Statistical Analysis
Data analyses were performed using SAS 9.4 software (SAS institute, Cary, NC). Figures of odds and hazard ratios were prepared using Prism 7 (GraphPad Software, Inc., La Jolla, CA, USA). Data are presented for the overall cohort, and the two groups stratified by timing of onset of deep sternal wound infection after index cardiac procedure. Categorical data were summarized as the number and percentage of patients. For continuous data, the mean value and standard deviations were reported for normally distributed data while the median value and interquartile ranges were reported for non-normally distributed data. Continuous variables were compared between the two groups using either Student’s t-test or the Wilcoxon rank sum test, depending on data normality. Categorical variables were compared using the chi square or Fisher’s exact test, depending on expected values. Unadjusted survival was assessed by Kaplan-Meier methods for the overall cohort and various strata. Between strata comparisons were made by log-rank. Multivariable Cox proportional hazard modeling was used to model predictors of death in the overall cohort.
Multivariable logistic regression was performed to determine which preoperative variables were predictive of onset of DSWI within 30 days of index cardiac procedure. Early onset DSWI was modeled as the dependent variable and the following independent variables were included in the model based on univariate analysis, the medical literature, and biologic plausibility: sex, smoking (yes vs. no), previous cardiac surgery to the index operation for the DSWI (yes vs. no), positive wound culture at time of initial DSWI operation (yes vs. no), and urgency of initial DSWI operation (yes vs. no). The C-statistic and Hosmer-Lemeshow goodness-of-fit test were reported for the model.
To estimate the impact of timing of onset of deep sternal wound infection after cardiac surgery on subsequent mortality, the multivariable regression model was used to calculate a subject’s propensity for having an early onset DSWI. Propensity scores were then entered into a separate Cox proportional hazard regression model to calculate adjusted hazard ratios for estimates of mortality.