Operating Characteristics of a Performance Index for Benchmarking Hospital Performance: Illustrated on Riksstroke, the Swedish Stroke Register

• Activity Number: 504
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Anita Lindmark*+ and Bart van Rompaye and Els Goetghebeur and Eva-Lotta Glader and Marie Eriksson
• Companies: Umeå University and USBE/Umeå University and Ghent University and Umeå University and Umeå University
• Keywords: benchmarking ; performance index ; dichotomous outcome ; hospital performance ; stroke

Abstract

We describe operating characteristics of a performance index for benchmarking hospitals in terms of a dichotomous outcome. The performance measure is based on direct standardization, i.e. what the (average) outcome for the entire population would have been, had all patients received the level of care of one particular hospital.

The estimate involves predictions from logistic regression models with fixed hospital effects, and patient-specific baseline covariates for case-mix adjustment. The index considers the probability that the performance measure for a given hospital exceeds the current population risk by more than a tolerated percentage. If this probability exceeds a threshold, the hospital is labelled as having deviating performance.

The effect of the tolerated risk increase, and the level of evidence demanded (or probability threshold)on sensitivity and specificity is illustrated through simulations based on data from Riksstroke, the Swedish stroke register. The outcome used is dead or dependent in activitied of daily living (ADL) 3 months after stroke.