Embedded Files
different settings and with different populations (McHorney, 1996; Ware, 1997). Hence, new and refined instruments, and those applied in different settings or with different populations require evidence of validity. Both qualitative and quantitative methods can be used to assess validity. Face and content validity require appraisal of item content, and assessment of its relationship to the instrument’s proposed purpose and application (Fitzpatrick et al., 1998). Methods of item generation and instrument development may influence this assessment. Literature reviews, theoretical propositions, and interviews or focus groups with patients or health-care professionals may all inform this process. However, for patient-reported instruments to have content validity and relevance to the recipients of care, patients should be directly involved in item generation, usually via one-to-one interviews or focus groups (Fitzpatrick et al., 1998). The quantitative assessment of validity requires comparison of the scores produced using patient-reported health instruments with those derived from other measures of health, clinical, and socio-demographic variables. Patientreported instruments measure hypothetical constructs which are by definition non-observable, for example, HRQL and pain, and address a more general hypothesis than that supported by a specific behaviour (Nunnally & Bernstein, 1994). However, by reference to established evidence and the instrument’s underlying theoretical base and item content, quantifiable relationships with a range of other instruments and clinical and socio-demographic variables can be expected (Ware, 1997; Fitzpatrick et al., 1998). Expected correlations between variables should be presented to allow validity to be disproved (McDowell & Jenkinson, 1996). The strength of correlation between variables, be they small (less than 0.30), moderate (less than 0.50), 8 or large (greater than 0.70), indicates that the instrument measures the construct in a manner founded on theory or established evidence (McHorney et al., 1993). For example, two patient-reported measures of functional disability with similar content would be expected to correlate strongly. Construct validity may also be assessed using ‘extreme groups’, which theorises that one group will possess more or less of a construct (Streiner & Norman, 2008). For example, compared to the general older population, older people who are hospitalised following a hip fracture may be expected to report greater pain and worse HRQL. The dimensionality or internal construct validity of a multi-item instrument can be assessed using factor analysis or principal component analysis. Principal component analysis can be used to assess the underlying structure of a multiitem instrument through the identification of components, or domains, into which items may group (McDowell, 2006). This form of analysis adds empirical weight to a hypothesised domain structure. For example, principal component analysis has supported the hypothesised eight-domain structure of the SF-36 (McHorney et al., 1993). Responsiveness is considered a necessary measurement property of instruments intended for application in evaluative studies measuring longitudinal changes in health (Beaton et al., 2001; Liang et al., 2002). The numerous approaches to evaluating responsiveness have been reviewed by a number of authors (Liang, 1995; Wyrwich et al., 2000; Beaton et al., 2001; Liang et al., 2002; Terwee et al., 2003). Responsiveness has been described as the ability of an instrument to measure clinically important change over time, when change is present (Fitzpatrick et al., 1998). It has also been argued that responsiveness can be viewed as longitudinal validity or as a measure of treatment effect (Terwee et al., 2003). Patient-reported health instruments have had by far the greatest application in clinical trials and most of the literature on responsiveness relates to the measurement of change in health for groups of patients (Fitzpatrick et al., 1998). There are two broad approaches to assessing responsiveness: distributionbased and anchor-based (Wyrwich et al., 2000; Norman et al., 2001). Distribution-based approaches relate changes in instrument scores to some measure of variability, the most common method being the effect size statistic. The three widely-reported effect size statistics use the mean score change in the numerator, but have different denominators (Fitzpatrick et al., 1998). The effect size (ES) statistic uses the standard deviation of baseline scores (Liang, 1995). The standardised response mean (SRM) uses the standard deviation of the change score to incorporate the response variance in change scores. However, both the ES and SRM may be influenced by natural variance in the underlying state and by measurement error. The modified standardised response mean (MSRM), or responsiveness index, addresses the inherent natural variance that may occur in patients who otherwise report their health as unchanged, and non-specific score change by using the standard deviation of change in patients who are defined as stable (Deyo et al., 1991). In 9 demonstrating responsiveness to clinically important change, instruments should detect change above the non-specific change incorporated in the MSRM (Deyo et al., 1991). It has been
Google Sites
Report abuse