PROJECT ABSTRACT Measurement of hospital performance through risk-adjusted mortality rates is a cornerstone of the United States health system, factoring heavily into value-based purchasing and other strategies to incent quality improvement. Although risk-adjusted mortality captures one of the key functions of a hospital – its ability to prevent death – it makes no distinction between deaths that result from suboptimal care and those that result from informed decisions to forego life-sustaining treatment and focus on palliation in anticipation of death. This results in two critical problems. First, by rewarding hospitals for fewer deaths, these measures create an incentive to prolong life with aggressive interventions when palliation in anticipation of death may better align with patient preferences. Second, by failing to distinguish between different types of deaths, these measures do not provide insight into the specific targets on which a hospital should focus its quality improvement efforts, since the mechanism underlying the high mortality rate is not known. Ideally, a mortality-based hospital performance measure would simultaneously (1) reward reductions in deaths that result from suboptimal disease-directed care or medical errors, (2) reward increases in high-quality palliation in anticipation of death when consistent with patient preferences, and (3) provide hospitals with actionable information for quality improvement by offering insight into how patients are dying. A critical barrier to the development of such a measure is our current inability to distinguish different types of deaths. Missing is a detailed epidemiology of the typologies of hospital deaths, i.e., “how” hospitalized patients are dying. The goal of this project is to identify typologies of hospital deaths that can inform new strategies in hospital performance measurement to improve multidimensional quality of hospital care. We will study patients hospitalized with chronic obstructive pulmonary disease, acute myocardial infarction, heart failure, pneumonia, and cerebrovascular accident because they are the subject of public mortality reporting and pay-for-performance initiatives and represent acute and chronic diseases for which in-hospital decisions regarding end-of-life care may be different. We will focus particular attention on patients with dementia who may be at disproportionate risk for both under- treatment (i.e., failure to rescue due to provider bias) and over-treatment (i.e., rescue when quality of life is already below the patient’s minimal acceptable function). We will use innovative data mining techniques to identify key clinical variables through mixed-methods analysis of the decision-making processes of institutional mortality reviewers. Findings from this R21 will be used to support an R01 application designed to externally validate the typologies with the ultimate goal of developing innovative composite performance measures that can simultaneously reward ...