Göran Isacsson, Department of Clinical Neuroscience, Division of Psychiatry, Karolinska Institute, S-141 86 Stockholm, Sweden. E-mail: Goran.Isacsson@ki.se
Objective: Ecological studies have demonstrated a substantial decrease in suicide in parallel with an increase in the use of antidepressants. Causality cannot, however, be inferred from such studies. The aim of this study was to test on the individual level the hypothesis that treatment with antidepressant medication has been a substantially contributing cause of the decrease in suicide.
Method: Time trends in the detection of antidepressants and five ‘control medications’ in the forensic toxicological screening of 16 937 suicides and 33 426 controls in Sweden 1995–2005.
Results: The expected number of antidepressant-positive suicides in 2005 was 409 if the hypothesis was true and 603 if it was false. The observed number in 2005 was 420. The control medications were detected to the extent that was expected if not preventing suicide.
Conclusion: The observed trend in the number of suicides with antidepressants was well predicted by the hypothesis that the increased use of antidepressants has been a substantially contributing cause of the decrease in suicide.
• The frequency at which antidepressants are detected in forensic toxicological screening of suicides depends on the use of antidepressants in the population and, importantly, whether or not they prevent suicide.
• Antidepressant medication was detected in the toxicology of suicides to the extent that was expected if they prevent suicide.
• Control medications were detected as expected if not preventing suicide, which validates the design of the study.
• An observational study cannot exclude the possibility of unknown confounding factors.
• Because of too few cases, the hypothesis could not be tested separately in young suicides.
There is accumulating evidence supporting the hypothesis that the treatment of depression with antidepressant medication prevents suicide (1–15). Most of this evidence consists of ecological studies from different populations, demonstrating that there are substantial decreases in suicide in parallel with an increasing use of antidepressants. Ecological studies cannot, however, establish causal relationships. To draw definite conclusions about causality, a randomized placebo-controlled trial would be ideal. Such a trial would demonstrate how many individuals in the treatment group and in the placebo group committed suicide. A randomized controlled trial on the effects on suicide of antidepressant treatment cannot, however, be performed for both ethical and practical reasons. Conclusions with regard to suicide can neither be drawn from studies of suicidal ideation and deliberate self-harm.
Individual-based naturalistic studies may therefore be the best choice in the investigation of a possible causal relationship between the treatment of depression and a decrease in the number of suicides. We have in several previous studies utilized information from a comprehensive Swedish national forensic toxicological database to compare frequencies of detection of different antidepressants in individuals who have committed suicide (16–19). Such a database can reveal which suicides were exposed to antidepressants at the time of suicide and which were unexposed. Exposure may approximate treatment. When comparing risks, however, a major problem is that the exposed individuals in the general population are not comparable with the unexposed. Instead of being randomized, exposed individuals are selected for depression. This obstacle can to some extent be overcome by comparing the exposed and unexposed populations over a period of time (i.e. the study of trends instead of rates).
In a previous study, when we compared the numbers of toxicological detections of different antidepressants in Swedish suicides between 1992 and 2000, we found that the respective detection rates correlated strongly with their relative use in the general population (19). The remaining variance was assumed to reflect an interaction between the drug and the risk of suicide. We found that the selective serotonin reuptake inhibitors (SSRIs) were less associated to suicide than other types of antidepressants, but the design of the study meant that we could not establish whether this should be interpreted as indicating that SSRIs reduced the risk of suicide more, or increased the risk less, than the other antidepressants. We noticed, however, that, over the 9-year study period, the annual proportion of suicides exposed to antidepressants increased little despite a four-fold increase in the use of antidepressants in the general population (Isacsson G, Holmgren A, Ahlner J, unpublished data). We interpreted this finding as suggesting that all antidepressants decrease the risk of suicide, because preventing suicides in the exposed part of the population will reduce the number of exposed suicides. The increased use of antidepressant medication in Sweden during 1992–2000 might thus explain the concomitant decrease in the number of suicide.
Aims of the study
We decided to extend the time series, mentioned above, to prospectively investigate the hypothesis that the increased use of antidepressant medication in Swedish during 1995–2005 may have been a substantially contributing cause of the concomitant decrease in suicide, by analyzing secular trends in the toxicological detection of antidepressants in individual suicides.
Material and methods
Forensic pathologists investigate apparently unnatural deaths in Sweden. Their investigation leads to a conclusion as to whether the cause of death was homicide, suicide, accident or natural. In some cases, the cause of death cannot be determined. The majority of these undetermined cases are, if investigated further, often found to be probable suicides, mostly because of intoxication (13). In epidemiological research, undetermined cases may preferably be considered as suicides (20). A routine procedure in the forensic investigation is to search for foreign substances in the body. All forensic toxicological analyses are performed at the same laboratory. In the analyses, around 200 substances are screened. All antidepressants can be detected at therapeutic concentration levels (16, 17). All analyses are entered into a database from which we have obtained toxicological data relating to all suicides during the period from 1992 to 2005 (n = 21 813). Undetermined cases were included. We observed that the annual number of suicides investigated with forensic toxicological analyses increased, according to official statistics, from 83% of the total number of suicides in Sweden in 1992 to 99% in 1995 (Diagram 1). Therefore, to avoid bias as a result of this time trend, we chose to exclude the years 1992–1994 from the study. The remaining number of suicides in 1995–2005 constituting the study group was 16 937 (mean age 50.6 years, 70.8% males). This restriction of the time did not affect the result in a way that might suggest different conclusions. There were minor deviations from the official statistics for 1995–2005 because the official statistics, in contrast to the forensic investigations, exclude foreign citizens and include Swedish citizens who committed suicide abroad.
From the same database, we also obtained toxicological data for the same period on all 21 245 cases considered to be natural deaths, and all 12 181 cases considered to be accidental deaths. Together, these cases constituted the control group (n = 33 426, mean age 57.3 years, 70.8% males).
The exposure to antidepressants in suicides as well as in controls may reflect the increasing trend in the use of antidepressants in the general population. However, the number of suicides exposed to antidepressants might be decreased as a consequence of the decreasing trend in all suicides and thus proportional to this (i.e. the null hypothesis, H0). Alternatively, a decrease in exposed suicides might be a cause of the decrease in the total number of suicides and thus numerically equal to this (i.e. the hypothesis H1).
To test the validity of these assumptions, we decided also to analyse the detections of five ‘control drugs’ in suicides and controls. We chose three new drugs which like antidepressants were increasingly being used in Sweden during the period: the hypnotics zolpidem and zopiclone, and the analgesic tramadole. As the trends for the two hypnotics were found to be very similar, we merged them to gain statistical power. We also chose two other drugs with decreasing use: the cardiovascular drug verapamil and the analgesic dextropropoxyphene. We proposed null hypotheses and hypotheses for each of the control drugs in the same manner as for the antidepressants, and we made the same calculations.
Thus, if antidepressant medication has been a substantially contributing cause of the decrease in suicide that occurred in Sweden during the study period, H1 should be accepted and H0 rejected, while the null hypothesis should be accepted for the control drugs if the design is valid. For details of the calculations, see the Results section.
The time trends of suicides and controls with toxicological detections of antidepressants and control drugs were linear (Table 1). We, therefore, compared the last year of the period, 2005, to the first year, 1995 (tramadole 1997), for the calculated data. Chi-squared tests (df = 1) were used for comparing observed data with expected data (exposed vs. unexposed suicides in 2005) according to the hypothesis for antidepressants and according to the null-hypothesis for the comparison drugs (Table 2). In the text and tables are the numbers rounded, while all calculations were done with the exact values.
Table 1. Toxicological detections of medications in suicides and controls 1995–2005
Bold-faced numbers were involved in the calculations, i.e. 2005 compared with 1995 (or in the case of tramadole to 1997).
Table 2. Observed and expected values according to the respective hypothesis h1, and to the null hypothesis h0
2 × 2 chi-squared test of observed values vs. expected values, i.e. Eh1 for antidepressants and Eh0 for the other drugs (bold values).
The total number of suicides (including undetermined deaths) decreased from 1783 cases in 1995 to 1451 cases in 2005 (i.e. −332 cases, −18.6%). The proportion of these suicides exposed to antidepressant medication increased by 46% (from 354 to 420 individuals; from 20 to 29%). The trend was linear (r2 = 0.91). Among the controls, however, the proportion of individuals increased by 109% (r2 = 0.86).
The exposure to zopiclone or zolpidem increased by 141% in the suicides (r2 = 0.61). The increase in the controls was 215% (r2 = 0.83). Among the suicides exposed to zopiclone or zolpidem, 41% were also exposed to antidepressants.
The trend with regard to dextropropoxyphene-exposed suicides decreased after a maximum in 1997 (−64%, r2 = 0.84) and this decrease was about the same as in the controls (−61%, r2 = 0.78).
Tramadole was not available until 1997. The increase in the exposure to tramadole (1997–2005) did not differ greatly between suicides (+167%, r2 = 0.98) and controls (+190%, r2 = 0.92).
During the 11-year period, only 83 suicides and 158 controls were exposed to verapamil and these were too few for further analysis (data not shown).
These calculations aim at assessing whether the actually observed number of exposed suicides in 2005 (n = 420) is consistent with the hypothesis that the observed decrease in Swedish suicides since 1995 (n = 332) has occurred in that part of the population that was exposed to antidepressants. Three known variables influence the number of exposed suicides: the use of antidepressants in the population, the total number of suicides (H0) and the extent to which antidepressants may prevent suicide (H1).
The first necessary step was to estimate the influence of the increased use of antidepressants in the population. Assuming that the total number of suicides had remained at the 1995 level, the expected ‘base’ number (Eb) of suicides exposed to antidepressants in 2005 was calculated as the number of exposed suicides in 1995 times the relative increase in exposed controls:
The influence of the decrease in the number of exposed suicides was then calculated, first assuming the null hypothesis to be true (i.e. exposed and unexposed suicides decreased in equal proportions), as Eb times the observed relative decrease in suicide:
and then second assuming H1 to be true (i.e. the decrease in suicides occurred only in those exposed to antidepressants), as Eb plus the observed absolute decrease in suicides:
The observed number of suicides with antidepressants in 2005 was:
It is evident from Table 2 that the observed number of suicides exposed to antidepressants (O = 420) was well predicted by the hypothesis (EH1 =409, χ2 = 0.17, N.S.), and less well by the null hypothesis (EH0 =603, χ2 = 50.0, P < 0.001).
Finally, the corresponding values of Eb,EH0,EH1 were calculated for each of the control drugs (Table 2).
The observed numbers of suicides exposed to the control drugs were all best predicted by the null hypothesis (EH0). Although this was true also for zopiclone–zolpidem, the observed number of suicides (O = 196) differed numerically (P < 0.01) also from the expected number (EH0 =257). The hypothesis (EH1) predicted negative values of the control drugs, which may be understood as meaning that their assumed use in suicides (Eb) was too small to explain the observed decrease in suicide (−332) even if they had had a 100% suicide-preventive effect.
We studied all the suicides, natural deaths and accidental deaths, which had been investigated by forensic toxicological screening in Sweden during 1995–2005. Our original aim was to include also the years 1992–1994 from which we had toxicological data. As this would have inferred bias by an irrelevant time trend, we restricted the study period to 1995–2005 (Fig. 1). All the individuals had been analyzed in an identical way, before the cause of death was established after the complete forensic investigation. The exposure to antidepressants and to dextropropoxyphene, verapamil, zolpidem, zopiclone and tramadole was determined from the toxicological screening, and all the calculations were based on these individual data. The accidental and natural deaths were studied to control the increasing use of antidepressants in the general population. The latter five drugs were studied to validate the design, i.e. to control the possible influence of unknown factors on the differential exposure to antidepressants in suicides and controls.
The empirical findings were that, during this period the total number of suicides decreased by 18.6%; that the proportion of suicides exposed to antidepressants increased by 46%; and that the proportion of exposed controls increased by 109%. We found a similar discrepancy with regard to exposure to zolpidem/zopiclone, which increased by 141% among the suicides and by 215% among the controls. Dextropropoxyphene decreased to the same extent among the suicides (−64%) as among the controls (−61%). Tramadole increased similarly in suicides (+167%) and controls (+190%).
Assuming that the influence of the increased prescribing of the antidepressants and the increased or decreased prescribing of the control drugs should be as large in suicides as in the controls, the expected numbers predicted by the hypothesis (Eh1), and by the null hypothesis (Eh0) were calculated and compared with the observed numbers (O) (Table 2). It is obvious from this comparison that the observed number of suicides exposed to antidepressants was best predicted by the hypothesis (H1), i.e. that suicides were prevented by antidepressants. Thus, the hypothesis (H1) was verified, while the null hypothesis (H0) was rejected. The observed values of the control drugs, however, were best predicted by the null hypothesis (H0), which indicates that the basic assumption was valid: If antidepressants had not prevented suicide, the increase in suicides exposed for antidepressants could be predicted from the increased exposure for antidepressants in the controls.
Numerically, the difference between the observed and the expected `base' numbers of suicides with antidepressants detected (O − Eb=−321), almost perfectly explains the actual decrease in suicide during the period (1451 − 1783 = −332). This suggests that antidepressants reduced the risk of suicide by 43% (1 − 420/741), which is very close to the 46%, as estimated 12 years ago from ecological data (21). It is further consistent with the actual decrease in Swedish suicide rates during 1971–1988 (22). The observed numbers of suicides who tested positive for zopiclone–zolpidem (O = 196) were lower than the number expected from the null hypothesis (EH0 =257), and this might suggest that these hypnotics also have some suicide-preventive effect. This might be a true pharmacological effect (i.e. the treatment of insomnia decreases the risk of suicide), or a false effect because 41% of the suicides exposed to zopiclone–zolpidem also tested positive for antidepressants (as did 35% of the controls, N.S.). If all suicides and controls that tested positive for both antidepressants and zopiclone–zolpidem were to be excluded from the analysis, the observed number of suicides positive for zopiclone–zolpidem would not differ significantly from the number expected from the null hypothesis (O = 116 and H0 =127, N.S.).
It is an open question whether antidepressants exert their apparent suicide-preventive effect in the beginning of the disease process, when the depressions are milder, or in a severe end stage, when the patients are very close to committing suicide. As the most widely used antidepressants, the SSRIs, are believed to be less effective in severe depression, it appears most probable that they prevent suicide by preventing mild and moderate depressions from deteriorating into severe suicidal depressions. The unspecific effects of contact with a doctor may to some extent also explain the suicide-preventive effect of antidepressants, but this effect does not explain the differences found between the antidepressants and the control drugs.
As this was a naturalistic study, the influence of unknown factors can never be excluded.
Possible bias by indication. The suicides may have suffered more severe depressions than the depressed controls, and it is not known whether the use of antidepressants in the population increased to the same extent in severe as in mild depressions. It is probable that, already in the beginning of the study period, severely depressed individuals were diagnosed and treated to a greater extent than those with mild depression. This does not, however, imply that the relative increase in the use of antidepressants should be different in severe and mild depressions. Regardless of this, it is, as discussed above, not known whether an increased use of antidepressants in mild, moderate or severe depressions is crucial to reduce suicide.
Possible bias by ‘saturation’. The use of a drug cannot increase infinitely; it must gradually level off as an increasing proportion of those individuals for whom the drug is indicated become treated (‘saturation’). In the last year of the study, 29% of the suicides and 10% of the controls were exposed to antidepressants, which represented increases since 1995 of 46% and 109% respectively. Could the lower increase in suicides be explained by the suicide population being more ‘saturated’ with antidepressant treatment? Investigations of consecutive suicides have found depression in up to 87% of the cases (23); therefore, the finding of antidepressants in 29% appears to be far from ‘saturation’. Antidepressants were found in 10% of the controls, which also appear to be far from ‘saturation’ because antidepressants may be indicated in at least 15% of the general population (roughly: acute depression 5%, continuation treatment 5%, anxiety disorders 5%), and probably to an even greater extent in the controls because these were probably sicker than the general population. The same conclusion may be drawn from a recent Swedish epidemiological investigation which showed that in 2001 only 24% of the individuals in the general population with major depression were treated with antidepressants (24). The trend regarding exposed suicides found in this study was clearly linear (r2 =0.91). The trend in the controls was also well described as being linear (r2 =0.86) although it appeared to level off somewhat, which tends to strengthen the significance of our results.
Thus, with regard to both indication and ‘saturation’ it seems relevant to expect an increase of antidepressants in suicides as large as in the controls if not preventing suicide, which was the basic assumption on which this study was based (i.e. the null hypothesis).
Possible bias by other factors. Besides antidepressant medicines, there might be several other factors that influenced the suicide rate in the population. It is possible that the net effect of all such factors might tend to increase the suicide rate or to decrease it or to cancel each other out, but to ascribe the whole observed decrease in suicide to such factors is the null hypothesis of this study, and this does not explain the observed data well. The level of alcohol consumption and the unemployment rate are often proposed to influence suicide rates. Bias by these factors is, however, unlikely. The annual per capita alcohol consumption increased substantially in Sweden during the period from 8.6 l in 1996 to 10.2 l in 2005 (Swedish National Institute of Public Health), and this would tend to increase the suicide rate. The unemployment rates during this period have both decreased and increased and these data do not correlate with the decreasing suicide rate (1, 25).
Relation to other studies
In 2000, Isacsson demonstrated a strong correlation between the increasing use of antidepressants and the decreasing suicide rate in Sweden, Norway, Denmark and Finland, and proposed the hypothesis that treatment of depression with antidepressants provides an effective suicide-preventive strategy on the population level (1). This hypothesis has since then been tested in several ecological studies in different parts of the world (2, 9–15). The latest is an analysis by Nakagawa et al. (6) showing that the odd trend of increasing suicide in Japan during the 1990s was broken when SSRIs were introduced in 1999, about 10 years later than in Western countries. One Italian (26) and one Icelandic (27) study have failed to demonstrate such a correlation. These studies have been commented on by Isacsson and Rich (2). Furthermore, Castelpietra et al. (28) have later demonstrated a clear correlation also in Italy. Ludwig and Marcotte (3) carried out an ecological analysis of trends in suicide and in the use of antidepressants in 27 countries (Sweden was not included). They found that a 13% increase in the use of antidepressants was on average associated with a decrease of 2.5% in the suicide rates (3).This implies that the 109% increase in the use of antidepressants observed among controls in this study would be paralleled by a 21% decrease in suicide in Sweden, which is very close to the observed decrease (19%) and supports the validity of both studies. Gibbons et al. (29) demonstrated that the decreased use of antidepressants in young people, as seen in USA in 2004 after the FDA warnings might be the cause of a concurrent drastic increase in suicide in that age group. Recently, was reported by Bridge et al. (30) that the suicide rates among the young in USA stayed at a high level in 2005. Katz et al. (31) reported that also in Canadian youth had the use of antidepressants decreased and paralleled by an increase in suicide.
There are further some other individual-based studies, which have attempted to investigate whether the use of antidepressants prevents suicide. Søndergård et al. (32) used four Danish registers to identify 438 625 patients who had purchased antidepressants, and compared them with 1 199 057 population-based control persons. They found a more pronounced reduction in suicide among persons treated with antidepressants than among persons not treated with antidepressants. They also compared patients who had filled one single prescription of an antidepressant with those who had filled two or more prescriptions. The latter was assumed to indicate continued treatment (33). They found that the decrease in suicide rates was more pronounced among those with two or more prescriptions.
Tiihonen et al. (4) identified from a national register all the individuals who had been hospitalized because of a suicide attempt in Finland during 1997–2003. They found during a mean follow-up time of 3.4 years that, among subjects who had at some time used an antidepressant, the current use of medication was associated with a 32% decreased risk of suicide, compared with no current use of medication (paradoxically, the risk of attempted suicide was increased by 39%).
Jick et al. as well as Martinez et al. (34, 35), identified from the British General Practice Research Database 159 810 and 146 095 patients, respectively, with a first prescription of a tricyclic or SSRI antidepressant during the periods 1993–1999, and 1995–2001 respectively. When patients who received tricyclics were compared with those who received SSRIs, no differences were found regarding the risk of suicide (or non-fatal self-harm). Jick et al. found, however, that, compared with 90 days or more after the first prescription, the odds ratio for suicide was 38.0 times higher during the first 1–9 days, 5.1 times higher during days 10–29, and 2.0 times higher during days 30–89, which may suggest that the risk of suicide decreased during the treatment with antidepressants (similar results were found with regard to non-fatal self-harm).
Angst et al. (36) did a follow-up of 406 mood-disorder patients with and without long-term medication for 40–44 years. They found that the standardized mortality ratio for suicide was reduced by 59% among those treated with antidepressants (lithium 65% and neuroleptics 41%).
Leon et al. (37) found that antidepressants were detected in the toxicology of suicides in New York 2001–2004 in 20%, 24%, 27% and 21% of the cases per year. Isacsson et al. found in San Diego 1981–1982 that the toxicology was positive for antidepressants in 8% of the suicides (38), in Sweden 1990–1991 and 1992–1994 16% and 16.5%, respectively (16, 17), and in Mobile, Alabama 1990–1995 15% (39). In Finland 1987–1988, Ohberg et al. (40) found antidepressants in toxicology in 19% of the female suicides and in 4.8% of the male. These latter studies included no data regarding the use of antidepressants in the population.
As with this study, the individual-based studies mentioned above were all naturalistic, which is unavoidable because experimental controlled studies cannot be carried out. Biasing factors cannot thus be excluded, but these studies are all consistent with, or clearly support, the conclusion of this study.
In conclusion, antidepressant medication, in contrast to the control drugs, was detected in the toxicology of suicides to the extent that was expected if they prevent suicide. This supports the a priori hypothesis that an increase in the use of antidepressant medication has been a substantially contributing cause of the decrease in suicide rates demonstrated repeatedly in ecological studies in Sweden and elsewhere. The current evidence-based standpoint must be that an increase in the use of antidepressants on the population level is a powerful suicide-preventive strategy. It is of outmost importance that this is acknowledged when the possible risks of using antidepressants are weighed against their benefits.
This study was financially supported by the Karolinska Institute and the Stockholm County Council, Psychiatry South-West (ALF); the National Board of Forensic Medicine.
Declaration of interest
Göran Isacsson has received fees for being a speaker at symposiums arranged by pharmaceutical companies like H. Lundbeck, Eli Lilly, GSK, Pfizer. Johan Ahlner has been involved in a research project funded by H. Lundbeck. Urban Ösby has been involved in advisory boards for BMS and Pfizer, received fees as a consultant from Eli Lilly, Astrazeneca and BMS, and has received research grants from BMS and Jansen-Cilag. Anita Holmgren declares no conflicts of interest.