Reducing Benzodiazepine Prescribing in Older Veterans: A Direct-to-Consumer Educational Brochure
The EMPOWER cluster randomized controlled trial (RCT) demonstrated the effectiveness of EB exposure in a Canadian population of elderly patients who were long-term BZD users.17 Randomized controlled trials are the gold standard for clinical trials because they can establish causal inference.23-25 Given ethical and practical concerns, however, RCTs cannot be applied to all clinical scenarios. Although EMPOWER is reported to be an effective tool in reducing BZD use in older adults, its application in a real-world, large, integrated health care system remains untested. Observational studies are often conducted as an alternative to RCTs but are subject to selection bias because of their lack of randomization.26 Therefore, robust research methods are needed to generate unbiased estimates of the impact of an intervention on an outcome. Propensity score matching simulates an RCT by balancing the covariates across treatment groups.21,22,27 Observed patient characteristics are used to estimate PS, the probability that treatment will be received. Logistic or probit regression is used to balance the potential confounding covariates between the treatment groups.Once PSs are known, mean treatment effect can be estimated without the mean model.28 In other words, PSM methods can be used to generate an unbiased estimate of the treatment.
Propensity Score Analysis
Baseline characteristics were compared using Student t test (continuous variables) and χ2 test (discrete variables). Results are presented as means and standard deviations (continuous variables) and frequency and percentage (discrete variables).
The main outcome was BZD discontinuation 9 and 12 months after the index date. A postindex lag of 6 months was used to capture any tapering (Figure 2). Discontinuation, defined as 3 consecutive months of no BZD prescription on hand, was measured for 2 periods: 6 to 9 months and 6 to 12 months after the index date. A secondary outcome was discontinuation 1 to 12 months after the index date. An estimate was made of the difference in the proportions of BZD discontinuers who received the EB and BZD discontinuers who did not receive the EB, where mean treatment (risk difference) was presented as the absolute risk difference with a 95% CI. Standard errors and 95% CIs for the risk differences were generated with biased-corrected CIs from 1,000 bootstrap samples.
Sensitivity Analyses
Naïve multivariate logistic regression analysis was performed to evaluate the association between EB exposure and BZD discontinuation while controlling for potential confounders. Results are presented as odds ratios (ORs) and 95% CIs. Confounders identified were the same covariates used to generate the PSs.
Several analyses were performed to test the sensitivity of the methods applied using PSM by changing caliber size while maintaining the nearest neighbor approach without replacement. Linear regression analysis was performed with robust standard errors to estimate the risk difference of BZD discontinuation between EB-exposed and EB-unexposed veterans.
Statistical significance was set at P < .05. All statistical analyses were performed with Stata/SE Version 13 (College Station, TX).
Results
Quality Improvement Project 1
On a rolling basis from December 2014 to February 2016, the EB was mailed once to 3,896 VISN 22 veterans 2 to 8 weeks before a clinic appointment with their BZD prescribing physician. 
