Postoperative But Not Preoperative Depression Is Associated With Cognitive Impairment After Cardiac Surgery

Exploratory Analysis of Data From a Randomized Trial

Choy Lewis; Mehmet E. Dokucu; Charles H. Brown; Lauren Balmert; Nina Srdanovic; Ashwin Shaan Madhan; Sahej Singh Samra; John Csernansky; Jordan Grafman; Charles W. Hogue

Disclosures

BMC Anesthesiol. 2022;22(157) 

In This Article

Data Analysis

Descriptive statistics summarized all variables by preoperative depression. Fisher's exact tests for categorical variables, two sample t-tests for normal continuous variables, and Wilcoxon rank sum tests for non-normal continuous variables, were used to compare distributions of baseline and intraoperative variables by depression status. A simple paired t-test assessed changes in Beck Depression Inventory from preoperative to the one month postoperative time points. Primary analyses utilized multivariable linear regression models to assess associations between preoperative depression and change in each domain Z-score from baseline to four to six week follow-up, adjusting for variables including: baseline domain Z-score, sex, age, education, parent study blood pressure treatment arm, diabetes, obesity, prior stroke, hypertension, atrial fibrillation, and COPD. A false discovery rate (FDR) correction was applied to correct for multiple testing.[25] This approach allows for controlling the expected proportion of tests which are falsely rejected. Due to missing data, sensitivity analyses were conducted for primary models using multiple imputation with 20 imputed datasets via multivariate imputation by chained equations, using the mice package in R.[26] Additional sensitivity analyses replicated models with the continuous Beck Depression Inventory score as the independent variable of interest. Secondary analyses considered similar models to assess associations between postoperative depression and the postoperative cognitive domain Z-scores, with similar adjustment for potential confounders. Generalized linear models with Poisson distributions and robust standard errors assessed associations between preoperative and postoperative depression, in turn, with the neurocognitive dysfunction indicator, adjusting for age, sex, education, treatment arm, diabetes, obesity, prior stroke, hypertension, atrial fibrillation, and COPD. Additional secondary analyses then considered the SF-36 scales and State Anxiety Inventory outcomes, measured at four to six weeks postoperative testing period, in similar models. Specifically, separate multivariable linear regression models were considered for preoperative depression and postoperative depression with each SF-36 scale and State-Trait Anxiety scale, with adjustment for age, sex, education, baseline score, treatment arm, diabetes, obesity, prior stroke, hypertension, atrial fibrillation, and COPD, unless otherwise specified in the results section. Due to violation of distributional assumptions, Physical Health Limits and Emotional Health Limits scales were dichotomized at 0 and considered in multivariable logistic regression models. Lastly, associations between depression and clinical complications were assessed using generalized linear models with Poisson distributions and robust standard errors for binary outcomes and multivariable linear regression models for continuous outcomes adjusted for treatment arm, sex, age, and education. A false discovery rate (FDR) correction was applied to correct for testing of the multiple cognitive domains, multiple quality of life scales, and clinical endpoints, within each set of analyses.

processing....