The Relationship Between Costs and Quality in Veterans Health Administration Nursing Homes

Summary: The complex variations in interactions between cost and quality in health care systems require careful empirical analysis to better understand the nature and determinants of relationships. This paper addresses the cost-quality relationship in the context of nursing homes in the U.S. Department of Veterans Affairs (VA) health care system, where there are significant patient severity differences but the systems of care delivery to patients vary less than in other health care contexts, such as hospitals. There are two types of cost-related phenomena associated with improved quality: one, additional staff and cost-increasing redesigned processes that may be necessary to improve quality; and two, costs that are saved by avoiding the additional needs of residents who experience one or more of the quality indicator events that VA uses to evaluate their nursing homes. Robust to a number of sensitivity tests, we find that the second phenomenon predominates, that is, lower quality is associated with higher costs.

Empirical Approach: In this study, we use preliminary 2005 to 2007 data to examine the relationship between costs and quality in 112 nursing homes, called community living centers by the VA. We are extending this data through 2012 so that we can perform better longitudinal cost/quality comparisons. A set of 24 quality indicators (QIs), which measures unfavorable events (e.g. falls) or patient states (e.g., depression), has been developed from the Minimum Data Set (MDS) to measure the quality of nursing home care. Four of the QIs are stratified into high and low risk categories, resulting in a total of 28 QIs. We used a multivariate normal distribution to model these rates with a 28 dimension multivariate normal distribution with mean vector γ and covariance matrix Σ. We specified a non-informative multivariate normal prior for the mean vector γ and used a Wishart distribution for the inverse of the covariance matrix T = Σ-1 WinBUGs. The data consisted of the following variables for each of the 112 facilities for each of the 3 fiscal years in our preliminary analysis: total patient days in the year, total costs during the year, quality score based on the last MDS assessment done in the year and Resource Utilization Groups (RUGs) score, calculated as “number of days in each RUGs category during the year times the RUG category relative weight” divided by the “total number of days”. In order to adjust for facility clustering over time, we used generalized estimating equations (GEE) with an exchangeable correlation matrix. We considered two link functions: a linear link with a normal error and a log link with a gamma error.

Preliminary Results: Robust to a number of sensitivity tests, we find that increases in the MDS-based quality score (interpreted as lower quality) are associated with higher costs. Holding everything else constant, a 0.022 increase in quality score (about 1 SD, and interpreted as a decrease in quality since, as noted, these are adverse events) is associated with a 6.56% increase in cost.