Stata se 14.2 statistical software

We determined ARDS incidence for each risk factor, and used 10-iteration multiple imputation to account for missing data (detailed in Supplemental Digital Content 2). We estimated associations between demographic, injury, and clinical variables with development of ARDS in bivariate analyses using generalized linear Poisson regression models, clustered by facility. We included all clinically important variables with bivariate p<0.005 in a backwards selection model-building process to develop a multivariable generalized linear Poisson regression model, using p<0.005 for variables to remain in the model. We developed a second multivariable model for examining associations between chest injury characteristics and ARDS. Multivariable models were adjusted for admission year and transfer status. We conducted all analyses using Stata/SE 14.2 statistical software (StataCorp LP, College Station, TX).

We calculated rates of each outcome for patients with and without ARDS. We estimated associations between ARDS and each outcome in bivariate analyses clustered by facility using generalized linear Poisson regression for death and tracheostomy, multinomial logistic regression for discharge disposition, and linear regression for duration of ventilation and LOS. We repeated each analysis as a multivariable model with the six pre-determined covariates, as well as admission year, transfer status, and facility trauma level designation, to estimate adjusted associations between ARDS and each outcome. We determined annual rates of mortality and post-discharge care among patients with and without ARDS, and estimated associations between year and discharge disposition in bivariate analyses and as multivariable models with the pre-determined covariates. All models were complete case analyses; no covariates had >5% missing data. We conducted all analyses using Stata/SE 14.2 statistical software (StataCorp LP, College Station, TX).

We determined the frequency, timing, and reason for unplanned 30-day readmission after trauma. We estimated associations between patient, injury, and hospitalization characteristics with risk of a binary outcome of unplanned readmission or not in bivariate analyses using generalized linear Poisson regression models with robust standard errors for the entire cohort and stratified by whether patients had undergone operative versus non-operative management during their index admission. ISS was included a priori in all models. We included other variables with bivariate p<0.2 in a backwards selection model-building process to develop multivariable generalized linear Poisson regression models with robust standard errors stratified by operative status, using p<0.05 for variables to remain in the models. We then used the same process to develop secondary multivariable models examining associations between risk factors and risk of readmission for each of the most common reasons for readmission. We conducted all analyses using Stata/SE 14.2 statistical software.
The study was approved by the University of Washington Institutional Review Board with a waiver of informed consent.