Aims: To illustrate the importance of access to register data on determinants and predictors of study participation to assess validity of population-based studies. In the present investigation, we use data on sociodemographic conditions and disease history among individuals invited to the Swedish cardiopulmonary bio-image study (SCAPIS) in order to establish a model that predicts study participation.
Methods: The pilot study of SCAPIS was conducted within the city of Gothenburg, Sweden, in 2012, with 2243 invited individuals (50% participation rate). An anonymous data set for the total target population ( n = 24,502) was made available by register authorities (Statistics Sweden and the National Board of Health and Welfare) and included indicators of invitation to and participation in SCAPIS along with register data on residential area, sociodemographic variables, and disease history. Propensity scores for participation were estimated using logistic regression.
Results: Residential area, country of birth, civil status, education, occupational status, and disposable income were all associated with participation in multivariable models. Adding data on disease history only increased overall classification ability marginally. The associations with disease history were diverse with some disease groups negatively associated with participation whereas some others tended to increase participation.
Conclusions: The present investigation stresses the importance of a careful consideration of selection effects in population-based studies. Access to detailed register data also for non-participants can in the statistical analysis be used to control for selection bias and enhance generalizability, thereby making the results more relevant for policy decisions.
Keywords: Bias correction; inverse probability weighting; population-based study; propensity score; register data; residential area; validity.