Learning Objective 1: To understand the implications of sample size and risk adjustment for ranking methods
Learning Objective 2: To identify the impact of different ranking methods on a ranking of acute care trusts in inpatient cancer care
Methods: Seven items relevant to inpatient nursing cancer care from the National Cancer Patient Experience Survey of 158 acute care trusts in England for 2010 (n=24,687) and 2011 (n=23,857) were considered. Items were assessed with regard to their ability to differentiate between providers by intraclass correlations (ICCs) and rankability estimates. Acute care trusts were ranked by their mean, the risk adjusted standardised rate, expected ranks and empiricial bayes estimates. The association between rankings using different methods and between years were assessed by Kendall’s Tau correlation coefficients.
Results: For 2010 ICCs ranged from .007 to 0.030 indicating weak levels of clustering in the data. Rankability ranged from 39% to 69% confirming this picture. Kendall’s Tau coefficients for associations between ranking methods ranged between .79 and .98, indicating strong associations between all ranking mechanisms. The risk and sample size adjusted empirical bayes and expected rank mechanisms correlated highest (Tau= .98) while standardized rates (.88 and .89) and mean-based rankings (.78 and .79) were less strongly correlated. Rankings from 2010 and 2011 correlated only modestly (.30 to 0.34).
Conclusion: Results show strong association between all ranking methods but far from complete agreement. Modest agreement between years raises questions about the extent to which the measures are evaluating a stable characteristic of the trusts.
See more of: Research Sessions: Oral Paper & Posters