Problems using aggregate scores of the Medical Outcomes Study 36-item Short Form Health Survey in knee and hip osteoarthritis: possible solutions

Abstract

Recently, Rannou et al.1Rannou F. Boutron I. Jardinaud-Lopez M. Meric G. Revel M. Fermanian J. et al.Should aggregate scores of the Medical Outcomes Study 36-item Short-Form Health Survey be used to assess quality of life in knee and hip osteoarthritis? A national survey in primary care.Osteoarthritis Cartilage. 2007; 15: 1013-1018Abstract Full Text Full Text PDF PubMed Scopus (14) Google Scholar performed an exploratory factor analysis on the eight subscale scores of the short form-36 (SF-36) in a sample of over 4000 patients with osteoarthritis (OA) and concluded that evidence did not support the use of the aggregate physical component score (PCS) and mental component score (MCS). This conclusion was based on the authors' interpretation that the factor structure they found was not consistent with the two-factor orthogonal factor structure used to develop the Medical Outcomes Study (MOS) SF-36 composite scoring system2McHorney C.A. Ware Jr., J.E. Raczek A.E. The MOS 36-item Short-Form Health Survey (SF-36): II Psychometric and clinical tests of validity in measuring physical and mental health constructs.Med Care. 1993; 31: 247-263Crossref PubMed Scopus (5337) Google Scholar. The authors pointed out that two subscales linked with the MCS (emotional role and social functioning) loaded on both the physical and mental component factors and that the general health perception subscale loaded on the mental component factor rather than the physical component factor. In fact, the authors of the original MOS SF-36 scoring system hypothesized that three subscales would load moderately to strongly on both the physical and mental component factors, and, indeed, in the original general population sample, four subscales had factor loadings ≥0.30 on both factors2McHorney C.A. Ware Jr., J.E. Raczek A.E. The MOS 36-item Short-Form Health Survey (SF-36): II Psychometric and clinical tests of validity in measuring physical and mental health constructs.Med Care. 1993; 31: 247-263Crossref PubMed Scopus (5337) Google Scholar – not unlike the results reported by Rannou et al. This in itself is not problematic. That is because the rotated factor component matrix interpreted by Rannou et al. is not used directly to compute orthogonal factor scores. Rather, the orthogonal factor scores are generated from the factor component score coefficient matrix. Orthogonal factor scores computed by this method eliminate the problem of multicollinearity when using both component scores in multiple regression, for instance, but result in potentially difficult to interpret residualized scores. This orthogonal scoring method does not, as suggested by Rannou et al., however, preclude individual subscales loading on both components in the factor component matrix. Nonetheless, there are two significant problems with the use of MOS SF-36 scoring in patients with chronic illness, and the results reported by Rannou et al. illustrate these problems. One problem relates to the assumption that the MOS SF-36 system will generate orthogonal factor scores. The validity of the computational formula for the PCS and MCS for patients with chronic illness is predicated on the assumption that the relationship between the eight subscales in this population is the same as in the general population data from where the formulas were derived. To the contrary, a robust relationship exists between physical and mental health, particularly in the context of physical or mental illness, and this relationship would be expected to alter aggregate scores as computed by the MOS SF-36 methods3Simon G.E. Revicki D.A. Grothaus L. Vonkorff M. SF-36 summary scores: are physical and mental health truly distinct?.Med Care. 1998; 36: 567-572Crossref PubMed Scopus (203) Google Scholar. Consistent with this, the correlation between the PCS and MCS reported by Rannou et al. was not zero, but rather 0.14. Thus, rather than residualized factor scores, one is left with partially or mostly residualized factor scores, the interpretation of which is somewhat unclear. The second problem relates to the use of negative scoring coefficients in the MOS SF-36 method to calculate summary scores. Farivar et al.4Farivar S.S. Cunningham W.E. Hays R.D. Correlated physical and mental health summary scores for the SF-36 and SF-12 Health Survey, V.I.Health Qual Life Outcomes. 2007; 5: 54Crossref PubMed Scopus (253) Google Scholar recently demonstrated that when physical subscale scores are well below the mean and mental subscale scores somewhat less below the mean, this scoring method will result in an artifactual migration of the aggregate PCS score away from the mean and a migration of the aggregate MCS score toward the mean. Indeed, in Table II of the report by Rannou et al., OA patient scores on the physical subscales (physical functioning, physical role, bodily pain, and general health perception) are 1.3–2.3 standard deviations lower than the Swedish general population, and the PCS score is approximately 1.8 standard deviations below the standardization population mean. More significantly, scores on the mental subscales (mental health, emotional role, vitality, and social functioning) are 0.8–1.3 standard deviations lower than the Swedish general population scores, but the MCS is only 0.3 standard deviations lower than the standardization sample mean. This is consistent with most reports on patients with chronic illness, in which MCSs tend to be close to the general population mean despite a high prevalence of depression, for instance. An important implication of this scoring artifact is that improvements in quality of life resulting from medical treatment may be evident in the subscale scores, but may not move the aggregate scores enough to be detected3Simon G.E. Revicki D.A. Grothaus L. Vonkorff M. SF-36 summary scores: are physical and mental health truly distinct?.Med Care. 1998; 36: 567-572Crossref PubMed Scopus (203) Google Scholar. Thus, as Rannou et al. argue, there are substantive problems with using the MOS SF-36 aggregate PCS and MCS in chronically ill patients. There are, however, possible solutions. The RAND-36 method, for instance, provides an alternative scoring algorithm that generates correlated, rather than orthogonal, summary scores that can be compared to a US standardization sample5Hays R.D. Sherbourne C.D. Mazel R.M. The RAND 36-item Health Survey 1.0.Health Econ. 1993; 2: 217-227Crossref PubMed Scopus (1983) Google Scholar. In addition, Farivar et al.4Farivar S.S. Cunningham W.E. Hays R.D. Correlated physical and mental health summary scores for the SF-36 and SF-12 Health Survey, V.I.Health Qual Life Outcomes. 2007; 5: 54Crossref PubMed Scopus (253) Google Scholar recently demonstrated that group-specific scoring algorithms can be generated, which, although not comparable to population norms, eliminate many of the problems that have been reported when the MOS SF-36 aggregate scores are used. We encourage Rannou et al. to consider testing these alternatives in their large sample of patients with OA. The authors have no conflicts of interest to disclose.