Huang-Tz Ou1, Chung-Ying Lin2, Steven R Erickson3, Rajesh Balkrishnan4. 1. Institute of Clinical Pharmacy and Pharmaceutical Sciences, College of Medicine, National Cheng Kung University, Tainan, Taiwan. huangtz@mail.ncku.edu.tw. 2. Department of Rehabilitation Sciences, Faculty of Health and Social Sciences, The Hong Kong Polytechnic University, Hung Hom, Hong Kong. 3. Department of Clinical Pharmacy, College of Pharmacy, University of Michigan, Ann Arbor, MI, USA. 4. Department of Public Health Sciences, School of Medicine, University of Virginia, Charlottesville, VA, USA.
Abstract
PURPOSE: To refine two subscales of the health-related quality of life comorbidity index (HRQoL-CI) into a single index measure. METHODS: The 2010 and 2012 Medical Expenditure Panel Surveys were utilized as development and validation datasets, respectively. The least absolute shrinkage and selection operator was applied to select important comorbidity candidates associated with HRQoL. Exploratory factor analysis and confirmatory factor analysis (CFA) were used to assess dimensionality in comorbidity. Statistical weights were derived based on standardized factor loadings from CFA and regression coefficients from the model predicting HRQoL. Prediction errors and model R(2) values were compared between HRQoL-CI and Charlson CI (CCI). RESULTS: Eighteen comorbid conditions were identified. CFA models indicated that the second-order multidimensional comorbidity structure had a better fit to the data than did the first-order unidimensional structure. The predictive performance of the refined scale under a multidimensional structure utilizing statistical weights outperformed the original scale and CCI in terms of average prediction error and R(2) in the prediction models (R(2) values from refined scale model are 0.25, 0.30, and 0.28 versus those from CCI of 0.10, 0.09, and 0.06 for general health, SF-6D, and EQ-5D, respectively). CONCLUSION: The dimensionality of comorbidity and the weight scheme significantly improved the performance of the refined HRQoL-CI. The refined single HRQoL-CI measure appears to be an appropriate and valid instrument specific for risk adjustment in studies of HRQoL. Future research that validates the refined scales for different cultures, age groups, and healthcare settings is warranted.
PURPOSE: To refine two subscales of the health-related quality of life comorbidity index (HRQoL-CI) into a single index measure. METHODS: The 2010 and 2012 Medical Expenditure Panel Surveys were utilized as development and validation datasets, respectively. The least absolute shrinkage and selection operator was applied to select important comorbidity candidates associated with HRQoL. Exploratory factor analysis and confirmatory factor analysis (CFA) were used to assess dimensionality in comorbidity. Statistical weights were derived based on standardized factor loadings from CFA and regression coefficients from the model predicting HRQoL. Prediction errors and model R(2) values were compared between HRQoL-CI and Charlson CI (CCI). RESULTS: Eighteen comorbid conditions were identified. CFA models indicated that the second-order multidimensional comorbidity structure had a better fit to the data than did the first-order unidimensional structure. The predictive performance of the refined scale under a multidimensional structure utilizing statistical weights outperformed the original scale and CCI in terms of average prediction error and R(2) in the prediction models (R(2) values from refined scale model are 0.25, 0.30, and 0.28 versus those from CCI of 0.10, 0.09, and 0.06 for general health, SF-6D, and EQ-5D, respectively). CONCLUSION: The dimensionality of comorbidity and the weight scheme significantly improved the performance of the refined HRQoL-CI. The refined single HRQoL-CI measure appears to be an appropriate and valid instrument specific for risk adjustment in studies of HRQoL. Future research that validates the refined scales for different cultures, age groups, and healthcare settings is warranted.
Entities:
Keywords:
Comorbidity index; Dimensionality; Factor analysis; Health-related Quality of life; Least absolute shrinkage and selection operator; Risk prediction model