Using smoothing spline anova to examine the relation of risk factors to the incidence and progression of diabetic retinopathy.

Smoothing spline ANOVA (ANalysis Of VAriance) methods provide a flexible alternative to the standard parametric GLIM (generalized linear models) methods for analysing the relationship of predictor variables to outcomes with data from large epidemiologic studies. These methods allow the visualization of relationships not readily fit by simple GLIM models, and provide for the ability to visualize interactions between the variables. At the same time, they reduce to GLIM models if the data suggest that the added flexibility is unwarranted. Using this method, we investigate risk factors for incidence and progression of diabetic retinopathy in a group of patients with older onset diabetes from the Wisconsin Epidemiological Study of Diabetic Retinopathy. We carry out four analyses to illustrate various properties of this class of methods. Some of the results confirm previous findings with use of standard methods, while others allow the visualization of more complex relationships not evident from the application of parametric methods.