228 Exposure-Response Modeling: Methods and Practical Implementation
Consequently, due to heterogeneity, trea tment e ffects in different patient p op-
ulations (thos e with high and low exposures) are also different. Therefo re,
to assess treatment effects using the fitted model, not only β
c
, but also β is
important. For models with a linear predictor , e.g., GLM, GLMM and the
Cox model, the same approaches for confounding adjustment can easily be
adapted by including potential confounding factors and/or their interaction
with the exposure in the model.
For nonlinear models , confounding factors may have an impact in multi-
ple places in the model. One may be able to determine a structured model
including the co nfo unders. Consider the model
y
i
= g(c
i
,