UCLA

In quality-of-life research, the Patient’s Assessment of Own Functioning Inventory (PAOFI) was developed by neuropsychologists to reflect capabilities in memory, language and communication, higher-level cognition, and sensorimotor functioning. A statistical perspective on applied research using the PAOFI raises questions about how ordinal item scores have been dichotomized into binary outcomes given what is known about the loss of information from dichotomizing continuously-scaled measures.

Drawing on a sample of breast-cancer survivors in a study of how breast-cancer treatment affected quality of life, and using an information-theory-based aggregate measure of entropy obtained by summing contributions of individual items, the score based on dichotomized PAOFI items was about 70% lower than using ordinal PAOFI items. Furthermore, investigation of PAOFI domain scores across breast cancer treatment groups revealed sensitivity of inferences to dichotomization cut-points, suggesting that avoiding dichotomization and analyzing PAOFI scores on the original ordinal scale might be preferable.

Previous investigation of the PAOFI on a diverse sample of individuals identified 4 prominent factors; however, in the breast-cancer-survivor sample, a 5-factor solution provided a more natural interpretation. Analyses of domain scores across breast cancer treatment groups revealed significant differences in factor scores with some degree of sensitivity to whether the factor analysis used ordinal or dichotomized items. After using item response theory to select 2 items per domain, it was still possible to detect significant differences in domain scores across breast cancer treatment groups.

To reflect the full range of associations between background characteristics and quality-of-life domains, we implemented 3-part models for PAOFI outcomes, modeling (1) an indicator for any serious problem versus no serious problem within a given domain, (2) the number of problems experienced by a patient on a given domain, and (3) either the average severity or the aggregate impact of problems experienced on a given domain. Using the "Memory: Absent-Mindedness" domain for illustration, we found diverse sets of significant predictors of the different outcomes. Overall, the dissertation reveals a number of ways to improve upon previous statistical analysis approaches that dichotomized ordinal PAOFI items to enhance understanding of breast-cancer quality of life.