A Structural Equation Model (SEM) for Pharmacist Competencies in Improving Quality of Life of Cancer Patients: Effect of Missing Values on the SEM

Abstract

Objective: With the goal of improving health-related quality of life (HRQOL) in cancer patients, we previously reported a structural equation model (SEM) of subjected QOL and qualifications of pharmacists, based on a series of questionnaires completed by patients and pharmacists. However, several patients and pharmacists were excluded from the previous study because it was not always possible to obtain all the data intended for collection. In order to reveal the effect of missing data on the SEM, we established SEMs of HRQOL and the competency of pharmacists, using correlation matrices derived by two different statistical methods for handling missing data. Method: Fifteen cancer patients hospitalized for cancer and were receiving opioid analgesics for pain control, and eight pharmacists were enrolled in this study. Each subject was asked four times weekly to answer questions presented in a questionnaire. SEMs were explored using two correlation matrices derived with pair-wise deletion (PD matrix) and list-wise deletion (LD matrix). The final models were statistically evaluated with certain goodness-of-fit criteria. Results: Data were intended to be collected four times weekly for each patient, but there were some missing values. The same SEMs for HRQOL were optimized using both the LD and PD matrices. Although the path diagrams of the SEMs were not identical in the “competency of pharmacists,” the two models suggested that a higher competency of a pharmacist lowered the “severity” of condition and increased the “comfort” of patients, resulting in an increase in the subjected QOL. Conclusion: In collecting data for clinical research, missing values are unavoidable. When the structure of the model was robust enough, the missing data had a minor effect on our SEM of QOL. In QOL research, the LD matrix as well as the PD matrix would be effective, provided the model is sufficiently robust.