INTRODUCTION
Several studies have found that members of some occupations are at greatly increased risk of suicide (Bedeian, Reference Bedeian1982; Boxer et al. Reference Boxer, Burnett and Swanson1995; Kelly et al. Reference Kelly, Charlton and Jenkins1995; Kposowa, Reference Kposowa1999). Studies have reported an elevated risk for suicide among medical doctors (Schernhammer & Colditz, Reference Schernhammer and Colditz2004), nurses (Hawton & Vislisel, Reference Hawton and Vislisel1999), veterinary surgeons (Mellanby, Reference Mellanby2005), pharmacists (Kelly & Bunting, Reference Kelly and Bunting1998), dentists (Alexander, Reference Alexander2001), farmers (Malmberg et al. Reference Malmberg, Simkin and Hawton1999), police officers (Hem et al. Reference Hem, Berg and Ekeberg2001), social workers (Stack, Reference Stack2004), artists (Stack, Reference Stack1996), craftsmen (Hamermesh & Soss, Reference Hamermesh and Soss1974), cooks (Stark et al. Reference Stark, Belbin, Hopkins, Gibbs, Hay and Gunnell2006), publicans (Kelly & Bunting, Reference Kelly and Bunting1998), automobile mechanics (Schwartz, Reference Schwartz1987), fishery, forestry and construction workers (Notkola et al. Reference Notkola, Martikainen and Leino1993; Liu & Waterbor, Reference Liu and Waterbor1994), and more generally in low paying occupation such as labourers, semi-skilled and unskilled workers (Platt & Hawton, Reference Platt, Hawton, Hawton and van Heeringen2000; Stark et al. Reference Stark, Belbin, Hopkins, Gibbs, Hay and Gunnell2006). In part, the elevated risk of some occupational groups may reflect occupational stress or access to lethal means of suicide such as medicines or firearms. In others the risk may be due to economic factors, such as the transient nature of some occupations predisposing to periods of unemployment and low income. Some studies have reported a reduced suicide risk among professionals (Lostao et al. Reference Lostao, Joiner, Lester, Regidor, Atach and Sandin2006), clerical workers (Hamermesh & Soss, Reference Hamermesh and Soss1974; Kposowa, Reference Kposowa1999) and military personnel (Rothberg et al. Reference Rothberg, Bartone, Holloway and Marlowe1990).
Most previous studies of this issue have been limited in scope or methodology. For example, studies often lack information on persons at risk (denominator data) from the same source as the occupation at death (numerator data). For this reason, proportional mortality ratios have been used as the measure of risk in a number of publications (Kelly et al. Reference Kelly, Charlton and Jenkins1995; Kelly & Bunting, Reference Kelly and Bunting1998; Stark et al. Reference Stark, Belbin, Hopkins, Gibbs, Hay and Gunnell2006); this measure may distort suicide risk in occupations at increased or decreased risk of death from other causes (Rothman & Greenland, Reference Rothman and Greenland1998). In addition, few studies have controlled for potentially confounding socio-economic and demographic factors (Kposowa, Reference Kposowa1999). Finally, while suicide is often regarded as a concomitant of psychiatric illness (Maris, Reference Maris2002), few, if any (Stack, Reference Stack2001), studies have considered the association between occupation and suicide among people who suffer from a psychiatric disorder or the possible interplay with psychiatric illness. Furthermore, it is fairly common that studies have been confined to specific occupations and relatively few reports have systematically investigated the risk across several occupational groups.
The increased risk of suicide among medical professionals has been most extensively studied (Platt & Hawton, Reference Platt, Hawton, Hawton and van Heeringen2000; Schernhammer & Colditz, Reference Schernhammer and Colditz2004; Schernhammer, Reference Schernhammer2005). As far as we are aware, no study has hitherto investigated the suicide risk among doctors who suffer from a psychiatric illness or investigated the possible mediating and confounding effects of psychiatric illness, socio-economic and demographic factors.
In this article, we aimed to answer the following three questions: (1) Are there differences in suicide rates across the broad range of occupation? (2) Are differences in suicide rates due to differences in psychiatric morbidity? (3) Are these differences explained by socio-economic and demographic factors?
METHOD
Source of data and population-based registers
Data were obtained by linking Danish population-based registers using the unique personal identification number assigned to all persons living in Denmark and used across all registration systems (Malig, Reference Malig1996). The Danish Medical Register on Vital Statistics contains dates and causes of all deaths in Denmark recorded from the Cause-of-Death Certificates since 1976 and for suicide since 1970 (Danish National Board of Health, 1992). Suicide was defined as ICD-8 (WHO, 1967) codes E950–E959 and ICD-10 (WHO, 1992) codes X60–X84. Undetermined deaths (open verdicts) were not included as this category might comprise accident victims and other non-suicidal deaths. ICD-9 has never been used in Denmark. Suicidal death by poisoning by medical compounds was defined as E950.0–E950.3 and X60–X64. The Integrated Database for Longitudinal Labour Market Research (IDA; Statistics Denmark, 1991) covers the entire population from 1980 onwards. It contains information on, for example, occupation since 1991, unemployment, marital status and income, and is updated annually by Statistics Denmark with information from administrative registers on individuals who were living in Denmark on 1 January. The Danish Psychiatric Central Register (DPCR) includes all admission and discharge dates and diagnoses according to the World Health Organization (WHO), 8th and 10th classification, and covers all psychiatric in-patient facilities in Denmark since 1969 (Munk-Jorgensen & Mortensen, Reference Munk-Jorgensen and Mortensen1997). There are no private psychiatric hospitals in Denmark, and all treatment is free of charge.
Study population and statistical design
All subjects who died by suicide in the period 1991–1997 while aged 25–60 years and were therefore of working age were identified from the Danish Medical Register on Vital Statistics. Using a nested case–control design (Clayton & Hills, Reference Clayton and Hills1996), each person who died by suicide was matched to a random sample of 20 controls of the same gender who were born the same year and who were alive when the case died by suicide (King & Zeng, Reference King and Zeng2002). Cases and controls were only included if they were resident in Denmark throughout the two preceding years.
Classification of occupations
Information on occupation according to the Danish version of the International Classification of Occupations (DISCO-88) (ILO, 1990; Statistics Denmark, 1996) was obtained from IDA for the main occupation held by the subject the year prior to the index year. DISCO-88 is a pyramidal hierarchical nested four-digit occupational nomenclature that divides subjects into 10 major groups at the top level of aggregation; these are subdivided into 27 submajor groups, 111 minor groups, and 372 unit groups according to skill level and skill specialization (ILO, 1990; Statistics Denmark, 1996). To conduct a reliable inference (Altman & Royston, Reference Altman and Royston2000), subjects were grouped, if possible, upwards in the hierarchy until there were more than 20 suicides in each occupational group. Since 1994, it is mandatory for employers with more than 10 employees to report occupational codes using DISCO-88. For employers with less than 10 employees, and prior to 1994, occupational reporting was based on occupational descriptions in text format and these have subsequently been converted to DISCO-88 codes. Statistics Denmark validates the DISCO codes against information on, for example, union membership, education and information from other routine registers, and therefore a person can be registered with an occupation while being unemployed. Any misclassification among suicides and controls will be non-differential (Rothman & Greenland, Reference Rothman and Greenland1998) as they are matched on year of death and the occupational classification used in that year. However, the misclassification might vary between occupational groups and therefore bias our assessment of differences between occupations.
Socio-economic, demographic and psychiatric factors
Information on the individual-level demographic and socio-economic factors was obtained from IDA and information on psychiatric admission was derived from the DPCR. The adjustment factors were: time since psychiatric hospital discharge (currently admitted, week, month, half-year, year, longer) and diagnoses (depression, schizophrenia, mania, alcohol, all others), employment status (fully employed, self-employed, unemployed, otherwise), marital status (married, cohabitating, single) and gross income in quartiles. In our adjusted models we initially controlled separately for psychiatric illness, income and employment status to investigate whether the observed associations arose because of lower income and job insecurity in certain occupations or because some occupations might attract people with previous mental illness (e.g. elementary or unskilled occupations) or that certain occupations might increase the risk of psychiatric illness and this might lie on the causal pathway between occupation and suicide risk. Further details on these factors are published elsewhere (Agerbo, Reference Agerbo2005a, Reference Agerbob; Agerbo et al. Reference Agerbo, Qin and Mortensen2006).
Data modelling
The data were analysed using conditional logistic regression (King & Zeng, Reference King and Zeng2002) with each case forming a separate stratum. As the controls were selected randomly within the appropriate risk sets, the estimated measures of relative risk are called rate ratios (RRs) (King & Zeng, Reference King and Zeng2002). Primary education teaching professionals (primary school teachers) were chosen as reference because this occupation is relatively large, well-defined and universal. Wald tests were used to test for interactions between occupational group and gender, gender composition within occupations among controls, trend with age and year, and suicide methods (stratified by poisoning by medical compounds or by use of firearm versus other cause) (Lunn & McNeil, Reference Lunn and McNeil1995). To avoid chance findings, only interactions that are significant at the 1% level are considered. Subsequently, data were stratified by whether the subjects had been admitted to a psychiatric hospital and odds ratios were calculated by unconditional logistic regression analysis. Analyses were performed using SAS version 9.1 (SAS Institute Inc., Cary, NC, USA).
RESULTS
In total, 3195 suicides (898 females) and 63 900 controls were identified. The distribution of cases and controls and occupation specific RRs are shown in Table 1, the ratios are sorted in descending order of magnitude.
Table 1. Age- and sex-adjusted suicide rate ratios in all occupational groups estimated from the conditional logistic regression model
DISCO-88, Danish version of the International Classification of Occupations; RR, rate ratio; CI, confidence interval.
a People with no registered occupation.
b Largely unskilled manual workers, for example domestic and related helpers, cleaners, launderers and manufacturing labourers (see Method section).
Suicide across the broad range of occupations
The highest risk of suicide is among medical doctors, followed by subjects without occupation and nurses. Several low-paid occupations had raised suicide RRs, including elementary occupations (largely unskilled manual workers, such as domestic and related helpers, cleaners, launderers and manufacturing labourers); plant and machine operators and assemblers; bricklayers and stonemasons; drivers and mobile-plant operators; painters and related workers; cooks; and carpenters and joiners. The lowest suicide risks are among (1) architects, engineers and related professionals; (2) armed forces; (3) finance and sales associate professionals (e.g. technical and commercial sales representatives); and (4) corporate managers. No gender or year interaction was found but suicide risk seems to increase more with age among finance and sales associate professionals (p=0·009) and to decrease in the group without occupation (p=0·004) compared with the general increase with age. A contextual effect associated with gender composition with occupation appeared as an interaction with gender: females working in male dominated occupations were at higher risk (RR 1·68, 95% CI 1·06–2·68) compared with females working in female-dominated occupations. The risk of suicide by poisoning by medical compounds was particularly pronounced among doctors (RR 11·12, 95% CI 5·98–20·66, p<0·0001) and to a lesser extent among females, although this did not reach conventional levels of statistical significance (RR 2·72, 95% CI 1·48–4·99, p=0·23). The suicide rate by firearms was increased only among farmers (market-oriented crop and animal producers) (RR 2·50, 95% CI 1·32–4·75, p=0·006) and not among soldiers (RR 0·64, 95% CI 0·09–4·76). The risk of suicide was estimated in some selected occupations with less then than 20 suicides: police officers (RR 0·85, 95% CI 0·43–1·68, number of cases/controls=10/311; by firearms: RR 1·69, 95% CI 0·39–7·33), social workers (RR 2·07, 95% CI 1·06–4·03, number of cases/controls=11/141), and waiters and bartenders (RR 1·84, 95% CI 0·92–3·69, number of cases/controls=10/146). The numbers of cases (controls) among pharmacists, dentists, veterinary surgeons, psychologists, and artists were 3 (123), 4 (111), 0 (49), 4 (54) and 3 (30) respectively, thus a reliable risk estimate could not be obtained.
Adjustment for psychiatric, socio-economic and demographic factors
Table 2 presents suicide RRs before and after controlling for psychiatric admissions and socio-economic indicators in the 20 highest and five lowest risk occupational groups. Except among doctors and nurses, the risks associated with occupations decreased markedly after adjustment for income and employment status. Considering RRs that changed by more than 10% (Greenland, Reference Greenland1989), the risks associated with occupation showed a tendency to increase after adjustment for psychiatric admission (Table 2, fifth column), with the exception of the residual group without occupation, elementary/unskilled occupations, cooks and sales and services elementary occupations. Generally, the risk of suicide in all occupational groups diminished in the final model, where we controlled for psychiatric admission, employment status, marital status and gross income (Table 2, last column). The only exception was medical doctors, where the risk further increased. The adjusted risks of suicide among doctors by medicines and other causes were 13·38 (95% CI 6·29–28·45) and 2·00 (95% CI 1·07–3·75) respectively and the analogous rates among nurses were 3·26 (95% CI 1·59–6·67) and 1·93 (95% CI 1·09–3·41).
Table 2. Suicide, 20 highest and five lowest occupational rate ratios with 95% confidence intervals
DISCO-88, Danish version of the International Classification of Occupations; RR, rate ratio; CI, confidence interval.
a Adjusted for age, gender and year.
Suicide risks across occupations among the (never) psychiatrically admitted individuals
Approximately 47·5% of individuals who died by suicide and 5·1% of the controls had been psychiatric in-patients. Table 3 shows the risk of suicide in the 10 highest and five lowest occupational groups stratified by psychiatric disorder. Among individuals who have been admitted with a psychiatric disorder, only moderate differences in suicide risk are observed between the various occupations (p heterogeneity=0·05); the notable exception is medical doctors, who are at almost fourfold greater risk than other occupations. Among individuals who have never been admitted, the suicide risk in a specific occupation is generally slightly larger or similar in magnitude to the analogous risk in the unstratified analysis (Table 2, last column) with the possible exception of nurses, where the risk, although statistical indistinguishable, is approximately 50% higher (p heterogeneity<0·0001).
Table 3. Suicide, 10 highest and five lowest occupational rate ratios stratified by admission to psychiatric hospital
RR, Rate ratio; CI, confidence interval.
a Adjusted for age, gender, year, income, employment and marital status.
b No cases are exposed.
DISCUSSION
This study reveals that the risk of suicide varies considerably across occupational groups. Such findings might raise awareness of suicide risk in certain occupations and aid our understanding of the aetiology of suicide by prompting further investigation of reasons for an occupational group's elevated risk. Compared to primary school teachers, the suicide rate is more than 50% higher among doctors, nurses, people in elementary/unskilled occupations, plant and machine operators and assemblers, bricklayers, drivers, painters, cooks, construction and maintenance workers, agricultural and fishery workers, plumbers, carpenters, social workers, waiters and bartenders, and in the group without occupation. The risk of suicide is low in certain occupations, including architects, engineers, armed forces, finance and sales associate professionals and corporate managers. The suicide risk decreased in all occupations after adjustment for psychiatric admission, employment status, marital status and gross income, with the exception of medical doctors and perhaps nurses, where the rate increased further. The most important determinants of occupational differences in suicide risk, as indicated by the reduction in risks in adjusted models (Table 2), appeared to be income and employment status. A similar result, however, is not seen among people who never have been admitted with a psychiatric disorder. Among people who have previously been admitted with psychiatric disorders, modest associations between suicide and occupation are observed except for doctors, who are at almost fourfold greater risk.
Epidemiological considerations: mediation, confounding and selection
A limitation of our analysis is that we cannot determine whether elevated suicide risks associated with particular occupations arise as a result of the characteristics of the jobs, the people who take up the jobs or a combination of the two. Several circumstances may elevate risk in a specific occupation. First, access to, and specialist knowledge of, lethal means of suicide, such as medicines among doctors and nurses or firearms among farmers, police officers and soldiers, might be contributing factors. This is supported by the elevated risks of method-specific suicide in some of these groups in our analyses. Second, an elevated risk could partly reflect occupational stress and economic factors, such as recurrent or persistent unemployment spells and low income, or that psychiatric illness is more common in certain jobs (Wieclaw et al. Reference Wieclaw, Agerbo, Mortensen and Bonde2005). In this study, we adjusted for psychiatric admission, employment status, marital status and gross income although it is not certain that these factors merely confound the association between suicide and occupation or lie on its causal pathway. It is plausible that these factors also act as intermediate conditions at different times, and therefore constitute time-dependent confounders (Robins et al. Reference Robins, Hernan and Brumback2000). For example, doctors are at higher risk of being admitted with an affective disorder (e.g. depression) (Wieclaw et al. Reference Wieclaw, Agerbo, Mortensen and Bonde2005), which suggests that these psychiatric illnesses lie on the causal pathway between occupation and suicide, but affective illnesses also tend to occur later in life. Third, the healthy worker effect (Li & Sung, Reference Li and Sung1999), which characterizes the two-sided selection of healthy people into occupation and the exclusion of the unhealthy from the workforce, should be kept in mind when interpreting the results concerning the group without occupation and elementary/unskilled occupations, where the excess risk might partly, if not almost entirely, be caused by unemployment, lower income, higher incidences of singleness and psychiatric illness. Fourth, it is possible that people at increased risk of suicide are attracted to particular occupations. For example, some studies have shown that publicans are at increased suicide risk (Kelly & Bunting, Reference Kelly and Bunting1998); it is possible that people at risk of alcoholism, a known risk factor for suicide, are attracted to occupations where alcohol is freely available.
Psychiatric, socio-economic and demographic factors
Apart from people working in elementary/unskilled occupations and those not registered with an occupation, a suppressor effect (MacKinnon et al. Reference MacKinnon, Krull and Lockwood2000) is found for most occupations as the suicide risk increased after adjustment for psychiatric admission. This indicates that levels of psychiatric admission are lower in these occupations than they are among teachers (Wieclaw et al. Reference Wieclaw, Agerbo, Mortensen and Bonde2005). It is plausible that people who suffer from a psychiatric illness are more likely to have unskilled occupations or no occupation. In keeping with the social drift hypothesis (Kreitman et al. Reference Kreitman, Carstairs and Duffy1991; Tohen et al. Reference Tohen, Bromet, Murphy and Tsuang2000), it is likely that these people drift into unemployment or an unskilled job from the occupations where the suppression effect is seen. In all occupations, except among doctors, the risk of suicide showed a tendency to decrease after further adjustment for employment status, marital status and gross income, which, in keeping with previous studies (Kposowa, Reference Kposowa1999; Stack, Reference Stack2001), suggests that these factors account for some of the differentials in suicide risk. Although these factors account for some of the variation, it is possible that other factors, such as internal occupational stress (Stack, Reference Stack2001), could account for the risk excess observed in some occupations. A skewed gender composition could be an internal stressor, as the results suggest that females (males) who are working in a male- (female)-dominated occupation are at increased risk of suicide. The additional risk increase among doctors after adjustment might be caused by doctors being less exposed to, for example, lower income, singleness and unemployment.
Psychiatrically hospitalized individuals
Except for medical doctors, among individuals who have been hospitalized with a psychiatric disorder, only moderate disparities in suicide risks are observed across the various occupations; thus occupation has little association with suicide among individuals who suffer from a psychiatric illness. A straightforward conjecture is that psychiatric illness lies on the causal pathway between occupation and suicide; or, in keeping with a past study on suicide and unemployment (Agerbo, Reference Agerbo2003, Reference Agerbo2005a), that occupational inequalities are less important once a person suffers from a psychiatric illness. However, there are relatively few exposed cases in each occupational group, and the statistical power is limited and perhaps not sufficient to distinguish the suicide risks, which are also rather limited in magnitude. This pattern, however, differs among doctors, who are at particular high risk of suicide when they have been admitted with psychiatric disorder. Anecdotal reports suggest that if doctors are treated, the treatment may be compromised because of collegial relationships and deference from the treating clinician, who may give more freedom to the doctor-patient to control the therapy and to self-medicate (Center et al. Reference Center, Davis, Detre, Ford, Hansbrough, Hendin, Laszlo, Litts, Mann, Mansky, Michels, Miles, Proujansky, Reynolds and Silverman2003). In keeping with a study on suicide in richer people with a psychiatric illness (Agerbo et al. Reference Agerbo, Mortensen, Eriksson, Qin and Westergaard-Nielsen2001), doctors might feel more stigmatized and have a higher loss of self-esteem than other people with psychiatric illnesses.
The never psychiatrically hospitalized individuals
In the section of the population who have never been admitted with a psychiatric disorder, the adjusted occupational risk of suicide tends to be slightly higher, or similar in magnitude, when compared with the general population. This is not too surprising because of the rarity of psychiatric illness in the general population. Perhaps nurses are an exception as their risk of suicide is considerably higher, which could be due to unobserved psychiatric illness. However, nurses are at increased risk of being admitted with a psychiatric disorder (Wieclaw et al. Reference Wieclaw, Agerbo, Mortensen and Bonde2006), and an odds ratio is always larger than the corresponding risk ratio (Newman, Reference Newman2001). A priori expectations suggest that the largest aberration should be found in the group without occupation, as this group has a particularly high risk of admission (Agerbo et al. Reference Agerbo, Byrne and Mortensen2004; Wieclaw et al. Reference Wieclaw, Agerbo, Mortensen and Bonde2005).
Availability of means for suicide
Restriction of access to means of suicide, such as lethal medicines or firearms, has been suggested as an important strategy for suicide prevention (Hawton, Reference Hawton and Hawton2005). In this connection, we found a particularly high risk of suicide by firearms among farmers and perhaps among police officers, as well as by medicines among doctors and nurses despite an excess risk of suicide by other means among doctors and nurses. By contrast, in keeping with previous findings (Hem et al. Reference Hem, Berg and Ekeberg2001), the risk of suicide by firearm is not increased among soldiers and might therefore indicate a healthy soldier effect (Bross & Bross, Reference Bross and Bross1987).
ACKNOWLEDGEMENTS
This study was supported financially by the Stanley Medical Research Institute. Psychiatric epidemiological research at the National Centre for Register-based Research is in part funded through a collaborative agreement with the Centre for Basic Psychiatric Research, Psychiatric Hospital in Aarhus, Denmark.
DECLARATION OF INTEREST
None.
