A methodological appraisal of the impact of different classification procedures used in three different phases of the australian rheumatoid arthritis twin survey