations conversely other individuals are inappropriately taking such
There are a number of limitations to this analysis. Utilisation is medications. Whether research projects, such as OSPI-Europe
only a proxy for rates of what occurs at the individual patient level: that seek to foster a better quality of care, starting at the primary
we do not know if people take the medication they obtain, or if care level, focusing on improved awareness of depression and risk
they are taking less or more than the standard DDD. The real of suicide, appropriate antidepressant and other treatment
rates of treatment of depression in Europe could conceivably be prescribing and monitoring, might produce such an effect will
lower than the high DDD/1000/day would suggest, taking in require empirical demonstration.
account the multiple indications of antidepressants, frequent use of
higher dosages than the DDD, non-compliance, and co-therapy Acknowledgments
with a second antidepressant [38]. Nor do we know the proportion
of individuals taking these medications that complete suicide. In We are especially grateful to IMS Health, Inc. (http://www.imshealth.
addition, we do not know the gender and age distribution of com) for providing data on drug sales.
antidepressant use, and so have not attempted to look at the We are also grateful to Per Zahl (Department of Suicide Research and
Prevention, Division of Mental Health, Norwegian Institute of Public
impacts of antidepressants on suicide rates by age or gender. We
Health, Oslo, Norway) and Goran Isacsson (Karolinska Institutet,
also do not know the distribution of utilisation in relation to Department of Clinical Neuroscience, Centre for Psychiatry Research,
severity of depression and anxiety disorders although some surveys Stockholm, Sweden) for gently sharing data on antidepressant utilization in
in Europe suggest the gap is greater in the lower end of severity, Nordic countries, and Pedro Aguiar (Epidemiology and Statistics, Escola
and a recent meta-analysis suggests the value of antidepressants for Nacional de Saúde Pública, Universidade Nova de Lisboa and Eurotrails
light and moderate as well as severe depression [80]. Scientific Consultants, Portugal) and Steve Marcus (School of Social Work
It should also be acknowledged that antidepressants are and Policy, University of Pennsylvania, USA) for their invaluable statistical
prescribed for other mental health problems in addition to advice. Marco Sarchiapone (University of Molise, Health Science
depression (e.g. anxiety disorders, anorexia and bulimia nervosa, Department, Italy) provided insights that we are grateful to have received.
ADHD), as well as for physical health problems (e.g. migraine
headaches, fibromyalgia, chronic pain) [81–83]. We also know Author Contributions
that poor physical health can be a risk factor for suicide [84]. Conceived and designed the experiments: RG SQ. Performed the
Our analysis is also limited by focusing only on completed experiments: RG SQ JC. Analyzed the data: RG SQ DM EA CVA CC
suicides, but a proportion of undetermined deaths will also be due AV PV JC UH. Contributed reagents/materials/analysis tools: RG SQ
to suicide; potentially including these data in our analysis might DM EA CVA CC AV PV JC UH. Wrote the paper: RG SQ DM EA CVA
impact on findings, particularly in countries such as Portugal and CC AV PV JC UH. Obtained permission for use of cell line: RG SQ.

References
1. Organisation for Economic Cooperation and Development (2012) Health at a Europe (OSPI Europe): an evidence-based multi-level approach. BMC Public
glance : Europe 2012. Paris: OECD Publishing. 154 p. Health 9: 428.
2. Carta MG, Kovess V, Hardoy MC, Brugha T, Fryers T, et al. (2004) 8. Fournier JC, DeRubeis RJ, Hollon SD, Dimidjian S, Amsterdam JD, et al.
Psychosocial wellbeing and psychiatric care in the European Communities: (2010) Antidepressant drug effects and depression severity: a patient-level meta-
analysis of macro indicators. Soc Psychiatry Psychiatr Epidemiol 39: 883–892. analysis. JAMA 303: 47–53.
3. Bertolote JM, Fleischmann A, de Leo D, Wasserman D (2003) Suicide and 9. Teicher MH, Glod CA, Cole JO (1993) Antidepressant drugs and the emergence
mental disorders: do we know enough? Br J Psychiatry 183: 382–383. of suicidal tendencies. Drug Saf 8: 186–212.
4. Fernández A, Haro JM, Martinez-Alonso M, Demyttenaere K, Brugha TS, et 10. Teicher MH, Glod C, Cole JO (1990) Emergence of intense suicidal
al. (2007) Treatment adequacy for anxiety and depressive disorders in six preoccupation during fluoxetine treatment. Am J Psychiatry 147: 207–210.
European countries. Br J Psychiatry 190: 172–173. 11. Hammad TA, Laughren T, Racoosin J (2006) Suicidality in pediatric patients
5. Mojtabai R, Olfson M (2008) National patterns in antidepressant treatment by treated with antidepressant drugs. Arch Gen Psychiatry 63: 332–339.
psychiatrists and general medical providers: results from the national 12. Hirschfeld RM (2000) Suicide and antidepressant treatment. Arch Gen
comorbidity survey replication. J Clin Psychiatry 69: 1064–1074. Psychiatry 57: 325–326.
6. van der Feltz-Cornelis CM, Sarchiapone M, Postuvan V, Volker D, Roskar S, et 13. Quitkin FM, Rabkin JG, Gerald J, Davis JM, Klein DF (2000) Validity of clinical
al. (2011) Best Practice Elements of Multilevel Suicide Prevention Strategies. trials of antidepressants. Am J Psychiatry 157: 327.
Crisis: 1–15. 14. Möller H-J (2006) Is there evidence for negative effects of antidepressants on
7. Hegerl U, Wittenburg L, Arensman E, Van Audenhove C, Coyne JC, et al. suicidality in depressive patients? A systematic review. Eur Arch Psychiatry Clin
(2009) Optimizing suicide prevention programs and their implementation in Neurosci 256: 476–496.

PLOS ONE | www.plosone.org 13 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

15. Möller H-J, Baldwin DS, Goodwin G, Kasper S, Okasha A, et al. (2008) Do 50. Viola R, Benko R, Nagy G, Soós G (2008) National trend of antidepressant
SSRIs or antidepressants in general increase suicidality? WPA Section on consumption and its impact on suicide rate in Hungary. Pharmacoepidemiol
Pharmacopsychiatry: consensus statement. Eur Arch Psychiatry Clin Neurosci Drug Saf 17: 401–405.
258 Suppl 3: 3–23. 51. Ciuna A, Andretta M, Corbari L, Levi D, Mirandola M, et al. (2004) Are we
16. Parker G (2009) Antidepressants on trial: how valid is the evidence? going to increase the use of antidepressants up to that of benzodiazepines?
Br J Psychiatry 194: 1–3. Eur J Clin Pharmacol 60: 629–634.
17. Bridge JA, Iyengar S, Salary CB, Barbe RP, Birmaher B, et al. (2007) Clinical 52. Zahl P-H, De Leo D, Ekeberg Ø, Hjelmeland H, Dieserud G (2010) The
response and risk for reported suicidal ideation and suicide attempts in pediatric relationship between sales of SSRI, TCA and suicide rates in the Nordic
antidepressant treatment: a meta-analysis of randomized controlled trials. JAMA countries. BMC Psychiatry 10: 62.
297: 1683–1696. 53. Edwards LJ (2000) Modern statistical techniques for the analysis of longitudinal
18. Gibbons RD, Brown CH, Hur K, Marcus SM, Bhaumik DK, et al. (2007) data in biomedical research. Pediatric pulmonology 30: 330–344.
Relationship between antidepressants and suicide attempts: an analysis of the 54. Milane MS, Suchard MA, Wong M-L, Licinio J (2006) Modeling of the
Veterans Health Administration data sets. Am J Psychiatry 164: 1044–1049. temporal patterns of fluoxetine prescriptions and suicide rates in the United
19. Bridge JA, Axelson DA (2008) The contribution of pharmacoepidemiology to States. PLoS Med 3: e190.
the antidepressant-suicidality debate in children and adolescents. Int Rev 55. Gibbons RD, Hur K, Bhaumik DK, Mann JJ (2005) The relationship between
Psychiatry 20: 209–214. antidepressant medication use and rate of suicide. Arch Gen Psychiatry 62: 165–
20. Rehkopf DH, Buka SL (2006) The association between suicide and the socio- 172.
economic characteristics of geographical areas: a systematic review. Psychol Med 56. Gibbons RD, Hur K, Bhaumik DK, Mann JJ (2006) The relationship between
36: 145–157. antidepressant prescription rates and rate of early adolescent suicide.
21. Zhang J, Ma J, Jia C, Sun J, Guo X, et al. (2009) Economic growth and suicide Am J Psychiatry 163: 1898–1904.
rate changes: A case in China from 1982 to 2005. Eur Psychiatry. 57. Helgason T, Tómasson H, Zoega T (2004) Antidepressants and public health in
22. Skog OJ, Teixeira Z, Barrias J, Moreira R (1995) Alcohol and suicide–the Iceland. Time series analysis of national data. Br J Psychiatry 184: 157–162.
Portuguese experience. Addiction 90: 1053–1061. 58. Isacsson G, Holmgren P, Wasserman D, Bergman U (1994) Use of
23. Brådvik L, Mattisson C, Bogren M, Nettelbladt P (2010) Mental disorders in antidepressants among people committing suicide in Sweden. BMJ 308: 506–
suicide and undetermined death in the Lundby Study. The contribution of 509.
severe depression and alcohol dependence. Arch Suicide Res 14: 266–275. 59. Varnik P, Sisask M, Varnik A, Arensman E, Van Audenhove C, et al. (2012)
24. Sher L (2006) Alcohol consumption and suicide. QJM 99: 57–61. Validity of suicide statistics in Europe in relation to undetermined deaths:
25. Barth A, Sögner L, Gnambs T, Kundi M, Reiner A, et al. (2011) Socioeconomic developing the 2–20 benchmark. Inj Prev 18: 321–325.
factors and suicide: an analysis of 18 industrialized countries for the years 1983 60. Möller-Leimkühler AM (2000) [Men and depression: gender-related help-
through 2007. J Occup Environ Med 53: 313–317. seeking behavior]. Fortschr Neurol Psychiatr 68: 489–495.
26. Stack S (2000) Suicide: a 15-year review of the sociological literature. Part II: 61. Knapp M, McDaid D, Mossialos E, Thornicroft G, editors (2007) Mental
modernization and social integration perspectives. Suicide Life Threat Behav 30: Health Policy and Practice Across Europe. Buckingham: Open University Press.
163–176.
62. Rihmer Z, Belsö N, Kalmár S (2001) Antidepressants and suicide prevention in
27. Lewis G, Sloggett A (1998) Suicide, deprivation, and unemployment: record
Hungary. Acta Psychiatr Scand 103: 238–239.
linkage study. BMJ 317: 1283–1286.
63. Johannessen HA, Dieserud G, Claussen B, Zahl P-H (2011) Changes in mental
28. Blakely TA, Collings SCD, Atkinson J (2003) Unemployment and suicide.
health services and suicide mortality in Norway: an ecological study. BMC
Evidence for a causal association? J Epidemiol Community Health 57: 594–600.
Health Serv Res 11: 68.
29. Baldessarini RJ, Tondo L, Strombom IM, Dominguez S, Fawcett J, et al. (2007)
64. Pirkola S, Sund R, Sailas E, Wahlbeck K (2009) Community mental-health
Ecological studies of antidepressant treatment and suicidal risks. Harv Rev
services and suicide rate in Finland: a nationwide small-area analysis. Lancet
Psychiatry 15: 133–145.
373: 147–153.
30. Wheeler BW, Metcalfe C, Martin RM, Gunnell D (2009) International impacts
65. Pirkola S, Sohlman B, Heilä H, Wahlbeck K (2007) Reductions in postdischarge
of regulatory action to limit antidepressant prescribing on rates of suicide in
young people. Pharmacoepidemiol Drug Saf 18: 579–588. suicide after deinstitutionalization and decentralization: a nationwide register
study in Finland. Psychiatr Serv 58: 221–226.
31. Ludwig J, Marcotte DE (2005) Anti-depressants, suicide, and drug regulation.
J Policy Anal Manage 24: 249–272. 66. Desai RA, Dausey DJ, Rosenheck RA (2005) Mental health service delivery and
32. Ludwig J, Marcotte DE, Norberg K (2009) Anti-depressants and suicide. J Health suicide risk: the role of individual patient and facility factors. Am J Psychiatry
Econ 28: 659–676. 162: 311–318.
33. Schwartz S (1994) The fallacy of the ecological fallacy: the potential misuse of a 67. Lewis G, Appleby L, Jarman B (1994) Suicide and psychiatric services. Lancet
concept and the consequences. Am J Public Health 84: 819–824. 344: 822.
34. Morgenstern H (1982) Uses of ecologic analysis in epidemiologic research. 68. Shah A, Bhandarkar R, Bhatia G (2010) The relationship between general
Am J Public Health 72: 1336–1344. population suicide rates and mental health funding, service provision and
35. Van Praag HM (2003) A stubborn behaviour: the failure of antidepressants to national policy: a cross-national study. Int J Soc Psychiatry 56: 448–453.
reduce suicide rates. World J Biol Psychiatry 4: 184–191. 69. Pompili M, Serafini G, Innamorati M, Dominici G, Ferracuti S, et al. (2010)
36. van Praag HM (2002) Why has the antidepressant era not shown a significant Suicidal behavior and alcohol abuse. Int J Environ Res Public Health 7: 1392–
drop in suicide rates? Crisis 23: 77–82. 1431.
37. Isacsson G, Bergman U, Rich CL (1996) Epidemiological data suggest 70. Wasserman D, Värnik A (1998) Suicide-preventive effects of perestroika in the
antidepressants reduce suicide risk among depressives. J Affect Disord 41: 1–8. former USSR: the role of alcohol restriction. Acta Psychiatr Scand Suppl 394:
38. Isacsson G (2000) Suicide prevention–a medical breakthrough? Acta Psychiatr 1–4.
Scand 102: 113–117. 71. Szanto K, Kalmar S, Hendin H, Rihmer Z, Mann JJ (2007) A suicide prevention
39. Po W, Li A (1993) Fluoxetine and suicide: Meta-analysis and Monte-Carlo program in a region with a very high suicide rate. Arch Gen Psychiatry 64: 914–
simulations. Pharmacoepidemiology and Drug Safety 2: 79–84. 920.
40. WHO Regional Office for Europe (2011) European mortality database (MDB). 72. Rihmer Z, Rutz W, Pihlgren H (1995) Depression and suicide on Gotland. An
Copenhagen, Denmark. intensive study of all suicides before and after a depression-training programme
41. Värnik P, Sisask M, Värnik A, Laido Z, Meise U, et al. (2010) Suicide for general practitioners. J Affect Disord 35: 147–152.
registration in eight European countries: A qualitative analysis of procedures and 73. Althaus D, Hegerl U (2003) [Concept and results of an awareness campaign: the
practices. Forensic Sci Int 202: 86–92. ‘‘Nuremberg Alliance against Depression’’]. MMW Fortschr Med 145: 42–44.
42. WHO Regional Office for Europe (2011) European health for all database 74. Mann JJ, Apter A, Bertolote J, Beautrais A, Currier D, et al. (2005) Suicide
(HFA-DB). Copenhagen, Denmark. prevention strategies: a systematic review. JAMA 294: 2064–2074.
43. World Health Organization (2007). WHO Global Information System on 75. Hegerl U, Wittmann M, Arensman E, Van Audenhove C, Bouleau JH, et al.
Alcohol and Health (GISAH). (2007) The ‘European Alliance Against Depression (EAAD)’: A multifaceted,
44. Organisation for Economic Cooperation and Development (2009) Society at a community-based action programme against depression and suicidality.
Glance 2009: OECD Social Indicators. World J Biol Psychiatry [epub ahead].
45. WHO Collaborating Centre for Drug Statistics Methodology WHO-CCFDSM 76. Wittchen H-U, Jacobi F (2005) Size and burden of mental disorders in Europe–a
(2009) Guidelines for ATC classification and DDD assignment, 2010. Oslo: critical review and appraisal of 27 studies. Eur Neuropsychopharmacol 15: 357–
WHO Collaborating Centre for Drug Statistics Methodology. 376.
46. OECD OECD Health Data: Pharmaceutical market. 77. Wittchen HU, Jacobi F, Rehm J, Gustavsson A, Svensson M, et al. (2011) The
47. Kapusta ND, Niederkrotenthaler T, Etzersdorfer E, Voracek M, Dervic K, et al. size and burden of mental disorders and other disorders of the brain in Europe
(2009) Influence of psychotherapist density and antidepressant sales on suicide 2010. Eur Neuropsychopharmacol 21: 655–679.
rates. Acta Psychiatr Scand 119: 236–242. 78. King M, Nazareth I, Levy G, Walker C, Morris R, et al. (2008) Prevalence of
48. Rihmer Z, Appleby L, Rihmer A, Belso N (2000) Decreasing suicide in common mental disorders in general practice attendees across Europe.
Hungary. Br J Psychiatry 177: 84. Br J Psychiatry 192: 362–367.
49. Kadusevicius E, Mikucionyte L, Maciulaitis R, Milvidaite I, Sveikata A (2006) 79. Knapp M, McDaid D, Amaddeo F, Constantopoulos A, Oliveira MD, et al.
Trends in the consumption of antidepressant drugs in Lithuania in 2002–2004. (2007) Financing mental health care in Europe. Journal of mental health 16:
Medicina (Kaunas) 42: 1020–1029. 167–180.

PLOS ONE | www.plosone.org 14 June 2013 | Volume 8 | Issue 6 | e66455

 Antidepressant Utilization and Suicide in Europe

80. Gibbons RD, Hur K, Brown CH, Davis JM, Mann JJ (2012) Benefits From 82. Pomerantz JM, Finkelstein SN, Berndt ER, Poret AW, Walker LE, et al. (2004)
Antidepressants: Synthesis of 6-Week Patient-Level Outcomes From Double- Prescriber intent, off-label usage, and early discontinuation of antidepressants: a
blind Placebo-Controlled Randomized Trials of Fluoxetine and Venlafaxine. retrospective physician survey and data analysis. J Clin Psychiatry 65: 395–404.
Arch Gen Psychiatry 69: 572–579. 83. Stone KJ, Viera AJ, Parman CL (2003) Off-label applications for SSRIs. Am
81. Milea D, Verpillat P, Guelfucci F, Toumi M, Lamure M (2010) Prescription Fam Physician 68: 498–504.
patterns of antidepressants: findings from a US claims database. Curr Med Res 84. Hawton K, van Heeringen K (2009) Suicide. Lancet 373: 1372–1381.
Opin 26: 1343–1353.

PLOS ONE | www.plosone.org 15 June 2013 | Volume 8 | Issue 6 | e66455