Association between consistent purchase of anticonvulsants or lithium and suicide risk: A longitudinal cohort study from Denmark, 1995–2001

Methods Longitudinal, retrospective cohort study of all individuals in Denmark purchasing anticonvulsants (valproic acid, carbamazepine, oxcarbazepine or lamotrigine) ( n = 9952) or lithium ( n = 6693) from 1995–2001 who also purchased antipsychotics at least once (to select out nonpsychiatric anticonvulsant use). Poisson regression of suicides by medication purchased (anticonvulsants or lithium) was conducted, controlling for age, sex, and calendar year. Confounding by indication was addressed by restricting the comparison to individuals prescribed the same medication: individuals with minimal medication exposure (e.g., who purchased only a single prescription of anticonvulsants) were compared to those individuals with more consistent medication exposure (i.e., purchasing ≥ 6 prescriptions of anticonvulsants). Results Demographics and frequency of anticonvulsant, lithium, or antipsychotic use were similar between lithium and anticonvulsant purchasers. Among patients who also purchased antipsychotic at least once during the study period, purchasing anticonvulsants more consistently (≥ 6 prescriptions) was associated with a substantial reduction in the risk of suicide (RR = 0.22, 95% CI = 0.11–0.42, p < 0.0001), similar to patients consistently purchasing lithium (RR = 0.27, 95% CI = 0.12–0.62, p = 0.006). Absolute suicide risks of consistent anticonvulsant and consistent lithium purchasers were similar. Limitations Lack of information about diagnoses and potential confounders, as well as other covariates that may differ between minimal and consistent medication purchasers, are limitations to this study. Conclusions In this longitudinal study of anticonvulsant purchasers likely to have psychiatric disorders, consistent anticonvulsant treatment was associated with decreased risk of completed suicide. Keywords Suicide Anticonvulsants Lithium Pharmacoepidemiology Longitudinal research Confounding by indication 1 Introduction Anticonvulsant medications have replaced lithium as the most popular mood stabilizing medications prescribed for bipolar disorder in many settings ( Goodwin et al., 2003; Collins and McFarland, 2008 ). However, in contrast to lithium, there is relatively little data concerning the effects of anticonvulsants on completed suicide. Randomized ( Thies-Flechtner et al., 1996 ) or observational ( Goodwin et al., 2003; Collins and McFarland, 2008 ) studies have failed to observe substantial protective effects of anticonvulsants on suicide when compared to lithium. An observational study of two U.S. health maintenance organizations found that risk of suicide death was significantly higher among individuals receiving divalproex treatment for bipolar disorder compared to lithium (Hazard Ratio = 2.7, p = 0.03) ( Goodwin et al., 2003 ). A recent observational study of U.S. Medicare patients that analyzed an extremely small sample of suicide deaths ( n = 4) found that differences in the hazard of completed suicides between divalproex and lithium were not significant, although the hazard was slightly larger for divalproex purchasers (Hazard Ratio OR = 1.5, p = 0.1) ( Collins and McFarland, 2008 ). Neither of these observational studies permit determination of the degree to which treatment by divalproex may reduce the suicidal risk of patients, rather they simply establish that suicide risk among individuals receiving divalproex continues to be higher than those receiving lithium. A study design that might address this question, although with limitations, are studies that compare suicide risk during periods on anticonvulsant versus not on anticonvulsant, similar to what exists for lithium (reviewed in Baldessarini et al., 1999; Tondo and Baldessarini, 2000 ). One limitation of this design is the possibility that comparisons may be exaggerated by an elevated risk of suicide during the period immediately after discontinuing lithium ( Tondo and Baldessarini, 2000 ), such as the 20-fold increase of risk of suicidal acts in the year after lithium discontinuation observed in one study ( Baldessarini et al., 1999 ). Important confirmation of this possibility of discontinuation-associated risk was provided by a recent study which observed an 11–24× increased risk of suicidal behavior or prophylactic hospitalization upon discontinuation of either lithium, divalproex, or carbamazepine ( Yerevanian et al., 2007a ). Studies also exist comparing the risk for nonfatal suicide behaviors (or prophylactic hospitalization) between anticonvulsants and lithium, with observational studies either reporting similar risk for users of the two types of mood stabilizers ( Yerevanian et al., 2003; Yerevanian et al., 2007a ) or reduced risk for lithium purchasers ( Goodwin et al., 2003; Collins and McFarland, 2008 ), and one randomized study reported a trend of decreased suicide attempt risk for lithium recipients compared to carbamazepine recipients ( Kleindienst and Greil, 2000 ). The importance of studying completed suicide as an outcome separately from suicide attempts or ideation has become increasingly recognized, however ( Beautrais, 2001; Arensman and Kerkhof, 1996; Kerhof and Arensman, 2001; Heeringen, 2001 ). One sufficiently large database for the study of completed suicide has been established by linking data from the Danish Medicinal Product Statistics database ( Danish National Board of Health (Laegemidelstyrelsen), 2007 ) and data from the Medical Registry of Vital Statistics ( Danish National Board of Health (Sundhedsstyrelsen), 2007 ). This linked database has the advantages of large size and extensive generalizability, in that it contains information about every resident of Denmark. However, information about specific diagnoses is inconsistently available, a limitation in studying anticonvulsant use, given the routine use of these medications for nonpsychiatric diagnoses (e.g., epilepsy, pain). We devised an approach to use these linked Danish databases to examine the effects of anticonvulsants and lithium on suicide among likely psychiatric patients by using antipsychotic medication use as a proxy for substantial psychiatric illness. We then compared the suicide risk of anticonvulsant/antipsychotic purchasers who received only a single prescription of anticonvulsants in the study period to those who received 6 or more prescriptions (often many more) in the study period. This approach helps address confounding by indication (the possibility anticonvulsant purchasers may be at a distinctly increased suicide risk due to baseline factors that may make them good candidates for anticonvulsants (e.g. rapid cycling) by comparing suicide risk only among individuals who have been prescribed the same medication. This “internal comparison” approach has been used to address confounding by indication in other studies of the association between psychiatric medication purchase and suicide risk. (Examples are recent studies investigating broad, population-based cohorts of patients purchasing lithium ( Kessing et al., 2005 ) or antidepressants ( Sondergard et al., 2006 ) or more narrowly-defined cohorts of patients with hospital-confirmed diagnoses of bipolar disorder receiving lithium or anticonvulsants ( Sondergard et al., 2008 )). The primary hypothesis of our study was that consistent purchase of anticonvulsants would be associated with reduced risk of completed suicide relative to a comparison group of anticonvulsant purchasers also prescribed the medication, but who purchase the medication rarely or infrequently. As a secondary analysis, to provide context for the degree of any suicide risk reduction observed, we applied the same procedures to lithium purchasers to determine to what extent a medication with a much better established evidence base for reducing suicidality was associated with reduced risk of suicide in this study design. 2 Method 2.1 Registries Data were obtained by linking the Danish Medicinal Product Statistics Registry ( Danish National Board of Health (Laegemidelstyrelsen), 2007 ) with the Danish Medical Register on Vital Statistics (Cause of Death Registry) ( Danish National Board of Health (Sundhedsstyrelsen), 2007 ) using the unique personal identification numbers assigned to all 5.3 million persons living in Denmark ( Malig, 1996 ). In Denmark, all medication prescribed by doctors (including anticonvulsants, lithium, and antipsychotics) is purchased only at pharmacies. The Medicinal Product Statistics Registry contains data on all prescribed medication purchased at pharmacies from January 1, 1995, onward. The Danish Medical Register on Vital Statistics contains dates and causes of all deaths in Denmark since 1976, recorded from death certificates completed by doctors at the time of death. Suicide was identified when the cause of death was coded as intentional self-harm ( International Classification of Diseases, 10th Revision (ICD-10) codes X600 to X840). Information about age, gender, and socioeconomic status was present in the registries. The study period was from January 1, 1995, to December 31, 2001. 2.2 Sample All persons aged 18 years or older who purchased lithium or anticonvulsants at least once in the study period from January 1, 1995, to December 31, 2001, and who had also purchased antipsychotics at least once during this study period were identified in the Danish Medicinal Product Statistics Register. “Anticonvulsants” were restricted to those most commonly used for psychiatric purposes: valproic acid, carbamazepine, oxcarbazepine, and lamotrigine. “Antipsychotics” included: olanzapine, risperidone, quetiapine, ziprasidone, clozapine, aloperidol, perphenazine, fluphenazine, chlorpromazine, sertindole, zuclopenthixol, penfloridol, sulpiride, and amisulpride. The requirement that individuals purchasing anticonvulsants or lithium also purchase antipsychotics at least once during the study period was imposed as a selection procedure with the aim of greatly enriching our anticonvulsant sample in individuals with mental illness rather than epilepsy. Because diagnostic information does not exist in the Danish Medicinal Product Statistics register, this procedure was chosen to exclude most of the anticonvulsant purchasers who receive the medication solely to control epilepsy or pain. Because application of our enrichment criteria (at least one antipsychotic prescription purchase) would be expected to select, on average, psychiatric patients with increased illness severity (i.e. severe enough to warrant antipsychotics), we applied the same criteria to lithium purchasers. 2.3 Statistical analysis The sample was initially categorized into two groups, one including persons who had purchased anticonvulsants and antipsychotics (either simultaneously or sequentially) and the second one including persons who had purchased lithium and antipsychotics. Entry into the cohort was dynamic, i.e., a person became a member of one of these groups at the moment when the second required drug was first prescribed, no matter the order in which the drugs were given. The association between both treatments and completed suicide was studied by Poisson regression analyses ( Clayton and Hills, 1993 ) with censoring occurring at one of four time points: 1) death by reasons other than suicide, 2) emigration, 3) the end of the study period (December 31, 2001), or 4) change in type of mood stabilizer medication — as soon as individuals purchased a prescription for lithium (if they entered as an anticonvulsant purchaser), or vice versa, their follow-up time was terminated. Patients on both types of mood stabilizers throughout the study period were not included in the cohort. Once the overall risk for suicide, adjusting for covariates, was determined for the groups purchasing anticonvulsants or lithium and an antipsychotic, further analysis was performed to test our primary hypothesis that consistent use of anticonvulsants would be associated with a reduction in suicide risk. During this analysis we addressed confounding by indication through restriction by the “internal comparison” method of Kessing et al. (2005) . This method involves comparing the risk of completed suicide only among users of the same medication, contrasting the suicide risk of individuals with minimal exposure (a single prescription) to individuals with greater (2, 3,4, or 5 prescriptions) to much greater exposure (at least 6 prescriptions, often many more). Subjects were designated as purchasing “1 prescription” of anticonvulsants if they purchased no additional anticonvulsant prescriptions after the index prescription which entered them into the study cohort (which could be either anticonvulsant or antipsychotic, depending on which medication was purchased second). The risk of suicide was then investigated by comparing the risk of the patients receiving greater anticonvulsant exposures (2,3,4,5, ≥ 6 prescriptions) to patients prescribed anticonvulsants but receiving a minimal exposure (1 prescription), adjusting for age, sex, and calendar time. Statistical significance was judged by the likelihood ratio test while confidence intervals were computed based on Wald's test. 3 Results A total of 9952 persons aged 18 or older who purchased at least one prescription of both an anticonvulsant and an antipsychotic from January 1, 1995 to December 31, 2001 were identified in the Danish Medicinal Products Statistics register. This number of people constitutes approximately 0.24% of the Danish adult population and 16% of the purchasers of anticonvulsants during the study period. A total of 6693 persons purchased lithium and antipsychotics at least once in the study period, approximately 0.16% percent of the Danish adult population and 40% of the purchasers of lithium during the study period. Table 1 presents the characteristics of persons who purchased an anticonvulsant (and antipsychotic) versus the characteristics of persons who purchased lithium (and antipsychotic). Individuals purchasing anticonvulsants appear to be more likely to die during the study period; the greater age of the anticonvulsant cohort relative to the lithium cohort (median age of 56 versus 53 years old, and 75th percentile age of 73 versus 67 years old) might at least partly explain this difference. The unadjusted suicide rate among patients receiving at least one prescription of anticonvulsant (0.56%) was similar to those receiving at least one prescription of lithium (0.60%), however the average follow-up times differ, suggesting the overall suicide rate is higher in the anticonvulsant group. After adjusting for the effect of age, gender, socio-economic status, and calendar year within a Poisson regression, the overall sample of individuals purchasing anticonvulsants did show an increased risk of completed suicide compared to those purchasing lithium, but the difference was marginally nonsignificant (RR = 1.49, 95% CI = 0.97–2.29, p = 0.064). Table 2 presents patterns of anticonvulsant, lithium and antipsychotic use in the samples. A large majority of cohort members (72–85%) used anticonvulsants or lithium repeatedly. The median number of antipsychotic prescriptions purchased was identical and substantial for both groups, suggesting that the cohort members on average had sufficiently severe mental illness to warrant repeated antipsychotic prescriptions. Table 3 presents the rate of completed suicide among individuals receiving varying numbers of prescriptions of anticonvulsants or lithium during the study period, compared to individuals receiving only a single prescription. Significant reductions in the rate of completed suicide occurred for individuals purchasing ≥ 6 anticonvulsant prescriptions compared to individuals purchasing only a single anticonvulsant prescription (RR = 0.22, 95% CI = 0.11–0.42, p < 0.0001). Similar reductions were observed for individuals purchasing ≥ 6 prescriptions of lithium versus only a single prescription of lithium (RR = 0.27, 95% CI = 0.12–0.62, p < 0.006). The absolute rate of suicide was also very similar between consistent purchasers (i.e., ≥ 6 prescriptions) of anticonvulsants and lithium (94 and 96/100,000 person-years, respectively, data not shown). However, this comparison between the two medications does not address confounding by indication, unlike the “internal comparison” analysis comparing single prescription purchasers to those purchasing ≥ 6 prescriptions. 4 Discussion Our principal finding is that among individuals judged likely to have psychiatric disorders (based on having purchased antipsychotic medication in addition to anticonvulsants), those who consistently purchased anticonvulsants (≥ 6 prescriptions after entering the cohort) appeared to have a markedly lower risk of completed suicide (RR = 0.22) compared to individuals receiving only a single anticonvulsant prescription. A similar reduction of suicide risk was observed in consistent versus single purchasers of lithium (RR = 0.27). This population-based investigation, along with our study based on a population of patients with bipolar disorder in hospital care settings ( Sondergard et al., 2008 ) are some of the first reports in which repeated purchase of anticonvulsants has been associated with substantially decreased suicide risk. Our findings are similar to those reported for lithium from “on–off” exposure studies ( Tondo and Baldessarini, 2000; Schou and Weeke, 1988; Muller-Oerlinghausen et al., 1992; Nilsson, 1995 ). These studies have the analytic advantage of controlling for confounding by indication but disadvantages in the estimate of effect size, given that individuals may be most likely to discontinue medication when they are experiencing an affective episode or, even if experiencing stable mood, are likely to subsequently experience a mood episode ( Tondo and Baldessarini, 2000 ), and thus be at higher suicide risk. In the present study, the pattern of decreased risk of suicide observed in association with the consistent use of anticonvulsants is similar to that observed in association with consistent use of lithium. The remaining absolute suicide risk is also highly similar between consistent anticonvulsant purchasers and lithium purchasers. In previous studies, lithium has been reported to be associated with significantly ( Goodwin et al., 2003 ) or nonsignificantly lower risk ( Collins and McFarland, 2008 ) of completed suicide compared to anticonvulsant mood stabilizers. The findings in our study of similar suicide risk between consistent anticonvulsant and lithium purchasers may be noteworthy given the substantial numbers of randomized ( Cipriani et al., 2005 ) and nonrandomized studies ( Kessing et al., 2005; Baldessarini et al., 2006; Ahrens et al., 1995; Coryell et al., 2003; Schou, 1998 ) suggesting lithium treatment is associated with reduced suicidality. However, our observations comparing anticonvulsant purchasers to lithium purchasers in this analysis are tentative, given that the comparison between anticonvulsant and lithium purchasers does not address possible confounding by indication, unlike our primary comparison of suicide risk within anticonvulsant purchasers (consistent versus single purchasers) or lithium purchasers (consistent versus single purchasers). An analogous study from our group of patients who initiated care during hospitalization for bipolar disorder found both anticonvulsants and lithium were associated with reduced risk of suicide, but those individuals who switched to or added lithium experienced a further reduction in suicide risk ( Sondergard et al., 2008 ). This observation may be similar to the findings of a study of suicidal behavior in 405 U.S. veterans, which found that lithium, divalproex, and carbamazepine monotherapy were associated with statistically similar risks of suicidal behavior, but observed that lithium monotherapy was associated with rates of suicidal behavior that were numerically the lowest ( Yerevanian et al., 2007a ). The absolute rate of suicide we observed for consistent lithium purchasers was intermediate between that observed in a recent metaanalysis of 31 studies from a variety of countries (155/100,000) ( Baldessarini et al., 2006 ) and the US retrospective cohort study of Goodwin et al. (2003) (70/100,000 person-years). Our absolute rate of suicide among consistent anticonvulsant purchasers was lower than that reported in Goodwin et al. (2003) for all anticonvulsant purchasers in that cohort (170/100,000 person-years). Strengths of our study include a longitudinal design following individual patients from a broad population base — our sample was composed of all purchasers of anticonvulsants or lithium within the country of Denmark who also used antipsychotics at least once. The combination of the large sample size and the length of follow-up increased our statistical power to study a relatively rare event such as suicide. Our study included substantially more suicides (approximately 3-30-fold more suicides) than previous research examining the psychiatric use of anticonvulsants ( Goodwin et al., 2003; Collins and McFarland, 2008 ), although our diagnostic categories were broader. We did have information to allow adjustment for several potential important confounders: age, sex, and calendar year (date), to control for possible secular trends. We were also able to determine the patterns of mood stabilizer and antipsychotic use by our cohort. Finally, we adopted an approach to reduce confounding by indication. This approach compared the suicide risk of purchasers of mood stabilizer who did not purchase any further prescriptions of anticonvulsant or lithium to individuals who purchased additional prescriptions (especially those who purchased ≥ 6 prescriptions). Those prescribed only a single prescription of either anticonvulsants or lithium presumably reached some threshold diagnostic profile necessary for the prescriber to decide to initiate treatment with that medication, but lacked any further medication exposure compared to those purchasing ≥ 6 prescriptions. While our “consistent” purchasers had to purchase only a minimum of 6 prescriptions to qualify for this stratum, we believe the term “consistent” is accurate because most members purchased considerably more than 6 prescriptions (overall median number of prescriptions for the entire cohort: 11 prescriptions for anticonvulsant purchasers and 17 for lithium purchasers). Limitations of our study include the fact it is an observational, nonrandomized study. Such studies are useful in establishing associations rather than drawing causal inferences. As with most observational studies of medication use, we were unable to precisely ascertain whether purchasers were actually taking their medication at the time of their suicide. Other limitations include our lack of more extensive information about potential confounders — in particular it would have been helpful to have information about the presence and number of past suicide attempts and psychiatric hospitalizations. In addition, since we relied on doctor-identified cases of suicide, the total number of suicides may be understated. Because of the limited number of suicides, we also combined different types of anticonvulsants (valproic acid, carbamazepine/oxcarbazepine, and lamotrigine) that may work through different mechanisms and have different anticipated effects on suicidality. 64.5% of anticonvulsant purchasers used carbamazepine or oxcarbazepine, 21% valproate and 14.5% lamotrigine, thus our anticonvulsant findings mainly pertain to carbamazepine/oxcarbazepine. Subanalyses of the association between continued treatment with individual anticonvulsants and suicide would suffer from poor statistical power and were not computed. Also, although our results are likely highly generalizable to Danish populations, the results may be less generalizable to other populations, due to differences in race and ethnicity, age structure, prescribing patterns, and other factors. A final limitation of our study was the lack of clear diagnostic information on our sample. Our group has recently published an analysis of a substantially smaller group of individuals ( n = 5926 versus n = 16645 in this study) with hospital-confirmed diagnoses of bipolar disorder, which reported generally similar findings to this study ( Sondergard et al., 2008 ). We also did not analyze whether suicide risks varied among patients purchasing lithium or anticonvulsants depending on the degree to which they continued to consistently purchase antipsychotics, or whether they also purchased antidepressants. These two types of medications found to be substantially associated with the risk of suicide behavior in patients receiving mood stabilizers by Yerevanian et al. (2007b,c) . Thus it remains possible that differences in the rate of polytherapy between the lithium and anticonvulsant samples (or differences in intensity of non-medication treatments) may have influenced our results. In addition, three other nonmedication factors may potentially influence our results. First, both our cohorts are of individuals with a heterogeneous set of psychiatric diagnoses, and it is likely that the anticonvulsant sample is more heterogeneous, despite our selection procedures, since anticonvulsants may be more frequently used for impulse control disorders, developmental disorders, etc. However, our design of primarily comparing risk within anticonvulsant users (comparing purchasers of 1 prescription with ≥ 6 prescriptions) helps to limit, although not eliminate, the potential impact of this heterogeneity. Second, it has been recently observed in a metaanalysis of both randomized and observational studies that consistent users of a wide variety of medications have reduced all-cause mortality ( Simpson et al., 2006 ), although suicide has not been studied specifically. It is difficult to know to what extent the reductions in suicide observed in our study could be a variant of what has been termed a “healthy adherer” effect ( Simpson et al., 2006 ). However, in contrast to our findings and those for newer antidepressants ( Sondergard et al., 2006 ), consistent purchase of certain other psychiatric medications (tricyclic and other older antidepressant medications) does not seem to be associated with reductions in suicidality ( Sondergard et al., 2006 ). Also, data suggesting that lithium adherence is associated with reduced risk of nonsuicide mortality, which a “healthy adherer” effect might suggest, is mixed ( Angst et al., 2002 ). Finally, our approach to controlling confounding by indication, the “internal comparison” approach, may introduce other biases. Specifically, clinical characteristics of single purchasers, separate from their medication use, could put them at increased risk of suicide, such as greater impulsivity and substance abuse, and their rate of suicide might also be more influenced by any discontinuation-associated risk ( Baldessarini et al., 1999; Tondo and Baldessarini, 2000; Yerevanian et al., 2007a ). In conclusion, this report and its analogous paper ( Sondergard et al., 2008 ) reporting on the Danish nationwide database are some of the first large observational studies to observe a substantial reduction in the risk of completed suicide in association with the consistent purchase of anticonvulsants. The reduction in suicide risk associated with consistent purchase of anticonvulsants is supported both by our “internal comparison” method that helps address confounding by indication, and by our comparison with lithium purchasers in this study. While our findings are limited by the lack of diagnostic information, our findings are highly similar to findings from a more limited (hospitalized) subgroup that has diagnostic information available ( Sondergard et al., 2008 ). We observed a sizeable association between consistent anticonvulsant or lithium purchase and reduced suicide risk among a broad, population-based sample of individuals likely to be using the medications for psychiatric purposes; further research is desirable to establish the generalizability of this finding and to provide further clarity concerning the potential reasons for these observed associations. Role of funding source This study was supported by a pilot Grant to Eric Smith, MD, MPH from the American Foundation for Suicide Prevention. The sponsors of the study had no role in study design, data collection, data analyses, data interpretation, or writing the report. The sponsors had no access to the data, the results, or the manuscript prior to submission. Conflict of interest All authors declare they have no conflict of interest. Acknowledgement This study was supported by a pilot Grant to Eric Smith, MD, MPH from the American Foundation for Suicide Prevention. The sponsors of the study had no role in study design, data collection, data analyses, data interpretation, or writing the report. The sponsors had no access to the data, the results, or the manuscript prior to submission. The views and opinions expressed in this report are those of the authors and should not be construed to represent the views of any sponsoring organization. References Ahrens et al., 1995 Ahrens B. Grof P. Moller H.J. Muller-Oerlinghausen B. Wolf T. Extended survival of patients on long-term lithium treatment Can. J. Psychiatry 40 1995 241 246 Angst et al., 2002 Angst F. Stassen H.H. Clayton P.J. Angst J. Mortality of patients with mood disorders: follow-up over 34–38 years J. Affect. Disord. 68 2002 167 181 Arensman and Kerkhof, 1996 Arensman E. Kerkhof J.F. Classification of attempted suicide: a review of empirical studies, 1963–1993 Suicide Life Threat. Behav. 26 1996 46 67 Baldessarini et al., 1999 Baldessarini R.B. Tondo L. Hennen J. Effect of lithium treatment and its discontinuation on suicidal behavior in bipolar manic–depressive disorders J. Clin. Psychiatry 60 Suppl. 2 1999 77 84 Baldessarini et al., 2006 Baldessarini R.J. Tondo L. Davis P. Pompili M. Goodwin F.K. Hennen J. Decreased risk of suicides and attempts during long-term lithium treatment: a meta-analytic review Bipolar Disorders 8 2006 625 639 Beautrais, 2001 Beautrais A.L. Suicides and serious suicide attempts: two populations or one? Psychol. Med. 31 2001 837 845 Cipriani et al., 2005 Cipriani A. Pretty H. Hawton K. Geddes J.R. Lithium in the prevention of suicidal behavior and all-cause mortality in patients with mood disorders: a systematic review of randomized trials Am. J. Psychiatry 162 2005 1805 1819 Clayton and Hills, 1993 Clayton D. Hills M. Statistical Models in Epidemiology 1993 Oxford University Press New York, NY Collins and McFarland, 2008 Collins J.C. McFarland B.H. Divalproex, lithium and suicide among Medicaid patients with bipolar disorder J. Affect. Disord. 107 2008 23 28 Coryell et al., 2003 Coryell W. Solomon D. Turvey C. Keller M. Leon A.C. Endicott J. Schettler P. Judd L. Mueller T. The long-term course of rapid-cycling bipolar disorder Arch. Gen. Psychiatry 60 2003 914 940 Danish National Board of Health (Laegemidelstyrelsen), 2007 Danish National Board of Health (Laegemidelstyrelsen). Medicinal Product Statistics. Available at: www.laegemiddelstyrelsen.dk . Accessed December 13, 2007. Danish National Board of Health (Sundhedsstyrelsen), 2007 Danish National Board of Health (Sundhedsstyrelsen). Cause of Death in Denmark. Available at: www.sundhedsstyrelsen.dk . Accessed December 13, 2007. Goodwin et al., 2003 Goodwin F.K. Fireman B. Simon G.E. Hunkeler E.M. Lee J. Revicki D. Suicide risk in bipolar disorder during treatment with lithium and divalproex JAMA 290 2003 1467 1473 Heeringen, 2001 Heeringen K.V. The suicidal process and related concepts K.V. Heeringen Understanding Suicidal Behaviour: The Suicidal Process Approach to Research, Treatment, and Prevention 2001 Wiley Publishers New York, NY 3 14 Kerhof and Arensman, 2001 Kerhof J.F. Arensman E. Pathway to suicide: the epidemiology of the suicidal process K.V. Heeringen Understanding Suicidal Behaviour: The Suicidal Process Approach to Research, Treatment, and Prevention 2001 Wiley Publishers New York, NY 15 39 Kessing et al., 2005 Kessing L.V. Sondergard L. Kvist K. Andersen P.K. Suicide risk in patients treated with lithium Arch. Gen. Psychiatry 62 2005 860 866 Kleindienst and Greil, 2000 Kleindienst N. Greil W. Differential efficacy of lithium and carbamazepine in the prophylaxis of bipolar disorder: results of the MAP study Neuropsychobiology 42 Suppl. 1 2000 2 10 Malig, 1996 Malig C. (December) The Civil Registration System in Denmark 1996 International Institute for Vital Registration and Statistics Bethesda, MD Muller-Oerlinghausen et al., 1992 Muller-Oerlinghausen B. Muser-Causmann B. Volk J. Suicides and parasuicides in a high-risk patient group on and off lithium long-term medication J. Affect. Disord. 25 1992 261 270 Nilsson, 1995 Nilsson A. Mortality in recurrent mood disorders during periods on and off lithium treatment Pharmacopsychiatry 28 1995 8 13 Schou, 1998 Schou M. The effect of prophylactic lithium treatment on mortality and suicidal behavior: a review for clinicians J. Affect. Disord. 50 1998 253 259 Schou and Weeke, 1988 Schou M. Weeke A. Did manic depressive patients who commit suicide receive prophylactic or continuation treatment at that time? Br. J. Psychiatry 1988 324 327 Simpson et al., 2006 Simpson S.H. Eurich D.T. Majumdar S.R. Padwal R.S. Tsuyuki R.T. Varney J. Johnson J.A. A meta-analysis of the association between adherence to drug therapy and mortality BMJ 333 2006 15 Sondergard et al., 2006 Sondergard L. Kvist K. Andersen P.K. Kessing L.V. Do antidepressants prevent suicide? Int. Clin. Psychopharmacol. 21 2006 211 218 Sondergard et al., 2008 Sondergard L. Lopez A.G. Andersen P.K. Kessing L.V. Mood-stabilizing pharmacologic treatment in bipolar disorders and risk of suicide Bipolar Disorders 10 2008 87 94 Thies-Flechtner et al., 1996 Thies-Flechtner K. Muller-Oerlinghausen B. Seibert W. Walther A. Greil W. Effect of prophylactic treatment on suicide risk in patients with major affective disorders. Data from a randomized prospective trial Pharmacotherapy 29 1996 103 107 Tondo and Baldessarini, 2000 Tondo L. Baldessarini R.B. Reduced suicide risk during lithium maintenance treatment J. Clin. Psychiatry 61 Suppl. 9 2000 97 104 Yerevanian et al., 2003 Yerevanian B.I. Koek R.J. Mintz J. Lithium, anticonvulsants and suicidal behavior in bipolar disorder J. Affect. Disord. 73 2003 223 228 Yerevanian et al., 2007a Yerevanian B.I. Koek R.J. Mintz J. Bipolar pharmacotherapy and suicidal behavior. Part I: lithium, divalproex, and carbamazepine J. Affect. Disord. 103 2007 5 11 Yerevanian et al., 2007b Yerevanian B.I. Koek R.J. Mintz J. Akiskal H.S. Bipolar pharmacotherapy and suicidal behavior. Part II: the impact of antidepressants J. Affect. Disord. 103 2007 13 21 Yerevanian et al., 2007c Yerevanian B.I. Koek R.J. Mintz J. Akiskal H.S. Bipolar pharmacotherapy and suicidal behavior. Part II: the impact of antipsychotics J. Affect. Disord. 103 2007 23 28