Using control charts to monitor quality of hospital care with administrative data.

OBJECTIVE Nearly all hospital-specific comparative analyses, based on administrative data, are presented using cross-sectional displays. In this paper, we compare cross-sectional analyses with sequential monitoring using control charts. DESIGN ANALYSIS of administrative data to compare cross-sectional funnel plots with one type of control chart: the risk-adjusted, expected-minus-observed plot. SETTING Eighteen tertiary and base hospitals in Queensland, Australia, for the two financial years 2003-04 and 2004-05. PARTICIPANTS Patients admitted with acute myocardial infarction. MAIN OUTCOME MEASURE Risk-adjusted, 30-day, in-hospital, mortality rates. RESULTS There were no outliers on the cross-sectional funnel plots for either of the 2 years using three-sigma limits and three low-outliers and one high-outlier using two-sigma limits. One reasonable interpretation of these plots is that most of the variations are due to statistical noise and there is little to be learnt by seeking to understand the reasons for variation across hospitals. In contrast, for the control charts, 28% of hospitals signalled for a relative increase of 75% above that for all hospitals combined. CONCLUSION If the aim of clinical indicators based on administrative data is to provide a starting point for learning, then control charting provides potentially more useful information than the more commonly used cross-sectional analyses. Control charts provide an understandable and up-to-date overview that allows early detection of runs of good or bad outcomes that can help hospitals identify areas for more in-depth self-monitoring and learning.