Abstract:
|
In medical research and epidemiology, the functional form of relationships between outcomes and continuous risk factors / covariates may be nonlinear. Non-linearity is not easily assessed and is quite possibly ignored in many medical research studies. There are a variety of methods available for building nonlinear models with linear model parameters (e.g., splines, polynomial regression). Fractional polynomial (FP) regression can be applied in almost any regression context including Cox proportional hazards regression. In many cases, FP regression is easier to implement and provides a better fit than ordinary polynomial regression. In this study, we apply FP to a Cox regression model of hospital length of stay (LOS), where some of the potential risk factors (e.g., hospital occupancy and physician workload) vary from day to day, and patients are clustered within physicians.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.