PHP252 - IS FACTOR ANALYSIS AS CREDIBLE AS OTHER TESTS OF DIFFERENCE AND ASSOCIATION FOR EVALUATING THE VALIDITY OF QUALITY OF LIFE MEASURES? ASSESSING THE CONVERGENT AND DISCRIMINANT VALIDITY OF THE WOMEN’S HEALTH QUESTIONNAIRE AGAINST THE EQ-5D-5L AND SF-6D

Abstract

Validation of quality of life (QoL) instruments assesses the degree to which they measure what they are supposed to measure. Despite most validation exercises being based on statistical tests of difference/association, other alternative validation techniques including exploratory (EFA) and confirmatory factor analysis (CFA) have been proposed. Little is however known about whether validation using EFA/CFA alone is sufficient and/or whether EFA/CFA validation results are comparable to those of other statistical tests. This issue is addressed within the context of validating the Women’s Health Questionnaire (WHQ-23) against the EuroQoL 5 Dimensions 5 Level (EQ-5D-5L) and the Short Form 6 Dimensions (SF-6D). Data on 560 women from the Australian general population were analysed. Tests of differences used were Kruskall Wallis (for evaluating discriminant validity), spearman’s correlation and Krippendorff’s alpha (for evaluating convergent validity). In EFA/CFA, discriminant validity was assessed by comparing the variance extracted estimates (AVE) to the squared correlation (Corr2). Convergent validity using EFA/CFA was evaluated by the extent to which one instrument’s dimension and summary/utility scores loaded onto the same factors as the scores of other instruments. Tests of difference showed that all three instruments discriminated between respondent characteristics in the same direction. The AVE (range: 0.4122-0.5991) from EFA/CFA was greater than Corr2 (range: 0.0001-0.00245) also suggesting good discriminant validity. In terms of convergent validity, weak to moderate convergence was observed using tests of association with correlation higher between the WHQ-23 and SF6D (correlation range: 0.19-0.77 and Krippendorff’s’ alpha range: 0.17-0.69). Similarly, in EFA/CFA, the WHQ-23 and SF6D were found to load onto more common factors than the WHQ-23 and EQ-5D-5L. All factor loadings were ≥ 0.59 suggesting moderate convergence. These results suggest that EFA/CFA yields similar results as tests of difference/association and may therefore offer an appropriate alternative approach for the validation of QoL instruments.