Phenobarbital Treatment of Alcohol Withdrawal in the Emergency Department:

Abstract

Objective:

Despite frequent treatment of alcohol withdrawal syndrome (AWS) in the Emergency Department (ED), evidence for phenobarbital as an ED alternative therapy is mixed. We conducted a systematic review and meta-analysis comparing safety and efficacy of phenobarbital (PB) to benzodiazepines (BZD) for treatment of AWS in the ED.

Methods:

We searched articles and references published in English in PubMed, Web of Science, and Embase from inception through May 2022. We included randomized trials and cohort studies comparing treatment with phenobarbital to benzodiazepine controls, and excluded studies focused on non-AWS conditions. Review was conducted by two blinded investigators and a third author; 8 of 59 (13.6%) abstracts met inclusion criteria for review and meta-analysis using a random-effects model. Treatment superiority was evaluated through utilization, pharmacologic, and clinical outcomes. Primary outcomes for meta-analysis were the proportion of patients (1) admitted to the intensive care unit (ICU), (2) admitted to the hospital, (3) readmitted to the ED after discharge, and (4) who experienced adverse events.

Results:

Eight studies (2 randomized controlled trials, 6 retrospective cohorts) comprised data from 1,507 patients in 2,012 treatment encounters for AWS. All studies were included in meta-analysis for adverse events, 7 for hospital admission, 5 for ICU admission, and 3 for readmission to the ED after discharge. Overall methodological quality was low-moderate, risk of bias moderate-high, and statistical heterogeneity moderate. Pooled relative risk of ICU admission for those treated with PB vs. BZD was 0.92 (95% CI 0.54–1.55). Risk for admission to the hospital was 0.98 (95% CI 0.89–1.07) and for any adverse event was 1.1 (95% CI 0.78, 1.57); heterogeneity prevented meta-analysis for ED readmission.

Conclusions:

The current literature base does not show that treatment with phenobarbital significantly reduces ICU admissions, hospital admissions, ED readmissions, or adverse events in ED patients with AWS compared with benzodiazepines alone.

Introduction

Excessive alcohol consumption has been a major contributor to Emergency Department (ED) visits, factoring into an estimated 5 million or more ED visits annually in the United States^{1, 2} and increasing faster than the rate of overall ED utilization since the early 2000s. Prior to the COVID-19 pandemic, lifetime prevalence of alcohol use disorder (AUD) in the U.S. was 29%, with 5.4% of all individuals 12 and older meeting criteria in a given year³. There is early evidence that the problem has worsened since the pandemic, with increases in drinking⁴, alcohol-related deaths⁵, and ED and hospital presentations of alcohol withdrawal syndrome (AWS)^{6, 7}. AWS alone is responsible for some 500,000 hospital admissions annually⁸.

Treatment of AWS presentations in the ED is well-established and relies on symptom-triggered benzodiazepine protocols as first line⁹. Benzodiazepines themselves, however, carry potential for abuse and withdrawal, and recent disruptions in drug supplies have spurred enthusiasm for identifying alternative agents. Phenobarbital is an older drug that has experienced a recent revival as a treatment for AWS with recent evidence from the intensive care setting showing some benefit over benzodiazepines^{10, 11}. In the ED setting specifically, the literature base has not been fully evaluated, and controversy remains about whether phenobarbital is a safe and efficacious alternative to benzodiazepine-based protocols¹².

This is a systematic review and meta-analysis of randomized trials and observational cohort studies comparing phenobarbital to benzodiazepines for the treatment of AWS in the ED. Though narrative reviews have considered this question, none has focused on treatment delivered in the ED, which differs in important ways from inpatient or critical care settings, and no published meta-analysis compares phenobarbital with benzodiazepine treatment. We aimed to evaluate the safety profile and efficacy of phenobarbital in terms of utilization, pharmacologic, and clinical outcomes.

Methods

Protocol

We designed our protocol in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and registered it with PROSPERO (CRD42022313102). Protocol checklists are included in the appendix (Supplemental Tables S1 and S2).

Eligibility Criteria

Population:

We included studies of human adults with a diagnosis of alcohol withdrawal who received initial treatment with a benzodiazepine or phenobarbital in the ED. Eligibility criteria defined AWS diagnosis by provider assessment or appropriate diagnostic codes. We excluded studies whose inclusion criteria were defined by a primary non-AWS diagnosis, such as trauma or status epilepticus, as these subpopulations demonstrate different risk for our outcomes of interest from the general ED population seen and treated for alcohol withdrawal. Similarly, studies of treatment taking place outside the ED, as in the intensive care setting, were also excluded due to different risk and timeframe for complications from the ED population.

Intervention:

Phenobarbital administered in the ED alone or in combination with other drugs, by any route.

Comparator:

Any benzodiazepine administered in the ED alone or in combination with drugs other than phenobarbital.

Outcomes:

We organized outcomes for review into utilization, pharmacologic, and clinical outcomes. The primary outcome for meta-analysis was proportion of patients admitted to a critical level of care from the ED. Secondary utilization outcomes were the proportion admitted to inpatient care and readmitted to the ED for any reason after discharge. Pharmacologic outcomes included total dose of sedative-hypnotics administered¹³ and the proportion of patients requiring continuous benzodiazepine infusion. Review of clinical outcomes focused on the proportion of patients who experienced adverse events, including intubation, seizure, or death.

Studies:

Randomized controlled trials or cohort studies with an intervention and control group meeting criteria were included; studies without a benzodiazepine comparator were excluded. We also excluded case reports, case-control studies, and studies not published in English.

Search Strategy

In consultation with a UCSF research librarian (PT), we queried three databases (PubMed, Embase, and Web of Science) from inception to May 2022. Search strategies are detailed in the appendix (Supplemental Table S3). We also performed manual citation review of included studies to identify additional publications meeting inclusion criteria.

Selection and Extraction

Two authors (CL and DD) independently reviewed abstracts for inclusion using Rayyan.ai, a web-based citation manager for systematic reviews. Full texts of retrieved abstracts were again independently reviewed for inclusion. Conflicts were adjudicated by a third author (CM). We used a pre-specified form for extraction of study characteristics and outcomes. For studies meeting inclusion criteria which only reported results for admitted or discharged ED patients, we used study participant flow diagrams to derive rates of admitted and discharged patients in the sample (Rosenson) or contacted study authors to include these complete samples in our estimates of hospital admission rates (Lebin). Results were compiled using STATA v.16 (StataCorp LLC).

Risk of Bias in Included Studies

We relied on Cochrane Collaboration tools to assess risk of bias included articles via the risk-of-bias (RoB-2) tool for randomized trials¹⁴ and the risk-of-bias in nonrandomized studies of interventions (ROBINS-I) tool for cohort studies¹⁵. Two authors (CL and DD) independently assessed one randomized and one non-randomized study for bias and compared conclusions; the remainder were assessed by one author each.

Effect Measures and Synthesis

We summarized dichotomous outcomes using relative risk and the 95% confidence interval around the summary estimate. Results were combined using random effects models and the appropriate measure of effect according to the Dersimonian and Laird method¹⁶. We did not combine effects if there were fewer than three studies reporting the outcome.

In anticipation of the limited evidence base, the decision was made a priori to include both randomized and non-randomized trials in combined effect estimates if they met criteria for statistical homogeneity. While not routine, this emerging practice in meta-analysis has not been shown to meaningfully change pooled effect estimates when studies are not heterogeneous17–20.

Planned Analysis

We assessed heterogeneity using Cochrane’s Q test21 and proceeded with meta-analysis for I² < 55%. Publication bias was examined using funnel plots, Egger’s linear regression, and Begg’s rank correlation test. Planned subgroup analyses included comparison based on study design, severity of AWS, and phenobarbital monotherapy versus combined phenobarbital and benzodiazepines. Subgroup analyses required 3 data points to proceed.

Results

We identified 94 abstracts through initial screening and retrieved 8 for full-text review (Figure 1). All 8 were included in the final analysis (Table 1). Both reviewers independently agreed on all included studies; overall, agreement was substantial with Cohen’s kappa at 0.76. The two earliest studies meeting inclusion criteria were double-blinded randomized controlled trials; all subsequent studies were retrospective cohort studies. Overall, 1,507 patients contributed data to this meta-analysis, comprising 2,012 treatment encounters for AWS.

Figure 1:

PRISMA flowchart of study selection

Table 1:

Study characteristics

Reference	Design	Participants	Disposition	Intervention	Comparator	Primary Outcome
Hendey 2011	RCT	44 adults with AWS requiring treatment in ED	Discharge, admission to hospital or ICU	260 mg IV PB initial dose + 130 mg subsequent (mean 509 mg) without BZD	2 mg IV LZ initial dose + subsequent at discretion of provider	Change in CIWA score from baseline to discharge or admission from ED
Rosenson 2013	RCT	102 adults with anticipated need for hospitalization for AWS	Admission to hospital or ICU	10 mg/kg IV PB initial dose followed by symptom-triggered LZ protocol	Placebo IV dose followed by symptom-triggered LZ	Proportion admitted to ICU from ED
Nelson 2019	Cohort	300 adults treated for AWS based on SEWS score	Discharge, admission to hospital or ICU	Symptom-triggered IV PB without LZ (mean dose 454 mg) or with subsequent LZ when available (mean 257.4 mg)	Symptom-triggered IV DZ	Proportion admitted to ICU from ED
Sullivan 2019	Cohort	209 adults with primary ED diagnosis of AWS	Discharge, admission to hospital or ICU	IV PB per provider discretion (median dose 260 mg) with or without BZD	Any IV BZD per provider discretion	Proportion admitted to ICU from ED
Ibarra 2020	Cohort	40 adults with moderate-severe AWS receiving at least 4 mg LZ in first 24h hospitalized	Discharge, admission to hospital or ICU	Single dose of IV PB at provider discretion with symptom-triggered IV LZ	Symptom-triggered IV LZ	Total daily LZ dose through hospital day 3
Henson 2021	Cohort	205 patients diagnosed with AWS	Discharge from ED	PB per provider discretion with or without BZD	Any BZD per provider discretion	Proportion re-presenting to ED within 30 days for any reason
Lebin 2022	Cohort	470 adults with discharge diagnoses including AWS	Discharge from ED	IV or IM PB per provider discretion with BZD (median dose 357.5 mg) or without BZD (median 390 mg)	Any BZD by any route per provider discretion	Proportion re-presenting to ED within 3 days for any reason
Staidle 2022	Self-controlled cohort	137 adults comprising 642 visits treated for AWS in ED	Discharge, admission to hospital or ICU	IV PB 130–260 mg every 30 min by symptoms, with (mean dose 405 mg) or without BZD (mean 424 mg)	Symptom-triggered IV LZ or DZ	Disposition from ED and re-presentation to ED within 48h for any reason

PB = phenobarbital; LZ = lorazepam; DZ = diazepam; BZD = benzodiazepine; CIWA = Clinical Institute Withdrawal Assessment for Alcohol; SEWS = Severity of Ethanol Withdrawal Symptoms

Risk of Bias of Included Studies

Both blinded randomized controlled trials had an unclear risk of bias driven by treatment of randomized but excluded participants; while the number of excluded subjects in each was low, differential exclusion could have biased results in these small studies (Figure 2a). Unbalanced enrollment in Hendey et al. suggested that excluded subjects may have come from the lorazepam group, amounting to 17% of the comparator. These patients were excluded because they had only an initial symptom score for AWS without additional assessment, leaving unanswered questions about their subsequent clinical or mental status and the relative performance of the benzodiazepine group. In the Rosenson study, 10% of randomized subjects were excluded in a differential pattern, indicating greater clinical stability in the phenobarbital group driving an increased rate of discharge. We treated these patients as discharged in our analysis of utilization, but they did not contribute data to pharmacologic or clinical outcomes, likely biasing these toward the comparator. Both studies were conducted with a low risk of bias across all other domains.

Figure 2:

Risk of bias in included studies

All observational studies carried significant risk of bias due to confounding (Figure 2b) because provider discretion influenced both medication selection and utilization (disposition), pharmacologic (dosing) and clinical (adverse events, including interventions) outcomes. Patient comorbidities, severity of withdrawal on presentation, risk for progression, cointerventions, and local factors such as ICU criteria may therefore have caused directly or indirectly effects attributed to the interventions. This is a shared fundamental limitation, with one study of 6 (Lebin) attempting to control for suspected confounding factors. No studies controlled for the severity of AWS at presentation, a likely central factor in medication choice and in all outcomes. Selection bias was an additional significant risk in studies which conditioned participation on discharge after AWS treatment (Henson and Lebin) or on multiple presentations for AWS (Staidle). Measurement of intervention and outcome were at low risk of bias as these were measured systematically from EHR data. Few studies reported numbers or treatment of missing data, and some (Nelson, Sullivan) showed evidence suggesting that either deviations from interventions or missing data was related to intervention status.

Utilization Outcomes

All of the studies which included patients who were admitted (6 of 8) reported hospital admission rates, and 5 reported ICU admission rates. Each of these also described length of stay related to each intervention group, either in the ED specifically or over the hospital admission. Hendey and Sullivan found no significant difference in ED length of stay, while both Nelson and Staidle noted significant differences with combination therapy associated with the longest periods in the ED. Relationship with hospital length of stay was similarly mixed, with Rosenson and Ibarra reporting no difference while Sullivan identified significantly shorter admissions in the phenobarbital group of 3 vs. 4 days.

Three observational studies reported the all-cause ED readmission rate within various timeframes, though high heterogeneity prevented meta-analysis. Staidle et al. observed no significant difference in risk of return to the ED within 48 hours (phenobarbital alone 17%; dual therapy 14%; benzodiazepine alone 15% risk of return). Henson similarly observed no difference within 30 days (phenobarbital group 49%; benzodiazepine alone 50% risk of return). In contrast, Lebin et al. found that phenobarbital alone (adjusted OR 0.45, 95% CI 0.23–0.88) or phenobarbital with benzodiazepines (adjusted OR 0.33, CI 0.15–0.74) significantly decreased risk of repeat ED presentation within 3 days for discharged patients, though as noted above this result is vulnerable to selection bias.

Pharmacologic Outcomes

Combined estimation of the cumulative dose of sedative-hypnotics administered was limited by the ways in which this outcome was reported. Two studies (Sullivan and Levin) reported median rather than mean doses, and two (Rosenson and Ibarra) did not summarize the mean dose of phenobarbital given in initial treatment in the ED. From the remaining data, the mean combined sedative-hypnotic dose to treat withdrawal in subjects receiving benzodiazepines alone was 10.7 (SD 7.68) lorazepam equivalents. In the combined group of subjects receiving any phenobarbital, the mean combined sedative-hypnotic dose was 30.7 (SD 1.9) lorazepam equivalents.

Only two studies (Rosenson and Sullivan) reported the proportion of patients who required a benzodiazepine infusion, precluding meta-analysis. In Sullivan et al. the difference was not significant (0% in the phenobarbital group; 2% in the benzodiazepine group), while in Rosenson significantly more patients in the benzodiazepine group ended up receiving an infusion (4% in the dual therapy group; 31% in the benzodiazepine group). Ibarra et al. were the only authors to report median daily lorazepam requirements throughout hospitalization, which were not significantly different between treatment groups.

Clinical Outcomes

No studies reported any deaths during follow-up. Nonfatal adverse events observed included seizures, hypotension, respiratory depression, and alcohol relapse after discharge (Figure 4). A total of 27 intubations were reported across studies and one progression to DT was noted. In discharged patients, Henson observed similar rates of persistent withdrawal on re-presentation across treatment groups (16% of the benzodiazepine-only group re-presented in withdrawal vs. 14% of the phenobarbital group). Several studies reported differences in clinical symptom scores across groups, such as change from initial assessment to disposition from ED (Hendey), the maximum scores at certain timepoints (Sullivan, Ibarra and Rosenson), or the first and maximum scores recorded (Staidle). All of these measures showed scores to be statistically equivalent across treatment groups.

Figure 4:

Adverse clinical events compared across treatment protocols

Risk of Bias Across Studies

Both Egger’s and Begg’s tests returned non-significant results (p=0.5), and there was no evidence of small study effects on funnel plot. Because no pre-specified subgroups contained 3 datapoints for any of our outcomes, we did not perform subgroup analyses. The results were robust to sensitivity analyses, with no significant changes when each study was omitted in turn. In planned post-hoc analysis, there was no significant difference in pooled effect estimates from randomized compared with non-randomized studies.

Discussion

A decade ago, two well-conducted but small randomized trials suggested benefit or at least equivalence of phenobarbital compared with benzodiazepines alone for treating alcohol withdrawal in the ED. Since then, only nonrandomized studies considering this question have been published, which have been overall very vulnerable to confounding by severity of AWS, comorbidities, and other local factors affecting medication selection, with varying results. This review presents the first combined estimates of the effect of phenobarbital on clinical, pharmacologic, and utilization outcomes related to AWS treatment in the ED, demonstrating equivalence with the first-line treatment in the existing literature base. Specifically, phenobarbital was not associated with statistically significant differences in adverse events, need for hospital admission, or need for ICU admission compared to benzodiazepines alone. These estimates, however, must be interpreted in light of the substantial risk of bias of the observational studies. Given the importance of clarifying first- and second-line therapy for a common and serious complaint in the ED, we conclude that a large, rigorously conducted RCT is an important next step to address continued controversy.

Utilization outcomes were particularly vulnerable to confounding by local practice patterns, resource availability, clinician comfort with treatment and disposition, and other difficult-to-measure factors. The ICU admission outcome, frequently adopted as a primary outcome after the dramatic reduction in ICU admissions reported by Rosenson, is especially difficult to separate from other factors affecting intervention choice in observational studies. Other utilization outcomes, such as hospital or ED length of stay, were inconsistently reported; aside from hospital and ICU admission, none met our threshold for meta-analysis, though they were likely equally vulnerable to local confounding factors. To understand how treatment affects utilization, rigorous randomization with intention-to-treat analysis is necessary to reduce or eliminate this unmeasured confounding, clarifying the resource and level of care implications of treatment with phenobarbital.

We analyzed all adverse events as defined by the authors together, rather than parse the relative importance of outcomes that could have been affected by medication choice. This outcome had the lowest heterogeneity of all our combined estimates (I²=25.8%), suggesting studies were most similar in the ways they tracked and reported these short-term adverse effects. Phenobarbital and benzodiazepines appear to be equally safe in terms of nonfatal clinical events, with similar incidence of medication side effects and events indicating treatment failure. Notably, combined treatment with phenobarbital and benzodiazepines has not been associated with greater risk of side effects, challenging practice conventions that have avoided combination therapy. This includes adverse events in the ED setting after treatment, like intubation, DT, or seizure, although very few studies measured adverse events after discharge and caution may be warranted when using combination therapy in discharged patients. With the caveat that rare bad outcomes are difficult to measure in each small study, the existing evidence shows combination therapy to be as safe as either option alone.

Though we did not identify clear differences in utilization or clinical outcomes, we did observe a large difference in combined sedative-hypnotic dose when treatment included phenobarbital. If outcomes are comparable yet doses differ by a factor of three, one possible explanation would be that traditional pharmacological equivalence between these drugs does not translate to clinical equivalence in the treatment of AWS. Alternatively, if these conversion factors are accurate, it is possible that dosing as currently practiced for phenobarbital may be reduced without sacrificing efficacy. In the broader context of a movement to pursue benzodiazepine-sparing treatments for alcohol withdrawal, this relationship deserves further consideration.

We decided in advance not to pursue a cost analysis because the prices and availability of these treatments have varied substantially over the last decade. Nelson et al. designed their study of three different treatment combinations in response to shortages of benzodiazepines affecting availability across the U.S. during the mid-2010s. Given ongoing market forces, we believe cost differences deserve future study in the context of clearer clinical evidence from a randomized, multi-center study. If there is true equivalence between phenobarbital and benzodiazepine protocols, relative cost is an important consideration; if, on the other hand, there is a small difference in effect but the outcome of interest (i.e., hospital admission) is very costly, a cost-benefit analysis may be particularly informative.

With 6 of 8 included studies published since 2019, there is clearly an accelerating interest in phenobarbital as an alternative treatment for AWS in the ED. Because evidence from larger rigorous trials is still needed, this represents an opportunity to consider additional outcomes that are measurable, relevant, and have yet to be studied in comparing efficacy. Decreased need for physician reassessment, fewer nursing interventions, and faster or more prolonged symptom control would be meaningful and objective differences to consider, along with subjective patient or provider experience with treatment options. In order to resolve the ongoing debate, we will need high-quality evidence that also addresses the metrics that best reflect successful treatment of alcohol withdrawal in our field.

Supplementary Material

Supinfo

Table S1: PRISMA Checklist for Abstracts

Table S2: PRISMA Checklist for Systematic Review and Meta-analysis

Table S3: Search Strategy Used for the Systematic Review

Figure 3:

Hospital utilization for phenobarbital-containing protocols