“…With the full cohort, 3 generalized estimating equation models (univariate, adjusted for selected variables, and adjusted for all covariates) were fit using each of the 3 end points as a dependent variable. A set of potential confounders was chosen a priori based on biological plausibility and a priori knowledge 12, 15, 24, 30, 31, 32, 33, 34, 35, 36, 37, 38. These selected variables included the following: age, sex, hospital category (to adjust for the difference in the frequency of mCPR device use between each institution, hospitals were categorized on the basis of the numbers of patients on whom mCPR devices were used: low volume, <20 per year; moderate volume, 20–100 per year; and high volume, >100 per year), witnessed status, bystander CPR, first documented rhythm, presumed cardiac cause, airway management by EMS, prehospital administration of epinephrine by EMS, tracheal intubation during advanced cardiovascular life support, administration of epinephrine, defibrillation attempt, extracorporeal CPR performed in the ED, and time from call to EMS arrival at scene, time from EMS arrival at scene to EMS arrival at the patient's side, time from EMS arrival at the patient's side to CPR initiation, and time from CPR initiation to hospital arrival (while also adjusting for within‐institution clustering effects using a generalized estimating equation,36, 37, 38, 39 because several articles have suggested the existence of hospital‐related differences in survival after OHCA) 40, 41, 42, 43, 44.…”