The Sickness Impact Profile is a well known measure of functional status which has been validated as a measure of differences in function between patients, but not as a measure of within-patient change. To evaluate its performance as a measure of change, the Sickness Impact Profile was compared to a transition index that directly assesses intra-patient differences in maximal physical and emotional function. With the index, patients rate themselves as much better, slightly better, the same, slightly worse or much worse as compared to a previous assessment. When compared to the corresponding index ratings, Sickness Impact Profile subscale score changes for a rating of worse were of greater magnitude than those for a rating of better. The inherent variability of the Sickness Impact Profile score was estimated using the standard deviation of the score change for the index rating "the same". For patients to be regarded as changed they had to lie outside the probability limits of this score change. Defined in this way, only patients who stated they were worse on the index were detected. Receiver operator characteristic curves were constructed and also demonstrated that deteriorations in function were more accurately detected. Using different score cutoffs, the best true positive rate (71%) was obtained using a one point change on the psychosocial subscale. This was accompanied by a 39% false positive rate. No score value appeared sufficiently accurate to be useful. Its apparent inability to detect improvements and deteriorations equally, and the wide standard deviations for a given change in function, may limit the use of the Sickness Impact Profile for following individuals over time.(ABSTRACT TRUNCATED AT 250 WORDS)