Risk adjustment – never enough or too much?

Comparisons of health outcomes between people, services, institutions, and populations are used to assess the relative quality of health care. However, these non-experimental comparisons may be confounded by differences in casemix factors associated with the outcome. Casemix adjustment is used to create a fair comparison – but does it? Simulation experiments suggest that casemix adjustment can make the bias worse in non-experimental studies. How is this possible? This talk will explore the problem that casemix variables may have a different effect in different populations – this is the constant risk fallacy. Examples of non-constant risk in hospital comparisons will be presented and possible solutions will be explored.