ABSTRACT OF THE DISSERTATION
A Psychometric Analysis of Patient-Reported Outcomes in Chronic Kidney Disease
by
John Devin Peipert
Doctor of Philosophy in Public Health
University of California, Los Angeles, 2017
Professor Donald Morisky, Chair
Background: Survival is a critical outcome in chronic kidney disease (CKD), but it provides a limited view of how well patients are doing. Many aspects of patients’ health can only be obtained by patient reported measures (PRMs), such as health-related quality of life (HRQOL). This dissertation examines the psychometric properties of currently used PRMs in chronic kidney disease (CKD) and make recommendations for how collection and reporting of PRMs can be systematized in the CKD field.
Methods: This dissertation used data from three separate sources, each containing kidney patients’ responses to PRMs: treatment decision-making, medication adherence, and HRQOL. The treatment decision-making PRMs examined include patients’ Decisional Balance (perceived Pros and Cons) and Self-Efficacy to pursue both living and deceased donor kidney transplant (LDKT, DDKT; 6 measures in total). The 8-item Morisky Medication Adherence Scale (MMAS-8) was examined as a measure of medication adherence. Finally, the Kidney Disease Quality of Life (KDQOL)-36 was examined as a measure of HRQOL. For each measure, internal consistency reliability was estimated. Dimensional structure was examined with exploratory and confirmatory factor analysis (EFA, CFA) and item-scale correlations, corrected for item overlap. To determine the dimensionality suggested by the exploratory factor analysis, several criteria were used, including the scree “elbow” test, parallel analysis, and the Tucker-Lewis reliability coefficient. For CFA models, model fit was determined with the Satorra-Bentler chi-square, the comparative fit index (CFI), Tucker-Lewis Index (TLI) and root mean square error of approximation (RMSEA). Good model fit is evidenced by a non-significant Satorra-Bentler chi-square, CFI and TLI values of above 0.95, and RMSEA of 0.06 or less. Next, the measurement invariance for each measure was examined between Black and White patients. Finally, recommendations for improvements to each measure were made, including calculations to determine the number of items needed to achieve good (>0.80) and excellent (>0.90) reliability (reliability of >0.70 is considered adequate). Generally, excellent reliability is required for use with individuals.
Summary of Results: For both the LDKT and DDKT Pros and Cons measures, 2 correlated factors CFA models fit the data well, verifying the original dimensional structures for these measures. Additionally, the LDKT and DDKT Self-Efficacy measures were supported by single factor CFA models, also supporting their original dimensional structure. However, several of these scales only evidenced adequate internal consistency reliability (LDKT Pros, LDKT Cons, DDKT Cons). These scales would require the addition of up to 15-17 parallel items to achieve excellent reliability (>0.90). Regarding the MMAS-8, the original factor structure was not supported by EFA and CFA models. One item, “Did you take your medicine yesterday?,” was weakly correlated (r<0.243) with the others, and did not load highly (λ<0.40) in EFA models. A CFA model with this item removed fit the data well (RMSEA = 0.06; CFI = 0.99; TLI = 0.98). The internal consistency reliability of the modified scale was 0.78, and 18 items would need to be added to achieve excellent reliability. Finally, regarding the KDQOL-36, the original factor structure was supported, and the internal consistency reliability for each KDQOL-36 scale exceeded the criterion for “good” reliability (>0.80), though the addition of up to 11 items would be required to increase reliability to “excellent” (Symptoms/Problems scale). For all scales in the dissertation, recommendations were made for increasing reliability using classical test theory and item response theory.
