To the Editor:
Severe alcohol withdrawal syndrome (SAWS) is characterized by brain hyperexcitation and autonomic instability that may require ICU management (1). Patients with SAWS who receive high doses of first-line benzodiazepines are at risk for over-sedation, worsening delirium, and initiation of invasive mechanical ventilation (IMV) (2). Phenobarbital—a barbiturate that reduces neuronal excitation—is associated with shorter length of stay (LOS) and lower rates of IMV during SAWS in single-center implementation studies (3, 4). However, multicenter trends in phenobarbital use and outcomes are largely unknown. In this study, we sought to characterize phenobarbital practice patterns and associated outcomes for SAWS across United States (U.S.) ICUs.
Methods
We used the Premier Healthcare Database, 2016–2020 (5) (∼20% of U.S. hospitalizations) to identify patients with SAWS: adults admitted to an ICU on Day 1 of admission with International Classification of Diseases, Tenth Revision primary or admitting diagnosis codes for alcohol withdrawal and who received benzodiazepines or phenobarbital. We excluded patients who received IMV on Day 1, transferred from outside hospitals, and admitted to hospitals with <25 eligible patients (6).
Analyses of hospital practices examined 1) rates of patients and hospitals using phenobarbital (oral or parenteral); 2), the cumulative dose of phenobarbital on Day 1 and during the hospitalization, and 3) the durations of SAWS medications. We reported the absolute and relative change in the proportion of patients who received SAWS medications between 2016 and 2020. Multivariable hierarchical logistic regression models were used to identify hospital- and patient-level characteristics associated with phenobarbital use. The median odds ratio was calculated to quantify the degree to which admission hospital contributes to phenobarbital use variation (7).
We evaluated outcomes associated with phenobarbital use, including 1) IMV initiation (primary); 2), adjunct medication use (α-2-agonists, antipsychotics, or anti-convulsants) after Day 1; and 3) hospital LOS. We used differences-in-differences (DiD)—a quasi-experimental method that quantifies the difference in outcomes between intervention and control in “pre” (2016) and “post” (2020) periods—to measure the effects of hospital-level phenobarbital adoption and avoid potential confounding by indication for an alternate approach to SAWS treatment (8). In the DiD analyses, the phenobarbital adoption intervention was defined as admission to hospitals that newly began using phenobarbital between 2017 and 2019 and sustained use thereafter (“new phenobarbital-adopter” hospitals). Control groups were defined as admission to hospitals with phenobarbital use or non-use throughout the study period. We used multivariable hierarchical linear models including terms for year (2016 or 2020), hospital-level phenobarbital adoption, and an interaction term between year and phenobarbital adoption (covariates listed in Table 1). The interaction term quantified the difference in the probability of outcomes (IMV, use of adjunct SAWS medication) or the absolute change in outcome (log hospital LOS) between patients admitted to hospitals after phenobarbital adoption and patients admitted to control hospitals. An exploratory analysis, examined the association between individual-level exposure to phenobarbital on Day 1 and IMV initiation using targeted maximum likelihood estimation (including the same covariates as in the DiD analysis) (9), a doubly robust causal inference method that estimates the adjusted difference in mean outcomes of phenobarbital versus no phenobarbital.
Table 1.
Difference-in-Difference Comparison between Patients Admitted to “New Phenobarbital Adopter” Hospitals versus Patients Admitted to Control Hospitals
| Unadjusted Model—Difference-in-Difference Estimator (95% CI)* | Adjusted Model † —Difference-in-Difference Estimator (95% CI) | |
|---|---|---|
| Invasive mechanical ventilation initiation, % | −3.3 (−6.7, 0.1) | −3.6 (−6.9, −0.3) |
| Adjunct medication use, % | −3.9 (−11.0, 3.2) | −2.3 (−8.7, 4.2) |
| Log of hospital length of stay, % | −9.6 (−17.7, −0.7) | −7.6 (−15.4, 0.9) |
| Hospital length of stay relative reduction (Poisson) | 0.88 (0.82–0.93) | 0.91 (0.86–0.97) |
Definition of abbreviation: CI = confidence interval.
For invasive mechanical ventilation initiation and adjunct medication use, the difference-in-difference estimator is the difference in the probability of the outcome between patients admitted to hospitals after adoption of phenobarbital and patients admitted to control hospitals. For log of hospital length of stay, the estimator is the absolute percent change in log hospital length of stay between patients admitted to hospitals after adoption of phenobarbital and patients admitted to control hospitals. Sensitivity analysis: For hospital length of stay (Poisson), the estimator is the relative reduction in hospital length of stay between patients admitted to hospitals after adoption of phenobarbital and patients admitted to control hospitals.
Adjusted for age, sex, race, United States region, hospital bed count, hospital teaching status, hospital safety-net status, hospital average number of severe alcohol withdrawal cases per year, phenobarbital dose, use of benzodiazepines, use of anticonvulsants, use of α-2 agonists, use of antipsychotics, Elixhauser Comorbidity Score, history of non-alcohol drug use, history of seizure disorder, cardiovascular organ dysfunction, respiratory organ dysfunction, neurologic organ dysfunction, hematologic organ dysfunction, hepatic organ dysfunction, and renal organ dysfunction.
Results
Of 13,061 included patients (245 hospitals), 11.3% received phenobarbital (91% of first doses were parenteral) with a median cumulative phenobarbital dose of 360 mg (interquartile range [IQR], 180–720 mg) on Day 1 and 780 mg (IQR, 390–1538 mg) during the hospitalization. Among patients who received phenobarbital, 91.2% also received benzodiazepines. Median cumulative days of SAWS medications were as follows: phenobarbital 3 (IQR, 1–4), benzodiazepines 4 (IQR, 3–6), α-2-agonists 3 (IQR, 2–4), and anti-psychotics 2 (IQR, 1–4). Hospital rates of phenobarbital use ranged from 0% to 77% (median 4%). From 2016 to 2020, phenobarbital use increased from 6.1% to 18.4% (relative change +204%). Over the same period, benzodiazepine (99.8–98.1%, relative change −2%) and anti-psychotic (12.6–12.2%, relative change −3%) use remained stable, and α-2-agonist use increased (25.4–34.5%, relative change +36%).
In the multivariable practice pattern models, younger patients (age adjusted odds ratio [aOR] 0.86 per 1-SD increase, 95% confidence interval [CI], 0.80–0.93), males (aOR, 1.40; CI, 1.17–1.68), patients who identified as White (aOR, 1.60 versus Black; 95% CI, 1.14–2.25), and patients admitted to hospitals in the West (aOR, 3.85 versus Midwest; 95% CI, 1.89–7.85) and Northeast U.S. (aOR, 4.06 versus Midwest; 95% CI, 1.92–8.58) had higher odds of phenobarbital use. Patients receiving phenobarbital were more likely to receive concurrent α-2-agonists (aOR 1.44; 95% CI, 1.22–1.69) and antipsychotics (aOR, 1.38; 95% CI, 1.13–1.69). Admission hospital was a strong driver of phenobarbital use (median odds ratio 4.41; 95% CI, 4.01–4.86).
We included 4,390 patients in DiD analyses evaluating outcomes associated with hospital-level phenobarbital adoption. We identified 46 phenobarbital-adopter hospitals and 166 control hospitals. Characteristics of patients by hospital-level phenobarbital adoption are shown in Table 2. At baseline (2016), IMV outcomes occurred during 6.1% of hospitalizations for patients admitted to eventual phenobarbital-adopter hospitals and 5.0% of control hospitals (+1.1% difference, 95% CI, −1.7 to 3.4). At the end of the study (2020), IMV outcomes occurred during 5.0% of hospitalizations to phenobarbital-adopting hospitals and 6.7% of controls (−1.7%, 95% CI, −3.4 to 0.8), resulting in an adjusted DiD of −3.6% (95% CI, −6.9 to −0.3) representing a lower probability of IMV initiation for patients admitted to phenobarbital-adopter hospitals compared with those admitted to control hospitals. In the adjusted DiD of additional outcomes, there was a possible reduction in hospital LOS with admission to phenobarbital-adopter hospitals but not adjunct medication use (Table 1).
Table 2.
Baseline Characteristics of Patients in the Difference in-Difference Analysis Stratified Admission to “New Phenobarbital Adopter” Hospitals versus Admission to Control Hospitals
| Characteristic | Admission to Control Hospitals in 2016 (n = 1,740) | Admission to Control Hospitals in 2020 (n = 1,712) | Admission to “New Phenobarbital Adopter” Hospitals in 2016 (n = 395) | Admission to “New Phenobarbital Adopter” Hospitals in 2020 (n = 543) |
|---|---|---|---|---|
| Age, yr mean (SD) | 48 (11) | 48 (12) | 48 (12) | 47 (12) |
| Male, n (%) | 1,356 (77.9) | 1,366 (79.8) | 318 (80.5) | 428 (78.8) |
| Race, n (%) | ||||
| Black | 116 (6.7) | 98 (5.7) | 30 (7.6) | 37 (6.8) |
| White | 1,501 (86.3) | 1,481 (86.5) | 331 (83.8) | 438 (80.7) |
| Other/unknown | 123 (7.1) | 133 (7.8) | 34 (8.6) | 68 (12.5) |
| Seizure present on admission, n (%) | 276 (15.9) | 269 (15.7) | 88 (22.3) | 83 (15.3) |
| Elixhauser comorbidity score, mean (SD) | 5 (2) | 5 (2) | 5 (2) | 5 (2) |
| Prior nonalcohol drug use disorder, n (%) | 343 (19.7) | 351 (20.5) | 72 (18.2) | 96 (17.7) |
| Organ dysfunctions present on admission, n (%) | ||||
| Cardiovascular | 42 (2.4) | 70 (4.1) | 10 (2.5) | 14 (2.6) |
| Respiratory | 32 (1.8) | 79 (4.6) | 10 (2.5) | 32 (5.9) |
| Neurologic | 80 (4.6) | 72 (4.2) | 18 (4.6) | 17 (3.1) |
| Hematologic | 519 (29.8) | 540 (31.5) | 112 (28.4) | 165 (30.4) |
| Hepatic | 10 (0.6) | 13 (0.8) | 2 (0.5) | 3 (0.6) |
| Renal | 173 (9.9) | 261 (15.2) | 41 (10.4) | 75 (13.8) |
| Concurrent medication use, n (%) | ||||
| Benzodiazepines | 1735 (99.7) | 1691 (98.8) | 395 (100) | 524 (96.5) |
| Antipsychotics | 218 (12.5) | 219 (12.8) | 45 (11.4) | 52 (9.6) |
| α-2-agonists | 446 (25.6) | 614 (35.9) | 104 (26.3) | 162 (29.8) |
| Other anticonvulsants | 141 (8.1) | 163 (9.5) | 17 (4.3) | 42 (7.7) |
| United States census region, n (%) | ||||
| Midwest | 611 (35.1) | 520 (30.4) | 188 (47.6) | 268 (49.4) |
| Northeast | 219 (12.6) | 219 (12.8) | 56 (14.2) | 77 (14.2) |
| South | 653 (37.5) | 676 (39.5) | 97 (24.6) | 124 (22.8) |
| West | 257 (14.8) | 297 (17.3) | 54 (13.7) | 74 (13.6) |
| Hospital bed size, n (%) | ||||
| 0–99 | 130 (7.5) | 163 (9.5) | 22 (5.6) | 35 (6.4) |
| 100–199 | 328 (18.9) | 262 (15.3) | 107 (27.1) | 158 (29.1) |
| 200–299 | 335 (19.3) | 361 (21.1) | 67 (17.0) | 91 (16.8) |
| 300–399 | 297 (17.1) | 277 (16.2) | 57 (14.4) | 57 (10.5) |
| 400–499 | 228 (13.1) | 163 (9.5) | 41 (10.4) | 61 (11.2) |
| 500 + | 422 (24.3) | 486 (28.4) | 101 (25.6) | 141 (26.0) |
| Number of alcohol withdrawal cases per year, mean (SD) | 16 (11) | 15 (10) | 14 (10) | 17 (12) |
| Teaching hospital, n (%) | 711 (40.9) | 603 (35.2) | 206 (52.2) | 334 (61.5) |
| Safety-net hospital, n (%) | 528 (30.3) | 414 (24.2) | 180 (45.6) | 282 (51.9) |
In the exploratory analysis (n = 12,635) examining patient-level exposure to phenobarbital, the adjusted IMV initiation rate was 1.0% (95% CI, 0.6–1.3) for phenobarbital and 5.5% (95% CI, 5.1–5.9) without phenobarbital yielding an adjusted risk difference of −4.6% (95% CI, −5.1 to −0.4).
Discussion
We examined phenobarbital practice patterns and outcomes for SAWS across U.S. ICUs. Phenobarbital use for SAWS was highly variable across hospitals and increased by 200% from 2016 to 2020. Hospital-level phenobarbital adoption and individual receipt of phenobarbital were each associated with lower IMV rates during SAWS.
Our findings of decreased IMV initiation after hospital phenobarbital adoption are consistent with single-center studies of SAWS phenobarbital-protocol implementation (3, 4). These prior studies protocolized both phenobarbital and adjunct medications use. Thus, it is unclear if improved outcomes were due to care protocolization, phenobarbital, or use of adjuncts. In contrast, our results accounting for use of adjuncts and benzodiazepines suggest that improvement in outcomes may be due to adoption of phenobarbital itself. Unlike prior studies (4, 10), we found limited use of phenobarbital monotherapy.
Our results directly inform the design of future clinical trials for SAWS treatments. We found that phenobarbital use was associated with reduced IMV initiation and possibly hospital LOS. Given that patients may be initiated on IMV for airway control during severe agitation and over-sedation, mechanical ventilation may be an important outcome to measure in SAWS treatments clinical trials. LOS may be similarly valuable as a clinical trial outcome because count outcomes like LOS allow for smaller sample sizes (compared with dichotomous outcomes); the low risk of death in SAWS also means there is less competing risk in measuring LOS. Using the effect estimates from our study, approximately 4,600 participants would be needed for a clinical trial of phenobarbital to usual care to detect a similar difference in IMV and 2,500 participants to detect a difference in LOS. Given that there is wide variation in SAWS treatments across hospitals, phenobarbital appears to be generally used with benzodiazepines - but some evidence supports phenobarbital monotherapy (3, 4) – a multi-arm (e.g., benzodiazepines and phenobarbital, phenobarbital monotherapy, benzodiazepines monotherapy), multi-stage, and multi-center adaptive clinical trial may be best suited to identify optimal SAWS treatment regimens.
Our study has limitations. Continuous infusion (e.g., benzodiazepines and dexmedetomidine) information was not available in the database. Future studies are needed to investigate variation in non-phenobarbital medication doses. We were also unable to examine other important SAWS outcomes including seizure and long-term abstinence. We limited our cohort to patients who were not initially receiving IMV to decrease medication misclassification (i.e., medications used to treat SAWs may be used for IMV sedation). Thus, our results may not be generalizable to mechanically ventilated patients. Patient weights were unavailable from the majority of patients, and we were unable to calculate weight-based phenobarbital doses. We did not have access to specific hospital SAWS treatment protocols that could allow further comparison and evaluation of different SAWS protocols. Last, because not all patients in the “new-adopter hospitals” received phenobarbital, the results from the DiD analysis may not be attributable to individual receipt of phenobarbital; however, the exploratory analysis evaluating individual receipt of phenobarbital also suggests an association between phenobarbital use and lower rates of IMV initiation.
In conclusion, adoption of phenobarbital for SAWS increased by more than 200% from 2016 to 2020 and was associated with lower need for IMV. These results strongly motivate and inform the need for clinical trials testing phenobarbital for the management of SAWS.
Footnotes
Author Contributions: N.A.B. takes responsibility for the integrity of the work as a whole, from inception to published article. All authors substantially contributed to the conception and design of this study. N.A.B acquired the data. All authors were involved in the interpretation of data. N.A.B. and A.C.L. drafted the manuscript and all authors revised it critically for important intellectual content. All authors read and approved the final manuscript.
Originally Published in Press as DOI: 10.1164/rccm.202203-0466LE on July 14, 2022
Author disclosures are available with the text of this letter at www.atsjournals.org.
References
- 1. Awissi D-K, Lebrun G, Coursin DB, Riker RR, Skrobik Y. Alcohol withdrawal and delirium tremens in the critically ill: a systematic review and commentary. Intensive Care Med . 2013;39:16–30. doi: 10.1007/s00134-012-2758-y. [DOI] [PubMed] [Google Scholar]
- 2. Mirijello A, D’Angelo C, Ferrulli A, Vassallo G, Antonelli M, Caputo F, et al. Identification and management of alcohol withdrawal syndrome. Drugs . 2015;75:353–365. doi: 10.1007/s40265-015-0358-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Gold JA, Rimal B, Nolan A, Nelson LS. A strategy of escalating doses of benzodiazepines and phenobarbital administration reduces the need for mechanical ventilation in delirium tremens. Crit Care Med . 2007;35:724–730. doi: 10.1097/01.CCM.0000256841.28351.80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Bosch NA, Crable EL, Ackerbauer KA, Clark K, Drainoni M-L, Grim V, et al. 2021;18:1708–1716. doi: 10.1513/AnnalsATS.202102-121OC. [DOI] [PubMed] [Google Scholar]
- 5. Premier Healthcare Database White Paper 2020. [accessed 2020 Mar. 2] https://products.premierinc.com/downloads/PremierHealthcareDatabaseWhitepaper.pdf
- 6.CMS. https://www.cms.gov/Medicare/Quality-Initiatives-Patient-Assessment-Instruments/HospitalQualityInits/Measure-Methodology
- 7. Merlo J, Chaix B, Ohlsson H, Beckman A, Johnell K, Hjerpe P, et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Community Health . 2006;60:290–297. doi: 10.1136/jech.2004.029454. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Wing C, Simon K, Bello-Gomez RA. Designing difference in difference studies: best practices for public health policy research. Annu Rev Public Health . 2018;39:453–469. doi: 10.1146/annurev-publhealth-040617-013507. [DOI] [PubMed] [Google Scholar]
- 9. Benkeser D, Carone M, Laan MJV, Gilbert PB. Doubly robust nonparametric inference on the average treatment effect. Biometrika . 2017;104:863–880. doi: 10.1093/biomet/asx053. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Nisavic M, Nejad SH, Isenberg BM, Bajwa EK, Currier P, Wallace PM, et al. Use of phenobarbital in alcohol withdrawal management - a retrospective comparison study of phenobarbital and benzodiazepines for acute alcohol withdrawal management in general medical patients. Psychosomatics . 2019;60:458–467. doi: 10.1016/j.psym.2019.02.002. [DOI] [PubMed] [Google Scholar]