Selection bias in a population survey with registry linkage: potential effect on socioeconomic gradient in cardiovascular risk.

Non-participation in population studies is likely to be a source of bias in many types of epidemiologic studies, including those describing social disparities in health. The objective of this paper is to present a non-attendance analysis evaluating the possible impact of selection bias, when investigating the association between education level and cardiovascular risk factors. Data from the INTERGENE research programme including 3,610 randomly selected individuals aged 25-74 (1,908 women and 1,702 men), in West Sweden were used. Only 42% of the invited population participated. Non-attendance analyses were done by comparing data from official registries (Statistics Sweden) covering the entire invited study population. This analysis revealed that participants were more likely to be women, have university education, high income, be married and of Nordic origin compared to non-participants. Among participants, all health behaviours studied were significantly related to education. Physical activity, alcohol use and breakfast consumption were higher in the more educated group, while there were more smokers in the less educated group. Central obesity, obesity and hypertension were also significantly associated with lower education level. Weaker associations were observed for blood lipids, diabetes, high plasma glucose level and perceived stress. The socio-demographic differences between participants and non-participants indicated by the register analysis imply potential biases in epidemiological research. For instance, the positive association between education level and frequent alcohol consumption, may, in part be explained by participation bias. For other risk factors studied, an underestimation of the importance of low socioeconomic status may be more likely.