Page 173 - Read Online

Page 4 of 13 Kolba et al. Vessel Plus 2023;7:12 https://dx.doi.org/10.20517/2574-1209.2022.61

other patient health-related risk factors (e.g., smoking status). Over the past two years, patients’ profiles for
their study-based AVR or r-AVR encounters were compared to their historical encounters within the past
two years; using a two-year look-back period, therefore, the new diagnoses and procedures performed were
differentiated from the historical diagnoses and procedures performed to differentiate comorbidities vs.
complications. Additionally, other patient risk information was assessed, such as calculating patient’s
baseline comorbidity complexity. As an example, the Elixhauser comorbidity scores mortality and 30-day
readmission were calculated; these standardized comorbidity score algorithms were used to summarize a
[16]
patient’s comorbidity burden for these study-specific endpoints .
Outcome measures
The primary study has three clinical endpoints, including new onset post-procedural atrial fibrillation or
flutter (i.e., POAF/AFL), 30-day readmission (READMIT), and a composite endpoint (MM) comprised of
major morbidity and 30-day operative mortality based upon the Society of Thoracic Surgeons [STS]
definitions used in the Adult Cardiac Surgery Database (ACSD).

The STS definitions for 30-day operative mortality and major morbidity were established in 1979. This
endpoint included in-hospital deaths and all post-discharge deaths occurring within 30 days. For the MM
composite, records with either the STS-defined 30-day operative mortality (i.e., death in-hospital or within
30 days of surgery) or STS-defined set of major complications were identified; major complications
included repeat procedures (i.e., including repeat procedures for bleeding or impaired valve functionality),
perioperative stroke, new renal failure requiring dialysis, deep sternal wound infection (i.e., mediastinitis),
or prolonged use of ventilation (i.e., greater than 48 h on a ventilator) . Following October 15, 2015, new
[17]
ICD-10 complication codes were also used to differentiate STS major complications from pre-AVR patient
comorbidities.

The READMIT endpoint was evaluated based on the time from the date of discharge to the admission date
for a subsequent encounter. Secondary patient outcomes included AVR-relevant complications (e.g.,
bleeding, stroke, and myocardial infarction) and the primary outcomes’ sub-components (i.e., STS-defined
30-day operative death and the five major STS complications as secondary outcomes).

Statistical analyses
All analyses were performed using SAS 9.4 by an institutional data analytics team (the Stony Brook
University School of Medicine’s Biostatistical Consulting Core [BCC] lab). The BCC team’s data extraction
and analysis tasks occurred from January 2021 to July 2022. Statistical analyses included both bivariate
comparisons and multivariable logistic regression analyses. For categorical variables, bivariate comparisons
used chi-square tests with exact P-values from Monte Carlo simulation to examine the relationship between
categorical variables (e.g., polychotomous, or dichotomous baseline patient risk factors such as sex, race,
ethnicity, insurance, etc.) and study endpoints (i.e., POAF, 30-day readmission, and the MM composite) .
[18]
Correspondingly, Welch t-tests compared the unadjusted marginal differences in relationships between
continuous variables (e.g., age, Elixhauser readmission score, Elixhauser mortality score) and study
[19]
endpoints (i.e., POAF, 30-day readmission, and the MM composite) .

For multivariable logistic regression analyses, a stepwise descending selection approach was used . To
[20]
initially identify potential model eligible variables, the literature on cardiovascular surgery and mental
illness was reviewed. These literature-based variables were screened using bivariate comparisons (P < 0.10)
with each study endpoint, verifying the clinical appropriateness of these findings’ directionality. To prevent
potential collinearity, model eligible variables were further refined based on selecting only one domain-
specific variable for model inclusion. The final set of endpoint-specific model eligible variables were entered