model simplification with lme4, brier etc with lrm

Dear List,

I have arrived at the most likely model using the  AIC value, AIC weights, etc etc as model selection, the model has  5 fixed variables and two random effects.

I am wanting to use this model in a predictive sense, i understand lmer does not have a predict() function and from reading the literature i understand why. 

So my approach was to use the "most likely" model selected with lmer and run it as a linear model with lrm( package design) and leave out the random effects ( which are significant).

From here i validated the model with the AUC, Brier score etc and used the predict function.

However my r2 values, brier score etc were not very good and i was wondering if this is because i left out the random effects? And if it is is there a work around?

Thanks

Peter