Study population
Our study population consisted of all 80 212 individuals aged 40 to 64 years (from 146 different countries) who resided in Malmo, Sweden, during 2003. (Malmo is the largest city in the County of Scania, in southern Sweden, with 257 455 inhabitants as of January 1, 1999.) Of the 80212 people studied, 33% were foreign-born men (n = 40 471) and 31% foreign-born women (n = 39 748). We also identified 73% of the population (58 848/80 212) who had visited a physician at least once in 2003. Of this group, 32% were foreign-born men (n = 26 599) and 31% foreign-born women (n = 32 249).
The present analysis is based on the 2003 County of Scania Register for Resource Allocation, which includes, among other variables, age, gender, marital status, education, income, disability pension, and country of birth, as well as detailed information on health care utilisation for each individual in the county. The Regional Ethics Review Board in Lund, and the Data Safety Committee at Statistics Sweden, approved the use of this database. The information was handled in such a way as to preserve the anonymity of the subjects. Analyses were performed at the individual level, but results always presented in the aggregate.
Outcome variable
A disability pension may be granted for medical reasons to a person who has lost 25% to 100% of their working capacity. For our purposes, the beneficiary of a disability pension is any individual who has been granted such a pension, irrespective of their degree of disability.
Individual characteristics
We considered those people who were single, separated, or widowed as "living alone". This definition did not take into account unmarried couples or people sharing the same household – a factor that might produce misclassification and lead to underestimating the association between living alone and having a disability pension. Formal schooling of nine years or less was categorised as "low educational achievement". Those with further schooling were labelled as having "high educational achievement". Age was considered a contiguous variable and centred on the median. Since the association between age and having a disability pension may not be linear, age-squared was also included in the models. Utilisation of health care (in the form of visits to doctors, irrespective of the type of provider, but excluding hospitalisations) was classified as either "yes" or "no". Those who visited a physician at least once were categorized according to the type of provider they consulted: public general practitioner, public specialist, private general practitioner, or private specialist. An individual who visited different types of physicians during 2003 received multiple classifications.
Country of birth (i.e., contextual) variables
The socioeconomic characteristics of one's country of origin, over and above those of the individual, may have conditioned the likelihood of receiving a disability pension. We tested this hypothesis by using the World Bank Classification of Country Economies as a contextual variable [29] in which countries are classified according their gross national income (GNI) per capita, using the World Bank Atlas method. The GNI categories employed are low, lower middle, upper middle, and high income. We merged the first two into a single category designated "low income country", and used the high income country category as a reference in the comparisons.
Statistical and epidemiological methods
In order to account for a possible modification of effects due to gender, we stratified the study population into males and females.
Due to the hierarchical structure of the data, with individuals nested within countries of birth, and the possibility of intra-country correlation regarding the likelihood of having a disability pension, we applied multilevel logistic regression [26, 30, 31]. Of the 146 different countries of birth that made up our study population, 66 countries were represented by fewer than ten individuals. However, this disproportion can be satisfactorily handled by multilevel regression analysis [31].
We investigated disability pensions for the entire population and separately for those who had contact with physicians. We then established four consecutive multilevel models. In the first or empty model (A), the probability of having a disability pension was only a function of an individual's country of birth and was modelled by a random intercept. Our second model (B) included the individual variables of age, education, and marital status. The third model (C) considered an individual variable for utilisation of health care. Finally, the fourth model (D) expanded the third model to include the contextual variable "World Bank Classification of Country Economies". Models B and C also allowed us to study the interaction between country of birth and individual variables (educational achievement, marital status, and the variable for utilisation of health care) by allowing the regression coefficients of the individual level variables to be random at the level of country of birth. Such models indicate whether the bearing of individual variables on disability pensions differ by country of birth (i.e., for individuals from some countries, living alone or consulting a public general practitioner may be associated with a higher probability of having a disability pension, whereas in other countries the reverse may be true).
By this general strategy we were able to quantify differences between various countries of birth (model A) and estimate the role played by individual characteristics of people from each nation, as well as quantify possible cross-level interactions between marital status, education, and country of birth (model B). The third model (C) assessed the relationship between consulting different providers and having a disability pension, and also indicated possible cross-level interactions between individual patterns of visiting different types of physicians and one's country of birth. Finally, the last model (D) indicated whether the economic circumstances in one's country of birth, over and above individual conditions, were associated with being granted a disability pension in Sweden.
We appraised the association of the variables studied and having a disability pension by odds ratios (OR) with a 95% confidence interval (CI), as obtained from the regression coefficients (standard error).
In order to quantify the influence of country of birth for having a disability pension, we computed the median odds ratio (MOR) [32]. With this method, the variance at the second level (country of birth) is translated into the well-known OR scale. The MOR could represent how much (in median) the likelihood of having a disability pension would increase if an individual had been born in a country whose inhabitants had a greater probability of receiving disability pensions. If the MOR is equal to one, country of birth does not condition an individual's likelihood of having such a pension.
In the presence of slope variance for marital status or level of education (model B), or the variable for utilisation of health care (model C), the country of birth variance becomes a function of individual marital status, education, or physician utilisation (e.g., disability pension differences between countries of birth may be much greater for married individuals than for those living alone, or lower for those who visit a private general practitioner than for those who visit other physicians). Therefore, we calculated the variance between countries of birth as a function of individual variables [33–35], and expressed this variance on the OR scale by means of the MOR [36].
MLwiN software, version 2.0 [34], was used for the analyses. Parameters were estimated using the Markov Chain Monte Carlo (MCMC) procedure. The Deviance Information Criterion (DIC) was used as a measure of how well our different models fit the data. A lower value on the DIC indicated a better fit of the model [37].