Abstract
Abstract: Role functioning and its limitations due to one's health is an important aspect of health-related quality of life (HRQoL). The Medical Outcomes Study (MOS) SF-36 includes 2 role functioning scales: role limitations due to physical health problems (RP) or emotional problems (RE). Although they capture important concepts of HRQoL, these 2 scales have some limitations in their measurement properties. Using dichotomized sets of response choices, the scales are limited in their distributional properties (eg, higher standard deviation than other SF-36 scales) and ability to discriminate between clinically relevant groups. In this study, we ascertain the improvements to these 2 scales using 5-point ordinal response choices for each of the scale items. Two thousand one hundred sixty-two patients from the Veterans Health Study (VHS), an observational study of health outcomes in patients receiving ambulatory care, completed a health status questionnaire and a medical history. The health questionnaire included (1) the MOS SF-36, in which the RP and RE items used dichotomized yes/no responses; and (2) a set of modified RP and RE items that used 5-response choices for each of the items, ranging from "no, none of the time" to "yes, all of the time." We compared the original and modified RP and RE scales using internal consistency reliability and factor analysis. We tested item convergent and discriminant validity using multitrait scaling, and scale discriminant validity using ordinary least squares regression. Results indicate that the modifications to the original RP and RE scales accomplish important gains in the distributional properties of the scales. The floor and ceiling effects of the 2 scales have been reduced and the reliability of the RP scale has increased (0.87-0.95). Factor analysis and multitrait scaling tests indicate that the modified items have the same interpretation as the original items. Tests of discriminant validity indicate that the modified RP and RE scales have greater explanatory power for measures of disease burden, depression, and disease severity. The modified SF-36 role scales are clearly superior to the original versions. The modifications have increased the explained variability, suggesting greater explanatory power and more information obtained by the role functioning measures. The modified RP and RE are capturing a wider spectrum of disease severity, in part due to the lowering of the floor and raising of the ceiling of the scales. Additional work needs to test these improvements in other populations and to expand the analysis to track the responsiveness of the modified scales to clinically and socially important changes over time.