Page 99 - Read Online

Hawkins et al. Vessel Plus 2022;6:42 https://dx.doi.org/10.20517/2574-1209.2021.116 Page 3 of 11

Data use and business associate agreements are in place with all members, VCSQI and the database vendor
(ARMUS Corporation, Foster City, CA). The primary objective of the VCSQI is quality improvement,
including prior work on the prevention of POAF. As this analysis represents a secondary analysis of the
registry without Health Insurance Portability and Accountability Act identifiers, it is exempt from
Institutional Review Board review per University of Virginia IRB policy.

De-identified records for all isolated CABG, AVR, and CABG/AVR patients from January 2012 through
December 2020 were extracted from the VCSQI data registry. Patients were excluded for missing atrial
fibrillation status, preoperative risk scores, and missing or zero charge data. Patients with arrhythmias at
baseline were excluded. A subgroup analysis was performed, excluding patients with major complications
after surgery (STS major morbidity). All clinical variables utilize standard STS definitions, including
operative mortality (30-day or in-hospital mortality) and major morbidity (permanent stroke, prolonged
ventilation, reoperation for any reason, renal failure, and deep sternal wound infection) .
[8]

Statistical analysis
Categorical variables are presented as counts (%) and continuous variables as median [25th, 75th
percentile]. Cost data is presented as both median [25th, 75th percentile] and mean ± standard deviation
(SD) to best clarify total cost implications. Patients were stratified by POAF for univariate analysis using the
Chi-square test for categorical variables and Mann-Whitney U-test for continuous variables. Data
missingness was low, no imputation was used for this first set of analyses, and missing data points were
excluded from the corresponding analysis. Henceforth this group will be called the pre-match cohort.

To account for baseline and postoperative differences, patients were propensity-score matched by POAF
status. Data missingness was accounted for with simple imputations where variables with < 5% missing data
[9]
were imputed using the methodology described in the creation of the STS risk models . This includes the
lower risk category for categorical variables and the median for continuous variables, with gender-specific
medians for body surface area. Next, propensity scores were created using logistic regression and 35
variables, including demographics, preoperative risk factors, and postoperative complications
[Supplementary Table 1]. Patients were then matched using a greedy algorithm from 8 to 3 digits of the
propensity score, matching sequentially without replacement. The logistic regression and match were
optimized in an iterative manner using standardized mean differences (SMD) and propensity score
histograms. An SMD of < 0.1 was considered well balanced. A sensitivity analysis was performed by
matching the second cohort of patients without major morbidity or mortality. The matched cohort was
compared by unpaired univariate analyses, except for cost differences where differences between matched
pairs were computed and also compared by Wilcoxon signed-rank test. As P-value less than 0.05
determined statistical significance. All statistical analyses were carried out using SAS Version 9.4 (SAS
Institutive, Cary, NC), graphics were created with Prism 8.0 (GraphPad, San Diego, CA).

RESULTS
Patient and operative characteristics
A total of 37,676 patients underwent CABG and/or AVR, of whom 1344 (3.63%) had a history of atrial
fibrillation and were excluded. After additional exclusion of patients with other documented preoperative
arrhythmia, as well as those lacking data for STS predicted risk of mortality, cost data, year of surgery, or
POAF, 27,307 patients were identified for analysis. Of these 27,307 patients, 6315 (23.1%) developed POAF
[Supplementary Table 2].