The subjects were participants of the Survey on Health and Ageing in Europe (SHARE study). Survey on Health and Ageing in Europe is a longitudinal survey that aims to collect medical, social, and economic data on the population aged over 50 in ten European Union countries (Sweden, Denmark, The Netherlands, Germany, Austria, Switzerland, France, Italy, Spain, and Greece) (Borsch-Supan et al. 2005a; Borsch-Supan et al. 2005b). In the participating SHARE countries the institutional conditions with respect to sampling were so different that a uniform sampling design for the entire project was not feasible. Different registries of national or local level were used that permitted stratification by age. The sampling designs varied from simple random selection of households to complicated multistage designs. The first wave of data was collected by interviews between April and October 2004. The overall household response across the ten SHARE countries in which data collection took place in 2004 was 61.8%, although substantial differences among countries were observed (Borsch-Supan et al. 2005b). From the collected 22,177 individuals, we investigated 11,462 subjects who were between 50–65 years old. We excluded those individuals over age 65, since we have assumed workers normally retire when they become 65 years old. While this assumption has certainly limitations, given the complexity to define retirement at individual level, it was regarded as the best available definition to facilitate cross-national comparisons.
Labor force participation
The outcome of this study is work status, which was based on self-reported current economic status with six mutually exclusive categories: paid work, retired, unemployed, disabled, homemaker, or others. The definition of being employed in SHARE encompasses all individual who declared to have done any kind of paid work in the last four weeks, including self-employed work for family business. Unemployed were those who were laid off from their last job before being able to benefit from normal pension benefits, and therefore were forced to spend some time in unemployment before effectively being retired. Sickness or disability insurance applied to people who exited the labor force for reasons of recognized health problems (Borsch-Supan et al. 2005a). We excluded the disabled participants, because this category predominantly includes persons whose health problems at work were an eligibility criterion for receiving a disability pension.
The European version of self-perceived health, a 5-point scale question ranging between very good to poor, was used to define participant with a poor health (less than good). This frequently used question has been shown to be a good indicator for general physical health (Dwyer et al. 1999; Idler et al. 1997). A second general health measure was long-term illness. Survey on Health and Ageing in Europe has asked respondents whether they had a chronic disease diagnosed by a doctor in their lifetime and those with a positive answer were asked to report the disease from a limitative list. The questionnaire also included the EURO-D scale for depression diagnosis, which has been validated in an earlier cross-European study on depression (Copeland et al. 1999). The EURO-D scale of depression takes into account the following 12 items: depression, pessimism, suicidal, guilt, sleep, interest, irritability, appetite, fatigue, concentration, enjoyment, and tearfulness. A sum score over dichotomous answers was calculated, varying from 0 (not depressed) to 12 (very depressed). For the purpose of this study we defined a clinically significant depression as a EURO-D score greater than three (Borsch-Supan et al. 2005a). In the analysis we used tertile cut-off points with a score from 0 to 3 as reference group, a score of 4–8 as moderately depressed, and a score from 9 to 12 as heavily depressed.
Education was coded according to the 1997 International Standard Classification of Education (ISCED-97) and categorized as low (pre-primary, primary and lower secondary education), intermediate (upper secondary education) and high (post secondary education). Body mass index (BMI) was calculated by dividing body weight in kilogram by the square of body height in meters. According to the BMI, we defined persons as normal (BMI below 25), overweight (BMI from 25 to 30), or obese (BMI above 30). Marital status was used to categorize individuals into those who had a partner and those without. Smokers were subjects who were currently smoking; all others were categorized as non-smokers. Problematic drinking was defined by alcohol consumption of two or more glasses of alcoholic beverage at least 5 days a week in last 6 months (Health Council of the Netherlands 2006). Physical activity was used to categorize individuals with vigorous or moderate physical activity and those without, in compliance with the guidelines for physical activity in leisure time (Pate et al. 1995).
Logistic regression analysis was used to calculate the association between several determinants and the occurrence of early retirement, unemployment, and homemaker. The Odds Ratio was estimated as the measure of association. For the initial selection of potential variables to be included in the multivariate models, univariate associations were determined and variables with a significance level of P < 0.10 were considered for further analysis. In the final multivariate models for each category of labor force withdrawal variables were only included when statistically significant (P < 0.05) with either early retirement, unemployment, or homemaker.
In the first step self-perceived poor health and presence of long-term illness were investigated with adjustment for sex, age, country, education, and marital status as potential confounders. In the second step different chronic diseases were investigated as determinants of early retirement and unemployment, while adjusting for self-perceived health and other potential confounders. Finally, adjusted odds ratios for perceived poor health with retirement and unemployment were calculated within each country, with adjustment for significant lifestyle and sociodemographic variables. Since the number of male subjects was too small in the category homemakers (i.e. taking care of a household), the analysis on the association between health and homemaker was performed only in women.
In order to investigate the influence of national labor market conditions, the Pearson correlation coefficient was used to analyze the association between unemployment rates at national level and observed odds ratios for health with early retirement, unemployment, and being homemaker. The statistical analyses were carried out with SPSS version 11.0 for Windows.