Quality measurement and provider assessment in diabetes care.

摘要

Objectives. To examine how the quality of diabetes care can be better defined and measured, and methods of improving provider quality assessment.;Methods. A secondary data analysis was performed on electronic patient records of diabetic patients in the Department of Veterans Affairs (VA) Maryland Health Care System. Information about patient care from 2006 to 2007, including outpatient provider orders, in-visit care, patient follow-up actions, and patient-specific outcomes, was abstracted from patient records and analyzed with reference to quality-of-care indicators. The existence and dimensionality of a diabetes quality-of-care construct was assessed using tetrachoric correlations, factor analysis, and structural equation modeling. Agreement was evaluated between 4 diabetes quality-of-care indicators based on provider orders and patient follow-up actions: HbA1c testing, LDL-C testing, nephropathy screening, and eye exams. A 2-parameter IRT model was used to assess performance on the quality-of-care indicators.;Results. A single dimension for the quality of diabetes care can be identified that is distinct from other quality-of-care indicators. Factor loadings remained consistent across gender, co-morbid conditions, and high- and low-performing provider groups. Flu Shot and Aspirin Use quality indicators were not associated with diabetes or any other quality indicators. Kappa scores between the order and action quality indicators averaged 0.40 across all 4 indicators, and ranged from kappa=0.74 for LDL-C measurement to kappa=0.15 for Eye Exam. IRT models showed that quality measures based on patient actions successfully differentiate average care quality from low care quality but provide less reliable information about higher-quality care. IRT models constructed from indicators based on provider orders showed high parameter discrimination for the LDL-C and HbA1c indicators and provided the most information about average care quality.;Conclusions. Existing quality indicators measure a discrete aspect of technical diabetes care quality. However, indicators based on provider orders measure care differently than indicators based on patient actions, suggesting inaccuracies in attributing care to providers. IRT modeling could improve the measurement of quality of care. These findings highlight the potential of using EHR-based data to improve quality measurement and provider profiling, and thus improve the data used for incentive-based quality improvement programs.;Funding. This project was supported by grant number F31HS017399 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the author and does not necessarily represent the official views of the Agency for Healthcare Research and Quality.