The Brigham and Women’s Primary Care Practice-Based Research Network (BWPC PBRN) includes 16 hospital- and community-based practices and community health centers in eastern Massachusetts. The BWPC PBRN practices used a fully functional, Certification Commission for Healthcare Information Technology (CCHIT)-certified electronic health record (EHR), which included problem lists, medication lists, and prescriptions. By policy, all medicines were prescribed through the EHR. Medications not prescribed by affiliated clinicians were listed in the EHR without dosing information.
Sociodemographic information was collected during registration and was updated periodically. Billing codes were recorded in a separate, dedicated billing system. Partners HealthCare—an integrated health delivery system in eastern Massachusetts, of which Brigham and Women’s Hospital is a part—had an information system that captured outpatient visits, emergency room visits, and hospitalizations for all Partners HealthCare facilities.
Approval for the conduct of this study was obtained from the Partners HealthCare Institutional Review Board.
We used the Partners HealthCare Research Patient Data Repository, which aggregates data from throughout Partners HealthCare facilities, to identify all patients who made at least one visit to any of the ten BWPC PBRN practices that were participating in an unrelated clinical trial between July 1, 2011, and June 30, 2012.27
29 We extracted and combined sociodemographic and clinical information from the EHR with billing codes.
We included all coded benzodiazepine prescriptions and listings. From the EHR we extracted prescription details that included the name of the medication, dose, frequency, total number of units prescribed, number of refills, and prescribing clinician. Our data source included prescriptions; we could not measure prescription fills or actual benzodiazepine use by patients.
We extracted medical diagnoses from the EHR problem list and ICD-9 billing codes associated with individual encounters (see
online appendix). We extracted medical diagnoses defined by the Healthcare Effectiveness Data and Information Set (HEDIS; asthma, COPD, cardiovascular disease, depression, diabetes, hypertension, obesity, osteoporosis, and tobacco use),30 psychiatric diagnoses for which benzodiazepines are commonly prescribed (anxiety and insomnia), and diagnoses for which benzodiazepines are contraindicated or controversial (alcohol abuse, sleep apnea, and substance abuse).6
We also extracted data about antidepressant medication prescribing from the EHR, because these are commonly considered first-line agents for depression and anxiety. We included the antidepressants fluoxetine, sertraline, paroxetine, citalopram, escitalopram, fluvoxamine, mirtazapine, bupropion, venlafaxine, desvenlafaxine, duloxetine, nefazodone, amitriptyline, amoxapine, clomipramine, desipramine, doxepin, imipramine, nortriptyline, protriptyline, trimipramine, phenelzine, tranylcypromine, isocarboxazid, trazodone, and vilazodone.
We extracted medical encounters from encounter-level billing data, including primary care visits (both any visit to the primary care clinic and any visit with the PCP of record), specialist outpatient visits, emergency department (ED) visits, and hospitalizations, and length of stay for patients with one or more hospitalizations. We defined a patient’s PCP as the PCP of record from the EHR. Listed PCPs are nearly always primary care clinicians.
We calculated benzodiazepine dosing and days prescribed based on a combination of pill dose/strength, dosing frequency, and number of pills prescribed during the study period. We converted prescriptions of lorazepam, clonazepam, and alprazolam—which, together with diazepam, accounted for 97 % of benzodiazepine prescriptions—to “average daily diazepam-equivalent dosages.” Only days for which benzodiazepines were prescribed were included in the calculation of average daily dose.
High-dose benzodiazepine prescribing has been defined in the literature as a daily dose equivalent of ≥30 mg per day of diazepam.32 Although potency equivalence between benzodiazepine agents is not clearly established, we defined 30 mg diazepam equivalents as 3 mg/d alprazolam, 3 mg/d clonazepam, and 5 mg/d lorazepam.32
33 For patients with multiple benzodiazepine agents (3 % of patients receiving benzodiazepines) for which diazepam-equivalent dosing could be calculated, we added them together as though they were concurrent or consecutive prescriptions of a single diazepam-equivalent agent. Other benzodiazepine prescriptions and benzodiazepine prescriptions without complete prescribing information were not included in the comparison between high-dose and standard-dose prescriptions.
To determine which patients were most likely to receive benzodiazepine prescriptions, we compared patients who received at least one benzodiazepine prescription with those who did not. We assessed differences in demographic variables, medical diagnoses, and inpatient and outpatient encounters. Among benzodiazepine recipients, we made parallel comparisons between patients who did and did not receive high-dose prescriptions.
We used means, medians, percentages, odds ratios, and rate ratios with 95 % confidence intervals to compare patients who did and did not receive benzodiazepines and those who received high doses versus standard doses. We compared categorical variables using the chi-square test and continuous variables using Student’s t test. We performed the Mann–Whitney–Wilcoxon test to compare days dosed among categorical variables with two groups and the Kruskal-Wallis test for the same comparison among categorical variables with three or more groups. We calculated odds ratios using logistic regression, and we used Poisson regression to calculate rate ratios. We used SAS software (version 9.3; Cary, NC) for all analyses and considered p values < 0.05 statistically significant.