Between 2006 and 2010, excessive alcohol consumption was responsible for 88,000 deaths and an estimated 2.5 million years of potential life lost each year. Risky alcohol use is among the leading preventable causes of death in the United States [1], but it remains highly prevalent and poorly intervened among the emergency department (ED) population. Prevalence rates of risky alcohol use among ED patients exceed the national average, making EDs an ideal location for innovative alcohol intervention efforts [2-5]. Over 130 million patients visit the ED each year [6], with a large percentage of these patients having unrecognized alcohol-related treatment needs (eg, risky drinking, problem drinking, and alcohol dependence). Patients with untreated needs are more likely to be admitted to the hospital and repeatedly rely on ED services [2,7-9]. Furthermore, disadvantaged populations including minorities, immigrants, people without insurance, and low-income households comprise the underserved populations that currently rely disproportionately on EDs for primary care [10] and suffer from alcohol-related concerns at elevated rates [11-12]. ED-originated alcohol intervention efforts have the potential for considerable impact on public health by promoting change in both high-risk and hard-to-reach populations [4,10,13].

This potential has been acknowledged by the latest health care legislation and numerous health care agencies. The Affordable Care Act includes strong incentives for the integration of behavioral health and medical treatment [2,14]. Numerous studies have demonstrated the effectiveness of screening and brief intervention (SBI) programs as well as screening, brief intervention, and referral to treatment (SBIRT) programs aimed at addressing alcohol use problems and treatment needs among ED patients [2,15-18]. As a result, the US Preventive Services Task Force and the Substance Abuse and Mental Health Services Administration have recommended universal SBIRT for alcohol and other substances in general medical settings, including EDs [19,20]. The Centers for Medicare and Medicaid Services and the American Medical Association have authorized billing codes to reimburse SBIRT services for alcohol, tobacco, and illicit drug use [21], and the Centers for Disease Control and Prevention have called for increased alcohol SBI, including systems-level changes to include integration into the electronic health record system [22].

Despite the support of numerous studies and many health agencies [2,14-23], ED-originated alcohol screening rates remain low, with many hospitals only screening alcohol toxicology reports, which do not assess problem-level severity [2,24,25]. Factors likely contributing to these low rates include lack of specialized behavioral health training, competing demands on time and resources inherent to the ED setting, and a primary objective of acute medical care rather than treatment for chronic conditions [5,13,23]. The use of convenient and brief procedures requiring minimal specialized training and focusing on connecting patients with outpatient resources for continued treatment after their ED visit could maximize successful implementation of SBIRT. Behavioral intervention technology advancement (eg, computerized assessments, personalized feedback reports, faxed referrals, electronic health records), holds promise for facilitating the implementation of SBIRT in EDs and a variety of health care settings [26-33]. Such advancements have also allowed for the development of computer-assisted SBIRT models designed to diminish interruptions in clinical care and mitigate clinician burden without sacrificing effectiveness [13,23].

Questions remain concerning the most effective and appropriate SBIRT model. The objective of this study was to assess an innovative Web-based program’s ability to facilitate alcohol SBIRT. The Health Evaluation and Referral Assistant (HERA) is patient-administered on a tablet computer during the ED visit and is modeled after the face-to-face SBIRT screening approach. This study hypothesized that the HERA would improve initiation of specialized outpatient treatment for risky alcohol use and reduce risky alcohol use among ED patients at 3 months postvisit as compared with a minimal intervention control condition.

A complete description of the HERA development and randomized controlled trial (RCT) methods were previously published [26,34]. Although the HERA assesses and refers patients to treatment for multiple substances, only results pertaining to alcohol are reported and discussed in this paper. A previous publication reported results for tobacco use [23], and a subsequent paper will address the results pertaining to illicit drug use. This clinical trial was registered with ClinicalTrials.gov as the Dynamic Assessment and Referral System.

Patients were enrolled from 4 EDs (see Table 1) between 8 am and 7 pm, with shifts occurring every day of the week. Research assistants (RAs) approached all adult patients at their bedside during their ED visit. Patients aged 18 years and older with risky alcohol use were considered. Risky alcohol users were defined as having used alcohol above the AUDIT quantity or frequency guidelines, with or without tobacco use but with no illicit drug use in the past 12 months. This paper focuses only on alcohol users who may have been smokers but did not use illicit drugs. Exclusion criteria were severe illness or distress, cognitive insufficiency, in state custody or restraints, being held involuntarily, and language barriers. Patients who were actively involved in alcohol treatment were eligible for the study, but few agreed to participate. Patients were enrolled regardless of whether they were admitted or discharged. The study components, including baseline and intervention, were completed while patients were in the ED. Participants were randomized to either the intervention or control condition by a random number generator from the Java programming language standard library embedded within the HERA. Immediately after discharge or transfer from the ED, the enrolling RA completed a brief interview with the participant, either in person or by telephone within 48 h (postvisit interview). A trained RA not affiliated with the data collection sites contacted all participants by telephone at 1 and 3 months following the ED visit to assess alcohol treatment initiation and to reassess alcohol use. This study was approved by the Institutional Review Boards for all data collection sites, in accordance with the ethical standards of the Helsinki Declaration of 1975. All participants gave their informed consent and signed a written consent form before inclusion in the study.

Intervention and control conditions were treated the same in all aspects of the study procedures; however, the groups differed on the type of referral and availability of reports. Participants in the intervention condition (HERA) (1) were offered a dynamic referral, (2) received the patient feedback report with a tailored referral list, and (3) their treating physician received the health care provider report. Participants assigned to the minimal intervention control condition (control) were given a standardized, printed list of local treatment providers instead of dynamic referrals, and health care provider reports were not made available.

The RA who performed the outcome assessments was partially blinded. Because the HERA is heavily focused on the referral process, and not all patients received the same type of referrals, to avoid confusion, the follow-up questions were tailored to the referral type received at baseline (printed list vs dynamic referral). Despite blinding efforts, the presence of particular questions for the intervention group revealed some information about group assignment. For example, only patients who chose a dynamic referral were asked whether they had been contacted by an alcohol treatment provider.

Baseline characteristics (eg, demographics, alcohol use) were compared across intervention conditions using chi-square test of independence and independent samples t test to confirm randomization success, and the potential for differential retention rates across conditions was examined using chi-square test of independence. Our primary outcomes (ie, alcohol treatment provider contact, treatment initiation, alcohol use) were then compared across conditions at 1 and 3 months using generalized estimating equation (GEE) models. Post hoc chi-square test of independence was performed following a statistically significant GEE model in order to better isolate the observed differences at each follow-up point. Chi-square analyses were also used to make comparisons across conditions for outcomes collected only at a single follow-up point (eg, ED counseling).

We then performed a series of analyses comparing participants in 3 distinct groups: (1) the control condition, (2) the intervention condition that declined a dynamic referral to providers (tailored list only), and (3) the intervention condition that accepted a dynamic referral (dynamic referral group). Because this categorization allows for preexisting differences across groups (particularly between the tailored list and dynamic referral groups), these models included theoretically relevant covariates that might impact the outcomes of interest (baseline AUDIT scores and readiness to quit). Missing data or attrition at follow-up was addressed using standard intention-to-treat principles whereby the least favorable outcome (eg, no provider contact, no treatment completion) was assigned to missing data points. Specifically, if data were missing at both follow-up points for a case, the least favorable outcome was imputed. If data from the first follow-up indicated a favorable outcome (eg, quit attempt, initiated treatment) and data was missing at the second follow-up, a favorable outcome would be imputed as we were interested in the event occurring by a given time point. If data were missing at the first follow-up and present at the second follow-up, regardless of the outcome at the second follow-up, the least favorable outcome would be imputed at the first follow-up. Given the use of these principles, the frequencies presented in each table represent observed data, whereas the percentages represent intention-to-treat estimates. All analyses were performed using Statistical Package for the Social Science 22 (IBM, 2012), with an a priori alpha level of .05.

Of 319 alcohol users who met eligibility criteria and did not report any drug use, 212 individuals were enrolled (see Multimedia Appendix 3). A greater proportion of eligible females (75.7%, 78/103) enrolled in the study than males (62.0%, 134/216), χ²₁=5.9 P=.02, and enrolled individuals were younger, on average (mean 38.1 years; SD 13.4) than nonenrolled individuals (mean 42.3 years; SD 13.9), t₃₁₇=2.63, P=.009. There were no differences in percentage of enrolled eligible patients across sites, concomitant tobacco use, and insurance status. Of the 212 participants enrolled, 115 were assigned to the control condition and 97 to the intervention condition. At baseline, there were no differences between the two conditions on demographics (see Multimedia Appendix 4), data collection site (P=.06), mental health diagnoses (P=.19 to .83), AUDIT scores (P=.24), or readiness to change (P=.10).

Of the analyzed participants, 196 out of 212 (92.5%) completed the postvisit interview, 157 out of 212 (74.1%) completed the 1-month follow-up, and 157 out of 212 (74.1%) completed the 3-month follow-up (see Multimedia Appendix 3). There were no differences between retained individuals and those lost to follow-up on age (P=.99; .17, respectively), baseline AUDIT scores (P=.62; .34), mental health diagnoses (P=.27 to .81; .09 to .74), or readiness to change (P=.58; .21). However, at the 3-month follow-up, there were more control individuals retained (95/115; 83.0%) than experimental (62/97; 64%), χ²₁=9.6, P=.002. There were also more female retained at the 3-month follow-up (64/78; 82%) than male (93/134; 69.4%), χ²₁=4.1, P=.04.

ED-originated alcohol interventions have potential for substantial public health impact by offering widespread SBIRT for risky alcohol use within a population that is both high risk and difficult to reach [4,10,13]. However, many challenges continue to impede the adoption of interventions into routine clinical care, including competing time demands and priorities, a focus on acute care, and insufficient specialized training of providers in risky alcohol use interventions. Technology facilitated intervention models that maximize efficiency and relieve clinician burden may offer a solution.

The results of this clinical trial exploring the benefits of using a single administration, stand-alone computerized intervention were mixed. All participants scored positive for risky alcohol use, and therefore received a patient feedback report with personalized information and referrals. Those who reported not currently being in treatment and who reported some desire to change their drinking were offered a dynamic referral. Although overall no significant differences were observed between conditions for contact with a treatment provider, treatment initiation, treatment engagement, and treatment completion, a closer look at the data suggests that the dynamic referral may still hold promise for promoting treatment engagement. Subanalyses revealed that among the experimental participants, those who accepted a dynamic referral were more likely to make contact with a treatment provider and have higher rates of treatment initiation than control participants. However, these effects did not lead to continued engagement in treatment or changes in alcohol use over the 3-month period following the ED visit. Moreover, some of the trends for attempting to change, such as reporting any attempt to reduce use, favored the control condition, rather than the intervention condition, though these differences were not statistically different. Additional research is needed to probe this pattern to establish if there may be an iatrogenic impact of providing personalized information and referrals in dampening self-change.

There are several factors that may have hampered the HERA’s impact on treatment and alcohol use behavior. One factor may be a lack of adoption and implementation by ED clinical staff. Although the clinical staff members who received the health care provider reports were trained to interpret the findings, they were not specifically trained or mandated to provide counseling or additional intervention materials to patients as a result of reviewing the report. Analyses indicated that clinical staff did not provide additional counseling or intervention materials to participants in the intervention group, which could be interpreted as weak clinician adoption or support of the intervention. Although the HERA is designed to offer brief intervention and referral to treatment as a stand-alone service, a cooperative approach which includes protocols for clinician involvement in response to a positive screen on the health care provider report may prove a stronger intervention than a stand-alone automated referral. Furthermore, the sample was heterogeneous, with only a minority scoring in the severe range on the AUDIT (low to moderate risk, n=173/212 [82%], moderate to high risk, n=13/212 [6.1%], and high to very high risk, n=26/212 [12.3%]). This undoubtedly dampens the level of interest in specialized treatment.

An additional factor limiting clinical impact could be the low-intensity nature of the HERA intervention. The HERA was designed as a one-time, brief interaction due to the fast-paced ED environment filled with competing demands for time and resources. Minimizing the intervention for this purpose could have adversely affected the HERA’s potential for clinical impact. The brief encounter with the HERA, while efficient and time-saving for clinicians, may not be powerful enough to support long-term changes in alcohol use behavior. Future technology-facilitated interventions may need to integrate motivational tools for behavior change, such as Web-based multimedia content or longitudinal interaction beyond the ED visit.

A final factor that may have impeded continued treatment and change in risky alcohol use behavior are barriers related to patient follow-through, including a lack of transportation or childcare, fees for services, and schedule conflicts. Although the dynamic referral was designed to connect patients with a “best match” treatment facility based on personal characteristics, the scope is limited to general characteristics such as location, insurance provider, and desire for telephone or in-person treatment. Motivated patients, who initiated contact with a nearby treatment provider compatible with their insurance carrier, may still have been unable to attend treatment due to the aforementioned circumstances [41]. Additionally, individuals were offered free alternatives to fee-for-service treatment models to help address cost barriers, but free treatment providers may not have been conveniently located or available during hours conducive to every patient’s schedule.

Several limitations exist that impact interpreting the results. First, because a minimal treatment control group was used, rather than true treatment as usual, the assessment and resource list provided to the minimal treatment control group may have had an intervention effect and artificially inflated treatment contact and behavior change in the control group. Second, the use of a modified AUDIT allowed for time-sensitive brief assessment of alcohol use, but assessed use over a shorter period than other methods, such as the Timeline Follow Back [42], and may not have allowed for a large enough assessment window to detect risky alcohol use in some individuals. Third, results may have been skewed by the nature of self-report due to the ambiguity inherent in measuring alcohol consumption by number of drinks. Although standard measurements have been outlined by the AUDIT [35], and clear examples of “one alcoholic drink” were provided during the assessment, participant understanding of the size and volume of “one alcoholic drink” varies considerably. Finally, by focusing solely on alcohol users, who used alcohol above the AUDIT quantity or frequency guidelines and who had not used illicit drug in the past 12 months, the generalizability of the results is limited. Future research should examine the efficacy of automated referral systems for alcohol treatment among all alcohol users.

The fact that very few participants accepted the dynamic referral highlights a potential limitation of the HERA model itself. Although participants who accepted a dynamic referral were more likely to contact a treatment provider and demonstrated higher rates of treatment initiation than control participants, impact will be minimal unless more patients begin accepting the referral. Future studies of similar models should aim to identify and overcome barriers to referral acceptance. A final limitation is that participants who failed to follow-through with treatment after receiving the referral were not questioned as to what factors contributed to their failure to follow-through. Costs associated with fee-for-service treatment options may have been a barrier to treatment initiation and engagement, although potentially alleviated by the inclusion of free treatment options in addition to the fee-for-service selections. Barriers to patient follow-through in systems like the HERA should be explored in future studies.

The HERA aims to satisfy clinical practice mandates for SBIRT for risky alcohol users in the ED setting. For those who accepted the dynamic referral, the HERA was effective at promoting contact with an alcohol treatment provider and initiating risky alcohol use treatment. Unfortunately, when employed as a stand-alone intervention, the HERA did not lead to sustained treatment engagement or changes in alcohol use during the 3 months following the initial ED visit. These results raise two questions: (1) Do stand-alone, brief, automated interventions lack the power to sufficiently motivate sustained alcohol use treatment engagement and behavior change? and (2) Is SBIRT for risky alcohol use satisfactory for all populations, particularly those unable or unwilling to pay fees associated with treatment services or underserved populations with limited access to health care, as represented in this study? This study highlights the need for developing and studying interventions that work alongside alcohol treatment linkage strategies. The prototype of the HERA was called the Dynamic Assessment and Referral System for Substance Abuse (DARSSA). The name was changed to reflect our long-term plans to expand the system to provide SBIRT for other nonsubstance problems, like depression and interpersonal violence.