The reliability of revision rates following primary shoulder arthroplasty as a quality indicator to rank hospital performance: a national registry analysis including 13,104 shoulders and 87 hospitals.

Background: To assess the extent of between-hospital variation in revision following primary shoulder arthroplasty (SA), both overall and for specific revision indications to guide quality improvement initiatives, and to assess whether revision rates are suitable as quality indicators to reliably rank hospital performance.

Methods: All primary SAs performed between 2014 and 2018 were included from the Dutch Arthroplasty Register to examine 1-year revision and all primary SAs performed between 2014 and 2016 for 1- and 3-year revisions. For each hospital, the observed number (O) of revisions was compared with that expected (E) based on case-mix and depicted in funnel plots with 95% control limits to identify outlier hospitals. The rankability (ie, the reliability of ranking hospitals) was calculated as the percentage of total hospital variation due to true between-hospital differences rather than chance and categorized as low (<50%), moderate (50%-75%), and high (>75%).

Results: A total of 13,104 primary SAs (87 hospitals) in 2014-2018 were included, of which 7213 were performed between 2014 and 2016. Considerable between-hospital variation was found in 1-year revision in 2014-2016 (median 1.6%, interquartile range 0.0%-3.1%), identifying 3 outlier hospitals having overall significantly more revisions than expected (O/E range 1.9-2.3) and for specific indications (cuff pathology and infection). Results for 2014-2018 were similar. For 3-year revision, 3 outlier hospitals were identified (O/E range 1.7-3.3). Rankabilities for all outcomes were low.

Conclusions: Considerable between-hospital variation was observed for 1- and 3-year revision rates following primary SA, where outlier hospitals could be identified based on large differences in revision for specific indications to direct quality improvement initiatives. However, rankabilities were low, meaning that much of the other (smaller) variation in performance could not be detected, rendering revisions unsuitable to rank hospital performances following primary SA.