Why not use logistic regression? Or, if you have time-to-event data, Cox regression.

While I'm not an R maven, it seems to me that R is probably interpreting your predictor (risk.factor) as a variable that has a length corresponding to the number of risk factors, and so cannot use it in the model

Thanks Ronan

firstly, let me congratulate you on the excellent use of the word 'maven'!

In terms of the stats suggestions, I completely agree with your suggestion. I'm planning to use logistic regression to create models predicting disease presence, and Cox to assess death.

For this analysis, I'm interested in assessing the additive benefit of individual markers, or a combination of markers to an already accepted base model. This could be done manually although the number of variables and possible combinations I have would make this mind-numbingly awful. Step-wise regression is another option, although there is no guarantee that the 'best' combination will actually be assessed. The benefit of using the R package 'MuMIn' is that it automatically assesses every covariate combination, bypassing the issues of stepwise regression, and overcoming brain drain associated with manual input. In theory, this is meant to be a short-cut!

In terms of the length issue - your suggestion is a possibility but am still hitting the books to get to the bottom of this.

Cheers

Joe