Introduction
Globally, over 800 000 people die each year because of suicide. Nevertheless, with the current suicide incidence rate of 15 per 100 000 for men and 8 per 100 000 for women worldwide (World Health Organization, 2014), longitudinal cohort studies examining risk factors for suicide are almost not feasible because of enormous costs and the necessary long follow-up period. A common way to circumvent the difficulty of examining risk factors for suicide is to examine risk factors for attempted suicide. A suicide attempt (SA) is obviously one of the strongest predictors of completed suicide (Fawcett et al., Reference Fawcett, Scheftner, Fogg, Clark, Young, Hedekker and Gibbons1990; Harris and Barraclough, Reference Harris and Barraclough1997; Suominen et al., Reference Suominen, Isometsä, Suokas, Haukka, Achte and Lönnqvist2004). Moreover, there is evidence for overlap between risk factors for suicide and SAs (Beautrais, Reference Beautrais2003; Nock et al., Reference Nock, Borges, Bromet, Cha, Kessler and Lee2008). Therefore, research on risk factors for attempted suicide can provide useful tools for suicide prevention.
Mental disorders represent a major source of risk for suicidality (Nock et al., Reference Nock, Borges, Bromet, Cha, Kessler and Lee2008). This is especially true for a major depressive disorder (MDD) (Kessler et al., Reference Kessler, Borges and Walters1999). A population-based hospital study reported that MDD was associated with a 20-fold increased risk of suicide (Ösby et al., Reference Ösby, Brandt, Correira, Ekbom and Sparén2001) and almost 16% of the persons in the community with a lifetime MDD disorder reported an SA somewhere in their lifetime (Chen and Dilsaver, Reference Chen and Dilsaver1996). The finding that MDD is one of the major risk factors for suicidality is not surprising given the fact that suicidal ideation is one of the defining symptoms of MDD. Moreover, some (recent) research suggests that patients with MDD primarily attempted suicide during an active major depressive episode (Holma et al., Reference Holma, Melartin, Haukka, Holma, Sokero and Isometsa2010; Riihimäki et al., Reference Riihimäki, Vuorilehto, Melartin, Haukka and Isometsä2014).
There is evidence however that factors other than clinical, depression-related factors such as young age (Bolton et al., Reference Bolton, Belik, Enns, Cox and Sareen2008), low income (Bolton et al., Reference Bolton, Belik, Enns, Cox and Sareen2008), lack of a partner (Sokero et al., Reference Sokero, Melartin, Rytsälä, Leskelä, Lestelä-Mielonen and Isometsä2005), adverse childhood experiences (Molnar et al., Reference Molnar, Berman and Buka2001), (cigarette) smoking (Mann et al., Reference Mann, Waternaux, Haas and Malone1999; Oquendo et al., Reference Oquendo, Galfalvy, Russo, Ellis, Grunebaum, Burke and Mann2004), (illicit) drugs use disorders (Bolton et al., Reference Bolton, Belik, Enns, Cox and Sareen2008, Reference Bolton, Pagura, Enns, Grant and Sareen2010), alcohol use disorders (Mann et al., Reference Mann, Waternaux, Haas and Malone1999; Sokero et al., Reference Sokero, Melartin, Rytsala, Leskela, Lestela-Mielonen and Isometsa2003), somatic chronic diseases (Duggan et al.,Reference Duggan, Sham, Lee and Murray1991), poor sleep quality (Ağargün et al., Reference Ağargün, Kara and Solmaz1997) and personality traits (Mann et al., Reference Mann, Waternaux, Haas and Malone1999; Oquendo et al., Reference Oquendo, Galfalvy, Russo, Ellis, Grunebaum, Burke and Mann2004) also contribute to suicidal behaviour in persons with MDD. However, our knowledge about prospective-independent risk factors for attempted suicide in MDD is limited (Oquendo et al., Reference Oquendo, Currier and Mann2006). Most information about associations between SAs and their possible risk factors is derived from retrospective and cross-sectional studies which are unable to unravel causal relationships. Furthermore, almost all prospective studies on risk factors for attempted suicide in MDD have been conducted in clinical samples (Oquendo et al., Reference Oquendo, Currier and Mann2006), although there are some recent studies conducted among individuals with MDD in a nationally representative sample (Bolton et al., Reference Bolton, Pagura, Enns, Grant and Sareen2010) and in a sample of primary care patients with mostly MDD (Riihimäki et al., Reference Riihimäki, Vuorilehto, Melartin, Haukka and Isometsä2014). Both studies examined sociodemographic and clinical risk factors for future SAs. They reported that a lifetime SA and mental illness comorbidity were significant predictors for future SAs. Bolton et al. (Reference Bolton, Pagura, Enns, Grant and Sareen2010) found that young age and low income were independently associated with future SAs whereas Riihimäki et al. (Reference Riihimäki, Vuorilehto, Melartin, Haukka and Isometsä2014) found strong evidence that severity of depression predicted SAs. Both studies, however, did not examine social, lifestyle, health and personality factors. There is an urgent need to assess risk factors from different domains in a comprehensive way to predict future suicidal acts (Oquendo et al., Reference Oquendo, Currier and Mann2006). Moreover, it is not clear whether there is an elevated risk for future SAs when a person is exposed to an increasing number of risk factors. This knowledge is necessary for designing effective suicide prevention programs.
The main aim of the present study is to examine in a comprehensive way a wide range of putative non-clinical (demographic, social, lifestyle and personality) and clinical (depression and suicidal-related) risk factors for future SAs among treated and untreated persons with MDD. Moreover, we examine whether there is a linear relationship between the number of significant predictors and the incidence of a future SA.
Methods
Sample
‘The Netherlands Study of Depression and Anxiety (NESDA) is an ongoing longitudinal cohort study designed to examine the long-term course and consequences of depressive and anxiety disorders in adults (18–65 years). One of the most important objectives of NESDA is to obtain a full understanding of the course of depressive and anxiety disorders. Therefore, the sample is stratified for setting (community, primary care and specialized mental health) and set up to include a range of psychopathology: those with no symptoms or disorders, those with earlier episodes or at risk because of subthreshold symptoms or family history, and those with a current first, recurrent or chronic depressive or anxiety disorder. In the period between September 2004 and March 2007, NESDA included 2981 respondents from the mentioned different healthcare settings and in different clinical developmental stages of illness. The respondents were recruited when newly enrolled at mental health care organizations (807 persons), through a screening approach conducted in primary care (1610 persons) and through the community (564 persons). Clinical exclusion criteria were a primary clinical diagnosis of other psychiatric disorders (e.g. psychotic disorder, obsessive-compulsive disorder, bipolar disorder or a severe addiction disorder) and not being sufficiently fluent in Dutch. Research assistants received 1 week of intensive training by the fieldwork coordinator. Quality control of interviewing was further conducted by providing feedback based on an evaluation of a random selection of 10% of all taped interviews. During training and feedback, the emphasis was given to question wording and probing the behavior of interviewers. The study protocol was approved by the ethical review board of the participating centres and all respondents signed a written informed consent. For more details about the rationale, objectives and methods of NESDA see Penninx et al., Reference Penninx, Beekman, Smit, Zitman, Nolen, Spinhoven, Cuijpers, de Jong, van Marwijk, Assendelft, van der Meer, Verhaak, Wensing, de Graaf, Hoogendijk, Ormel and van Dyck2008.
Follow-up assessments were conducted after 2 years (N = 2596, 87.1%), 4 years (N = 2402, 80.6%) and 6 years (N = 2256, 75.7%), including the same face-to-face interview and questionnaires as the baseline assessment. The present study used these various data cycles over 6 years. Of the 2981 respondents, 1925 respondents had a lifetime MDD. MDD (DSMIV diagnoses 296.2× and 296.3×) was assessed with the Composite Interview Diagnostic Instrument (CIDI), version 2.1, which classifies diagnoses according to DSM-IV criteria (American Psychiatric Association, 2001). We only included the lifetime MDD respondents with at least one follow-up measurement with SA information, which resulted in a sample size of 1713 respondents. Those with no follow-up measurements were less educated (11.2 v. 12.0 years, p = 0.001), had more manic or hypomanic symptoms (M = 6.1 v. M = 5.2, p = 0.007, had more depressive symptoms (M = 31.0 v. M = 25.9, p < 0.001), had more often a current depressive and anxiety disorder (54.9% v. 36.7%), p < 0.001) and had more often chronic symptoms (34.1% v. 25.8%), p < 0.001) than those with at least one follow-up measurement. No differences between groups were found for any of the other demographic characteristics (gender, age, employment status, income) and clinical characteristics (age of onset depression, level of insomnia, antidepressant and psychotherapy use, current suicidal thoughts and history of attempted suicide) at baseline.
Future SA
SAs were assessed in NESDA in all face-to-face follow-up assessments. The question is derived from the WHO/Euro multicentre study on parasuicide (Platt et al., Reference Platt, Bille-Brahe, Kerkhof, Schmidtke, Bjerke, Crepet, De Leo, Haring, Lonnqvist, Michel, Philippe and Pommereau1992). To assess a history of SAs the following question was asked: ‘Have you since the last interview made a serious attempt to end your life, for instance by harming or poisoning yourself or by getting into an accident? No/yes’. If the respondent answered ‘yes’ to this question, the number of SAs and the year of the last SA were assessed.
Time to the first occurrence of SA was calculated. Since no exact date but only the overall year of the reported SA at follow-up was reported, we had to estimate the time to the first onset of an SA. In case the reported year of SA lies entirely within a 2-year follow-up period, the date was set in the middle of the year (1 July). In case the year of the reported attempt lies in the year of a measurement, the date of the attempt was set in the middle of the period between 1 January and the date of previous/current interview. In case of multiple attempts at a particular follow-up assessment only the year of the last attempted was enquired and the year of the first attempt was not clear. When a respondent reported e.g. two SA's at a particular follow-up assessment we used the date lying in the middle between the previous interview and the second attempt to determine the year of the first attempt.
Predictor variables
Six clusters of variables were assessed as possible predictors of future SAs. All predictor variables were assessed at the baseline interview.
Demographics
Gender, age, education in number of years, employment status (employed; occupational disabled/on sickness benefit; unemployed; other) and net income of the household in Euros per month were assessed during the baseline interview.
Social factors
Social factors consisted of marital/partner status (not married and no partner; married or partner), loneliness, childhood trauma and childhood life events. Loneliness was measured with the 11-item self-report Loneliness scale (De Jong Gierveld and kamphuis, Reference de Jong Gierveld and Kamphuis1985). Childhood trauma and childhood life-events were assessed retrospectively using the childhood trauma interview (CTI) (de Graaf et al., Reference de Graaf, Bijl, Smit, Vollebergh and Spijker2002). As childhood trauma indicators the occurrence and frequency of emotional neglect, psychological abuse, physical abuse and sexual abuse before age 16 were assessed. A cumulative index (0–8) was calculated by adding the frequencies of occurrence (0 = absent; 1 = once or sometimes; 2 = regularly, often or very often) of the different childhood trauma indicators. Finally, the cumulative childhood trauma index was categorized into five categories (0–4). Presence of a childhood life event was assessed by asking respondents if the following events had occurred before age 16: parental loss, divorce of parents, placed in a juvenile prison, raised in a foster family or placed in a child home.
Lifestyle and health
Lifestyle and health factors consisted of cigarette smoking status, use of illicit drugs, alcohol intake, number of chronic somatic diseases. Cigarette smoking status was classified into current cigarette smokers and non-smokers. Use of illicit drugs status was assessed in the past month (yes; no). Alcohol intake was assessed with the Alcohol Use Disorders Identification Test (AUDIT) (Saunders et al., Reference Saunders, Aasland, Babor, De la Fuente and Grant1993) and was categorized as non-drinker, mild/moderate drinker (1–21 glasses/week for men and 1–14 glasses for women) and heavy drinker (>21 glasses/week for men and >14 glasses/week for women). Treated somatic chronic diseases (lung disease, cancer, ostheoarthritis, intestinal disorders, liver disease, epilepsy, thyroid, gland disease, cardiovascular disease and diabetes) were measured with an instrument that is derived from the chronic disease questionnaire of Statistics Netherlands (CBS, 1989). The selection of these chronic diseases was based on prevalence and functional consequences.
Personality
Personality was measured by the NEO-FFI, a short form of the NEO-PI-R by Costa and McCrae (Reference Costa and McCrae1992). Neuroticism, extraversion, openness, agreeableness and conscientiousness were the domains which were measured with the questionnaire. Mastery or locus of control was measured by a translated five-item abbreviated version of the Pearlin Mastery Scale (Pearlin and Schooler, Reference Pearlin and Schooler1978; Ormel et al., Reference Ormel, Kempen, Wolffensperger, Steverink, van Eijk and Brilman1992). Higher scores on this scale indicate more control of events/situations. Aggression was measured by a subscale of the Leiden Index of Depression Sensitivity-Revised (LEIDS-R) (van der Does, Reference van der Does2002) which measures cognitive thoughts during sad mood.
Depression-related characteristics
Like the MDD, dysthymia and anxiety disorders (social phobia, generalized anxiety disorder, panic disorder and agoraphobia) were assessed with the CIDI, version 2.1, which classifies diagnoses according to DSM-IV criteria (American Psychiatric Association, 2001). In NESDA specially trained clinical staff conducted the CIDI. The CIDI is a reliable and valid instrument for assessing depressive and anxiety diagnoses (Wittchen, Reference Wittchen1994). A diagnostic status variable was constructed which classified respondents as having 1 = remitted depressive and/or anxiety disorder, 2 = current depressive disorder only, 3 = current anxiety disorder only and 4 = current depressive and anxiety disorder (current comorbid-anxiety depression). A 6-month disorder was seen as a current disorder.
Depression severity was assessed with the 30-item Inventory of Depressive Symptomatology self-report version (Rush et al., Reference Rush, Gullion, Basco, Jarrett and Trivedi1996). The lifetime severity of manic or hypomanic symptoms was assessed with the 13-item Mood Disorder Questionnaire (Hirschfeld et al., Reference Hirschfeld, Williams, Spitzer, Calabrese, Flynn, Keck, Lewis, McElroy, Post, Rapport, Russell, Sachs and Zajecka2000). The duration of symptoms was assessed with the LifeChart method (Lyketsos et al., Reference Lyketsos, Nestadt, CWI, Heithoff and Eaton1994). This method assessed the percent of time depressive symptoms were present in the last 4 years using a calendar event recall method. Furthermore, the age of onset of the depressive disorder was assessed in the CIDI interview. Insomnia was assessed with the five-item self-report Women's Health Initiative Insomnia Rating Scale (IRS) (Levine et al., Reference Levine, Kripke, Kaplan, Lewis, Naughton, Bowen and Shumaker2003). Antidepressant use in the past month was obtained by inspection of the medication boxes of all medication used in the past month. It was classified according to the Anatomical Therapeutic Chemical (ATC) classification system: tricyclic antidepressants (TCAs) (ATC-code N06AA), selective serotonin re-uptake inhibitors (SSRIs) (ATC-code N06AB) and other antidepressants (ATC-code N06AF/N06AX). Receiving psychotherapy in the last 6 months (either formal psychotherapy or counselling) was assessed by using the TIC-P questionnaire (Hakkaart-van Rooijen, Reference Hakkaart-van Roijen2002).
Suicidal characteristics
Suicidal thoughts at baseline were assessed with the five-item Scale for Suicide Ideation (SSI, (Beck et al., Reference Beck, Kovacs and Weissman1979)). Each item had three answer categories in ascending order of severity (0–2). A positive score (1 or 2) on one of the five items was used to define a suicidal ideator (0 = no ideation, 1 = ideation). A history of lifetime SAs before baseline was assessed with the same question as that for future SAs, but at baseline referring to the lifetime period before baseline.
Statistical analyses
In subjects with a lifetime MDD, baseline characteristics of those with and without a future SA over a follow-up period of 6 years were summarized using descriptive statistics. Cox proportional hazard regression analyses were used to examine a wide range of possible predictors (demographic, social, lifestyle, personality and clinical) for future SAs. For respondents who attempted suicide after the baseline measurement time-to-event was calculated as the time between baseline and follow-up when the outcome occurred for the first time. Those who did not attempt suicide after the baseline measurement were censored at the date of their last follow-up measurement. The proportionality of hazards assumption was checked by including a time-to-event-by-predictor in separate models and was met for all predictors (p > 0.05).
To identify which predictors were independent risk factors for future SAs, predictors that showed a significant association (p < 0.10) in univariable Cox regression analyses were entered into multivariable analyses per cluster of predictors. Next, non-clinical predictors (demographic, social, lifestyle and personality characteristics) with a significant association (p < 0.10) in the per cluster analyses were entered into multivariable analyses by the forced enter method. In a first step of the final analyses, the remaining significant non-clinical predictors (p < 0.10), and the significant clinical depression-related predictors (p < 0.10) of the per cluster analyses, were entered into multivariable analyses by the forced enter method and in a second step the significant clinical suicidal predictors (p < 0.10) of the per cluster analyses were added to the final multivariate model. The cut-off of p < 0.10 was used to minimize the chance of missing any relevant variables in multivariable analyses. By constructing the models in a way that variables concerning a similar subject were analysed first within this specific cluster gives the reader a comprehensive picture of the predictive value of the different domains of risk factors (demographic, social, lifestyle, personality, depression and suicide-related risk factors). In addition, we assessed the relationship between the number of significant predictors derived from the final multivariate model and incidence of SAs. Multicollinearity was checked by evaluating the VIF scores, and no multicollinearity problems were found. All statistical analyses were performed using the Statistical Package for Social Sciences (SPSS), version 20.0 for Windows and for all analyses two-tailed tests were used with α = 0.05. Some of the baseline predictors had missing data, ranging from 0.1 to 10.9% (see also Table 1). For the Cox regression analyses the missing values for the continuous variables were replaced with the mean value and for the categorical variables, the few missing values were replaced by the mode.
Baseline characteristics of persons with a lifetime MDD without v. with an incident suicide attempt
a Based on descriptive statistics.
Results
The mean age of the study sample at baseline (N = 1713) was 42.1 years (s.d. = 12.3), 68.9% was female and the mean years of education of the respondents was 12.0 years (s.d. = 3.2). Fifty-eight respondents (3.4%) attempted suicide at some time during the follow-up period (incidence rate: 5.7 cases per 1000 person-years) and more than half (53.4%) of these respondents attempted suicide in the first 2 years after the baseline interview. Twenty-two (37.9%) respondents attempted suicide more than once (median = 3 attempts, range = 2–28 attempts). Table 1 summarizes the characteristics of persons with a lifetime MDD without a future SA v. the persons with a lifetime MDD with a future SA.
Table 2 shows the results of univariable and multivariable Cox regression analyses of the non-clinical predictors. In these analyses, only the demographic characteristics younger age, lower education and employment status were significantly associated with a higher hazard for future SAs in multivariable analyses. Univariate and per cluster analyses showed an association between social factors (loneliness and childhood life events), lifestyle & health factors (cigarette smoking and illicit drugs use) and personality factors (higher neuroticism, lower extraversion, lower agreeableness, lower control of events and higher aggression) on the one hand and future SAs on the other hand. However, in the multivariate analyses, no significant relationship was found between these factors and future SAs. No significant relationship at all was found with partner status, childhood trauma and alcohol use.
Predictive value of non-clinical risk factors for an incident suicide attempt in respondents with a lifetime MDD at baseline in univariable and per cluster analyses
a Based on univariate Cox regression analyses.
b Based on multivariate Cox regression analyses, including all the variables within the same cluster that had an association of p < 0.10 in univariate analyses.
c Based on multivariate Cox regression analyses, including all the variables that had an association of p < 0.10 in the per cluster analyses.
d HR per s.d. increase; s.d. for Loneliness = 3.5; s.d. for Neuroticism = 8.0; s.d. for Extraversion = 7.0; s.d. for Openness = 6.0; s.d. for Agreeableness = 5.3.
s.d. for Conscientiousness = 6.1; s.d. for Control of events = 4.0; s.d. for Aggression = 4.3.
Table 3 shows the results of univariable and multivariable Cox regression analyses of the clinical predictors. In the depression-related domain a higher level of manic symptoms, higher severity of depression, a current or comorbid depression, longer duration of the depressive disorder, a higher insomnia score, use of antidepressants and receiving psychotherapy increased the risk of a future SA in univariable analyses. In the multivariable analyses within the depression-related domain, only severity of depression, insomnia and use of antidepressant medication remained significantly associated. The use of SSRIs, as well as TCAs and other antidepressants, is approximately twice as common in those with future SAs than in non-attempters. The association with antidepressants was consistent across different types of medications, as there is no differential risk in respondents using SSRIs compared with respondents using TCAs (HR 1.37, 95% CI 0.47–3.98, p = 0.56) or other antidepressants (HR 1.17, 95% CI 0.52–2.64, p = 0.70). Furthermore, current suicidal ideation, as well as a previous SA, strongly predicted future SAs in univariable and multivariable analyses within the suicidality-related domain.
Predictive value of clinical psychiatric risk factors for an incident suicide attempt in respondents with a lifetime MDD at baseline in univariable and per cluster analyses
a Based on univariate Cox regression analyses.
b Based on multivariate Cox regression analyses, including all the variables within the same cluster that had an association of p < 0.10 in univariate analyses.
c HR per s.d. increase; s.d. for IDS = 13.3; s.d. for MDQ = 3.1; s.d. for WHI-IRS = 4.9.
The final multivariable Cox regression analyses of all the remaining significant predictors derived from the per cluster analyses are shown in Table 4. Both non-clinical (younger age, lower education and unemployment) and depression-related clinical factors (severity of depression, insomnia and antidepressant use) predicted a future SA. When adding the clinical suicidality characteristics (current suicidal thoughts and a previous SA) to the model, all predictors except a higher severity of depression predicted a future SA significantly. The overall explained variance of our model measured with generalized R-squared, is 0.061.
Predictive value of risk factors for an incident suicide attempt in respondents with a lifetime MDD at baseline in multivariable analyses
a Based on multivariate Cox regression analyses, including all the variables that had an association of p < 0.10 in the multivariate analyses of the non-clinical variables in Table 2 and the clinical psychiatric related variables in Table 3.
b Based on multivariate Cox regression analyses, including all the variables that had an association of p < 0.10 in the first multivariate analyses in Table 4 and the significant suicidality characteristics.
c HR per s.d. increase; s.d. for IDS = 13.3; s.d. for WHI-IRS = 4.9.
Finally, we assessed the relationship between the number of significant predictors derived from the final multivariate model and incidence of SAs (see Fig. 1). The predictor's age (younger than 44 or older than 43) and years of education (<12 years of education or more than 11 years of education) were dichotomized by median split. Insomnia was dichotomized based on the IRS cut-off point of 9. The other four predictors were already dichotomous. Figure 1 shows that 0.0% of the respondents with no predictors attempted suicide (N = 75). The SA incidence was slightly higher in those with one, two or three risk factors (0.6, 1.0 and 2.9%, respectively). Incidences were substantially higher among those with four or five risk factors (8.0 and 9.7%) reaching 32.3% in the combined group of those with six or all seven risk factors present. In Cox's survival analyses, we found that – as compared with persons with 0–1 risk factors – the risk for a future SA was 3.92 (95% CI 0.91–16.95, p = 0.07) for persons with 2–3 risk factors, 18.56 (95% CI 4.43–77.76, p < 0.001) for persons with 4–5 risk factors and 77.14 (95% CI 16.90–352.19, p < 0.001) for persons with 6–7 risk factors.
Relationship between number of significant predictors* and incidence of suicide attempts (N = 1713). *The seven significant predictors are: younger age (younger than 44), lower education (<12 years of education), unemployment, insomnia (higher Insomnia Rating Scale score than 8), current antidepressant use, current suicidal thoughts and a history of suicide attempts before baseline.
Discussion
To our knowledge, this is one of the few studies that prospectively examined a wide range of possible risk factors for subsequent SAs in a large sample of persons with a lifetime MDD. Over a period of 6 years, 3.4% of the respondents attempted suicide. Younger age, lower education and unemployment were the only non-clinical factors that predicted a future SA independently. Furthermore, clinical factors, such as insomnia, anti-depressant use and both current suicidal ideation as well as a previous SA predicted future SAs independently. Overall, the seven independent risk factors showed a linear increase in predicting future SA incidence. Those with zero risk factors had no incidence of SA, but in those with 6 or 7 risk factors, the incidence of SA was 30%.
The incidence of SAs in the 6-year follow-up period in our study (3.4%) was in line with the incidence of 2.7% reported in a comparable community study of individuals with lifetime MDD (Bolton et al., Reference Bolton, Pagura, Enns, Grant and Sareen2010) although the follow-up period was only 3 years in that study. In a prospective study of suicidal behaviour in primary care patients with mostly MDD (Riihimäki et al., Reference Riihimäki, Vuorilehto, Melartin, Haukka and Isometsä2014) the percentage of suicide attempters was much higher compared with our study. In this study, 10% of the patients attempted suicide within 5 years. A possible explanation for this large difference could be that the respondents in this study had a current depressive disorder at baseline instead of a lifetime disorder, putting them at higher risk for SA.
Socio-economic factors such as lower education and unemployment were predictive of SAs independent of clinical factors. Our finding that social disadvantage predicts SAs is consistent with another study in depressed persons where lower income predicted future SAs independent of depression severity (Bolton et al., Reference Bolton, Pagura, Enns, Grant and Sareen2010). Moreover, results from studies in other psychiatric samples (persons with a borderline personality disorder, persons admitted to psychiatric hospital) showed also that attempters were characterized by low socio-economic status and low educational achievement (Mann et al., Reference Mann, Waternaux, Haas and Malone1999; Soloff and Chiappetta, Reference Soloff and Chiappetta2012). These results suggest that persons with mental disorders and social economic disadvantages are particularly at risk for attempted suicide. No other significant relationships were found between non-psychiatric factors and future SAs.
Not having a partner was not associated with future SAs and the association with loneliness disappeared in multivariate analyses. These results are partly in line with studies in depressed persons that reported that marital status is not predictive of SAs in multivariate analyses (Bolton et al., Reference Bolton, Pagura, Enns, Grant and Sareen2010) and with a systematic review that reported that suicide was not associated with marital status in persons with depression (Hawton et al., Reference Hawton, Casaňas i Comabella, Haw and Saunders2013). However, other studies in a sample of MDD patients reported that lacking a partner and poor social support were one of the most robust predictors for SAs (Sokero et al., Reference Sokero, Melartin, Rytsälä, Leskelä, Lestelä-Mielonen and Isometsä2005; Holma et al., Reference Holma, Melartin, Haukka, Holma, Sokero and Isometsa2010). So further research is needed to determine how important proxies for (emotional) social support are for predicting suicide risk in depressed persons. Surprisingly, we did not find an association with childhood abuse and future SAs. This is not in line with a recent meta-analyses that reported that childhood maltreatment is positively associated with suicidal behaviour (Liu et al., Reference Liu, Fanga, Gong, Cui, Menga, Xiao, He, Shen and Luo2017) and a meta-analyses that reported evidence of an association between childhood sexual abuse and incident suicidal behaviour (Devries et al., Reference Devries, Mak, Child, Falder, Bacchus, Astbury and Watts2014). However, none of the studies in the meta-analyses of Devries et al. (Reference Devries, Mak, Child, Falder, Bacchus, Astbury and Watts2014) controlled for a diagnosis of major depression and Liu et al. (Reference Liu, Fanga, Gong, Cui, Menga, Xiao, He, Shen and Luo2017) reported that in some of the subgroup analyses no significant effect was observed between childhood maltreatment and suicidal behaviour in individuals with mood disorders.
These inconclusive results warrant more research concerning the relationship between childhood abuse and attempted suicide in persons with an MDD especially later in life.
Heavy alcohol use did not predict future SAs in our study. This is in line with studies in depressed persons which did not find an association between alcohol dependence/abuse and future SAs (Sokero et al., Reference Sokero, Melartin, Rytsälä, Leskelä, Lestelä-Mielonen and Isometsä2005) or did not find an independent association in multivariate models (Bolton et al., Reference Bolton, Belik, Enns, Cox and Sareen2008, Reference Bolton, Pagura, Enns, Grant and Sareen2010; Holma et al., Reference Holma, Melartin, Haukka, Holma, Sokero and Isometsa2010) with SAs. However, other studies reported strong independent associations between alcohol use disorders and future SAs in depressed persons (Isometsä, Reference Isometsä2014; Riihimäki et al., Reference Riihimäki, Vuorilehto, Melartin, Haukka and Isometsä2014;). Further studies are necessary to examine the complicated relationship between substance (mis)use and SAs. Cigarette smoking was only associated with future SAs in univariate analyses and per cluster analyses, but not in multivariate analyses. Other studies in depressed persons reported the importance of cigarette smoking (Mann et al., Reference Mann, Waternaux, Haas and Malone1999; Oquendo et al., Reference Oquendo, Galfalvy, Russo, Ellis, Grunebaum, Burke and Mann2004; Bolton et al., Reference Bolton, Pagura, Enns, Grant and Sareen2010) as a predictor for SAs. There are indications that this association is partly explained by more aggression/impulsivity traits among depressed smokers compared with non-depressed smokers (Oquendo et al., Reference Oquendo, Currier and Mann2006). In previous research Oquendo et al. (Reference Oquendo, Galfalvy, Russo, Ellis, Grunebaum, Burke and Mann2004) also reported the importance of the presence of the aggression/impulsivity traits for predicting SAs. Although we found a univariate relationship between aggression and future SAs, it lost significance when controlling for the other significant personality domains. Unfortunately, impulsivity was not assessed at the NESDA baseline measurement, so we were not able to test this possible important predictor in full force. While finding some bivariate associations, no independent associations were found with the other measured personality domains. This is partly in line with a recent study in depressed persons which reported that the relationship between personality features and SAs is not direct, but is mediated by the time spent in major depressive episodes (Jylhä et al., Reference Jylhä, Rosenström, Mantere, Suominen, Melartin, Vuorilehto, Holma, Riihimäki, Oquendo, Keltikangas-Järvinen and Isometsä2016). We found no association between somatic chronic diseases and future SAs. This result is discrepant with a study in a relatively small number of depressed inpatients (Duggan et al., Reference Duggan, Sham, Lee and Murray1991) but in line with some larger studies in depressed populations (Sokero et al., Reference Sokero, Melartin, Rytsälä, Leskelä, Lestelä-Mielonen and Isometsä2005; Holma et al., Reference Holma, Melartin, Haukka, Holma, Sokero and Isometsa2010; Riihimäki et al., Reference Riihimäki, Vuorilehto, Melartin, Haukka and Isometsä2014). Moreover, another study reported that that the association between chronic psychical disease and completed suicide is explained by clinical depression (Webb et al., Reference Webb, Kontopantelis, Doran, Qin, Creed and Kapur2012).
In the present study also, some depression and suicidality-related factors predicted future SAs independently. A lifetime history of SAs and suicidal ideation at baseline were predictive for future SAs in the present study independent of depression severity. There is a large body of research about the predictive value of previous SAs for future suicidal acts in depressed populations (Fawcett et al., Reference Fawcett, Scheftner, Fogg, Clark, Young, Hedekker and Gibbons1990; Bolton et al., Reference Bolton, Pagura, Enns, Grant and Sareen2010; Holma et al., Reference Holma, Melartin, Haukka, Holma, Sokero and Isometsa2010; Riihimäki et al., Reference Riihimäki, Vuorilehto, Melartin, Haukka and Isometsä2014). In the literature, the importance of suicidal ideation as a predictor for SAs is however unclear. Most studies found no relationship (Riihimäki et al., Reference Riihimäki, Vuorilehto, Melartin, Haukka and Isometsä2014) or found a univariate relationship between suicidal ideation and future SAs but failed to find an independent relationship (Bolton et al., Reference Bolton, Pagura, Enns, Grant and Sareen2010; Holma et al., Reference Holma, Melartin, Haukka, Holma, Sokero and Isometsa2010). A possible explanation could be that in most of these studies lifetime suicidal ideation was assessed while in our study suicidal ideation in the past week was assessed. Other studies that reported that the presence of suicidal ideation was a risk factor for future suicidal acts assessed indeed current suicidal ideation (Fawcett et al., Reference Fawcett, Scheftner, Fogg, Clark, Young, Hedekker and Gibbons1990; Oquendo et al., Reference Oquendo, Galfalvy, Russo, Ellis, Grunebaum, Burke and Mann2004). Insomnia was a consistent predictor of future SAs independent of depression severity. This finding concurs with a recent population-based study in older adults which showed that poor subjective sleep quality increased the risk of death by suicide 10 years later independent of depression severity (Bernert et al., Reference Bernert, Turvey, Conwell and Joiner2014) and earlier findings in depressed persons that insomnia/poor subjective sleep quality predicted suicide (Fawcett et al., Reference Fawcett, Scheftner, Fogg, Clark, Young, Hedekker and Gibbons1990) or was associated with higher suicidality (Ağargün et al., Reference Ağargün, Kara and Solmaz1997). Both insomnia and suicidal ideation are symptoms of an MDD and are part of the depression severity scale used in this study. This is probably an explanation why depression severity failed to predict future SAs in our final multivariate model. It suggests that insomnia and suicidal ideation are key symptoms in the relationship between MDD and attempted suicide. Concluding on the bases of these results that depression severity is an unimportant predictor for future SAs would therefore probably be erroneous.
Besides our analyses with single predictors, we have also examined the relationship between the number of significant predictors and the incidence of a future SA. Our results showed that the number of significant risk factors linearly predicted the incidence of future SAs, suggesting an elevated risk for a future SA when a person is exposed to an increasing number of risk factors. This knowledge could potentially be useful for clinicians when evaluating suicide risk and may contribute to designing effective suicide prevention programs.
In univariate analyses, both receiving psychotherapy and antidepressant use predicted future SAs. Contrary to the association with antidepressant use the association with psychotherapy lost its significance after controlling for other depression-related severity characteristics. It seems therefore that depression-severity is not the main explanation for finding an association between antidepressant use and future SAs. However, in observational epidemiologic studies of drug effects/safety confounding by indication is frequently encountered, because the allocation of treatment is not randomized and the indication of treatment may e.g. be related to severity, prior history and risk of side-effects (Signorello et al., Reference Signorello, McLaughlin, Lipworth, Friis, Toft Sørensen and Blot2002), resulting in possible biased results. Although it was possible to control for depression severity, suicidal ideation and a history of SA and therefore reducing the risk for channelling bias we cannot rule it out completely. That is that the patients at increased risk for (attempted) suicide were more likely to get antidepressants prescribed. It is still possible that confounding-by-indication had an influence on our results, because we had in our model no (sufficient) information about depression chronicity, a number of depressive episodes in the past, the severity of the suicidal thoughts, or a family history of suicidality. Moreover, it is possible that SA occurred long after discontinuing all antidepressant use. Considering the limitations of this design for examining drug effects/safety these results should be interpreted with caution and primarily serve as a first identifying step to examine antidepressant effects/safety in more suitable designed studies.
Strengths of this study were that the depressed individuals were from different healthcare settings and in different clinical development stages of the MDD. So our results are probably better suited to generalize to a wider population of persons with MDD, than studies in clinical samples. Moreover, the large sample size and longitudinal design of the study were major strengths. Furthermore, the NESDA study had a relatively wide range of predictors, allowing meaningful analyses. Finally, the attrition rate was rather low. We interviewed 89% of the sample more than once and the use of Cox proportional hazard regression analyses allowed us to make use of all these respondents despite their different lengths of time in the study. While the risk for reporting errors is rather high when assessing a lifetime history of SAs (Eikelenboom et al., Reference Eikelenboom, Smit, Beekman, Kerkhof and Penninx2014), this risk is probably much lower when assessing SAs in shorter follow-up periods. However, some limitations should also be mentioned. First, despite the relatively large cohort and the 6-year follow-up period, the number of respondents who attempted suicide at follow-up was not very high. The rule of thumb that Cox models should be used with a minimum of 10 outcome events per predictor variable (EPV) was therefore in some of our multivariate models violated. However, the number of EVP was in none of the models lower than 7. Moreover, a simulation study showed that systematically discounting results from any model with 5–9 EVP does not appear to be justified (Vittinghof and McCulloch, Reference Vittinghof and McCulloch2006). Nevertheless, although the main findings are robust there is a chance for type-II errors, meaning failing to detect a significant (small) effect that is present. Therefore, the results should be interpreted with some caution. Second, there was no information about the intent to die and the lethality of the reported SA. Moreover, the information about SAs was based on a report during the face-to-face interview. It is imaginable that given the ongoing debate among experts about the meaning of the term SA (O'Carroll et al., Reference O'Carroll, Berman, Maris, Moscicki, Tanney and Silverman1996; Silverman et al., Reference Silverman, Berman, Sanddal, O'Carroll and Joiner2007) respondents and also trained research assistants are unlikely to have a clear idea as to which behaviours fall under the definition of a SA. This could result in inaccurate incidence rates of SAs. So there were no gold standard methods of the SA measure providing external validity. However, such a gold standard is not likely to exist currently in suicide research. Third, it is possible that our findings do not completely apply to completed suicide although attempted suicide is closely related with completed suicide (Isometsä, Reference Isometsä2014). Fourth, certain risk factors including personality disorders, post-traumatic stress disorder and impulsivity were not studied. Fifth, because we have excluded those who were not fluent in Dutch our results are only generalizable to our native Dutch population not necessary to e.g. migrant populations.
In conclusion, our results suggest that focusing on depression treatment is important but will not be enough in significantly reducing the risk for SAs. Our results give evidence that especially young currently depressed persons with a history of attempted suicide, suicidal ideations, insomnia, antidepressant use and socio-economic disadvantages (low education and unemployed) are at high risk for attempting suicide in the (near) future.
Acknowledgements
The infrastructure for the NESDA study (http://www.nesda.nl) is funded through the Geestkracht program of the Netherlands Organization for Health Research and Development (ZonMW, grant number 10–000–1002) and financial contributions by participating universities and mental healthcare organizations (VU University Medical Center, GGZ inGeest, Leiden University Medical Center, Leiden University, GGZ Rivierduinen, University Medical Center Groningen, University of Groningen, Lentis, GGZ Friesland, GGZ Drenthe, Rob Giel Onderzoekcentrum).
Financial support
Penninx has received research funding from Jansen Research and Boehringer Ingelheim.
