An ecological study was done, in which the informations on SES, lifestyle, and mortality due to CSD collected from 66 administrative sub-regions of Poland was analyzed. Sub-regions were defined according to the Nomenclature of Territorial Units for Statistics NUTS-3 in 2006 [25]. Information for analysis in the period 2010–2014 was obtained from the Central Statistical Office and the Social Diagnosis survey conducted every 2 years since 2000 on the same sample of households [26].
Mortality
For the analysis, data on deaths during the period 2010–2014 were used. The causes for these deaths have been encoded according to the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) as CSD (codes: I00–I99), including IHD (I20–I25) and CD (I60–I69).
In statistical analyses, the age standardized number of deaths was used. A direct standardization method considering demographic structure in 5-year age groups separately for each of the 66 sub-regions of Poland was used. As standard population, we assumed populations of males and females in Poland in 2010, for which standardized mortality coefficients were determined [27]. The coefficients were calculated for the population aged over 14. In addition to the standardized mortality coefficients, 95% confidence intervals were presented.
When describing the dynamics of changes in mortality rates in the period 2010–2014, the deaths prevented or postponed (DPP) index was used [28]. The index value is the percentage obtained as the difference between the expected number of deaths in 2014, with the assumption that the mortality rate was the same as in 2010, and the actual number of deaths in 2014 relative to the expected number of deaths in 2014:
Area-level SES index
The SES index was constructed using the following sub-region characteristics: percentage of people with university education (%), percentage of people employed in finance and real estate (%),average monthly salary (PLN), unemployment rate (%) and percentage of people on social support due to poverty (%) [29]. To calculate the SES index, its component variables were standardized via linear transformation such that their expected value was equal to 0 and the standard deviation was equal to 1. In addition, in the case of destimulants (variables: unemployment, social support users), their sign was reversed. Finally, the SES value was assumed as the arithmetic mean of the transformed components. The index was calculated based on the averaged values of component variables in the period 2010–2014. In addition, the 66 sub-regions were divided into three tercile groups based on the SES value: group with the lowest values of the index and thus the highest deprivation level (deprived), group with the highest index values (affluent), and middle group.
Other explanatory variables
Tobacco consumption and body mass index (BMI) in the 66 sub-regions were calculated based on the Social Diagnosis survey from 2011. This survey included 26,453 people aged 16 and above, who were residents of 12,386 households. The households were selected based on the two-stage stratified sampling method, first at the voivodeship level, followed by the class of residential location (large cities, small cities, and villages) [26].
Statistical analysis
To determine the relationship between mortality and SES, the Poisson regression model for count data was used. As this model aimed to represent the coefficient of mortality per 100,000 residents, and not the number of deaths, the age-standardized number of deaths in individual sub-regions was taken as a dependent variable, including the number of people in sub-regions as the offset. Calculations were realized using generalized linear models with Poisson distribution as probability distribution and logarithm as a link function.
Due to repeated measurements in the same statistical units (mortality in sub-regions in subsequent years in the period 2010–2014), we used generalized estimating equations (GEE) for obtaining generalized linear models for correlated data [30]. In GEE models, an exchangeable structure was assumed for the working correlation matrix.
The Poisson models were presented using the regression coefficients exponentially transformed, so they can be interpreted as the expected relative change of dependent variable (here: number of deaths standardized for the age) calculated for the increase of independent variable by 1 unit, i.e., relative risk. The presented coefficients can also be interpreted as a relative change of mortality coefficients and not only death number. Furthermore, 95% confidence intervals were presented for the coefficients of the regression model and their p-values of corresponding Wald’s tests.
The created regression models included three causes (CSD, IHD, CD) of death in men and women. As independent variables, the models included calendar year (for assessment of time effect on the dependent variable) and the aforementioned index—the level of socioeconomic development in the sub-regions. In addition, lifestyle characteristics were included: BMI and frequency of smoking tobacco in the sub-region (respectively in female and male population) and population density in the sub-region with logistic transformation at base 2 applied.
Statistical calculations were performed using the IBM® SPSS® Statistics for Windows, version 20.0 statistical package (IBM Corporation, Armonk, NY, USA). The level of statistical significance was assumed at α=0.05.