Developing a Comorbidity Index for Health-related Quality of Life Studies

University Of Michigan At Ann Arbor, Ann Arbor MI

Investigators

Linked publications & trials

Abstract

[unreadable] DESCRIPTION (provided by applicant): The overall goal of this study is to derive a co-morbidity index that includes diseases that have the greatest association with health-related quality of life (HRQL) as measured by the SF-12, a general HRQL instrument, in the MEPS dataset. The secondary goal is to validate the new index using data with a unique set of patients with the same measures. Another secondary goal is to compare the performance of the new HRQL co-morbidity index with the mortality-based co-morbidity index, the Charlson Index. This study will use the 2003 full year dataset of the MEPS to develop the HRQL-specific co-morbidity index using the SF-12 PCS (physical-related HRQL) and MCS (mental health related HRQL) as the HRQL dependent variable. The index will be validated using the 2005 full-year dataset within the MEPS. Further validation testing of the new index will be conducted using the same procedures using the Charlson Co-morbidity index. The initial step in index development will be to identify respondents who themselves completed the SF-12 and who were 18 years of age or older. Using a unique respondent identifier variable in MEPS, the Clinical Classification Codes (CCCs) within the MEPS HC-078 Conditions File that meet the following criteria will be merged with the Consolidated Full Year data file. CCCs that will be excluded include conditions that are gender specific, acute trauma, dementia or mental retardation, and administrative codes. A relatively new statistical model selection technique will be used to search through the model space for important models and predictors, not just singling out one specific model. These include the LASSO (Least absolute shrinkage and selection operator) and CART (classification and regression trees) techniques. These techniques will be applied separately for the SF-12 PCS and MCS. Once the top prediction models are identified for the SF-12 PCS and MCS HRQL scores, we will assess them via several selection criteria such as adjusted R-square, Mallow's Cp, Akaike Information Criteria (AIC), and Bayes Information Criteria (BIC) to arrive at the final selected model. [unreadable] [unreadable] [unreadable] [unreadable]