Embedded Files
suggested that statistical measures of responsiveness are an insufficient basis for assessing responsiveness and that patients’ views on the importance of the change should inform testing (Liang et al., 2002; Terwee et al., 2003). Anchor-based approaches assess the relationship between changes in instrument scores and an external variable (Norman et al., 2001). This includes health transition items or global judgements of change used to estimate the Minimal Important Difference (MID), the instrument change score corresponding to a small but important change (Jaeschke et al., 1989; Juniper et al., 2002). The MID can inform sample size calculations but consideration must be given to specific groups of patients and specific settings (Terwee et al., 2003). Score interpretation may be improved through the provision of evidence relating to score variation (Terwee et al., 2003) or a score range against which real change may be assessed (Streiner & Norman, 2008; Beaton et al., 2001). External variables including transition ratings have also been compared to instrument score changes using correlation. This form of longitudinal validity (Kirshner and Guyatt, 1985; Terwee et al., 2003) assesses the extent to which changes in instrument scores concord with an accepted measure of change in patient health (Deyo et al., 1991; Fitzpatrick et al., 1998). Precision refers to the ability of an instrument to distinguish clearly and precisely between respondents in relation to reported health or illness (Fitzpatrick et al., 1998). Ideally, items within an instrument should capture the full range of health states to be measured, supporting discrimination between respondents at clinically important levels of health (Fitzpatrick et al., 1998). Precision is influenced by several factors including response categories and item coverage of the defined concept of health purportedly measured by the instrument. Limited response categories lack precision and detail, whereas increased gradations of response increase measurement precision (Streiner & Norman, 2008; Fitzpatrick et al., 1998). Modern psychometric methods, including Rasch analysis, are also used to assess item distribution. Where there is an uneven distribution of items across the proposed hierarchy of health, for example, item grouping in the middle range of functional ability, score change may be influenced by baseline scores and should be considered when interpreting changes in health. Item content and response format will inevitably influence data quality and scaling, in which floor and ceiling effects are key features. Where more than 20% of responders score at the maximum level of good or bad health, score distribution generally suggests ceiling or floor effects, respectively (Streiner & Norman, 2008; Fitzpatrick et al., 1998). The greater concern is for respondents with already poor health who score at the floor of the instrument range and are consequently unable to report further deterioration in health. 10 Evidence suggests that floor effects are more common with instrument completion by older, sick, or disadvantaged respondents (McHorney, 1996). Acceptability addresses the willingness or ability of patients’ to complete an instrument (Fitzpatrick et al., 1998). Although difficult to evaluate directly, this is most readily assessed through instrument completion, response rates, and missing values. Where items within an instrument are consistently omitted, or difficulty is encountered in providing an answer, perhaps due to perceived irrelevance, this would suggest poor acceptability (McHorney, 1996). The font style and size used in questionnaires may also influence completion. Ideally, patients’ should be interviewed for their views on instrument completion, content relevance and format during the pre-testing stage of instrument development (Fitzpatrick et al., 1998). Reading ability is a further consideration regarding instrument acceptability (Streiner & Norman, 2008). A reading level equivalent to that of a 12 year-old has been recommended for questionnaires applicable to the general population (Streiner & Norman, 2008). However, many instruments, including the widely used Nottingham Health Profile (NHP) and the SF-36 have higher reading level requirements (McHorney, 1996; Sharples et al., 2000). It must also be remembered that reading ability may decrease with age (McHorney, 1996). Lack of familiarity with a questionnaire may further reduce response rates in older people (McHorney, 1996). Instrument completion will also be influenced by mode of administration. Although cheaper than interview or telephone administration, postal administration often results in higher levels of missing values (McHorney, 1996; McColl et al., 2001). Evidence suggests that respondents are more willing to report less favourable health states when completing an instrument themselves than when the instrument is administered by interview (Fitzpatrick et al., 1998; Smeeth et al., 2001). Furthermore, response rates may be influenced by specific item content, for example, items relating to physical or emotional issues; the associated item relevance and appropriateness to the specific population (Bowling, 2005); and response formats, for example, visual analogue scales or Likert scaling (Fitzpatrick et al., 1998). The burden imposed by instrument length and time needed for completion is an important consideration for both
Google Sites
Report abuse