164-2010: Using the GLIMMIX Procedure to Model Hospital Quality Measured by CMS: Comparing City-Owned Hospitals, Other Safety Net Hospitals, and Hospitals for the Well-Insured

Health care quality, costs, and methods to control them are debated. Medicare (CMS) measures quality. We studied New York City hospitals, divided into three groups by funding source. PROC GLIMIX enabled us to study proportions of compliance of hospitals nested within funding type, with repeated measures over time. Public hospitals performed better on the CMS measures than the other safety net hospitals and, generally, better than those hospitals that are funded mostly by private insurance. INTRODUCTION The appropriate role of government in US health care has been a subject of considerable ongoing debate. Proponents argue that government, unlike private financial interests, lacks an inherent conflict of interest with patients. Opponents argue that government is inherently an inefficient, expensive way to provide mediocre health care. These opponents argue that government employees lack financial incentive to excel so that government programs can not achieve high quality. One role the federal government has recently assumed is to measure and publicly report quality of health care. In addition to the public reporting, CMS plans to tie future reimbursement levels to these quality measures: pay for performance, (P4P). In theory, P4P would apply capitalistic incentives to medical care, incentivizing improved performance. Werner, in a recent JAMA article, suggested that the federal program may have an unintended, perverse effect: reducing funding to safety net hospitals which care for a disproportionate share of poor patients. She proposed that, when P4P is implemented, it may have the unintended consequence of worsening disparity of care: safety net hospitals would enter a downward spiral of low scoreslower reimbursementfewer resources for improvementminimal improvementstill lower relative scores. Since most safety net hospitals are disproportionately government funded, the P4P program itself might be an example of inefficient government programs working at cross purposes. We wanted to see if our government operated safety net hospitals: 1) were at risk at entering the downward spiral proposed by Werner 2) performed at the same level as safety net hospitals which are not run by governmental agencies. The role of government in safety net hospitals varies greatly. Involvement may include actually running the hospitals as it does for the largest two government operated health care systems are the VA system, run by the federal government, and New York City Health and Hospitals Corporation (HHC), which is owned and operated by New York City. There are differences between states in terms of rates of Medicaid reimbursement, rules about what services and who should be covered, labor rates, union rules, and rules governing “obs units.” To eliminate the effect of these variables on quality outcome measures, we limit our study to one city, New York. Within that city, these rates and regulations are constrant. We divided the 53 acute care hospitals in New York City into 3 groups by safety net status: government run safety net hospitals (HHC, 11 hospitals), other safety net hospitals (OSN), and non-safety net hospitals (NSN) which care for a proportionately large number of well insured patients (NSN). To define safety net status, we chose an arbitrary cutoff of total hospital income of at least 30% Medicaid. After limiting the study to New York and dividing hospitals into groups of similar hospitals, there is still unwanted variability. Within each group, there are some consistent differences between individual hospitals (fig 1). For CQI purposes, the differences between hospitals are interesting, but they are not the focus of this paper. We are looking for differences between the three payment groups. For the hypotheses being considered here, differences between hospitals are nuisance variables. Time trends are interesting for two reasons: 1) an improvement over time bolsters the AHRQ hypothesis that collecting and publicly reporting data will spur improvement and 2) if we find a group*time interaction, with non-safety net hospitals improving at a relatively fast rate, that would support Werner’s concern; in this case, that public reporting alone was having the adverse effect she predicted. Until P4P is implemented, we can not fully test Werner’s hypothesis. Healthcare Providers & Insurers SAS Global Forum 2010