Stacy A. Sterling, DrPH, MSW

 Concomitant excessive alcohol consumption and benzodiazepine use is associated with adverse health outcomes. We examined associations of unhealthy alcohol use and other patient characteristics with benzodiazepine use.

 A cross-sectional analysis of 2,089,525 Kaiser Permanente of Northern California outpatients screened for unhealthy alcohol use in primary care between November 1, 2014, and December 31, 2016.

 We fit multivariable generalized linear models to estimate the associations between unhealthy alcohol use and benzodiazepine dispensation and, among patients who were dispensed a benzodiazepine, mean doses (in mean lorazepam-equivalent daily doses [LEDDs]) and prescription durations. We controlled for patient sex, age, race/ethnicity, estimated household income, Charlson Comorbidity Index (CCI) score, anxiety disorder, alcohol use disorder, insomnia, musculoskeletal pain, and epilepsy.

 In the 12 months centered around (6 months before and 6 months after) the first alcohol-screening visit, 7.5% of patients used benzodiazepines. The following characteristics were independently associated with higher rates of benzodiazepine use, higher LEDD, and longer prescription duration: older age, white race/ethnicity, lower estimated household income, higher CCI score, and the presence of an anxiety disorder, insomnia, musculoskeletal pain, or epilepsy. Women and patients with an alcohol use disorder or unhealthy alcohol use, compared with men and patients with low-risk drinking or abstinence, were more likely to use a benzodiazepine; however, their LEDDs were lower and their prescription durations were shorter.

 Benzodiazepine use in primary care was associated with older age, female sex, white race/ethnicity, lower socioeconomic status, and unhealthy alcohol use. These findings may be applied to develop policies and interventions to promote judicious benzodiazepine use.

Among patients in an integrated healthcare delivery system who were screened in primary care for unhealthy alcohol use, we examined cross-sectional benzodiazepine use patterns.

Unhealthy alcohol users, patients with an alcohol use disorder, and women were more likely to use benzodiazepines, but at lower doses and for shorter durations, compared with low-risk alcohol users and men.

Patients who were older, white, and of lower socioeconomic status were more likely to use benzodiazepines and had higher doses and prescription durations.

These findings may inform the development of systemwide interventions to address disproportionate benzodiazepine use among specific patient groups, especially among unhealthy alcohol users.

Benzodiazepines have been increasingly used in the United States, leading to increases in overdose mortality and healthcare utilization due to their potent sedative properties. Between 1996 and 2013, the percentage of US adults who filled a benzodiazepine prescription increased from 4.1% to 5.6%, and overdose mortality involving benzodiazepines increased from 0.58 to 3.07 per 100,000 adults.

 In a large primary care sample, 15% of patients were prescribed a benzodiazepine, one-third of whom received high-dose benzodiazepines (ie, a daily dose equivalent of ≥30 mg/day of diazepam); high-dose recipients were more likely than low-dose recipients to have multiple medical comorbidities and high healthcare utilization.

 Furthermore, between 2003 and 2015, the percentage of outpatient medical visits including a benzodiazepine prescription doubled, and more than half of these visits were to primary care physicians.

Between 7.5% and 20% of primary care patients acknowledge unhealthy alcohol use,

 defined as use exceeding recommended weekly or daily limits.

 Excessive alcohol use is associated with numerous individual and societal consequences, such as development or exacerbation of medical and mental health comorbidities, decreased medication adherence, increased HIV risk behaviors, motor vehicle crashes, and interpersonal violence.

 Unhealthy alcohol use may precede development of an alcohol use disorder

 and frequently goes undetected in routine clinical care.

 In this context, the US Preventive Services Task Force recommends screening all adult primary care patients for unhealthy alcohol use and offering patients with unhealthy alcohol use brief behavioral counseling.

When benzodiazepines and alcohol are used concurrently, their sedative effects significantly increase the risk of adverse events, including fatal overdose.

 In 2010, alcohol was involved in 27.2% of benzodiazepine-related visits and 21.4% of benzodiazepine-related deaths in US emergency departments.

 Long-term consequences of combined benzodiazepine and alcohol use include cardiovascular, gastrointestinal, hepatic, kidney, and neurologic injury and exacerbation of psychiatric conditions.

Nationally representative survey data from 2015-2016 suggest that past-year alcohol use or an alcohol use disorder is associated with an increased risk of concurrent benzodiazepine use, misuse, and use disorder.

 Similarly, among Medicare beneficiaries in 2002, presence of drug or alcohol use or dependence was associated with higher odds of benzodiazepine use.

 However, to our knowledge, no health system, primary care—based studies have examined patient characteristics associated with benzodiazepine dispensation, mean dose, and prescription duration. This information may inform the development of tailored interventions to reduce exposure to benzodiazepines among patients with unhealthy alcohol use.

In this study, we examined the prevalence of benzodiazepine use and its associations with patient demographic and clinical characteristics, including unhealthy alcohol use, among Kaiser Permanente of Northern California (KPNC) adult outpatients who were screened for alcohol use in primary care clinics. We hypothesized that people drinking at unhealthy levels, compared with low-risk drinkers and abstainers, would be less likely to receive benzodiazepines and would receive prescriptions of lower doses and shorter durations. Although this contrasts with previous national trends, we postulated that the combination of recent warnings about rising benzodiazepine use in the United States, concerns about concurrent use with other substances, and a primary care initiative in our healthcare system to detect and reduce unhealthy alcohol use would sensitize clinicians to reconsider benzodiazepine prescribing in this patient subgroup.

We conducted a retrospective, cross-sectional study using electronic health record (EHR)—derived data among adult outpatients screened for alcohol use in KPNC primary care clinics. KPNC is an integrated healthcare delivery system with more than 4 million members that covers approximately 30% of the population in the served geographic areas. For these analyses, we identified KPNC members 18 years or older who were screened for unhealthy alcohol use with 1 or more completed alcohol screenings as part of an ongoing primary care–based alcohol screening and brief intervention protocol (Alcohol as a Vital Sign [AVS])

 between November 1, 2014, and December 31, 2016. We defined the study period as the 12-month period centered around (6 months before and 6 months after) each patient’s clinic visit in which the first AVS screening occurred. We excluded patients with noncontinuous KPNC coverage during the study period (

]). Use of these data was approved by the institutional review boards of KPNC and the University of California, San Francisco.

We included outpatient dispensations for benzodiazepines filled at KPNC pharmacies in the 12 months centered around a patient’s first AVS screening visit in a primary care clinic. We extracted data for oral formulations of all benzodiazepine-containing medications (ie, alprazolam, chlordiazepoxide, clobazam, clonazepam, clorazepate, diazepam, estazolam, flurazepam, lorazepam, midazolam, oxazepam, temazepam, and triazolam) dispensed during the study period. For each patient, we assessed the following prescription attributes: (1) whether a benzodiazepine had been dispensed; (2) the duration of the benzodiazepine prescriptions (calculated by summing the total number of days supplied in the study period

); and (3) the mean daily benzodiazepine dose, expressed in mean lorazepam-equivalent daily dose (LEDD), which is an average over the 12-month study period (eAppendix Methods and 

Patient sex, age, and race/ethnicity (Asian, black/African American, Hispanic, white, and other/unknown) were extracted from the EHR at the time of the first AVS screening. We estimated household income from the United States Census 2010 median household income data by geocoding patients’ residential addresses to Census blocks.

International Classification of Diseases, Ninth Edition 

International Classification of Diseases, Tenth Edition, Clinical Modification 

codes, we assessed for the presence of the following encounter-associated codes in the study period: anxiety disorders, insomnia, musculoskeletal pain, alcohol use disorders (including alcohol-related psychoses, dependence, and abuse; excluding conditions in remission), and epilepsy (

). In addition, we estimated patients’ medical comorbidity burden using the Charlson Comorbidity Index score, which estimates 1-year mortality risk based on a weighted score of 17 conditions.

To determine whether a patient engaged in unhealthy alcohol use, we used EHR-documented alcohol screening data collected as part of the AVS protocol during the study period. During the primary care visit “rooming” process, medical assistants asked all patients to estimate within the past 90 days (1) the mean number of days per week they consumed alcohol and (2) how many alcoholic drinks they consumed “on a typical drinking day.” Using these variables, the EHR calculates the mean number of alcoholic drinks consumed weekly (ie, mean number of drinking days per week multiplied by the number of drinks consumed on a typical drinking day).

 Based on guidelines established by the National Institute on Alcohol Abuse and Alcoholism,

 we defined drinking severity as “low-risk alcohol use” (0-14 drinks per week for men ≤65 years or 0-7 drinks per week for all women and for men >65 years) or “unhealthy alcohol use” (>14 drinks per week for men ≤65 years or >7 drinks per week for all women and for men >65 years).

 and Wilcoxon rank-sum tests to examine differences in categorical and continuous patient demographic and clinical characteristic variables (including presence of unhealthy alcohol use) by whether the patient filled a prescription for a benzodiazepine in the 12 months centered around their first AVS screening visit. We fit multivariable logistic regression models to estimate the adjusted odds ratios (AORs) and corresponding 95% CIs of being dispensed a benzodiazepine by patient characteristics. Among patients who were dispensed a benzodiazepine during the study period, we fit log-linear and negative binomial regression models (accounting for positive skew and overdispersion of the data) to estimate the adjusted rate ratios (ARRs) of mean LEDD and duration of their prescriptions, respectively, by patient characteristics. All analyses were performed using SAS version 9.4 (SAS Institute Inc; Cary, North Carolina). Significance was assessed at 2-sided 

Approximately 87% of eligible KPNC members during the defined study period completed 1 or more AVS screenings, and 70% of the benzodiazepine prescriptions were ordered by a primary care provider (ie, internal, family, preventive, or adolescent medicine or pediatrics). Among KPNC members included in these analyses, approximately 97% had a continuous prescription benefit for KPNC pharmacies during the 12-month study period. The analytic cohort consisted of 2,089,525 patients (median [interquartile range] age = 49.0 [29.0] years; 53.7% female; 50.0% white; 4.0% unhealthy alcohol users) after excluding patients without continuous KPNC coverage in the 12 months centered around the first AVS screening (n = 803,991) and patients with missing Census-derived estimated household income data (n = 1390) (

In the study period, 157,449 patients (7.5% of the study sample) filled a prescription for a benzodiazepine (Table 1). Patients with unhealthy alcohol use had slightly higher odds of using benzodiazepines compared with patients with low-risk alcohol use (AOR, 1.15; 95% CI, 1.12-1.19) (

). From the same multivariable analysis, women had 64% higher odds of using a benzodiazepine compared with men (AOR, 1.64; 95% CI, 1.62-1.66). Conversely, nonwhite patients had significantly lower odds of using benzodiazepines compared with white patients. Among racial/ethnic groups, Asian patients were the least likely to use benzodiazepines, with an AOR of 0.38 (95% CI, 0.37-0.38), representing 62% lower odds of using benzodiazepines compared with white patients. Older patients had significantly higher odds of using benzodiazepines compared with younger patients aged 18 to 39 years, with AORs of 1.82 (95% CI, 1.79-1.84) and 1.77 (95% CI, 1.73-1.80) for patients aged 40 to 65 and older than 65 years, respectively. Anxiety disorders, insomnia, musculoskeletal pain, epilepsy, and alcohol use disorder were all significantly associated with benzodiazepine use.

Mean Daily Dose and Duration of Benzodiazepine Prescriptions

Among patients with benzodiazepine prescriptions during the study period, the mean (SD) LEDD was 0.58 (1.51) mg/day and mean (SD) duration was 87.9 (114.0) days. Patients with unhealthy alcohol use had 40% lower LEDDs (ARR, 0.60; 95% CI, 0.55-0.66) and 18% shorter durations (ARR, 0.82; 95% CI, 0.80-0.84) compared with patients with safe alcohol use (

). Women had ARRs of 0.85 (95% CI, 0.83-0.87) and 0.92 (95% CI, 0.90-0.93), representing a 15% lower LEDD and an 8% shorter duration of benzodiazepine prescriptions, respectively, compared with men. Asian patients had the lowest LEDD and shortest prescription durations of any racial/ethnic group compared with white patients. Whereas LEDD was highest among patients aged 40 to 65 years compared with patients aged 18 to 39 years (ARR, 1.49; 95% CI, 1.43-1.55), prescription durations were the longest among patients 65 years or older compared with patients aged 18 to 39 years (ARR, 2.04; 95% CI, 2.00-2.08). Higher LEDD and longer prescription durations were also significantly associated with low (compared with high) Census-derived estimated household income and the presence of an anxiety disorder, insomnia, musculoskeletal pain, and epilepsy.

This study is the first to our knowledge to examine the association of unhealthy alcohol use with benzodiazepine use, dosage, and prescription duration in a large primary care sample. Notably, 7.5% of patients in this cohort were dispensed a benzodiazepine in a 12-month period.

 In addition, women, older adults, and white patients were more likely to use benzodiazepines.

The rate of unhealthy alcohol use of 4.0% in our sample is lower than estimates from other health maintenance organization—based primary care samples (7.5%-20%)

 but is within the range estimated in population-based survey samples (0.3%-20%).

 This discrepancy may be attributable to different populations, years of data collection, and definitions of unhealthy alcohol use. Similarly, the prevalence of filled benzodiazepine prescriptions in our sample (7.5%) is consistent with a cross-sectional analysis of US retail pharmacies (5.2% of adults filled a prescription for a benzodiazepine in 2008)

 and with data from the Medical Expenditure Panel Survey (the prevalence of filled benzodiazepine prescriptions between 1996 and 2013 among noninstitutionalized adults increased from 4.1% to 5.6%).

 However, among adults in a primary care practice—based research network in 2011 and 2012, 15% of patients were prescribed a benzodiazepine.

 This larger percentage may reflect ordered prescriptions instead of filled prescriptions; the number of ordered prescriptions is likely to be greater than the corresponding number of filled prescriptions.

Comparing our results regarding the concomitant use of benzodiazepines and alcohol with other samples is challenging given the methodologic heterogeneity among previous studies. For instance, among psychiatric outpatients (n = 93), roughly 40% used a benzodiazepine and simultaneously consumed alcohol; furthermore, alcohol use severity was associated with higher odds of benzodiazepine use (OR, 2.4; 95% CI, 1.3-4.2).

 In a study examining nationally representative survey data from 1999 to 2002, Jalbert and colleagues

 found that 8% to 10% of adult respondents reported both benzodiazepine and unhealthy alcohol use. Consistent with our findings, Kroll and colleagues

 examined data from a large primary care sample and found that patients with a diagnosis of alcohol abuse (compared with those without) had higher odds of benzodiazepine use (OR, 1.5; 95% CI, 1.3-1.7). In the emergency department setting, Jones and colleagues

 found that 27.2% of benzodiazepine-related visits involved alcohol. However, in smaller studies involving older patients

 the prevalence of concomitant benzodiazepine and unhealthy alcohol use (approximately 3% in these 2 samples) was significantly lower than what we found in this sample.

Clearly, previous analyses of concomitant alcohol and benzodiazepine use vary widely in their study populations, periods, and definitions of unhealthy alcohol consumption and benzodiazepine use. However, in general, the existing literature supports our conclusion that unhealthy alcohol use is associated with a greater likelihood of concomitant benzodiazepine use.

 Our study adds to the literature by examining not only benzodiazepine use but also mean doses and prescription durations among subpopulations of primary care patients. Interestingly, we found that although patients with unhealthy alcohol consumption were more likely to use benzodiazepines, the benzodiazepines were prescribed to them at lower doses and for shorter durations. This pattern persisted even when accounting for the presence of an alcohol use disorder, for which benzodiazepines may be prescribed to manage alcohol withdrawal. These findings fail to confirm the first part of our hypothesis (ie, patients with unhealthy alcohol use are less likely to use benzodiazepines) but do confirm the second part of our hypothesis (ie, that these patients would use benzodiazepines for shorter periods and at lower doses). One possible explanation for this pattern of findings is that those drinking at unhealthy levels may be more likely to use multiple abusable substances,

 but that clinicians, aware of the contraindications of concomitant use, limit the dose and duration of benzodiazepines for these patients. It is also possible that patients with unhealthy alcohol use may voluntarily limit their use of benzodiazepines to avoid functional impairment associated with concurrent alcohol and high-dose benzodiazepine consumption.

We also found, similar to results of surveys

 that women were more likely, and nonwhite patients less likely, to use benzodiazepines. Furthermore, we found that women and nonwhite patients who used benzodiazepines used them at lower doses and for shorter durations compared with men and white patients who used benzodiazepines.

 Women may be more likely to seek treatment for an anxiety disorder.

 Therefore, women may be more likely than men to report anxiety relief and to require lower doses of benzodiazepines in part because they are simultaneously receiving nonbenzodiazepine treatments (eg, antidepressants, psychotherapy).

 Simultaneously, women may use benzodiazepines at lower doses and for shorter amounts of time compared with men in part because clinicians may believe sex moderates benzodiazepine metabolism, despite limited evidence to support this contention.

 Previous research demonstrates that nonwhite patients receive fewer benzodiazepine prescriptions than white patients; however, white patients were more likely to have a benzodiazepine dependence diagnosis.

 Nonwhite patients may be less trusting of the healthcare system, particularly as it relates to seeking help for mental health problems.

 White patients may be more assertive in requesting symptom relief

 compared with nonwhite patients, who may be less likely to seek medical help for psychiatric symptoms such as anxiety.

 Furthermore, some Asian patients may be more likely to seek nonallopathic remedies for mental health concerns in primary care.

 Symptoms of anxiety in nonwhite patients may also be underrecognized or undertreated by clinicians, resulting in fewer benzodiazepine prescriptions for nonwhite patients. These sex- and ethnicity-associated disparities in benzodiazepine prescribing may also be the result of implicit clinician biases.

These results should be interpreted in the context of several limitations. As with all observational studies, causality cannot be determined. Pharmacy data reflect dispensations and we are unable to measure patient medication consumption; nonetheless, dispensations are commonly used in pharmacoepidemiologic studies

 and, in the current sample, approximately 84% of prescribed benzodiazepines were filled in KPNC pharmacies. Additionally, our results may not be generalizable to uninsured populations. However, some of our findings are consistent with other, nationally representative samples (eg, lower benzodiazepine use among nonwhite patients

), suggesting that other findings in our study may also be applicable to the general US primary care adult population.

We are also unable to determine whether benzodiazepine use was inappropriate or unhealthy; subsequent studies may identify potentially inappropriate benzodiazepine use using algorithms, such as those used in geriatric samples.

 Likewise, by restricting our analyses to benzodiazepine use, we did not examine concurrent opioid use, especially among patients with chronic pain. Compared with patients who only use opioids, patients who use both opioids and benzodiazepines are at higher risk of overdose

; therefore, national guidelines recommend against prescribing benzodiazepines with opioids.

 Future studies may examine the association among benzodiazepine and opioid use with unhealthy alcohol consumption.

We included conditions frequently associated with benzodiazepine prescriptions; however, benzodiazepines may be prescribed for other disorders, such as anxious depression.

 Finally, statistically significant associations may be a product of the large sample size of our study; therefore, we have focused our conclusions on associations with larger AORs and ARRs, which are most likely to be clinically meaningful and successfully replicated in other primary care samples. If these associations are replicated in subsequent studies, they may be applied to develop and validate clinical decision support systems, such as a quantitative risk-stratification tool to identify patients at high risk of concomitant benzodiazepine and unhealthy alcohol use.

In this large primary care cohort, unhealthy drinking was associated with higher rates of benzodiazepine use. However, benzodiazepine mean dose was lower and duration of use was shorter among patients with unhealthy drinking. Nonetheless, these results suggest that concomitant benzodiazepine and excessive alcohol use among primary care patients should receive increased vigilance, and health system—wide efforts to reduce this potentially lethal combination should be considered.

 Department of Psychiatry and the UCSF Weill Institute for Neurosciences, University of California, San Francisco (MEH), San Francisco, CA; The Permanente Medical Group, Department of Psychiatry (MEH), Oakland, CA; Division of Research, Kaiser Permanente of Northern California (MEH, VAP, AHK-S, CIC, SAS), Oakland, CA; Indiana University School of Medicine (KK), Indianapolis, IN; Regenstrief Institute, Inc (KK), Indianapolis, IN; VA Health Services Research and Development, Center for Health Information and Communication (KK), Indianapolis, IN.

 This research was supported by a grant from the National Institute on Alcohol Abuse and Alcoholism (grant number R01-AA025902-01).

 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 (MEH, VAP, KK, SAS); acquisition of data (VAP); analysis and interpretation of data (MEH, VAP, KK, CIC, AHK-S, SAS); drafting of the manuscript (MEH, VAP, CIC); critical revision of the manuscript for important intellectual content (MEH, VAP, AHK-S, KK, CIC, SAS); statistical analysis (VAP, AHK-S); provision of patients or study materials (SAS); obtaining funding (SAS); administrative, technical, or logistic support (SAS); and supervision (SAS).

 Matthew E. Hirschtritt, MD, MPH, Department of Psychiatry and the UCSF Weill Institute for Neurosciences, University of California, San Francisco, 401 Parnassus Ave, Box 0984, San Francisco, CA 94143. Email: matthew.hirschtritt@ucsf.edu.

1. Bachhuber MA, Hennessy S, Cunningham CO, Starrels JL. Increasing benzodiazepine prescriptions and overdose mortality in the United States, 1996-2013. 

. 2016;106(4):686-688. doi: 10.2105/AJPH.2016.303061.

2. Kroll DS, Nieva HR, Barsky AJ, Linder JA. Benzodiazepines are prescribed more frequently to patients already at risk for benzodiazepine-related adverse events in primary care. 

. 2016;31(9):1027-1034. doi: 10.1007/s11606-016-3740-0.

3. Agarwal SD, Landon BE. Patterns in outpatient benzodiazepine prescribing in the United States [erratum in 

. 2019;2(3):e191203. doi: 10.1001/jamanetworkopen.2019.1203]. 

. 2019;2(1):e187399. doi: 10.1001/jamanetworkopen.2018.7399.

4. Fleming MF, Manwell LB, Barry KL, Johnson K. At-risk drinking in an HMO primary care sample: prevalence and health policy implications. 

. 1998;88(1):90-93. doi: 10.2105/ajph.88.1.90.

Articles

Among outpatients who were screened for alcohol use, those with unhealthy alcohol use, women, and those who were older, white, and of lower socioeconomic status were more likely to use benzodiazepines.

Benzodiazepine and Unhealthy Alcohol Use Among Adult Outpatients

 Behavioral health conditions and social problems are common yet underrecognized among emergency department (ED) patients. Traditionally, ED-based behavioral health screening is limited. We evaluated the feasibility of expanded behavioral health screening by a trained nonclinician.

 Prospective observational study of a convenience sample of ED patients.

 A research assistant (RA) approached a convenience sample of adult ED patients within an integrated healthcare delivery system. Patients completed a paper screening instrument (domains: mood, anxiety, alcohol use, drug use, sleep, intimate partner violence, and chronic pain) and reviewed responses with the RA, who shared positive screening results with the treating ED physician. We abstracted behavioral health and medical diagnoses from the electronic health record (EHR), comparing the screened cohort with the eligible population. We used χ

 tests to assess differences in demographics and comorbidities between screened patients and the eligible group and differences between self-reported symptoms and EHR diagnoses among screened patients.

 Among 598 screened patients, the prevalence of self-reported symptoms was higher than that of associated EHR diagnoses in the year prior to the ED visit (anxiety, 45% vs 19% [

 A dedicated RA was able to integrate screening into patient idle times in the ED visit. The prevalence of behavioral health problems was higher than indicated in the EHR.

Behavioral health problems (depression, anxiety, and alcohol and drug misuse) are common yet underrecognized in emergency department (ED) patients. These conditions contribute to other health conditions, such as diabetes and hypertension, but if they are undiagnosed, they go untreated. Improving our detection of these conditions can accelerate referral to and initiation of treatment and improve downstream health outcomes and costs.

Leveraging a nonclinician to perform screening in the ED was successful and nonintrusive.

The prevalence of behavioral health conditions was higher than identified in the electronic health record.

Further studies linking screening to pathways for referral to treatment and assessing quality outcomes would be helpful.

Research findings suggest that behavioral health conditions such as depression, anxiety, and alcohol and drug misuse, as well as intimate partner violence (IPV), are common among emergency department (ED) patients

 and that their prevalence is higher than typically noted by ED physicians and staff.

 These conditions are a source of considerable morbidity (eg, disability-adjusted life-years lost due to depression/anxiety), are associated with other common health conditions and poorer health outcomes, and can drive healthcare utilization and costs.

 Although effective interventions for these conditions exist, identification is essential to providing patients with appropriate referral and treatment.

Historically, widespread screening among ED patients has been limited to acute risk for suicide and IPV. However, the typical ED visit is interspersed with periods of idle time during which patients wait for test results, re-evaluation by clinicians, and treatment effects. These intervals create opportunities for screening, brief interventions, and referrals to treatment or other resources.

Previous behavioral health screening pilot implementation studies in the ED have focused primarily on alcohol and drug abuse and suggest feasibility, albeit with concerns about workload for ED providers and intervention fidelity.

 The screening instruments used in US-based studies were lengthy, and the pilot periods were short (1 week). A brief screening instrument that covers several key domains has not been tested in the ED setting. Furthermore, IPV, chronic pain, and sleep problems are important contributors to depression, anxiety, and alcohol and drug use.

 Screening, intervention, and referral for these conditions in ambulatory care settings such as primary care has increased detection and treatment initiation rates.

 However, patients may miss opportunities for screening and referral to treatment if their main point of contact with health services is the ED.

This study examined the feasibility of introducing screening for common behavioral health problems in the ED setting using an embedded research assistant (RA) and a brief screening instrument. We also examined the rates of documented diagnoses compared with self-reported problems covered in the screener among patients presenting for ED services in an integrated healthcare system.

Kaiser Permanente Northern California (KPNC) is a nonprofit integrated healthcare delivery system providing comprehensive healthcare services to more than 4 million members in northern California. KPNC provides integrated medical and specialty psychiatric and chemical dependency treatment within the health system. Members are racially and socioeconomically diverse and representative of the regional population.

 The study ED is the site of 65,000 annual visits and is staffed by more than 60 full-time board-certified or board-eligible physicians and 220 nurses.

KPNC members 18 years or older with no health plan enrollment gaps of greater than 3 months in the year prior to their index ED visit (during the study period, October 10, 2015, to June 12, 2016) were eligible to participate. Patients were initially eligible if they presented with non—life-threatening and nonminor complaints and were defined as having an Emergency Severity Index (ESI) score of 2 or 3. The ESI calculator is a commonly used triage algorithm for stratification based on acuity and predicted resource needs that grades patients from level 1 (most urgent) to level 5 (least resource intensive).

 Patients who were seen in the “fast-track” area of the ED for minor complaints (ESI score of 4 or 5) were considered ineligible because their length of stay was typically too short to allow for screening without interrupting workflow.

Over the study period, an RA approached ED patients who met eligibility criteria in their treatment rooms, 5 evenings per week (Monday through Friday), from approximately 3 pm to 10 pm. Patients verbally consented to participate and completed a paper screening tool covering the following domains: depression, anxiety, alcohol use, drug use, sleep, IPV, and chronic pain. The RA reviewed the screener with patients for clarity and confirmed affirmative responses. The RA also suggested to patients that they discuss positive results with their care team. The ED attending physician was notified of positive screening results, and the care team addressed the identified concerns based on clinical judgment.

This screening pilot was supplemental to currently practiced screening for IPV and emergent psychiatric conditions, which were separate from the study protocol. Current ED practice for addressing IPV and emergent psychiatric conditions includes consultation with psychiatry and mental health, as well as social work referral. Because the screener conditions did not require emergent intervention and because our study was designed to assess screening feasibility, we did not have mandated referral pathways but instead relied on patient—clinician and RA–clinician communication for next steps.

Because the RA was in the ED only during limited hours, our screened cohort comprised a convenience sample of eligible patients during the study period. Our RAs did not document patients who declined to participate, so the cohort consists of only those patients who agreed to participate.

Feasibility of the proposed screening intervention was assessed during the prestudy period through qualitative discussions with ED physicians, nursing staff, social workers, administrative staff, and behavioral health leadership. We observed ED clinical workflow in relation to screening activities to ensure minimal disruption of the ED visit. We also identified treatment team members with whom the RA would communicate before approaching patients to confirm that the screening activity would not delay treatment.

The study was approved by the KPNC Institutional Review Board.

 Patient age, race/ethnicity, gender, and psychiatric and medical comorbidities based on 

International Classification of Diseases, Ninth Revision

 diagnosis codes were extracted from the electronic health record (EHR). We identified psychiatric diagnoses (eg, alcohol or drug use disorders, depression, anxiety, and psychotic disorders; specific codes available on request) and other chronic disease diagnoses (Charlson Comorbidity Index [CCI]) in the year prior to each patient’s index ED visit. To ensure accurate ascertainment of these demographic and comorbidity variables, we included only patients without prolonged health plan membership gaps.

 All participants completed the same screening tool (

]). Questions from the National Institute on Alcoholism and Alcohol Abuse’s validated clinician screening guide were used,

 including the number of heavy drinking days (≥4 drinks for women, ≥5 drinks for men) in the past 3 months, drinking days per week, and typical number of drinks per drinking day. Drug use questions were based on the National Institute on Drug Abuse (NIDA)—validated NIDA Quick Screen

 and assessed past-year frequency (never, less than monthly, monthly, weekly, daily/almost daily) of marijuana use and of illegal drug use or prescription drug misuse.

The Patient Health Questionnaire (PHQ-2) is a validated 2-item symptom-based screening instrument based on 

Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition

 depression criteria, asking patients about mood symptoms in the prior 2 weeks.

 The General Anxiety Disorder questionnaire (GAD-2) is a validated 2-item anxiety symptom screening instrument.

Questions about IPV were informed by validated instruments such as the HARK (Humiliation, Afraid, Rape, Kick) and the HITS (Hurt, Insulted, Threatened with harm, Screamed at them) and recommended by the health system’s Director of Family Violence Prevention. These questions included whether the patient was currently in a relationship in which their partner hit, slapped, kicked, choked, or hurt them, or had threatened them, and whether they had ever had a partner who physically hurt or threatened them.

Chronic pain items included whether they had experienced chronic pain during the previous 6 months, and if so, whether they felt able to manage their pain well. Sleep questions included the average number of hours of sleep per night and whether they felt that amount was adequate. Because validated tools to assess sleep, IPV, and chronic pain are limited, we relied on questions recommended by KPNC clinical leaders for these domains.

Chi-square tests were performed to assess baseline differences in demographics and comorbidities between screened patients and the full cohort. All analyses were performed using SAS 9.3 (SAS Institute, Inc; Cary, North Carolina).

Over the study period, 598 of 773 patients who completed the ED screener met eligibility criteria and had complete data. We also identified from the EHR a group of 14,919 patients evaluated in the ED during the study period who met eligibility criteria (age ≥18 years, membership gaps <3 months, ESI level 2 or 3) but were not screened (

). Screened patients were largely similar to the eligible group with regard to psychiatric comorbidities, as indicated by EHR data, although they had slightly higher rates of depression (22% vs 18%, respectively; 

 = .018), and several chronic medical conditions (

Self-reported Symptoms Compared With Diagnoses

Except for IPV, the prevalence of self-reported behavioral health symptoms among those screened was higher than that of diagnoses made in the year prior to the index ED visit (

). Notably, almost half (45%) of patients endorsed anxiety symptoms, although only 19% had an anxiety diagnosis in the year prior (

 <.001). Similarly, 40% of patients endorsed depressive symptoms, although only 22% had a related diagnosis 

 <.001). Self-reported drug use (7%) and hazardous alcohol use (12%) were approximately double the prevalence of associated diagnoses (4% [

 <.001], respectively). The largest disparity between endorsed symptoms and clinical diagnoses was recorded for sleep disorders and associated symptoms (sleep disorder diagnosis, 4.2% vs self-reported sleep problems, 47%; 

 <.001). Chronic pain was the most common past-year diagnosis of all conditions assessed (30%), although self-reported chronic pain was significantly higher, in 47% of patients (

The RA was able to approach patients and offer screening for common behavioral health conditions within the workflow and timing of the ED visit. Communication with clinical providers (nurses and physicians) was important in identifying suitable times to approach patients and ascertaining that patients were medically stable and appropriate for screening. The RA either left the screener with the patient, returning for the completed survey at a later time, or, for patients who preferred it, would read the questions to the patient. The RA reviewed the responses and discussed questions with unclear or unmarked answers, as well as affirmative answers, with the patient.

Because our study examined feasibility of screening, we did not collect data on occurrence, rates, and outcomes of physician-initiated referrals. The total time per patient screened was approximately 15 minutes, including the time from introducing the screener, allowing time for self-administration, and reviewing results with the patient.

In this pilot study, we assessed the feasibility of adding a behavioral health screening instrument to the ED workflow and examined self-reported behaviors and symptoms compared with diagnoses documented in the EHR. Similar screening for alcohol and drug use, as well as depression, is performed in the ambulatory setting, but this pilot was the first attempt in this health system to perform this screening in the ED.

 ED visits represent an important opportunity to identify behavioral health problems, yet systematic screening for these rarely occurs beyond IPV and suicide risk. Opportunities exist in the idle times that patients experience during ED visits to expand screening to more occult behavioral health problems without disrupting workflow.

The screener we tested was not a diagnostic assessment instrument, and endorsement of symptoms does not constitute a diagnosis. Nevertheless, a considerable proportion of screened patients endorsed these symptoms, which adds to our understanding of unrecognized behavioral health problems among ED patients.

 We incorporated validated screening tools that are widely used in healthcare settings and have relatively high sensitivity and specificity for depression and anxiety: PHQ-2 and GAD-2 have specificities of 76% and 81% and sensitivities of 89% and 76%, respectively. This suggests that a substantial proportion of patients with positive screening would meet diagnostic criteria for these disorders.

The ED offers an opportune context for screening for behavioral health issues: Many patients come to the ED in crisis and may be more willing to reveal symptoms of distress that otherwise might remain unidentified and untreated. Additionally, some patients’ only contact with the health system is the ED, leaving these encounters as the sole opportunity to screen them and refer for treatment.

Recognizing that the prevalence of behavioral health symptoms in ED patients is higher than is currently identified is important because these symptoms cause considerable distress and morbidity on their own, exacerbate chronic health conditions, and may result in higher health services utilization and costs. Examples include the adverse effects of alcohol consumption on hypertension and of depression on post—myocardial infarction outcomes, healthcare utilization, and chronic medical conditions.

 Early identification of these comorbid conditions can help facilitate clinical attention or specialty treatment initiation.

 Depression, anxiety, and alcohol or drug misuse are known to adversely impact health outcomes, costs, and utilization. The prevalence of these symptoms that we observed in the ED suggests that expanded screening, with referral and treatment as indicated, would be of value at the individual and health system levels.

Understanding how to effectively implement screening into the ED workflow is challenging. To that end, we examined feasibility and found that having an ED-embedded RA conduct screening was possible and created minimal disruptions to normal clinical workflow. Questions remain about the scalability of systematic screening, which we were unable to address in the absence of additional research funding. Implementing systematic screening into regular ED operations during all 24 hours per day of its operations would require existing staff to assume screening responsibilities or additional personnel, both of which are resource allocation concerns for ED leaders. Clinical response workflows for positive responses (eg, brief interventions, information, referrals to specialty care and/or community resources) are also necessary. As previous studies have shown, large-scale implementation of ED-based Screening, Brief Intervention, and Referral to Treatment for drug and alcohol abuse resulted in widespread adoption and high referral and treatment initiation rates in Massachusetts, but long-term outcomes are unclear.

Technology could be leveraged to facilitate the integration of behavioral health screening into the ED workflow. A systematic review of technology-based behavioral health interventions in the ED found high levels of acceptability and feasibility, but limited evidence on efficacy, especially with limited measuring or reporting of clinical outcomes such as decreased IPV incidence or alcohol use.

 The optimal approach to behavioral health screening in the ED remains to be determined, but might include some mix of electronic screening with in-person interaction, intervention, and referral.

We found significantly higher levels of self-reported depression and anxiety symptoms, risky alcohol and drug use, sleep problems, and chronic pain compared with documented diagnoses. Although screening is by nature less specific than a formal assessment and diagnostic process, the magnitude of differences identified supports the findings of previous research suggesting that ED patients may have higher prevalence rates, and lower rates of detection, than the general population.

 Similar discrepancies between prevalence of diagnoses and positive screening for behavioral health conditions have also been found among primary care patients, especially among patients with lower utilization of health services.

 As such, the ED may be the only opportunity to screen for and detect behavioral health problems in this subset of patients with limited healthcare contact. Performing screening in the ED would leverage that healthcare contact to allow appropriate and timely referral to primary care, mental health, or drug and alcohol treatment.

Our screened cohort was not identical to the eligible population. The differences, however, were minor and to be expected with convenience sampling, limited hours, and our small sample (4% of similar ED patients). The 2 groups had small racial/ethnic and age differences, and the screened sample had more medical comorbidities, higher overall CCI scores, and higher prevalence of depression and panic disorders. Chronic medical conditions are known to be associated with increased prevalence of depression and anxiety disorders, which may partially explain the higher observed prevalence in the screened population.

 However, the observed rates of positive depression and anxiety screens are similar to those that have been reported in other studies of ED patients.

As this was a feasibility study, research staff hours were limited and we were unable to serially screen every eligible patient. We did not approach patients seen in the fast-track section of the ED, who represent up to 40% of ED patients, nor were patients with truly emergent medical problems screened. Because participation was voluntary, patients self-selected into or out of screening. All of these limitations introduce the potential for bias, and the screened sample may not be representative of all ED patients in both measured and unmeasured variables. Although we were unable to compare self-reported symptoms between the 2 groups, we did compare self-reported symptoms with their past-year diagnoses, and we found higher proportions of medical, depression, and panic disorder diagnoses among the screened group. It may be that eligible patients who were more severely ill spent longer in the ED and thus had more time to be approached for screening, or that their medical or mental health problems made them more willing to be screened.

Although this facility is subject to the same regulations governing all EDs in the United States and thus accepts all patients, for these analyses, we limited the sample to health system members in order to have access to their EHR data. Because this is a private, nonprofit healthcare delivery system, its population may not be representative of ED populations in public systems.

This feasibility study of expanded emergency department screening identified a high prevalence of behavioral health conditions. Screening was successfully integrated into emergency visit idle times.

Stacy A. Sterling, DrPH, MSW

Articles

Benzodiazepine and Unhealthy Alcohol Use Among Adult Outpatients

Feasibility of Expanded Emergency Department Screening for Behavioral Health Problems