Statistics from Altmetric.com
The Journal of Epidemiology and Community Health has a tradition of publishing articles and commentaries on associations between unemployment and health; work that has helped shaped the agenda for future research in the field.1 2 3 4 5 The article by Lundin and colleagues in the current issue, (see page 22),6 is a welcome continuation of that tradition. Coming early in the current recession the article is a timely reminder of the individual health consequences of economic downturn.
Sweden’s military capacity is built on conscription, and until the early 1990s and the end of the Cold War nearly all men reaching the age of military service were conscripted. At enlistment all conscripts undergo a comprehensive physical examination, an interview with a psychologist and provide personal information by questionnaire. These data have been used previously to investigate a number of health and health-related outcomes.7 8 However, little previous work has made use of these data to study labour market exposures later in life, or capitalised to such a degree on record linkage to other register-based data widely available in Scandinavian countries.
The personal ID issued to all Swedish citizens has enabled the authors to link information from the conscription examination to data from four national registries at the level of the individual. Censuses provide socioeconomic data from childhood and early mid-life, questionnaire data at conscription is used to determine health-related behaviour and contact with the authorities in late adolescence. Information on psychiatric diagnoses in late adolescence is supplemented by hospital discharge data, and health status in the 2 years prior to the unemployment observation window is provided by sickness absence data from national registers. The result is a large and impressive dataset that allows the authors to bring a life course approach to the association between unemployment and mortality, and to address a number of the issues that have bedevilled the long-running debate in the field on causation versus selection.
The use of sickness absence data in this way is particularly innovative. Health-related selection out of the labour force has long been considered a candidate confounder of the association between unemployment and health, and, at times of rising unemployment, a poor sickness absence record will mitigate both against retention in the work force and re-employment post-redundancy. Medically certified sickness absence has recently been shown to be a strong predictor of premature death.9 10 11 12 Adjustment for this measure produces the largest attenuation of the association between unemployment and health in the present study, and, together with the psychiatric health measures, explains 49% of the association, indicating a sizable role for health-related selection.
However, it is important, as we enter another period of high unemployment, that the arguments about direct and indirect selection as they apply to unemployment and health are rehearsed. In “direct” selection, the apparent association between unemployment and mortality is due to over-representation among the unemployed of people with diseases that are going to shorten their lives regardless. Lundin et al nominate as their “direct” selection factors: psychiatric diagnoses and sickness absence. The idea of “indirect” selection is that the confounding factor is not a disease, but some other characteristics that make an individual more prone both to unemployment and to early mortality. For example, it is known that smokers are more likely to experience unemployment and that smoking is a health risk. Lundin et al nominate as their examples of indirect selection factors all the other variables that are controlled in their models, for example, father’s social class, low income, smoking and low personal control.
When the question of the extent to which health selection influenced the association of unemployment to mortality was first raised during the recession of the 1980s, an elegant solution was proposed by Moser and colleagues in the UK.13 14 15 Moser et al argued that if associations between unemployment and mortality are a result of direct selection, relative risks in the unemployed should be higher in the earlier period of follow-up and then reduce. If the association is due to the over-representation among the unemployed of people with a disease that is going to result in early mortality (regardless of whether unemployment takes place or not), then eventually this “ill” group will all have died, and the mortality level in the group will return towards the population average. They then showed that in their dataset, in men who were out of work due to ill-health mortality was high at the beginning of follow-up, and then declined as those with more rapidly fatal illnesses died. Those who were unemployed, but healthy enough to be actively looking for another job showed the “healthy worker effect” in which mortality was actually lower than the average for the male working-age population at the beginning of follow-up, and then increased over time.
In acknowledgement of this, Lundin et al break down their mortality data into two periods: 1995–8 and 1999–2003. Here they find that, in contrast to the findings of Moser et al, excess mortality among the unemployed is in fact higher in the earlier period of follow-up. Excess mortality from suicide and external causes, which account for a very large proportion of all deaths, actually decreases (wears off) more than that from other causes.
However, comparing these studies shows that there are major methodological differences. In the late 1980s, Moser et al’s “baseline” was taken to be all men of working age, including employed, unemployed, early retired and permanently sick. Lundin et al calculate the hazard ratio for the men who experienced 90 or more days of unemployment in the economic crisis, taking as their baseline those in their rather complex sample who had no unemployment during that period. In addition, Moser et al’s definition of “unemployment” was quite different. Their data used for analysis consisted of employment status in 1971 or 1981 and mortality in the subsequent 9 years. In Lundin’s study, the period of unemployment could have taken place at any time in a 3-year period 1991–4, and deaths taking place during this period were not counted in their analysis. Moser et al only knew that their risk group had been unemployed on the day of a national Census; the “unemployed” in their article would have been more likely to be longer term repeatedly unemployed for the simple reason that the more unemployment they experienced, the higher the likelihood that a Census would take place while they were without work. Lundin et al’s measure of unemployment is more precise due to the superior data available to them, but they will have included more men with relatively short spells and spells that did not repeat. Moser et al included all men of working age in 1971 and 1981; Lundin et al include only men aged between 44 and 54 in 1995–2003 who had no spells without work during 1991–2.
What effect might these differences be expected to have on the results of the two studies? The Swedish study compares mortality in the “unemployed” with that in a group with a very favourable life course both before and after the economic crisis; the most vulnerable men having already been excluded from the subsample included in the analyses. In addition to the 2–3% deemed unfit for conscription; 2.8% did not survive to the start of the mortality follow-up, 2.2% were already on disability pension, 9.6% earned less than 50 000 kroner (4360 Euro; £3881; 5474 US$) and 4.1% were not in stable employment 1990–1991. The 14% of men in the latter two categories will mostly belong to the secondary labour market, subject to the adverse effects of job insecurity, employment insecurity and poverty wages or a mix of wages and benefits.15 This marginalised sector of the labour force are likely to have higher mortality rates than those who lose stable jobs to join the ranks of the unemployed (see Lundin et al, table 3). Job insecurity, precarious employment, temporary work and employment insecurity have all been shown to have adverse effects on health.17 18 19 20 Our interpretation of the changes in relative risk over time must be informed by an understanding of sample selection, which needs to be clearly explained in articles of this kind. Lundin et al do us a service by promoting this kind of discussion.
Perhaps the most striking part of the article is the way in which it shows us how adversities over the life course cast long shadows forward. All of the chosen risk variables, such as household overcrowding in childhood, risky alcohol use, contact with police and psychiatric diagnoses, show associations not only with unemployment during the recession years, but also with lack of stable employment and mortality up to 1995.
Unemployment spells do not happen at random, but tend to take place as part of a generally disadvantaged life course.5 21 Lundin et al add to these findings and point to ways in which use of the Nordic register system can improve upon them. However, their study also shows that, even in such a heavily selected sample and after controlling for numerous markers of this disadvantage across the life course, including sickness absence in the 2 years immediately prior to the observation window, unemployment is associated with a 57% excess risk of early death over the first 4 years in men aged 44–54. This would lead us to conclude that the potential health consequences of unemployment should be a source of concern in the present recession.
Funding M. B. acknowledges support by ESRC grant number RES 596-28-0001, International Centre for Life Course Studies in Society and Health.
Competing interests None declared.
Provenance and Peer review Commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.