Geographic Variation in Quality of Care for Commercially Insured Patients

Background: There has been little work assessing geographic variation in quality for a national sample of commercially insured enrollees.  With the continued push of payers towards tying reimbursement to claims-based quality measures, it is increasingly important to understand the relationship between different quality measures and their validity. 

Objective:To evaluate the magnitude of claims-based quality measures among the commercially insured, their association to one another, and their comparability to similar measures calculated for the Medicare population.

Data Source: Physician and hospital claims for 30 million privately insured employees and their dependents from the Truven Health MarketScan Commercial Claims and Encounters Database from 2007 to 2009 and publically available Medicare data compiled by the Dartmouth Atlas. 

Methods: We identified 10 commonly used quality measures captured well in claims including 7 HEDIS measures, 2 Prevention Quality Indicator indices, and 30-day hospital readmissions.    We calculated the rates of each quality measure across the 306 hospital referral regions.  HEDIS measures were not adjusted for patient level characteristics because they are based on an already restricted cohort, while Prevention Quality Indicator indices and 30-day readmission rates were adjusted for age, sex, and prior year health status.  We calculated summary statistics for each quality measure across the HRRs and ran pair-wise correlations to assess their associations.  To account for measurement error in smaller markets, we compute Empirical Bayes shrinkage estimates weighting HRR level estimates with the overall mean according to the ratio of within and between market variation.  We used publically available Medicare data compiled by the Dartmouth Atlas to compare commercial and Medicare estimates for three measures.

Results: The quality of care for commercially insured patients varies across the 306 HRRs, although the extent of variation depends on the measure, with the top quartile HRRs have 1.45 higher rates of acute PQIs than the lowest quartile.   However, both the level of compliance and the magnitude of variation differ substantially from measure to measure.  The average compliance on DMARD treatment for rheumatoid arthritis is 89% whereas depression treatment is 64%.   There is substantially more variation in the two PQI composites than among HEDIS measures.  HEDIS measures are somewhat more likely to be modestly correlated or uncorrelated with one another than to be negatively correlated.  We find a positive correlation between Medicare and commercial populations for 30-day readmission rates and mammography.

Conclusion: There is substantial variation in quality of care provided to the commercially insured population, although the measures are not particularly well correlated with one another.  While it is possible these measures reflect different dimensions of quality, other literature questions the validity of claims-based quality indicators.  There continues to be a push towards tying reimbursement to quality, much of which is derived from insurance claims.  However, our findings, combined with the substantial research on the Medicare population, suggesting policy makers and researchers need to assess the validity of current indicators and perhaps develop new measures that accurately reflect the underlying quality of care being provided to patients.