WEIGHTS in lme

["Hunsicker, Lawrence" <lawrence-hunsicker@uiowa.edu>]

Back to the trough again.

I am trying to analyze factors affecting slopes of estGFR following kidney transplantation, using lme in a repeated measures design.  The follow-up of the patients varies, to some extent simply because patients are “randomly” lost, but also to some degree because of informative censoring.  (When the patient’s loss of function is rapid, not surprisingly we have fewer measurements before the patient goes back on dialysis.)  Since lme effectively weights PATIENTS by the numbers of observations in the model, this systematically undervalues the patients with more negative slopes and biases the outcomes. 

Now I recognize that reweighting the observations so that each PATIENT gets a cumulative weight of 1 does not really undo the problem of informative censoring.  But it should help minimize the bias.  So I planned to use the WEIGHTS option in lme.  However it appears that WEIGHTS is expecting a function related to the within individual variance, not a predefined and assigned weight.  Is there any way to get it simply to accept a defined weight for each observation (which I would set to 1/(nobs-1) )?

As always, many thanks in advance to anyone that can help me with the above.

Larry Hunsicker