|
Abstract:
|
In studying genetically-caused susceptibility to lung cancer, potential confounders must be handled with more care than is commonly exercised in regression analyses of observational studies. Each of the following factors---family history of lung cancer, age, race, and smoking---presents unique challenges and different consequences for adjustment. Theories of causal analysis provide a framework for assessing causal effects in distinction to associations, in the presence of other causes and associations. Causal diagrams can help one develop and critique analysis strategies. In some circumstances, the causal calculus of J. Pearl provides consistent estimators for causal relationships, given a reasonably complete causal diagram. This is often true even with the presence of unmeasured factors in the diagram. Advantages and limitations of this theory are examined in the lung cancer study.
|
