Guneet K. Jasuja, PhD

 Use of nonbenzodiazepine sedative hypnotics, especially zolpidem, has grown substantially, raising concerns about safety. Here, we evaluated prescribing patterns of zolpidem in the Veterans Health Administration.

 A cross-sectional study of veterans receiving zolpidem in the outpatient setting from October 1, 2011, to September 30, 2016.

 The study population consisted of 500,332 zolpidem users (58,598 women and 441,734 men) and a random 10% sample (n = 631,449) of nonusers. We examined 2 outcomes related to inappropriate prescribing: high-dose zolpidem prescribing and overlap with benzodiazepines. We generated interrupted time series and logistic regression models to analyze these outcomes in men and women separately.

 In 2016, 29.7% of female veterans received an inappropriately high guideline-discordant dosage compared with 0.1% of male veterans (

 <.001 for all reported comparisons). Furthermore, more women than men had overlapping benzodiazepine and zolpidem prescriptions (18.8% vs 14.3%). In fully adjusted models, inappropriately high doses were more commonly received by younger women (adjusted odds ratios [AORs]: 2.75 for 21-39 years and 2.97 for 40-49 years compared with ≥80 years) and women with substance use disorder (AOR, 1.48). In the second inappropriateness outcome models, women with anxiety (AOR, 2.28) or schizophrenia (AOR, 2.05) and men with cancer (AOR, 1.42), anxiety (AOR, 2.66), or schizophrenia (AOR, 2.46) were more likely to receive an overlapping prescription of zolpidem and benzodiazepines.

 We found evidence of inappropriate zolpidem prescribing among veterans, particularly women. Greater understanding of the drivers of this inappropriate prescribing is necessary to develop interventions to promote safer, more guideline-concordant prescribing.

Gender-specific guidelines have been recommended for appropriate zolpidem prescribing to address patient safety concerns.

The present study suggests that a higher proportion of female veterans received an inappropriately high zolpidem dosage and had overlapping zolpidem and benzodiazepine prescriptions compared with male veterans.

Findings have implications both for patient safety and promoting guideline-concordant prescribing in female zolpidem users.

Managed care decision makers may wish to query adherence to guidelines and raise awareness to drive practice changes and reduce morbidity and mortality related to preventable adverse events in their systems of care.

Zolpidem, a nonbenzodiazepine sedative hypnotic, is extensively prescribed in the United States

 for short-term treatment of insomnia. However, there have been growing concerns associated with its use, including higher risks of falls and fractures,

 central nervous system—associated adverse drug reactions,

 In fact, it has been shown that the mortality implications of taking zolpidem are equivalent to those associated with a benzodiazepine.

 Clinically significant drug interactions of zolpidem with benzodiazepines are associated with increased hospitalizations.

 To add to the complexity, gender differences have been noted, with women reporting greater sensitivity than men to the sedative effects of zolpidem.

These safety concerns prompted the FDA in 2013 to recommend cutting doses prescribed to women in half, specifically from 10 mg immediate release (IR) daily to a maximum of 5&thinsp;mg/day and from 12.5 mg controlled release (CR) daily to 6.25 mg/day.

 Usually prescribed for a 30-day period, the IR formulation dissolves much more rapidly than the CR formulation to cause sedation facilitating the induction of sleep.

 Although there was also a recommendation to extend this dose reduction to men, the guidelines remained unchanged for men (5 mg or 10 mg IR and 6.25 mg or 12.5 mg CR). This mandate of a gender-specific guideline for zolpidem was due to the fact that women metabolize the same dose of zolpidem more slowly than men, resulting in 50% higher serum levels.

 Previous studies have found increased compliance among providers, specifically midlevel and physician providers, with reduced-dose zolpidem in women after the FDA-mandated labeling changes.

 Although adherence to guidelines was attributed to change in prescribing habits of the providers, it could not be ascertained whether it was driven by the drug safety communication from the FDA or individual patient factors. However, these studies included relatively small sample sizes; they did not evaluate the proportion of women who were still being prescribed an inappropriate dosage after the FDA guidance, nor did they examine predictors of this inappropriate outcome.

In line with this FDA recommendation, the national Veterans Health Administration (VHA) Pharmacy Benefits Management (PBM) Service issued a similar directive to lower the doses of zolpidem in women in 2013.

 However, little is known about zolpidem prescribing patterns in the VHA. Therefore, the goal of this study was to examine utilization trends and inappropriate prescribing of zolpidem in accordance with PBM guidelines in male and female veterans in the VHA. We expected significant differences in inappropriate zolpidem prescribing in male and female veterans by age, with women potentially being more likely to receive zolpidem inappropriately.

The study was approved by the Institutional Review Board of the Bedford Veterans Affairs Medical Center. We examined demographic, diagnostic, and prescription data of patients receiving outpatient prescriptions for zolpidem at any VHA site between October 1, 2011, and September 30, 2016.

We examined all patients who received zolpidem (n = 500,332; 441,734 men and 58,598 women) between fiscal year (FY) 2012 and FY 2016. We defined a zolpidem user as a patient who received 3 or more days of continuous zolpidem prescription during the study period. Additionally, we examined a random 10% sample of patients who did not receive zolpidem (n = 631,449). We also used the full study population (N = 1,131,781) as a basis for calculating prevalence rates.

The index date for zolpidem users was the earliest prescription of zolpidem they received in the entire study period.

We examined the sociodemographic factors gender, age, race, Hispanic ethnicity, marital status, VHA co-payment for prescription drugs based on VHA eligibility status, FY of zolpidem prescription, US region where the patient received care, and the urban versus rural status of the facility where the index prescription was written. These variables, and in particular gender and age, have been previously linked to healthcare utilization.

We also evaluated a number of physical and mental health conditions that have been shown to be associated with zolpidem use.

 The presence of these conditions was determined from inpatient and outpatient diagnoses using 

International Classification of Diseases, Ninth Revision

) codes, using a 1-year look-back period prior to the date of the index prescription (

]). We considered a condition to be present when there were at least 2 diagnostic codes for the condition separated by 7 or more days.

In this study, we evaluated 2 outcomes relating to appropriateness of zolpidem prescribing. The first measure was whether the daily dose prescribed to male and female veterans exceeded the 2013 PBM recommended dosage (men: IR ≤10&thinsp;mg or CR ≤12.5 mg; women: IR ≤5&thinsp;mg or CR ≤6.25&thinsp;mg). In our study, we calculated the daily dose by multiplying the inferred tablets per day (quantity divided by days’ supply) by the tablet strength given in the full description of the medication. We also examined a second measure of inappropriate prescribing in both men and women: the overlap in prescribing of zolpidem with the prescribing of a benzodiazepine. We defined overlapping prescribing as a benzodiazepine prescription for more than 3 days written within 30 days before or after the index zolpidem prescription.

As a first step, we generated descriptive characteristics stratified by gender. We used Pearson’s χ

 test to perform statistical comparisons between men and women. We grouped age and race variables into categories to yield a reasonable frequency distribution. Physical and mental health conditions were binary (yes/no) variables. We applied a sampling weight of 10 to the nonzolpidem cases when computing prevalence and initiation. We imputed missing data on all variables except race, marital status, and comorbidities to allow inclusion of all observations in multivariate analyses. Second, we conducted an interrupted time series analysis before and after the 2013 PBM guidelines (which recommended lower zolpidem doses for women) to determine whether these guidelines led to significant improvement in appropriate zolpidem dosing. For this analysis, we excluded 1849 men and 193 women because they had incomplete dose data on the index date of their zolpidem prescription. We calculated the unadjusted proportion of male and female veterans receiving high-dose zolpidem prescriptions from 2011 to 2016. Finally, we generated multivariate logistic regression models to analyze the determinants of the 2 inappropriate prescribing outcomes, specifically dose exceeding guideline recommendations and overlap with benzodiazepines, in separate models for men and women. We used SAS version 9.4 (SAS Institute Inc; Cary, North Carolina) for all analyses. We used 

 <.001 as the threshold for statistical significance for all analyses.

 reports descriptive characteristics of zolpidem users stratified by gender. Across the VHA, 7.1% of men and 10.0% of women received zolpidem at least once during the study period. Compared with male zolpidem users, female users were younger (45.7 years vs 56.6 years) and were more likely to have anxiety disorders (21.1% vs 13.1%), bipolar disorder (11.4% vs 6.1%), and major depression (44.0% vs 29.8%); 

 <.001 for all. Men receiving zolpidem had higher proportions of coronary artery disease (11.1% vs 1.8%), diabetes (21.0% vs 9.2%), and hypertension (38.5% vs 20.3%) compared with women; 

Trends in High-Dose Zolpidem Prescribing for Men Versus Women

Unadjusted trends in high-dose zolpidem prescriptions by gender before and after the PBM 2013 recommendations for suggested lower dose in women are shown in the 

(A and B). The observed proportion of women receiving the higher dose decreased considerably from 75.9% in 2011 to 29.1% in 2016 (

 <.001), and the 2013 PBM recommendations had a significant effect after adjustment for secular trends and covariates (

 <.001). For men, there was a decline in observed receipt of high-dose zolpidem from 1.7% in 2011 to 0.1% in 2016 (

 <.001), but the recommendations did not have a significant effect (

Patient Characteristics Associated With Inappropriate Zolpidem Prescribing Stratified by Gender

For our first outcome, inappropriate dose, the analysis was limited to men and women with an index prescription in January 2013 or later, resulting in a sample of 194,355 men and 28,181 women. As reported in 

, compared with 0.1% of men who received more than 10 mg of IR or 12.5 mg of CR, approximately 29.7% of women zolpidem users were receiving an inappropriate dosage, defined as exceeding 5 mg/day for IR or 6.25 mg/day for CR. Men with substance use disorder (adjusted odds ratio [AOR], 3.34; 

 <.001 for this and all other findings) and men living in the Midwest (AOR, 1.42) were more likely to receive an inappropriate zolpidem dose after adjusting for covariates. In fully adjusted models, younger women were more likely to receive an inappropriately high dose compared with women 80 years or older (AORs: 21-39 years, 2.75; 40-49 years, 2.97). Further, women with certain conditions, including major depression (AOR, 0.91), were less likely to receive an inappropriate dosage. Women in the West were less likely to receive an inappropriate dose (AOR, 0.80) compared with women residing in the Northeast. Similar to the men, women with substance use disorder were more likely to receive an inappropriately high dose of zolpidem (AOR, 1.48).

Of the 500,332 zolpidem users, 193,355 were also prescribed benzodiazepines at some point in the study period. 

results on the concomitant prescribing of zolpidem and benzodiazepines by gender. A higher proportion of female veterans had an overlapping benzodiazepine prescription along with their zolpidem prescription within a 30-day period (18.8% of women vs 14.3% of men). For men, there was a significant age effect with coprescribing of zolpidem and benzodiazepines, with men aged 40 to 79 years at an increased risk of this coprescribing compared with men 80 years or older (AORs: 40-49 years, 1.12; 50-59 years, 1.33; 60-69 years, 1.23; 70-79 years, 1.14) for all age categories (

 < .001 for all comparisons). Conditions including cancer (AOR, 1.42), anxiety (AOR, 2.66), posttraumatic stress disorder (PTSD) (AOR, 1.47), and schizophrenia (AOR, 2.46) were also associated with overlapping prescribing for male veterans. Further, in fully adjusted models for men, sites in the South (AOR, 0.96) and West (AOR, 0.87) were less likely to prescribe overlapping doses compared with sites in the Northeast. Similarly, for women, mental health conditions such as anxiety (AOR, 2.28), bipolar disorder (AOR, 1.73), PTSD (AOR, 1.44), and schizophrenia (AOR, 2.05) were associated with overlapping prescribing of zolpidem and benzodiazepines. Regions other than the Northeast tended to be associated with less overlapping prescribing.

In this study, we found that inappropriate prescribing of zolpidem is not uncommon in the VHA, particularly among female veterans. Although there was a drastic decrease in prescribing of an inappropriate zolpidem dose in women from before to after the PBM guideline was issued, 30% of female zolpidem users in the VHA continued to be prescribed an inappropriate dose in 2016, contrary to the 2013 PBM guidance.

 Further, these decreasing rates of inappropriate prescribing over time potentially suggest that the guidelines may have had some impact, although they might be slow to fully disseminate into practice. Patient characteristics associated with higher odds of receiving an inappropriate dose among women included younger age and requirement of a co-payment. For both male and female veterans, having a substance use disorder was associated with inappropriate high dose.

We also found that a higher proportion of female compared with male veterans were being coprescribed benzodiazepines along with zolpidem, another form of inappropriate prescribing. Because zolpidem has similar hypnotic effects to benzodiazepines and affects the same receptor,

 this coadministration of the 2 medications may potentially increase harm and incur a higher risk of drug interactions and adverse events; however, to our knowledge, this has not been studied directly. The significant association between mental health conditions, including anxiety, bipolar disorder, PTSD, and schizophrenia, and this inappropriate outcome measure for both men and women suggests the already highlighted risk of dependence on benzodiazepines

 in patients with psychiatric illnesses, as described in the literature.

Although some observers may assume that providers always follow dosing recommendations closely, the findings of this study suggest that they may not. A recent study conducted in the University of Colorado health system compared zolpidem prescribing practices before and after the FDA labeling change in 2013.

 The Colorado study found that providers changed their prescribing habits in response to FDA guidance, reflected by a significant increase in the overall percentage of young women (<65 years) receiving an appropriately low dose of zolpidem after the labeling change. This increase in appropriate prescribing, from 42% before the label change to 70% after the label change, nonetheless implies that 30% of women were still receiving the inappropriate higher dose of zolpidem. Findings from the Colorado study are very similar to those of our study, which also found that 30% of female veterans—88% of them younger than 60 years&mdash;continued to receive an inappropriately high dose of zolpidem after the PBM guidance in 2013.

 Compliance among midlevel providers and physicians with FDA-mandated dosing guidelines was also noted in a retrospective cohort study, but only 16% of the prescriptions in that study were in accordance with the FDA recommendation.

 Taken together, these findings potentially suggest a quality of care issue that needs to be addressed both within and outside the VHA. Contrary to our study finding of a higher proportion of women receiving an inappropriately high dose in the post—PBM recommendation period, women reported lower odds of high-dose zolpidem exposure (>10 mg IR; >12.5 mg extended release) compared with men in a recent study conducted among veterans of the Iraq and Afghanistan wars.

 This discrepancy in findings could be due to the focus of the study on a special veteran population characteristically different from the general veteran population.

Potential contributors to this inappropriate prescribing of zolpidem in women could include the fact that the majority of patients served by the VHA are men. Because VHA providers see such a preponderance of male veterans, they may be less familiar with issues specific to treating women, such as the PBM guidance for zolpidem dose reduction for women.

 Similar differential prescribing patterns and adherence to guidelines in male versus female veterans have been reported in previous studies.

 Additionally, clinician failure to adhere to recommended guidelines, either due to lack of awareness or unfamiliarity with the guidelines or due to other factors that render it difficult to follow the guidelines,

 could also explain this trend of inappropriate zolpidem prescribing in female veterans.

Our study finding of female veterans, specifically women with existing mental health conditions, as a high-risk group in both the receipt of a guideline-discordant high dose and overlapping prescribing of zolpidem and benzodiazepines compared with male veterans will potentially alert and motivate providers to follow clinical recommendations in their zolpidem prescribing practices with respect to this population. Further, findings may help guide and inform intervention efforts targeted toward female veterans with mental comorbidities to optimize zolpidem prescribing in the VHA. Efforts to reduce inappropriate prescribing among zolpidem users could include targeted dissemination and implementation strategies, including educational outreach, continuing medical education, and audit and feedback.

 The objectives of these activities would be to increase clinician awareness of the existing guidelines for prescribing zolpidem and to compare their prescribing practice with that of their peers and with guideline recommendations. Previous studies have suggested that such an approach may contribute to improved prescribing, especially if delivered in a targeted and personalized manner.

 The use of clinical decision support systems may also help guide providers to improve practice over time.

 Even further gains could potentially be made using a more proactive (if more invasive) approach, such as requiring pharmacy approval or a nonformulary consult for inappropriately high doses of zolpidem.

Our study has important strengths, specifically the detailed nature and large size of the database, which included medication dispensing records, 

 codes, and sociodemographic factors. However, we also acknowledge limitations. First, our study shares the limitations inherent in any analysis of electronic health record data. One such limitation is that diagnosis codes may not always be applied accurately by clinicians. We addressed this concern in part by requiring 2 

 codes to confirm the conditions. Second, VHA patients and the VHA system may not be typical of the general population or of other healthcare systems due to the disproportionately large population of men served. Hence, results from this study might not be generalizable to the population outside of the VHA. Third, although we used national VHA pharmacy data to account for zolpidem prescriptions, we did not include prescriptions obtained from Medicare Part D in veterans with dual VHA and Medicare Part D pharmacy use. We plan to focus on these dual zolpidem users in a future study. Finally, in this study, we focused only on patient-level predictors of inappropriate prescribing. We recognize the importance of provider- and system-level factors in the prescribing process and plan to focus on these factors in a separate study.

We found that after the VHA PBM formally recommended a lower maximum dose of zolpidem for women in 2013,

 a considerable number of female veterans continued to receive the higher dose or were coprescribed zolpidem and a benzodiazepine. This implies a quality of care problem with implications for patient safety, which may well be amenable to targeted interventions.

Center for Healthcare Organization and Implementation Research, Edith Nourse Rogers Memorial VA Medical Center

(GKJ, JIR, RSW), Bedford, MA; Department of Health Law, Policy and Management, Boston University School of Public Health (GKJ), Boston, MA; The Pulmonary Center (RSW) and Section of General Internal Medicine (AJR), Department of Medicine, Boston University School of Medicine, Boston, MA; Pharmacy Benefits Management Academic Detailing Services, Veterans Health Administration (MLC), San Diego, CA.

 The research reported/outlined here was supported by the Department of Veterans Affairs (VA), Veterans Health Administration, Health Services Research and Development (HSR&D) Service. Dr Jasuja is a VA HSR&D Career Development awardee at the Bedford VA (CDA 13-265). The views expressed in this article are those of the author(s) and do not necessarily represent the views of the VA. The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and the decision to submit the manuscript for publication.

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (GKJ, RSW, MLC, AJR); acquisition of data (GKJ, JIR); analysis and interpretation of data (GKJ, JIR, AJR); drafting of the manuscript (GKJ, JIR, MLC); critical revision of the manuscript for important intellectual content (GKJ, RSW, MLC, AJR); statistical analysis (JIR); and supervision (GKJ, RSW, AJR).

 Guneet K. Jasuja, PhD, Center for Healthcare Organization and Implementation Research, Bedford VA Medical Center, 200 Springs Rd, Bedford, MA 01730. Email: guneet.jasuja@va.gov.

1. Huedo-Medina TB, Kirsch I, Middlemass J, Klonizakis M, Siriwardena AN. Effectiveness of non-benzodiazepine hypnotics in treatment of adult insomnia: meta-analysis of data submitted to the Food and Drug Administration. 

. 2012;345:e8343. doi: 10.1136/bmj.e8343.

2. Kaufmann CN, Spira AP, Alexander GC, Rutkow L, Mojtabai R. Trends in prescribing of sedative-hypnotic medications in the USA: 1993-2010. 

. 2016;25(6):637-645. doi: 10.1002/pds.3951.

3. Kolla BP, Lovely JK, Mansukhani MP, Morgenthaler TI. Zolpidem is independently associated with increased risk of inpatient falls. 

4. Wang PS, Bohn RL, Glynn RJ, Mogun H, Avorn J. Zolpidem use and hip fractures in older people. 

. 2001;49(12):1685-1690. doi: 10.1111/j.1532-5415.2001.49280.x.

5. Chang CM, Wu EC, Chen CY, et al. Psychotropic drugs and risk of motor vehicle accidents: a population-based case-control study. 

. 2013;75(4):1125-1133. doi: 10.1111/j.1365-2125.2012.04410.x.

6. Mahoney JE, Webb MJ, Gray SL. Zolpidem prescribing and adverse drug reactions in hospitalized general medicine patients at a Veterans Affairs hospital. 

7. Heydari M, Isfeedvajani MS. Zolpidem dependence, abuse and withdrawal: a case report. 

Articles

We found inappropriate prescribing of zolpidem, in terms of both guideline-discordant dosage and coprescribing with benzodiazepines, with female veterans affected more than male veterans.

Gender Differences in Prescribing of Zolpidem in the Veterans Health Administration

 The aim of this study was to evaluate whether veterans in Massachusetts receiving opioids and/or benzodiazepines from both Veterans Health Administration (VHA) and non-VHA pharmacies are at higher risk of adverse events compared with those receiving opioids at VHA pharmacies only.

 A cohort study of veterans who filled a prescription for any Schedule II through V substance at a Massachusetts VHA pharmacy. Prescriptions were recorded in the Massachusetts Department of Public Health Chapter 55 data set.

 The study sample included 16,866 veterans residing in Massachusetts, of whom 9238 (54.8%) received controlled substances from VHA pharmacies only and 7628 (45.2%) had filled prescriptions at both VHA and non-VHA pharmacies (“dual care users”) between October 1, 2013, and December 31, 2015. Our primary outcomes were nonfatal opioid overdose, fatal opioid overdose, and all-cause mortality.

 Compared with VHA-only users, more dual care users resided in rural areas (12.6% vs 10.6%), received high-dose opioid therapy (26.3% vs 7.3%), had concurrent prescriptions of opioids and benzodiazepines (34.8% vs 8.2%), and had opioid use disorder (6.8% vs 1.6%) (

 <.0001 for all). In adjusted models, dual care users had higher odds of nonfatal opioid overdose (odds ratio [OR], 1.29; 95% CI, 0.98-1.71) and all-cause mortality (OR, 1.66; 95% CI, 1.43-1.93) compared with VHA-only users. Dual care use was not associated with fatal opioid overdoses.

 Among veterans in Massachusetts, receipt of opioids from multiple sources was associated with worse outcomes, specifically nonfatal opioid overdose and mortality. Better information sharing between VHA and non-VHA pharmacies and prescribers has the potential to improve patient safety.

Fragmented coordination of care has been shown to be associated with adverse opioid-related outcomes.

The present study suggests that veterans receiving opioids at both Veterans Health Administration (VHA) and non-VHA pharmacies are at higher risk of opioid overdoses compared with veterans receiving prescriptions at VHA pharmacies only.

Findings have implications for patient safety and the Veterans Choice program, which offers services from non-VHA providers to veterans living outside of VHA catchment areas.

Managed care decision makers should query the dual use of pharmacy services—specifically with regard to opioids&mdash;in their systems of care and its association with opioid overdoses.

Concomitant use of opioids with benzodiazepines significantly increases the risk of adverse outcomes. Specifically, rates of overdose death among those who were codispensed benzodiazepines and opioid analgesics were 10 times higher than among those who were dispensed opioid analgesics alone in a prospective cohort study.

 Similarly, among veterans using opioids, coprescribing of benzodiazepines was associated with an increased risk of opioid-related overdoses, injuries, and mortality.

 In another veteran-based study, nearly half of drug overdose deaths occurred among those concurrently prescribed benzodiazepines and opioids.

 Effective prescription monitoring programs (PMPs) have been shown to change such prescriber behavior by bringing about a reduction in opioid prescribing.

 Robust PMPs, as measured by assessing local and federal laws for prescribing controlled substances, have shown to be associated with fewer opioid overdose deaths than weaker PMPs.

 However, strong laws cannot overcome fragmented or uncoordinated care, which can still be a barrier to the effective use of PMPs.

Fragmented care results in increased medical errors, hospital readmissions, emergency department (ED) visits, and healthcare costs.

 Specifically, this holds true in the area of pain management, where the risk of opioid misuse may be compounded when patients obtain opioid prescriptions from multiple prescribers and/or pharmacies.

 Such fragmented or uncoordinated care has been shown to be associated with adverse opioid-related outcomes, including hospital admissions associated with opioid use

The impact of care coordination on opioid-related outcomes has not been fully examined among veterans who have “dual care use” of pharmacy benefits both within and outside of the Veterans Health Administration (VHA).

 Veterans may have a higher risk of overdose given their prevalence of chronic pain disorders, which are often treated with opioids,

 and that risk might be further aggravated with the coprescribing of benzodiazepines.

A recent cross-sectional study using the Kentucky PMP found that compared with those with VHA payments only, veterans with multiple payment sources for opioid prescriptions were more likely to receive risky opioid therapy, defined as combination opioid/benzodiazepine or high-dose opioid therapy.

 Besides the fact that it did not examine whether veterans were more likely to experience adverse events, the study also examined VHA as a payer, whereas our current study examines it as a dispenser.

Here, we fill this void in the literature by reporting on whether Massachusetts veterans receiving opioids and/or benzodiazepines from VHA and non-VHA pharmacies are at higher risk of nonfatal opioid overdose, fatal opioid overdose, or all-cause mortality compared with those receiving these prescriptions from VHA pharmacies only. Further, we report on whether risk is associated with the number of transitions between VHA and non-VHA systems.

The study was conducted on a cohort of veterans with opioid and/or benzodiazepine prescriptions recorded in the Massachusetts Department of Public Health (MDPH) Chapter 55 data set.

Chapter 55 Acts of 2015 mandated the analysis of data from several Massachusetts government agencies to report trends on fatal and nonfatal opioid overdoses. The database included 16 administrative sources covering approximately 98% of the Massachusetts population 11 years and older. In our study, we used 6 data sets: PMP, Massachusetts All-Payer Claims Database (APCD), Acute Care Hospital Case Mix Database, death records, Massachusetts Ambulance Trip Record Information System, and Office of the Chief Medical Examiner toxicology data. The PMP maintains information on filled prescriptions for Schedule II through V controlled substances from Massachusetts clinics, hospitals, and retail pharmacies, including out-of-state deliveries to Massachusetts residents. It also includes prescriptions by mail order pharmacies that deliver to patients residing in Massachusetts. Our study data span from October 2013, when VHA data were first systematically reported to the PMP, until December 2015. Other VHA data beyond those reported to the PMP (eg, demographics, diagnosis codes, laboratory test results, or pharmacy records) were not available to us because they have not been linked to the Chapter 55 data sets. The Chapter 55 initiative was mandated by law and conducted by a public health authority. No institutional review board (IRB) review was required by MDPH, and it was deemed exempt from research review by the Bedford VA Medical Center IRB.

We defined a veteran as anyone filling a prescription for any Schedule II through V substance at a Massachusetts VHA pharmacy (19,479 veterans). Prescriptions for Massachusetts residents filled outside the state were not included. Further, those with a non-Massachusetts residential zip code (n = 201) or insufficient prescription data to ascertain dual care status (n = 2412) were excluded, resulting in a sample of 16,866 veterans. No veterans were excluded due to age, race, or gender.

Veterans were categorized into those filling prescriptions for opioids and/or benzodiazepines at VHA pharmacies only (“VHA-only”; n = 9238) or those filling such prescriptions at both VHA and non-VHA pharmacies (“dual care users”; n = 7628). Opioids used for medication-assisted therapy for opioid use disorder, such as buprenorphine, were included (

]). For veterans with dual care use, we quantified the number of switches between VHA and non-VHA fills as a measure of the extent of discoordination in the patients’ care. The index date was defined as the earliest date during the study period that a veteran filled an opioid or benzodiazepine prescription, whether inside or outside of the VHA.

We examined 3 outcomes after the index prescription date: nonfatal opioid overdose, fatal opioid overdose, and all-cause mortality. Nonfatal overdose was identified in 2 ways. First, any individual who had an ambulance encounter related to opioid overdose was included. The algorithm that was used to identify opioid-related overdoses in the emergency medical system data was the result of collaboration between MDPH and the CDC.

 The second was an ED visit, outpatient observation, or inpatient hospital discharge with an 

 code containing a diagnosis code for opioid poisoning. Fatal opioid-related overdoses were defined using the 

International Classification of Diseases, Tenth Revision

 codes for mortality. These multiple cause-of-death fields were then used to identify an opioid-related death. All-cause deaths were identified using death certificates that are filed with the Massachusetts Registry of Vital Records and Statistics and contain the official cause of death and manner of death assigned by physicians and medical examiners. All outcomes were chronologically sequenced to ensure that none occurred before the index opioid use date.

Covariates included age, gender, high-dose opioid therapy (defined as exceeding 50 morphine milligram equivalents per day, on average, in at least 1 month),

 concurrent opioid/benzodiazepine use (defined by overlapping prescriptions of at least 1 day),

 and rural status of pharmacy and residence (defined as having a rural designation in Massachusetts). Demographics were gathered from all Chapter 55 data sources; when a conflict was found, it was resolved based on a hierarchy of reliability developed by the Chapter 55 team. The PMP data provided pharmacy location, generic codes of drugs, quantity, and dose.

 Additional information was obtained from the APCD on homelessness, which is almost certainly an undercount of all persons experiencing homelessness, and comorbidities, including the mental and physical components of the Elixhauser comorbidity index (eAppendix Table).

 Although the APCD contains health and pharmacy insurance claims data from across the state, it does not include service records from the VHA. Because of concerns that this may lead to differential completeness of data between the study groups, addition of homelessness and comorbidities in the form of an index (which would not have been available to us for VHA-only patients) was analyzed in a separate model.

Covariates and outcomes were compared between VHA-only and dual care users, using χ

 tests to detect differences. Logistic regression models were constructed for the 3 outcomes, including (1) an unadjusted model and (2) an adjusted model including terms for demographics, high-dose opioid therapy, and concurrent opioid/benzodiazepine use. A third model also included terms for homelessness and Elixhauser comorbidity index. These models were repeated, limiting to veterans who received opioids only. We also examined interaction terms between dual care use and homelessness because of a suspected heterogeneity of effect. Odds ratios (ORs) generated from these logistic regression models, although not entirely intuitive to some readers, are generally similar to more intuitive measures when examining rare outcomes. Analyses were performed using SAS Studio version 3.5 (SAS Corporation; Cary, North Carolina).

Descriptive characteristics of veterans were compared across groups (

). Compared with VHA-only users, more dual care users resided in rural areas (12.6% vs 10.6%), received high-dose opioid therapy (26.3% vs 7.3%), had concurrent prescriptions of opioids and benzodiazepines (34.8% vs 8.2%), had documented opioid use disorder (6.8% vs 1.6%), and were homeless (1.5% vs 0.7%), and they had higher mean physical (3.0 vs 1.3) and mental (0.6 vs 0.2) comorbidity scores (

 <.0001 for all). In terms of outcomes, dual care users had more nonfatal overdoses (160 [2.1%] vs 114 [1.2%]) and higher all-cause mortality (711 [9.3%] vs 390 [4.2%]) (

All 3 study outcomes were more common among dual care users in the unadjusted models (

). Adjustment for demographics, high-dose opioid therapy, and concurrent opioid/benzodiazepine use attenuated the estimates for nonfatal overdose (OR, 1.29; 95% CI, 0.98-1.71), fatal overdose (OR, 1.32; 95% CI, 0.64-2.70), and all-cause mortality (OR, 1.66; 95% CI, 1.43-1.93). Adjustment for homelessness and comorbidities further attenuated the point estimates (data not shown). There was no significant interaction between dual care use and homelessness. The number of switches between VHA and non-VHA prescriptions was not associated with any outcome. Similar effects were obtained for dual care users with opioid prescriptions only.

The study is the first collaborative effort of MDPH and VHA to examine VHA and non-VHA care coordination among veterans at risk of opioid overdose. Findings suggest that those receiving opioid and/or benzodiazepine prescriptions from both VHA and non-VHA pharmacies have higher odds of nonfatal opioid overdose and all-cause mortality compared with veterans receiving controlled substances from the VHA only. Although we found stronger effects of dual care use on nonfatal opioid overdose and all-cause mortality in unadjusted analyses, the multivariate adjustment process reduced this risk, as the variables controlled for may be inherently related to the mechanism of harm. For example, adjusting for rural residence of veterans, shown to be associated with both dual care

 (the outcome), might have lessened the risk of nonfatal opioid overdose in multivariate models. Although we found that the risks of opioid-related deaths among dual care users did not reach statistical significance, potentially due to the small number of events, there remains concern that this group is at higher risk of opioid-related mortality. Further, as is evident from the striking differences in baseline characteristics between the 2 groups in regard to receipt of high-dose opioid therapy (26.3% vs 7.3%), concurrent prescriptions for opioids and benzodiazepines (34.8% vs 8.2%), and opioid use disorder (6.8% vs 1.6%), it may be that individuals who seek dual sources of medications are often essentially different than those who do not (and that covariation cannot account for that) and that is the causal direction.

Our finding of association between dual care use and adverse outcomes in Massachusetts veterans is consistent with a previous report suggesting that having multiple payment sources, including VHA, cash, and noncash (Medicare, Medicaid, and private insurance), was associated with increased risky patterns of opioid therapy (overlap with benzodiazepines and/or high-dose opioid therapy) after controlling for age and sex in veterans in the Kentucky PMP.

 Our study went further, however, by examining outcomes and adjusting for a wider range of potentially confounding variables. Associations of dual pharmacy use with adverse opioid-related outcomes have also been reported in non-VHA populations. Multiple pharmacy use (defined as >2 pharmacies) predicted opioid overdose in Medicaid patients,

 and obtaining opioid prescriptions from multiple healthcare providers was associated with higher rates of hospital admission related to opioid use.

The proportion of dual care users between VHA and non-VHA pharmacies in our study was approximately 45%. In a recent study, the prevalence of veterans receiving prescription opioids from both VHA and Medicare Part D was reported to be 13.2%.

 Although fatal opioid overdose was not significantly associated with dual care use in our model, possibly due to limited statistical power, we found that a higher proportion of dual care users had opioid-related mortality compared with VHA-only users (0.30% vs 0.18%). In the context of these findings, our study highlights the importance of PMPs as a potential tool to reduce fractured care, especially regarding opioids, among multiple providers, pharmacies, and healthcare systems both within and across VHA and non-VHA facilities. Through the implementation of its Opioid Safety Initiative in 2013, VHA required contribution of its controlled substance prescribing and dispensing data to state PMPs.

 In turn, PMPs have been found effective in both reducing adverse events associated with opioid use

 and reducing prescriptions from multiple providers.

 Further, the statutory requirements for VHA under the Comprehensive Addiction and Recovery Act (CARA; PL 114-198) enacted in July 2016 mandate a designated Pain Management Team, consisting of healthcare professionals at each facility, responsible for coordinating and overseeing pain management therapy for patients experiencing acute and chronic pain that is not cancer-related. Pursuant to CARA and VHA Directive 1306,

 in October 2016 it was further required that state PMP databases are queried for VHA patients who are receiving prescriptions for controlled substances on a minimum of an annual basis and that the results of those queries are documented in the VHA medical record.

An important strength of our study was the use of a rich and unique data set, which captured prescribing of opioids both inside and outside of the VHA. To our knowledge, no other data resource would have supported such an analysis. Further, this study addresses a key issue, dual care use, which is leading to greater likelihood of uncoordinated care among veterans across the United States compared with commercially insured civilians who benefit from PMPs and other strategies known to work in anyone at high risk of overdose. However, like any state, Massachusetts has its own cultural, demographic, and socioeconomic environment, possibly limiting the generalizability of these findings to other states. More importantly, findings from this veteran study cannot be generalized to the general population. Further, we lacked complete information on medical utilization and comorbidities in our VHA-only group, which could confound the associations. Our attempts to include them in the analysis were limited by the recognition that differential completeness in the data between VHA-only patients and dual care use patients may result in bias and that their inclusion in the models might not lead to accurate results. Another limitation of the Chapter 55 data set was that pharmacy-specific variables such as distance to the pharmacy and number of non-VHA pharmacies could not be included in the model due to lack of patient address, geocoding of pharmacies, and unique pharmacy identifier. Further, in our study, we did not exclude patients with a cancer diagnosis. The findings of this study investigating opioid overdose may not be completely generalizable to the cancer population. A final limitation is that nonfatal overdoses may be underascertained in the study, as not all nonfatal opioid overdoses may involve an encounter with the healthcare system.

As the VHA is expanding its use of non-VHA care and providers through the Veterans Choice program,

 compounding the issue of coordination of care across systems, these findings are timely. The present study expands our understanding of opioid-related outcomes. Findings suggest the need not only to continue to share data between VHA and state PMPs but also to take further steps. These steps could include implementing prescription drug disposal, safe opioid prescribing education, aggressive dispensing of naloxone to veterans and their families, including specific alerts in PMPs regarding patients with previous histories of dual care use, and providing these dual care users more resources and closer care coordination, especially with regard to opioids.&ensp;

The authors acknowledge the participation and contributions of Dr Keith McInnes, Dr Avron Spiro, the late Dr James F. Burgess Jr, Dr Craig Coldwell, and Dr Karen Drexler. They acknowledge the Massachusetts Department of Public Health for creating the unique cross-sector database used for this project and for providing technical support for the analysis.

Center for Healthcare Organization and Implementation Research (CHOIR), Edith Nourse Rogers Memorial VA Medical Center (GKJ, DRM, DRB, DAS), Bedford, MA; Department of Health Law, Policy and Management, Boston University School of Public Health (GKJ, OA, DRB), Boston, MA; Executive Office of Health and Human Services, Department of Public Health, The Commonwealth of Massachusetts (TL, DB), Boston, MA; University of Massachusetts School of Medicine (TL), Worcester, MA; Department of Medicine, Section of General Internal Medicine, Boston University School of Medicine (AJR), Boston, MA; Department of Psychiatry, University of Massachusetts Medical School (DAS), Worcester, MA.

 The research reported/outlined here was supported by the Department of Veterans Affairs, Veterans Health Administration, Health Services Research and Development (HSR&D) Service. Dr Jasuja is a VA HSR&D Career Development awardee at the Bedford VA (CDA 13-265). The views expressed in this article are those of the author(s) and do not necessarily represent the views of the Department of Veterans Affairs.

The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and the decision to submit the manuscript for publication.

 Concept and design (GKJ, OA, DRM, TL, DB, AJR, DRB, DAS); acquisition of data (GKJ, OA, TL, DB, DRB, DAS); analysis and interpretation of data (GKJ, OA, DRM, TL, AJR, DRB, DAS); drafting of the manuscript (GKJ, OA, TL, DB, DAS); critical revision of the manuscript for important intellectual content (GKJ, OA, DRM, TL, DB, AJR, DRB); statistical analysis (OA, DRM, TL); administrative, technical, or logistic support (TL); and supervision (GKJ).

1. Dasgupta N, Funk MJ, Proescholdbell S, Hirsch A, Ribisl KM, Marshall S. Cohort study of the impact of high-dose opioid analgesics on overdose mortality [erratum in 

. 2016;17(4):797-798. doi: 10.1093/pm/pnw044]. 

. 2016;17(1):85-98. doi: 10.1111/pme.12907.

2. Gressler LE, Martin BC, Hudson TJ, Painter JT. Relationship between concomitant benzodiazepine-opioid use and adverse outcomes among US veterans. 

. 2018;159(3):451-459. doi: 10.1097/j.pain.0000000000001111.

3. Park TW, Saitz R, Ganoczy D, Ilgen MA, Bohnert AS. Benzodiazepine prescribing patterns and deaths from drug overdose among US veterans receiving opioid analgesics: case-cohort study. 

. 2015;350:h2698. doi: 10.1136/bmj.h2698.

4. Bao Y, Pan Y, Taylor A, et al. Prescription drug monitoring programs are associated with sustained reductions in opioid prescribing by physicians. 

. 2016;35(6):1045-1051. doi: 10.1377/hlthaff.2015.1673.

5. Pardo B. Do more robust prescription drug monitoring programs reduce prescription opioid overdose? 

. 2017;112(10):1773-1783. doi: 10.1111/add.13741.

6. Bodenheimer T. Coordinating care—a perilous journey through the health care system. 

. 2008;358(10):1064-1071. doi: 10.1056/NEJMhpr0706165.

Among veterans in Massachusetts, receipt of opioids from multiple sources, with or without benzodiazepines, was associated with worse opioid-related outcomes.

Guneet K. Jasuja, PhD

Articles

Gender Differences in Prescribing of Zolpidem in the Veterans Health Administration

Overdose Risk for Veterans Receiving Opioids From Multiple Sources