“…We used greedy 1:1 nearest neighbor matching, with exact matching on trial, without replacement, within caliper widths <0.2 (Austin, 2014 ). Due to the small number of eligible persons initiating EBP within 2 weeks of the start of the trial, the number of predictors included in the logistic regression models used to create the propensity scores was necessarily restricted to baseline variables known a priori to be associated with both treatment selection and follow-up PCL score: age, baseline PCL z -score, and comorbid conditions (alcohol abuse, alcohol dependence, depression, drug abuse, drug dependence, psychosis, pain, suicide, and TBI; Holder et al, 2020 ). The EBP initiator and non-EBP initiator groups were compared with standardized mean differences, where an absolute difference of <0.1 on confounders of age and comorbidities was indicative of group similarity (e.g.…”