@article {Hobart363,
author = {Hobart, J and Freeman, J and Lamping, D and Fitzpatrick, R and Thompson, A},
title = {The SF-36 in multiple sclerosis: why basic assumptions must be tested},
volume = {71},
number = {3},
pages = {363--370},
year = {2001},
doi = {10.1136/jnnp.71.3.363},
publisher = {BMJ Publishing Group Ltd},
abstract = {OBJECTIVES To evaluate, in people with multiple sclerosis, two psychometric assumptions that must be satisfied for valid use of the medical outcomes study 36-item short form health survey (SF-36): the data are of high quality and, it is legitimate to generate scores for eight scales and two summary measures using the standard algorithms. METHODS SF-36 data from 438 people representing the full range of multiple sclerosis were examined (mean age 48; 70\% women). Data quality (per cent missing data and computable scale and summary scores) were determined, six scaling criteria were tested to determine the legitimacy of generating the eight SF-36 scale scores using Likert{\textquoteright}s method of summed ratings, and two scaling criteria were tested to determine the appropriateness of the standard SF-36 algorithms for weighting scale scores to generate two summary measures. RESULTS Data quality was excellent except in the most disabled subgroup where missing responses reached a maximum of 16.5\% and summary scores could only be computed for 72\%. There was clear support for the generation of SF-36 scale scores. Item response distributions were symmetric, item mean scores and variances were equivalent, corrected item-total correlations were high (range 0.46{\textendash}0.85) and similar, and definite scaling success rates exceeded 96\%. Nevertheless, there were notable floor or ceiling effects in four of the eight scales. Assumptions for generating two SF-36 summary measures were only partially satisfied. Although principal components analysis suggested a two component model, these components explained less than 60\% of the total variance in SF-36 scales, and less than 75\% of the variance in five of the eight scales. Moreover, scale to component correlations did not support the use of scale weights derived from United States population data. CONCLUSIONS When using the SF-36 as a health measure in multiple sclerosis summary scores should be reported with caution.},
issn = {0022-3050},
URL = {https://jnnp.bmj.com/content/71/3/363},
eprint = {https://jnnp.bmj.com/content/71/3/363.full.pdf},
journal = {Journal of Neurology, Neurosurgery \& Psychiatry}
}