Reliability and validity in measuring physical and mental health construct of the Portuguese version of MOS SF-36.

Abstract

To identify and evaluate the Psychometric validity of the two general dimensions of the Short Form (SF)-36 Portuguese version.A representative sample of 1446 adults (60,4% women) resident in Porto, Portugal completed a structured questionnaire and a Portuguese version of SF-36, final data allowed the estimation of all 8 sub-dimensions for 1326 (91,7%) participants. The internal consistency was evaluated using the Cronbach's alpha. Principal components analysis (PCA) was used to test the dimensionality. To evaluate the construct validity of the two general dimensions the sample was divided into 4 groups, according to the presence of chronic diseases and depression (evaluated by Beck Depression Inventory). Logistic regression was use to measure how the new dimensions capture the theoretical differences between groups and bootstrapping to check the principal components loadings reliability. To evaluate loss of discrimination power when using only the 2 general dimensions, for the different groups, we compared the area under the ROC curve of the logistic regression with the original dimensions and the logistic regression with the 2 dimensions.The Cronbach s alpha was 0.82 for the physical and 0.87 for the mental dimension. The total variance explained by the extraction of 2 components was 70.4%. The 4 sub-dimensions from physical domain correlated more strongly with the first component (r=[0.69-0.83]), and the 4 sub-dimensions from the mental domain correlated more highly with the second component (r=[0.65-0.88]). The largest standard deviation obtained for principal components loadings of the bootstrapping was 0.05. The general dimensions capture the theoretical differences between groups. There were no significant differences between the areas under the ROC curve of the logistic regression for the original sub-dimensions and the general dimensions.The use of the two SF-36 summary measures, physical and mental health, allows us to analyse the results more efficiently without loss of information and capture different manifestations of health status.