TY - BOOK

T1 - Essays on the health economics of hospital quality

AU - Murphy, Shane

PY - 2016/11

Y1 - 2016/11

N2 - This thesis consists of three essays on hospital quality of inpatient care for patients with acute myocardial infarction (AMI) in the United States. First, it explores issues in the measurement of quality, particularly through the estimation of risk-adjusted mortality rates (RAMRs) for hospitals. This work then examines the relationship between hospital quality for AMI patients and the volume of AMI patients.Chapter 2 proposes using machine-learning techniques, particularly random forests, for risk adjustment of patient severity to predict patient mortality. This work shows that these methods greatly outperform other commonly-used methods in precision of patient risk estimates and also that a facility’s estimated RAMR is sensitive to the underlying patient risk-adjustment model.Chapter 3 asks whether a model which aggregates patient mortality risk for AMI patients matters when estimating RAMRs. To do this, it creates a simulation based on realistic assumptions about how patient case mix can vary by hospital quality and how hospital quality can vary by hospital volume. Because different methods of estimating patient mortality risk have different degrees of precision, the simulation considers variation in this precision and further allows precision to vary by hospital. Again, the ranking of hospitals is sensitive to the method used and this paper finds that common methods are not preferred in many important contexts. Both of the first two chapters pay particular importance to applications of their results to pay-for-performance schemes.Chapter 4 examines the relationship between quality, measured by RAMR, and volume in hospital health provision for AMI inpatients. The main contribution of the paper is estimate the causal effect of volume on quality. To do this, it uses a novel instrument, the volume of shock and of trauma patients. Previous work has found mixed results and has primarily used the volume of patients with the same condition within a certain radius of the hospital as an instrument for volume within the hospital. This paper argues that this instrument has a number of shortcomings that its instrument does not. This paper tests various specifications used in other work and finds robust results for its conclusion.

AB - This thesis consists of three essays on hospital quality of inpatient care for patients with acute myocardial infarction (AMI) in the United States. First, it explores issues in the measurement of quality, particularly through the estimation of risk-adjusted mortality rates (RAMRs) for hospitals. This work then examines the relationship between hospital quality for AMI patients and the volume of AMI patients.Chapter 2 proposes using machine-learning techniques, particularly random forests, for risk adjustment of patient severity to predict patient mortality. This work shows that these methods greatly outperform other commonly-used methods in precision of patient risk estimates and also that a facility’s estimated RAMR is sensitive to the underlying patient risk-adjustment model.Chapter 3 asks whether a model which aggregates patient mortality risk for AMI patients matters when estimating RAMRs. To do this, it creates a simulation based on realistic assumptions about how patient case mix can vary by hospital quality and how hospital quality can vary by hospital volume. Because different methods of estimating patient mortality risk have different degrees of precision, the simulation considers variation in this precision and further allows precision to vary by hospital. Again, the ranking of hospitals is sensitive to the method used and this paper finds that common methods are not preferred in many important contexts. Both of the first two chapters pay particular importance to applications of their results to pay-for-performance schemes.Chapter 4 examines the relationship between quality, measured by RAMR, and volume in hospital health provision for AMI inpatients. The main contribution of the paper is estimate the causal effect of volume on quality. To do this, it uses a novel instrument, the volume of shock and of trauma patients. Previous work has found mixed results and has primarily used the volume of patients with the same condition within a certain radius of the hospital as an instrument for volume within the hospital. This paper argues that this instrument has a number of shortcomings that its instrument does not. This paper tests various specifications used in other work and finds robust results for its conclusion.

M3 - Doctoral Thesis

PB - Lancaster University

ER -