Data were obtained from the PharMetrics Patient-Centric Database, which is comprised of facility, professional-service, and retail (i.e., outpatient) pharmacy claims from over 85 US health plans. The plans provide healthcare coverage to approximately 14 million persons annually throughout the US (Midwest, 35%; Northeast, 21%; South, 31%; West, 13%). All patient identifiers in the database have been fully encrypted, and the database is fully compliant with the Health Insurance Portability and Accountability Act of 1996 (HIPAA). Since there was no contact with either patients or their providers, all patient-identifying information was de-identified, and all analyses were retrospective in nature, Institutional Review Board (IRB) approval was not required.
Information available for each facility and professional-service claim includes date and place of service, diagnoses (in ICD-9-CM format), procedures (in ICD-9-CM [selected plans only], Current Procedural Terminology, 4th Edition [CPT-4], and HCPCS formats), and provider specialty. Data available for each retail pharmacy claim include the drug dispensed (in NDC format), the dispensing date, and the quantity dispensed and number of days of therapy supplied (selected plans only). All claims include both charged and paid amounts, the latter including patient deductibles, copayments, and/or coinsurance.
Selected demographic and eligibility information is also available, including age, gender, geographic region, coverage type, and the dates of insurance coverage. All patient-level data can be arrayed in chronologic order to provide a detailed, longitudinal profile of all medical and pharmacy services used by each plan member. The database for this study encompassed the period from January 1, 2003 through December 31, 2007 (“study period”).
We identified all patients with two or more paid claims for outpatient encounters on different days with a diagnosis of GAD (ICD-9-CM diagnosis code 300.02) between 1/1/2003 and 12/31/2007. Among these persons, we limited attention to those who were beginning a long-term course of treatment (≥90 therapy-days over 6 months) with a benzodiazepine anxiolytic (i.e., alprazolam, chlordiazepoxide, clonazepam, clorazepate, diazepam, lorazepam, oxazepam) (“study agents”). Date of treatment initiation was designated the “index date”. We excluded from the study sample all patients: (1) with any evidence of receipt of benzodiazepines prior to their index date; (2) with one or more days of ineligibility for medical and pharmacy benefits over the 6-month periods immediately preceding and following their index dates (“pretreatment” and “post-treatment”, respectively); (3) who were Medicaid beneficiaries; or (4) aged ≥65 years as of their index date and enrolled in a Medicare supplemental or fee-for-service plan (their claims histories may be incomplete). We then stratified patients according to whether they received a benzodiazepine anxiolytic alone (“monotherapy cohort”) or adjunctive to escitalopram, paroxetine, sertraline, or venlafaxine (“add-on” cohort), as described below.
To maximize the likelihood that patients in the monotherapy cohort were initiating treatment for GAD, we required that they have at least one healthcare encounter with a diagnosis code of GAD in the 90-day period immediately preceding (and including) their index date, and no evidence of receipt of any other GAD-related medication (i.e., escitalopram, paroxetine, sertraline, venlafaxine, imipramine, buspirone, hydroxyzine, trifluoperazine)
[9, 40] or any other agent from any corresponding class of medication (e.g., selective serotonin reuptake inhibitors [SSRIs]) during the 6-month pretreatment period.
Patients in the add-on cohort were required to have at least one claim with a diagnosis code of GAD as well as evidence of extended (≥90 days) receipt of escitalopram, paroxetine, sertraline, or venlafaxine (all recommended as first-line treatment for GAD in recent clinical guidelines
[9, 41]) during the 6-month period preceding initiation of benzodiazepine therapy (a longer time period was used for the add-on cohort than the monotherapy cohort, as GAD diagnoses may not be rendered as frequently in patients receiving long-term therapy). They also had to have evidence of receipt of these agents in the 90-day period following their index date.
Patients not meeting criteria for either the monotherapy or add-on cohorts were dropped from the study sample. For all remaining patients, we compiled all pharmacy, professional service, and facility claims during both the pre-index and post-index periods.
Measures and analyses
The demographic and clinical characteristics of study subjects, including prevalence of selected comorbidities (Appendix), were characterized on the basis of information during the 6-month pre-index period.
All healthcare encounters with ICD-9-CM diagnosis codes in the ranges 800.XX-959.XX, 990.XX-995.XX, E800.X-E849.X, E880.X-E889.X, E916.X-E929.X, and V155.5 were considered to be “accident-related,” following methods first employed by Oster et al.. Healthcare encounters for other potential adverse consequences of long-term benzodiazepine therapy were identified based on visits/admissions for known sequelae of benzodiazepine use [33–38], as follows: (1) drug dependence/addiction/poisoning (ICD-9-CM diagnosis codes 304.1X, 304.9X, 305.4, 969.4, V58.69, E853.2, E980.3); (2) withdrawal syndrome (292.0X); (3) drowsiness (780.09); (4) ataxia (781.2, 781.3); (5) dysarthria (784.5); (6) diplopia (368.2); (7) vertigo/dizziness (780.4); (8) mental confusion/disorientation (292.8X, 292.9X, 293.0, 780.1, 780.2); (9) cognitive impairment (780.02, 780.09, 780.93, 780.97); and (10) other adverse effects recorded that related to benzodiazepine therapy (E939.4, E950.3). These encounters were denoted, collectively, as “other possibly related” care.
We also examined patterns of healthcare utilization during the pretreatment and post-treatment periods in terms of the numbers of physician office visits, other outpatient office visits, emergency department (ED) visits, and hospitalizations. Use of these services was examined in terms of the numbers of patients receiving each type of service, as well as counts of the numbers of ervices used (e.g., office visits, hospitalizations). All healthcare encounters and corresponding costs were designated as “accident-related”, “other possibly related”, or “other”, depending on the ICD-9-CM diagnosis code(s) noted on each claim. Total healthcare costs were tallied in terms of: (1) prescription pharmacotherapy; (2) physician office visits; (3) other outpatient visits; (4) ED visits; (5) inpatient care; and (6) all other care. Reimbursed amounts (including any patient liability, such as co-pays and co-insurance) were used in all analyses of healthcare costs. Healthcare utilization and costs were assessed over the 6-month pretreatment and post-treatment periods, respectively.
Since patients served as their own controls in the analyses, the statistical significance of differences in healthcare costs between the pretreatment and post-treatment periods was assessed using Wilcoxon signed-rank tests; McNemar and Bowker’s tests were used to assess the statistical significance of differences in categorical variables. All tests of statistical significance were two-tailed with an alpha level of 0.05. All analyses were first performed by cohort and, within each cohort, on an overall basis as well as for patients aged <50 years versus ≥50 years, respectively. Finally, analyses were performed for the two cohorts combined, on an overall basis as well as by age stratum (<50 years vs ≥50 years). All analyses were conducted using SAS® Proprietary Software, Release 9.1 (SAS Institute Inc., Cary, NC).