Statistical Analysis
Data are shown as mean ± standard deviation (SD) for continuous
variables and n (%) for categorical variables. All relevant variables
were compared between procedures with and without a 30-day all-cause
readmission with Student’s T-test or Chi-square as appropriate. To
derive independent predictors of readmission we followed the 10EPV
guideline which allowed for 24 degrees of freedom for our adjusted
model13. Based on univariate associations and clinical
interest we used the following as potential predictors: elective status
at index admission, age, sex, disposition, hospital bed size,
cardiomyopathy, chronic pulmonary disease, congestive heart failure
(CHF), ischemic heart disease (IHD), obesity, hypertension (HTN),
peripheral vascular disease (PVD), diabetes mellitus (DM), renal disease
including chronic kidney disease (CKD) and end stage renal disease
(ESRD), fluid and electrolyte disorder, anemia, and peptic ulcer
disease. These variables were entered into a hierarchical logistic
regression model using a random intercept for site, with odds ratios and
95% confidence intervals. Next, we ran a selection
procedure using stepwise methods to obtain a parsimonious list of
independent predictors of readmission in our cohort. We then used the
beta estimates to create a simple integer scoring system to predict
readmission14. We then used the applied the scoring
system to our derivation cohort and tested the 30-day readmission rate
by score. Then using data from the 2014-2015 NRD as a validation cohort,
we applied the same integer scoring system to test discrimination with
30-day readmissions. Next, we determined the top causes of readmission
using the primary diagnosis code at readmission.14 For
general categorization we used the letter and first 2 numbers of the ICD
10 code. All analysis was done with SAS version 9.4 (Cary, NC) with a p-
value of 0.05 marking statistical significance.