Unemployment and Early Cause-Specific Mortality: A Study Based on the Swedish Twin Registry

Margaretha Voss, PhD, MPH; Lotta Nylén, MPH; Birgitta Floderus, PhD; Finn Diderichsen, MD, PhD; Paul D. Terry, PhD


Am J Public Health. 2004;94(12) 

In This Article


The study population comprised 18516 women and 18020 men, constituting in principle all same-sex twins born in Sweden between 1926 and 1958.[30] Data were based on a 1973 mailed questionnaire, which was answered by 15683 women and 14287 men (response rate=85% for women and 79% for men), and on information from the Swedish Causes of Death Registry. All responders with a job title indicating gainful employment were included in the study (i.e., 9500 women and 11132 men). Housewives, students, retired individuals, persons on disablement pension, and persons liable for military service at baseline were excluded from the analyses.

A short occupational history, including history of unemployment, was recorded in the 1973 questionnaire that included unemployment. Information about unemployment was based on answers to the following questions: "Are you employed at the present time?"; "Are you now or have you ever been unemployed?"; "For how long have you been unemployed?"

In our main analyses, we compared ever unemployed (875 women and 1309 men) with never unemployed according to the 1973 data. In some analyses, we distinguished between "short-term" and "long-term" unemployment. Short-term unemployment was defined as being unemployed in 1973 and experiencing less than 1 year of lifetime unemployment (144 women and 185 men). Long-term unemployment was defined as being unemployed for 1 year or more altogether during the life course (260 women and 311 men). Small numbers precluded separate analyses of mortality among the short- and long-term unemployed.

In the analyses, the following social, behavioral, health, and personality factors were considered: marital status, children, education, smoking and alcohol habits, use of sleeping pills and tranquilizers, stress, shift work, personality factors, long-lasting/serious illness, and socioeconomic status. The selection of potential confounding factors originates from our previous study.[29] The variables were dichotomized; "exposed" categories are listed in Table 1 (reference groups were those "not exposed.") The reference category for marital status was married and cohabitant; smokers and former smokers were compared with never smokers. Alcohol consumption was analyzed by consumed grams of absolute alcohol per month.[30] No alcohol consumption and consumption of more than 250 g of alcohol per month were compared with consumption of 1 to 250 g of alcohol per month.

"Stressful life" was defined by the question, "Do you experience your everyday life as being very stressful?" The personality factors "instability" and "introversion-extraversion" were based on 9 items each selected from the "neuroticism" and "extraversion" dimensions of the Eysenck personality inventory; these short scales were developed to be used in comprehensive questionnaire investigations.[31,32] Unstable personality (>4 points) was compared with stable personality (<5 points), and extravert personality (>4 points) with introvert personality (<5 points). The question "Have you ever had any long-term or serious illness?" was used as an indicator of ill health. Unskilled/skilled workers and low-level white-collar workers were compared with medium- and high-level white-collar workers.

All individuals of the study population were followed regarding mortality from January 1, 1973, to December 31, 1996. For deaths that occurred between 1973 and 1986, specific causes of death were taken from the International Classification of Diseases, Eighth Revision (ICD-8)[33]; for deaths between 1987 and 1996, the International Classification of Diseases, Ninth Revision (ICD-9)[34] was used. The 2 revisions were then harmonized. The following underlying causes of death were analyzed: malignant neoplasms (ICD-8 codes 140-209); malignant neoplasms of trachea, bronchus, and lung (162); diseases of the circulatory system (390-458); ischemic heart diseases (410-414); injuries, poisoning, and other external causes (E800-E999); suicide (E950-E959); injury, undetermined whether accidentally or purposely inflicted (external undetermined cause; E980-E989); other diseases (001-139, 210-389, 460-799); and alcohol-related diseases (303, 571, 577).

Differences in social, behavioral, health, and personality characteristics between individuals with and without experience of unemployment were analyzed with adjustment for age (5-year intervals), and the statistical precision was indicated by 95% confidence intervals.

We estimated the mortality rate ratio, referred to here as the relative risk, together with 95% confidence intervals by Cox proportional hazards regression,[35] using PHREG software (SAS 6.12; SAS Institute Inc, Cary, NC). The analyses included a full 24-year follow-up as well as a follow-up restricted to the first 10 years.

In these analyses, both twins in a pair were included and treated as independent individuals. To ensure that confidence intervals were not erroneously narrow owing to similarities within pairs, we performed proportional hazards regression analyses with variance estimates adjusted for correlated outcomes.[36,37,38] We accomplished this through the use of a SAS macro that stems from the same theoretical background and yields the same results as the published Fortran program of Lin.[39]

Synergistic effects were analyzed on the basis of odds ratios from logistic regression models, according to methods suggested by Rothman.[40] The factors showing a significant (P<.05) prevalence difference between ever- and never-unemployed individuals were considered to be of interest for an assessment of potential interaction with unemployment. Individuals unexposed to both unemployment and the other factor under study constituted the reference group. A synergy index was computed that indicated to what extent the presence of the factor influenced the effect of unemployment on mortality. A synergy index of 1 means no interaction, and a synergy index of 2 means an effect among those with combined exposures that is twice what would be expected from an additive effect of the 2 exposures. To calculate confidence intervals, we used methods of Hosmer and Lemeshow[41] and computer programs described by Lundberg et al.[42]

To control also for genetic and early childhood factors measured by the 1973 questionnaire, we analyzed mortality from all causes among 1067 twin pairs, 1 twin of whom had experienced unemployment while the other had not. We based the risk estimates (odds ratios) on conditional logistic regression for matched data using PHREG.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.