Monir Hossain
School of Public Health
Department of Epidemiology and Biostatistics
University of South Carolina
Hierarchical Generalised Linear Models with Time-dependent Clustering:
Assessing the Effect of Health Sector Reform on Patient Outcomes in New
Zealand
New Zealand has one of the most reformed health systems in the world. This
talk is primarily concerned with modelling the impact on hospital outcomes
of the reforms of the early 1990s, when as part of a major, health sector
wide reform process, the administration of public hospitals passed from
elected Area Health Boards (AHBs) to Crown Health Enterprises (CHEs)
operating under a competitive model of health care provision dominated by
the funder/purchaser/provider split. The impact of reform processes on
public hospitals is of particular interest since they consume 40%-50% of
public expenditure on health, and have been repeatedly restructured in an
attempt to contain the ever-expanding cost of health care. There is concern
among both health professionals and the general public that these
restructurings are reducing the quality of hospital services, and therefore
negatively effecting patient outcomes. Using data from a study of 34 New
Zealand public hospitals, we discuss the application of Bayesian
hierarchical generalised linear models to the analysis of trends in patient
outcomes over the period 1988-2001. The time-varying nature of the grouping
of hospitals within larger health authorities complicates the application of
HGLMs because the cluster structure of the data changes over the study
period. An approach to dealing with such 'time-dependent clustering' by
introducing period-specific authority level effects is developed. The
analysis does not support the proposition that higher level authorities had
an effect on outcome trends, or that the administrative changeover from AHBs
to CHEs impacted on 60-day post-admission mortality.
At the end of this talk, we would also look to other outcome measures,
length-of-stay and readmission.
Back to Colloquium Series