Article Text

Download PDFPDF

Changes in directly alcohol-attributable mortality during the great recession by employment status in Spain: a population cohort of 22 million people
  1. Ignacio Alonso1,
  2. Fernando Vallejo2,3,
  3. Enrique Regidor3,4,
  4. M José Belza1,3,
  5. Luis Sordo2,3,4,
  6. Laura Otero-García3,5,
  7. Gregorio Barrio1
  1. 1 National School of Public Health, Carlos III Health Institute, Madrid, Spain
  2. 2 National Epidemiology Center, Carlos III Health Institute, Madrid, Spain
  3. 3 Consortium for Biomedical Research in Epidemiology and Public Health (CIBERESP), Madrid, Spain
  4. 4 Department of Preventive Medicine and Public Health, Madrid Complutense University, Madrid, Spain
  5. 5 Departmental Nursing Section, Faculty of Medicine, Autonomous University of Madrid, Madrid, Spain
  1. Correspondence to Dr María José Belza, National School of Public Health, Carlos III Health Institute, Avenida Monforte de Lemos 5, Madrid E-28029, Spain; mbelza{at}


Background Macroeconomic fluctuations can impact differentially on alcohol-related problems across sociodemographic groups. We assess trend changes in directly alcohol-attributable (DAA) mortality in the population aged 25–64 during the post-2008 recession in Spain according to employment status and other sociodemographic factors.

Methods Nationwide cohort study covering 21.9 million people living in Spain in 2001. People were classified by employment status and other factors. The annual percentage change (APC) in mortality rates during 2002–2007 (precrisis) and 2008–2011 (crisis) was estimated by the Poisson regression. The period effect size was then calculated as the difference between crisis and precrisis APCs.

Results The age-adjusted APCs in DAA mortality were 6.9% in 2002–2007 and 3.7% in 2008–2011 among employed people, and −4.3% and −0.4%, respectively, among non-employed people. Statistically significant trend changes in such mortality during the crisis were found, which were favourable in certain employed subgroups (manual workers and employees aged 25–49), and unfavourable in the total non-employed population and certain non-employed subgroups (men, non-married and especially medium/high-wealth people). The greatest unfavourable change corresponded to non-employed people living in households of 72–104 m2 who had 2 or more cars. Favourable changes were also found in the remaining employed subgroups, especially women and non-married people, although they did not reach statistical significance.

Conclusions Our findings suggest that the post-2008 Spanish crisis had a heterogeneous impact across sociodemographic subgroups on DAA mortality, and that employment status seemed to have an important effect. The impact was especially unfavourable on the non-employed, particularly those with substantial material wealth.

  • Cohort studies

Statistics from


Alcohol consumption is responsible for considerable disease burden worldwide. The post-2008 economic crisis has aroused intense concern about its potential adverse impact on alcohol-related problems.1–3 Alcohol consumption can be affected by economic cycles through various mechanisms, such as changes in alcohol affordability (due to changes in income, taxes or fees), leisure time availability or psychological distress (due to changes in ability to cope with working conditions, expenses or debts).4 ,5 Moreover, alcohol-related mortality can be influenced by economic cycles through changes in social welfare and healthcare, whose accessibility may be limited during crises, shifting the financial burden to households.6

Although the effect of macroeconomic fluctuations on overall and cause-specific mortality is generally procyclical (ie, favourable changes during crisis),7 countercyclical changes have often been observed for alcohol-related mortality,8–12 as well as for alcohol use disorders13 or heavy or binge drinking.3 ,4 ,14–17 Depending on differences in the predominant mediator mechanism, vulnerability of different sociodemographic groups, crisis features and context, the effects of recession on alcohol-related problems may be heterogeneous.2 ,6 ,10 ,16 ,18 Thus, some evidence suggests that adverse effects are greater in men13 ,14 ,19 or people who are unemployed,12–14 ,20–25 less educated,13 ,14 ,18 ,25 ,26 unmarried14 ,25 ,26 or experiencing housing instability or rent/mortgage payment difficulties.24 ,27 We aimed to assess the impact of the post-2008 economic crisis on directly alcohol-attributable (DAA) mortality according to employment status and other sociodemographic factors, testing the hypotheses of especially unfavourable impacts on non-employed or socioeconomically disadvantaged people.

The post-2008 economic crisis in Europe was characterised by a fall in gross domestic product (GDP) and large growth in unemployment, putting many governments in a squeeze between falling tax revenues and rising costs of economic support measures. The Spanish crisis was more complex because it was accompanied by a sudden drop in housing prices. Thus, the annual mean growth in GDP dropped from 3.5% in 2002–2007 to −0.9% in 2008–2011, and housing prices went down from 9.9% in 2007 to −4.4% in 2008–2011, while the unemployment rate increased from 9.6% in the first quarter 2008 to 22.6% in the last quarter of 2011. Also, strict fiscal austerity measures were adopted to reduce the general government deficit (which went from 1.4% of GDP in 2004–2007 to −8.6% in 2008–2011),28 possibly aggravating the financial strain in individuals. The specific study objectives were to compare trends in DAA mortality rates in working-age people between 2002 and 2007 (precrisis) and 2008–2011 (crisis) in different sociodemographic population strata in Spain and to quantify the possible effect of the crisis period (changing trends) in these strata.


Study population

All residents in Spain on 1 November 2001 (Census date) were followed until December 2011 to determine vital status and cause of death. The National Institute of Statistics (INE) performed the record linkage between population and mortality registries using common identifiers, providing researchers with a data file without personal identifiers, census tract or death day to maintain confidentiality. Census participants not found in population or mortality registers (1.7%) were excluded. The risk contribution of 1% of participants was censored because they had moved abroad. The sociodemographic characteristics of excluded or censored participants did not differ from included participants. The analysis was restricted to participants of working age (25–64 years) in each calendar year during 2002–2011 (21.9 million people). This study was approved by the INE Institutional Review Board.


The main variable definitions and categories are included in online supplementary appendix 1. The outcome was DAA mortality, including underlying causes of death where alcohol is explicitly mentioned (International Classification of Diseases, 10th Edition). Other deaths where alcohol was a major contributing factor could not be identified, because multiple coding was not available. Independent variables were measured only once, at Census date (baseline), including employment status, labour market participation and occupation, which referred to the week before baseline. However, age was included as a time-varying variable in mortality analyses. Participants were initially classified as employed or non-employed, and subsequently subclassified by gender, age group (25–49, 50–64), marital status (married, non-married), educational attainment (primary, >primary) and material wealth (low, medium/high). This latter variable was obtained by combining information from household floor space and household car ownership. Employees were subclassified by occupational class (manual workers, non-manual workers), and non-employees by labour market participation (unemployed, inactive). Finally, for certain analyses, some resulting subgroups were subsequently stratified by other variables. Less than 0.1% of participants lacked information on classification variables. Details on definitions and categories of included variables can be seen in online supplementary tables S1–S4.

supplementary tables

Statistical analysis

We calculated mortality among participants in each year of the precrisis and crisis periods. The annual crude mortality rate (CMR) was obtained by dividing the number of deaths by person years (py) at risk in the same calendar year. Trend changes in mortality risk between the precrisis and crisis periods were then assessed within each aforementioned stratum, because evidence from elsewhere suggested that such changes could be heterogeneous. For that purpose, the annual change in the mortality rate adjusted by 5-year age group was modelled as a linear trend using the Poisson regression, where the annual death count was the dependent variable and the annual py the offset variable, first for the precrisis period and then for the crisis period. Subsequently, the annual percentage change (APC) was computed as 100×[exp(B)–1], where B was the calendar-year regression coefficient. The 95% CIs of APC were estimated as 100×[exp(B±1.96×SE)–1], where SE is the standard error of B. Finally, mortality trends during the crisis and precrisis periods were compared using the difference between the APCs in 2008–2011 and 2002–2007 (crisis period effect size), considering a negative effect size as favourable, and a positive one as unfavourable. The 95% CI of effect size was estimated using the SE of a percentage difference. Previously, the SE of each APC had been calculated by subtracting the point estimate from the lower limit of the 95% CI and dividing the absolute value of the difference by 1.96. Analyses were performed with Stata V.14.0 (Stata Corporation, College Station, Texas, USA).


General characteristics of participants at baseline

Employees compared with non-employees had a higher proportion of men and younger, unmarried, more educated and wealthier people. In particular, notable is the high representation of women among the inactive population (77.3%) and their low representation among manual workers (23.8%), as well as the large proportion of more educated people among non-manual workers (90.0%) (table 1).

Table 1

Classification of employed and non-employed population in a Census cohort of working-age population by different sociodemographic factors at baseline, Spain, 2001

DAA mortality in 2002–2011

The analyses included 226.6 million py at risk and 9029 DAA deaths. The CMR from DAA at working age during 2002–2011 was significantly lower among employees (2.9/100 000 py) than non-employees (5.9/100 000 py), as was proportional DAA mortality (1.5% vs 1.9%). The main specific cause of DAA mortality was alcoholic liver disease (80.3% of DAA deaths among employees and 77.9% among non-employees), followed by alcohol use disorder (13.4% and 16.2%) and acute alcohol poisoning (3.0% and 2.4%).

The CMR increased in the precrisis and crisis periods among employees, although the increase was apparently more intense in 2002–2007. The CMR was 1.6, 2.9 and 4.0/100 000 py in 2002, 2007 and 2011, respectively. Conversely, in working-age non-employees, the CMR clearly decreased during 2002–2007 and remained almost stable or declined very slightly during 2008–2011. The CMR was 7.7, 5.7 and 4.6/100 000 py in 2002, 2007 and 2011, respectively. However, this general pattern is not so clear in some employee subgroups (ie, women, non-manual workers) and non-employees (ie, medium/high wealth or the unemployed) (table 2).

Table 2

Crude rate of mortality directly attributable to alcohol in a Spanish Census cohort of working-age employed and non-employed population by sociodemographic factors

Trend changes in DAA mortality during the crisis

The DAA mortality trend during the whole study period was stable among all participants (APC=0.0; 95% CI −0.8 to 0.8), while it increased among employees (APC=4.8; 95% CI 3.7 to 6.0) and decreased among non-employees (APC=−2.7; 95% CI −3.8 to −1.6). Opposite changes in age-adjusted mortality rate were observed in working-age employees (favourable) and non-employees (unfavourable) during the crisis, although the former did not reach statistical significance. Favourable changes were found in all demographic and socioeconomic subgroups among employees, non-employed women and non-employed low-wealth people, although they only reached statistical significance in manual workers and employees aged 25–49. Conversely, significant unfavourable changes were found only in certain non-employed groups such as men, non-married and especially medium/high-wealth people, although non-significant unfavourable changes were found in the remaining non-employed groups, except women and low-wealth people (table 3).

Table 3

Calendar period effect in directly alcohol-attributable mortality in a Census cohort of working-age employed and non-employed population by sociodemographic factors, Spain, 2002–2011

A statistically significant unfavourable change was found among unemployed and inactive medium/high wealth people (figure 1), although after stratification of the two subgroups by gender, the change was not observed in medium/high wealth inactive women (annual effect size: −0.9%; 95% CI −8.5% to 10.3%). When stratifying by the original indicators of material wealth, the greatest unfavourable change was observed in non-employees living in households of 72–104 m2 and who owned two or more cars (table 4).

Figure 1

Time trends in directly alcohol-attributable mortality in precrisis and crisis periods among a census cohort of working-age unemployed and inactive people with medium/high wealth. Spain, 2002–2011. Lines represent age-adjusted mortality rates modelled with the Poisson regression. py, person-years of follow-up. Vertical bars represent 95% CIs of the modelled rates. Asterisks correspond to observed mortality rates. APC, annual percentage change. The effect size indicates the magnitude and direction of change in the linear trend of modelled rates between the crisis and precrisis periods and was calculated as the difference between the APC in 2008–2011 and 2002–2007. It was 9.8; 95% CI 1.5 to 18.3 for the unemployed (estimated from 648 deaths and 14.4 million person-years), and 8.1; 95% CI 1.9 to 14.3 for the economically inactive (estimated from 1467 deaths and 43.4 million person-years).

Table 4

Calendar period effect in directly alcohol-attributable mortality at age 25–64 in a Census cohort of non-employed population by selected categories of material wealth Spain, 2002–2011


Main findings

This study assessed trend changes in DAA mortality at age 25–64 during the post-2008 economic recession in Spain across sociodemographic groups. Although a statistically significant trend change during the crisis was not observed in participants overall, favourable changes were found in certain employed groups (manual workers and employees aged 25–49), and unfavourable changes in the total non-employee group and in certain non-employed subgroups (men, non-married and especially medium/high-wealth people). The greatest unfavourable change corresponded to non-employed people living in households of 72–104 m2 and who owned two or more cars.

Trend changes in alcohol consumption and DAA mortality in Spain during the study period

During the crisis, there was a possible acceleration of the downward trend in per capita alcohol consumption in Spain,29 mainly due to decreased affordability of alcoholic beverages, especially for on-premises consumption,29 as a consequence of decreased income and increased alcohol price (price annual variation: 3.2% in 2002–2007 and 2.2% in 2008–2011);30 although some preventive interventions, especially workplace drinking bans and traffic safety regulations,29 may also have contributed. Trends in DAA mortality in the total study population do not adequately reflect these changes in per capita consumption. However, in a sensitivity analysis, these changes were adequately reflected by mortality from some partially alcohol-attributable conditions, including unspecified liver cirrhosis/diseases, which showed a possible acceleration of the downward trend during the crisis (see online supplementary table S5). Thus, other possible mediating factors to explain trends in DAA mortality should be considered, such as unfavourable changes in distribution of alcohol consumption among drinkers or drinking patterns. Previous studies suggest a decrease in heavy drinking prevalence and an increase in binge drinking prevalence in Spain during the study period;15 ,31 however, data comparing trends in such indicators during the precrisis and crisis periods are lacking. When interpreting trends in DAA mortality, it must be considered that most such deaths were from alcoholic liver diseases (80.3%). It is known that these diseases are associated with regular or daily heavy drinking, but generally not with binge drinking,32–34 and also that mortality from hepatic cirrhosis often changes quite soon (in the same year) after changes in average drinking volume or excessive drinking,35–38 which seems attributable to worsening of disease prognosis, since a long exposure period is required to develop cirrhosis.39 ,40 In addition, it appears there would be a threshold of alcohol consumption (above which the risk hardly increases) for cirrhosis morbidity but not mortality.40

Explaining the divergent changes during the crisis in employed and non-employed people

When participants were classified by employment status at baseline, convergent trends in DAA mortality among employees (upward) and non-employees (downward) were observed. Given the positive association between non-employment and DAA mortality, which was stronger at the beginning of follow-up than later (RR=3.4 in 2002, 1.7 in 2007 and 1.3 in 2011), trends in DAA mortality during the whole study period in both subgroups could simply be explained by the bidirectional flow between employees and non-employees during follow-up. However, if changes in DAA mortality trends among employees (ie, slowing the increase) were mediated only by these flows, an accelerated upward trend would have been expected during the crisis, when job losses increased. As the upward trend slowed down, it can be hypothesised that in this subgroup, favourable changes in DAA mortality were mainly due to a predominance of the effect of decreasing alcohol affordability and other factors like preventive interventions. Decreased affordability has also been invoked elsewhere to explain favourable changes in alcohol consumption (ie, drinking prevalence or frequency or per capita alcohol consumption) during recessions.5 ,10 ,14 ,23

The unfavourable changes during the crisis in DAA mortality among non-employees (ie, slowing the decrease) could be explained by particularly unfavourable changes in distribution of alcohol consumption among drinkers or drinking patterns (ie, heavy drinking prevalence) in this subgroup, which would match well with the hypothesis of excessive drinking as a coping mechanism to relieve the psychological distress caused by unemployment.5 ,14 However, the predominance of flows between employment categories towards non-employment during the crisis may also have contributed to the unfavourable change. Our findings are consistent with previous studies indicating a positive association during the crisis between unemployment and alcohol-related mortality,12 ,20 ,21 ,25 or problem drinking14 ,22–24 and with a Spanish study indicating an increased frequency of alcohol-related disorders in primary healthcare during 2006–2010, and a stronger positive association between such disorders and unemployment in 2010.13 However, another Spanish study showed no heterogeneous changes in heavy or binge drinking prevalence by employment status during the crisis.15 In the aforementioned sensitivity analysis, unfavourable changes in mortality from partially alcohol-attributable conditions among non-employees were not found (see online supplementary table S5). All this suggests that DAA deaths would have occurred mostly among the heaviest drinkers, and that a polarisation in drinking trends probably occurred during the crisis, such that most people would have moved towards lower consumption (procyclical effect), while a small subgroup, mainly not employed, would have moved towards more risky or intensive consumption.

Rejecting the hypothesis of a greater adverse crisis effect in low socioeconomic positions

Our findings suggest that the adverse crisis effects on DAA mortality were not worse in the lower socioeconomic positions. Thus, no relevant heterogeneity in outcome by educational attainment was observed; significant favourable changes were found in employed manual workers but not non-manual workers, as well as significant unfavourable changes in medium/high-wealth non-employees (especially those living in households of 72–104 m2 with two or more cars), but not in low-wealth non-employees. As houses and cars in Spain are often purchased with loans, the unfavourable changes among non-employees living in relatively large houses with two or more cars could be partially explained by increased psychological distress linked to difficulties in repaying loans and coping with expenses or having to sell their property at a loss. These people probably made risky investments during economic expansion (with strongly increasing housing prices), and during the crisis were not able to find a job or sell their home (due to its reduced market value). Although some studies elsewhere have found more unfavourable crisis effects on alcohol-related problems or heavy or binge drinking in less-educated people,14 ,18 ,25 ,26 other studies have found particularly unfavourable effects on such alcohol outcomes in subgroups with high household income.14 ,16 A positive association has also been observed between housing troubles (difficulties in paying rent/mortgage, lost housing) and alcohol-related problems,24 ,27 but not always.41 These housing problems may have been more intense in Spain, where buying rather than renting homes is more widespread, than elsewhere. Thus, in 2010, household occupancy as an owner was 79.8% in Spain vs 70.7% on average in the European Union (EU-28), while the proportion of homeowners with mortgages or loans was 43.1% and 39.2%, respectively.28

The heterogeneity of changes during crisis across other sociodemographic groups

Findings suggest more unfavourable crisis effects on DAA mortality in men than women. Regarding non-employees, this was mainly due to lack of adverse effects among economically inactive women. Other studies have suggested more adverse crisis effects on problematic drinking, alcohol-related problems or binge drinking in men than women,13 ,14 ,24 ,42 highlighting women's greater resilience to the deleterious interaction between alcohol and financial distress.43 However, one study has found a positive association between unemployment and deaths due to alcohol abuse in women, and a negative association in men.20 This study also suggests greater sensitivity to changes in DAA mortality during the crisis, favourable (employees) and unfavourable (non-employees) at ages 25–49 than 50–64. These findings are consistent with previous studies on the prevalence of excessive drinking,15 ,24 although previous results are difficult to summarise due to differences in age categories.15 ,16 A higher sensitivity of DAA mortality to cyclical variation in economic conditions at younger ages is plausible, given the greater financial insecurity of young people.4 Finally, among employees, our study suggests more favourable crisis effects on the outcome in married than unmarried people, which is also consistent with previous studies.14 ,25 ,26

Strengths and limitations

This study has allowed examination of changes in DAA mortality during the post-2008 crisis in a European Mediterranean country across detailed sociodemographic groups in one of the largest population cohorts ever studied. However, it has limitations. Causal relationships cannot be established because of the study design. In addition, alcohol-use measures were not available, and some participants might not have been free of alcohol-related harm at baseline. Findings cannot necessarily be extrapolated to all alcohol-attributable deaths because only a few alcohol-related causes of death were examined. However, as stated above, a sensitivity analysis was performed including deaths from some partially alcohol-attributable conditions. It is very probable that in Spain, DAA deaths, especially from alcohol poisoning, were underestimated, thus trend changes among youth may have been less well captured. Such underestimation may be due to the very low proportion of deaths with forensic assessment (only 6% in Catalonia in 201344), the poor transfer of forensic data to mortality statistics (the retrieval of such data in a Catalonia study increased accidental poisoning deaths by 61.3%44) or even because illegal drugs are primarily coded when found together with alcohol. Changes in certification and coding routines of DAA deaths during the study period were not identified. However, as these processes depend on the level of knowledge of the deceased's alcohol exposure, changes in this knowledge over time could have influenced trends.45 It is also possible that some non-registered emigration abroad contributed to some underestimation of DAA deaths, especially among youth. Living circumstances were measured only at baseline and might have changed during follow-up in some individuals—especially employment, occupation and material wealth—which would primarily affect young adults who are likely to be entering the labour market for the first time and leaving their parents' home. However, the economic crisis can also be expected to have had a dampening effect on this transition. The possible influence of changes in employment status has been discussed above. On the other hand, several longitudinal studies have found that all-cause mortality in employees at baseline who then involuntarily lose their jobs is similar to that of those who remain employed.46 ,47 A possible cohort survival bias would not explain the changes in DAA mortality, because they were almost identical when considering all residents in Spain aged 25–64 during each calendar year. Finally, at the time of record linkage, the most recent mortality data referred to 2011. Consequently, only the first 4 years of the crisis were analysed. A longer time lag would better reflect the crisis impact on DAA mortality, because government austerity measures have continued after the study period, especially in 2011–2012.

Policy implications

In Spain, study findings suggest that the heterogeneous impact of the post-2008 economic recession on DAA mortality may have been conditioned by employment status, occupation and material wealth. Such findings are relevant because Spain is one of the European countries most adversely affected by the crisis and associated austerity measures in terms of social programmes, and labour and housing markets. Although we expected to find a more adverse effect on the outcome in people of lower socioeconomic status, who might have increased problematic drinking due to greater psychological distress, this was not the case. Thus, being employed could be a protective factor for adverse changes in alcohol-related health outcomes during recessions in all socioeconomic positions, and having a medium/high level of material possessions (perhaps bought on credit) could be a risk factor among non-employees. The adverse impact of future crises on alcohol-related problems will likely be lower if employment sustainability is improved and people avoid excessive indebtedness during economic boom periods. Strengthened interventions are also recommended to prevent excessive alcohol consumption and to promote social support for high-risk groups during economic crises.

What is already known on this subject?

  • Although macroeconomic fluctuations are generally accompanied by procyclical changes in overall mortality, alcohol-related mortality changes may vary depending on the context and socioeconomic group. For example, some studies suggest that adverse effects are greater in people who are unemployed,12–14 ,20–25 less educated or experiencing housing instability or rent/mortgage payment difficulties. Studies on this subject in Mediterranean Europe have not been carried out.

What this study adds?

  • Trend changes in directly alcohol-attributable mortality during the crisis were favourable in certain employed strata (ie, manual workers) and unfavourable in non-employed ones, especially those who owned a relatively large house and several cars (medium/high wealth people). Higher adverse effects on the outcome in low socioeconomic positions were not found.


The authors want to thank Pablo Fernández-Navarro (Group leader), Mario González-Sánchez and Javier Gonzalez-Palacios from the Bioinformatics and Research Group in Genetic and Environmental Epidemiology (BRG-GEE), Cancer and Environmental Epidemiology Unit (ISCIII), who kindly provided their computer equipment and experience for analysis of large databases.



  • Contributors GB coordinated the design of the study and writing of the article; IA and FV carried out the statistical analysis and reviewed the consistency of data included in the paper; LS and LO-G carried out the bibliographic searches and elaborated the theoretical framework. MJB wrote the first draft of the manuscript in collaboration with IA. ER performed the interpretation of the results and wrote the final version of article in collaboration with GB and MJB. All authors critically reviewed, approved the final version. Also all authors believe in the overall validity of the paper and take public responsibility for its contents.

  • Funding This work was supported by the Spanish Health Research and Development Strategy (PI15CIII/00022) and National Plan on Drugs (2015I040). Writing the paper was also partially supported by the ISCIII Network on Addictive Disorders (Networks for Cooperative Research in the Carlos III Health Institute) (grant numbers RD16/0017/0013 and RD12/0028/0018).

  • Competing interests None declared.

  • Ethics approval Institutional Review Board of the National Institute of Statistics (INE).

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Study data are currently not available for sharing.

Request Permissions

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.