Analysing hospital variation in health outcome at the level of EQ-5D dimensions – Centre for Health Economics, University of York – January 2012

Posted on February 13, 2012. Filed under: Clin Governance / Risk Mgmt / Quality, Health Economics, Surgery | Tags: |

Analysing hospital variation in health outcome at the level of EQ-5D dimensions – Centre for Health Economics, University of York – January 2012

by Nils Gutacker, Chris Bojke, Silvio Daidone, Nancy Devlin, Andrew Street


“The English Department of Health has introduced routine collection of patient-reported health outcome data for selected surgical procedures (hip and knee replacement, hernia repair, varicose vein surgery) to facilitate patient choice and increase provider accountability. The EQ-5D has been chosen as the preferred generic instrument and the current risk-adjustment methodology is based on the EQ-5D index score to measure variation across hospital providers.

There are two potential problems with this. First, using a population value set to generate the index score may not be appropriate for purposes of provider performance assessment because it introduces an exogenous source of variation and assumes identical preferences for health dimensions among patients. Second, the multimodal distribution of the index score creates statistical problems that are not yet resolved. Analysing variation for each dimension of the EQ-5D dimensions (mobility, self care, usual activities, pain/discomfort, anxiety/depression) seems therefore more appropriate and promising.

For hip replacement surgery, we explore a) the impact of treatment on each EQ-5D dimension b) the extent to which treatment impact varies across providers c) the extent to which treatment impact across EQ-5D dimensions is correlated within providers.

We combine information on pre- and post-operative EQ-5D outcomes with Hospital Episode Statistics for the financial year 2009/10. The overall sample consists of 25k patients with complete pre- and post-operative responses.

We employ multilevel ordered probit models that recognise the hierarchical nature of the data (measurement points nested in patients, which themselves are nested in hospital providers) and the response distributions. The treatment impact is modelled as a random coefficient that varies at hospital-level. We obtain provider-specific Empirical Bayes (EB) estimates of this coefficient. We estimate separate models for each of the five EQ-5D dimensions and analyse correlations of the EB estimates across dimensions.

Our analysis suggests that hospital treatment is indeed associated with improvements in health and that variability in treatment impact is generally more pronounced on the dimensions mobility, usual activity and pain/discomfort than on others. The pairwise correlation between the provider EB estimates is substantial, suggesting a) that certain providers are better in improving health across multiple EQ-5D dimensions than others and b) multivariate models are appropriate and should be further investigated.”


Liked it here?
Why not try sites on the blogroll...

%d bloggers like this: