In spite of well-known drawbacks, standard regression analysis such as cox-regression and logistic regression are used in observational studies to identify and disentangle risk factors for a disease or outcome of interest. When the purpose is to guide policy-making, the regression models are often used to adjust for confounding. There are limitations when interpreting regression coefficients as effects of risk factors. Apart from model assumptions, the causal relations between the independent variables in the models are not considered which may lead to fallacies. Also, a main drawback with regression is that the estimates cannot be used for comparisons to other populations or subpopulations, for examples comparisons between countries, regions or risk-groups.
These threats when using register-based studies to evaluate risk factors are investigated within the research field of causal inference in observational studies. Here, causal parameters, estimands, are studied and proposed. To improve studies of risk factors for decision making we propose research in three areas:
1. Estimands and identification of risk factors in a causal structure.
2. Estimands and variable importance measures.
3. Application of the methods to register studies of two chronic diseases: Type 1 Diabetes and End Stage Renal Disease.
The datasets in the registers contain a rich set of medical and socioeconomic variables through linkage to data from Statistics Sweden and the National Board of Health and Welfare.