Dear friends. We have a problem in designing a study of patient tolerance to manipulation of dialysate composition (control, intervention) during hemodialysis of patients on chronic dialysis.
The response variable is a qualitative measure of wellbeing on a visual-analogue scale. We suspect unknown patient characteristics to be of uttermost importance in determining the response to any intervention and would therefore prefer to base the analysis on within-patient differences.
We would therefore consider a cross-over design.
To get a reasonable estimate of the size of effect, we will have to use a fair numbers of dialysis sessions to measure on for each patient.
For various practical reasons, the most easy method would be to randomize each patient individually to a free sequence of treatments (probably of equal size, though). We are particularly keen to avoid the bias induced by the dialysis nursing staff knowing either the sequence or the timing of changing treatments.
The question is if this would be straightforward to analyse? We suspect that both carry-over and other period effects might occur. Extensive wash-out would however probably invalidate even within-patient comparisons and lead to disturbances caused by drop-outs and instability.
We appreciate that a parallel design would obviate some of these concerns but fear that between patient comparisons would be too weak.
Troels Ring, MD
Department of Nephrology
Aalborg Hospital
Aalborg Denmark
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|