There were 3 primary outcomes: (1) postoperative daily opioid
dose: we used log-transformed postoperative daily opioid
dose for modeling purposes; (2) postoperative pain scores; and
(3) pain-related patient satisfaction score. To analyze the trend
in postoperative daily opioid dose and pain score, we built two
separate linear regression models to assess the association between
year of surgery and the 2 outcome variables. We used
log-transformed daily opioid dose as an outcome to meet the
model assumption of normality. Because the percentage of “topbox”
responses on pain-related HCAHPS questions, pain-related
patient satisfaction score only had 3 possible outcomes (0%,
50%, or 100%), the association between year of surgery and the
pain-related patient satisfaction score was evaluated using a proportional
odds model. All models were adjusted for the prespecified
demographic, baseline and perioperative variables listed inTable 1.
The predictor variable of interest, year of surgery, was treated
as a continuous variable because we assumed that there was a
trend in the primary outcomes. Bonferroni correction was used
to adjust for multiple comparisons. Thus, 98.3% confidence
intervals (CIs) were reported, and significance criterion was
Pb .017 (ie, .05/3).
In addition, we separately plotted the three outcome variables
on the aggregate level (average across patients) over time
for information purposes.