ALCOHOLISM: CLINICAL AND EXPERIMENTAL RESEARCH Vol. 38, No.

10
October 2014

Predictors of Severe Alcohol Withdrawal Syndrome: A

Systematic Review and Meta-Analysis
Carrie M. Goodson, Brendan J. Clark, and Ivor S. Douglas

Background: Severity of alcohol withdrawal syndrome (AWS) is associated with hospital mortality
and length of stay. However, as there is no consensus regarding how to predict the development of
severe alcohol withdrawal syndrome (SAWS), we sought to determine independent predictors of
SAWS.
Methods: We conducted a systematic review and meta-analysis of studies evaluating hospitalized
patients with AWS versus SAWS—delirium tremens (DT) and/or seizures. Random-effects meta-analy-
sis [PRISMA guidelines] was performed on common baseline variables and predictive effects for devel-
opment of SAWS were calculated using RevMan v5.2. Funnel plots were constructed, and tests of
heterogeneity were performed.
Results: Of 226 studies screened, 17 met criteria and 15 were included in the meta-analysis. The pri-
mary findings were that an incident occurrence of DT or alcohol withdrawal seizures was significantly
predicted by history of a similar event (OR 2.58 for DT vs. no-DT, 95% CI 1.41, 4.7; OR 2.8 for seizure
vs. no-seizure, 95% CI 1.09, 7.19). Both a lower initial platelet count and serum potassium level were
predictive of an incident occurrence of DT (platelet count mean difference [MD] 45.64/mm3 vs.
no-DT, 95% CI 75.95, 15.33; potassium level MD 0.26 mEq/l vs. no-DT, 95% CI 0.45, 0.08),
seizures, and SAWS. Higher initial alanine aminotransferase was seen in patients with SAWS (MD
20.97 U/l vs. no-SAWS, 95% CI 0.89, 41.05). Higher initial serum gamma-glutamyl transpeptidase was
seen in patients with incident alcohol withdrawal seizures (MD 202.56 U/l vs. no-seizure, 95% CI 3.62,
401.5). Significant heterogeneity was observed, and there was evidence of publication bias. Notably,
neither gender nor comorbid liver disease was predictive.
Conclusions: The course of prior episodes of AWS is the most reliable predictor of subsequent epi-
sodes. Thrombocytopenia and hypokalemia also correlate with SAWS. We propose further research
into drinking patterns, gender, and medical comorbidities.
Key Words: Alcohol Withdrawal Syndrome, Delirium Tremens, Predictors.

F OURTEEN PERCENT OF Americans have an alco-

hol use disorder in their lifetime and approximately half
of those will experience withdrawal symptoms (Hall and
Zador, 1997). The majority of these symptoms are minor
and no medical treatment is sought, sometimes outpatient
treatment is appropriate, and a minority of cases require
inpatient treatment (Hall and Zador, 1997). Up to one-fifth
of hospitalized patients have an alcohol-related diagnosis
(Marik and Mohedin, 1996), frequently alcohol withdrawal
syndrome (AWS). AWS is a major reason for both general
inpatient and ICU admission (Marik and Mohedin, 1996)
and also complicates the hospital course of patients admitted
for other reasons. Severe alcohol withdrawal syndrome
(SAWS) is inconsistently defined in the literature, but
From the University of Colorado School of Medicine (CMG, BJC,
ISD), Denver, Colorado; and Denver Health Medical Center, Depart-
ment of Medicine (CMG, ISD), Denver, Colorado.
Received for publication March 20, 2014; accepted July 15, 2014.
Reprint requests: Ivor S. Douglas, MD, FRCP(UK), 777 Bannock
Street, MC 4000, Denver, CO 80204; Tel.: 303-602-5012;
Fax: 303-602-5055; E-mail: idouglas@dhha.org
Copyright © 2014 by the Research Society on Alcoholism.
DOI: 10.1111/acer.12529
2664
includes life-threatening complications such as withdrawal
seizures and delirium tremens (DT). DT develops in 24 to
33% of hospitalized patients treated for AWS and has a
mortality rate of 0 to 8% (Ferguson et al., 1996; Lee et al.,
2005), a decline from 20% reported in the 1970s (Wright
et al., 2006). Patients with SAWS are usually closely moni-
tored in an intensive care unit (ICU). Up to 70% of patients
in an ICU with AWS require mechanical ventilation (Secades
et al., 2008). The occurrence of SAWS prolongs hospital stay
by an average of 4 days (Kraemer et al., 2003; Wright et al.,
2006). Consequently, SAWS is a costly and resource-
intensive diagnosis for hospitalized patients (Ferguson et al.,
1996).
SAWS is not a uniform complication of AWS. Despite
numerous studies seeking reliable predictors of SAWS, no
standardized risk prediction model has been derived nor vali-
dated. Several possible reasons include lack of standard defi-
nitions, inadequate statistical power, or a true lack of
predictive power in the evaluated factors. Without reliable
risk prediction models for SAWS that can be applied at the
time of initial evaluation, appropriate resource allocation
and triage to ICU, general hospital ward, or outpatient treat-
ment is difficult. This contributes to additional handoffs,
avoidable cost, and potential risk when the patient’s level
Alcohol Clin Exp Res, Vol 38, No 10, 2014: pp 2664–2677
SEVERE ALCOHOL WITHDRAWAL PREDICTORS
of care changes. Early recognition of a propensity for
developing SAWS could have potentially significant impacts
on patient outcomes, healthcare resource utilization, and
improved prophylaxis and treatment strategies.
We conducted a review and synthesis of the existing pub-
lished literature reporting risk factors for SAWS to better
understand the strength of the evidence. Because there is no
standard definition of SAWS, we sought to reduce the heter-
ogeneity of the results while maximizing the amount of data
using 3 different pairwise outcomes: DT versus no-DT, sei-
zure versus no-seizure, and SAWS versus no-SAWS. SAWS
as an outcome in our meta-analysis is, therefore, a heteroge-
neous outcome. Our goal was to identify commonly avail-
able demographic and clinical criteria present at initial
evaluation of inpatients who were diagnosed with SAWS
during the index admission. We then performed a meta-
analysis of the reported results to calculate measures of
predictive strength.
MATERIALS AND METHODS
We applied the methods outlined in the Preferred Reporting
Items for Systematic Reviews and Meta-Analyses: The PRISMA
statement (Moher et al., 2009). Two authors (CMG and ISD)
performed a MeSH search in Ovid using the terms “substance
withdrawal syndrome” limited to “alcohol,” English language,
and epidemiologic studies published by December 2012. The last
search was conducted on April 21, 2013. Each of the resulting
references was reviewed for relevance. Bibliographies were ana-
lyzed to identify studies that may have been overlooked by the
initial search. The search was then duplicated in PubMed. The
abstracts of studies identified in the search were then screened by
1 author (CMG). Studies that described characteristics of inpa-
tients who experienced AWS were retained for full-text analysis.
Studies that did not compare patients with SAWS, DT as defined
by DSM, or seizure to patients with uncomplicated AWS were
excluded.
Summary baseline variables were extracted from each study by 1
author (CMG). The methods section of each study was reviewed
regarding inclusion and exclusion criteria, study setting, method of
analysis, and definitions such that the risk of bias of individual stud-
ies could be analyzed. Variables that were similarly defined by at
least 3 studies were entered into the meta-analysis (Review Man-
ager, RevMan version 5.2; The Nordic Cochrane Centre, The
Cochrane Collaboration, Copenhagen). We transformed units of
measurement to the most commonly reported measure, imperial
units. For measures of daily alcohol intake, we converted standard
drinks to grams (1 standard drink = 14 g ethanol [EtOH] unless
otherwise defined). For proportional variables, individual study
data were used to calculate a combined odds ratio (OR) with 95%
confidence intervals (CI) using the Mantel–Haenszel method. For
interval variables, we used the reported means and standard devia-
tions to derive a mean difference (MD) and 95% CIs. These analy-
ses were used to construct forest plots for point estimates and
confidence intervals using 3 outcomes: DT, seizure, and SAWS.
SAWS was defined by each individual study, and therefore, is a het-
erogeneous outcome compared with DT and seizures. In studies
that used both DT and seizures as outcomes, we used the DT data
in the SAWS analysis. For all variables, we used random-effects
models given the significant heterogeneity between studies (Boren-
stein and Higgins, 2013). We also constructed fixed-effect models as
a secondary analysis of the DT and seizure outcomes because these
meta-analyses were less heterogeneous. We calculated I2 values as a
2665
measure of heterogeneity and constructed funnel plots to evaluate
for publication bias.
RESULTS
We identified 226 epidemiologic studies of AWS published
in English. From an initial review of publication abstracts,
studies that did not include primary data regarding baseline
characteristics of inpatients with AWS were excluded. Forty-
three articles were determined to be of relevance and compre-
hensively reviewed. Ten articles were excluded for lack of a
standard definition of SAWS or for lack of an appropriate
control group. Details of these exclusions can be found in the
Data S1. Fifteen studies were excluded because the number
of patients was not reported or point estimates and/or ranges
of potential predictor variables were missing (Chan et al.,
2009; Coffey et al., 2007; Cushman, 1987; Dolman and Haw-
kes, 2005; Driessen et al., 2005; Findley et al., 2010; Frieling
et al., 2012; Hillemacher et al., 2006; Jacques et al., 2011;
Lukan et al., 2002; Palmstierna, 2001; Tadic et al., 2005; de
Timary et al., 2008; Wojnar et al., 1997, 2001). An addi-
tional study conducted exclusively in an outpatient setting
was also excluded (Ceccanti et al., 2006). Following these
exclusions, 17 studies were included in the qualitative review.
Of those 17 studies, 15 reported the primary findings using
comparable measures (e.g., means rather than medians) and
in sufficient detail to be incorporated in the meta-analysis
(Fig. 1).
One hundred and two variables were evaluated across
the 17 studies; 29 variables were comparably defined and
analyzed in at least 3 studies. For each of these variables,
summary statistics from each study were compiled to calcu-
late a combined OR or an MD with 95% CI. We analyzed
the data with regard to 3 outcomes: SAWS versus no-
SAWS, DT versus no-DT, and seizure versus no-seizure.
DT was defined according to DSM-III-R (American Psy-
chiatric Association, 1987) or DSM-IV (American Psychia-
tric Association, 1994) criteria. There was moderate
heterogeneity among the studies with regard to the criteria
for SAWS diagnosis. We therefore did not apply an exter-
nal standard for diagnosis and derived the criteria from
each individual study. Some studies defined SAWS as a
Cushman score ≥8 (Brousse et al., 2012; Mennecier et al.,
2008). DT correlates to a Cushman score of 15 (Mennecier
et al., 2008). SAWS was defined in 1 study according to a
modification of the Clinical Institute Withdrawal Assess-
ment of Alcohol Scale, Revised rating scale as an AWS
score ≥10 (Wetterling and Junghanns, 2000). Approxi-
mately half of patients who score ≥10 develop DT (Wetter-
ling et al., 1997). There was significant variation among the
studies with regard to the medical treatments administered
for the treatment of alcohol withdrawal. While treatment
was not standardized in a few studies (Ferguson et al.,
1996; Wright et al., 2006), symptom-triggered benzodiaze-
pines (Lee et al., 2005) or benzodiazepine-like clomethiazole
(Eyer et al., 2011; Monte et al., 2009; Wetterling et al.,
2666 GOODSON ET AL.

Fig. 1. Study flow diagram.

2676
sufficient to predict SAWS, DT, or seizures, and the avail-
able biochemical parameters are poorly calibrated to serve
as reliable biomarkers. It remains unclear if severity of
alcohol abuse disorder is a significant contributor to devel-
oping SAWS. The role of gender and medical comorbidi-
ties while incompletely evaluated thus far may yield
important insights into future prospective studies.
Taken together, retrospective analysis is insufficient to
derive a robust risk prediction model but provides strong
justification for future well-designed, prospective approaches
to derive and validate severity risk prediction and outcomes
models for SAWS and identify novel predictive biomarkers
for risk prediction, clinical stratification, and treatment
response measurement.
ACKNOWLEDGMENT
This research was funded by Denver Health Medical
Center Division of Pulmonary and Critical Care, Internal
Funding.
REFERENCES
American Psychiatric Association (1987) Diagnostic and Statistical Manual
of Mental Disorders. 3rd ed., text rev. Author, Washington, DC.
American Psychiatric Association (1994) Diagnostic and Statistical Manual
of Mental Disorders. 4th ed. Author, Washington, DC.
Ballenger JC, Post RM (1978) Kindling as a model for alcohol withdrawal
syndromes. Br J Psychiatry, 133:1–14.
Barrio E, Tome S, Rodriguez I, Gude F, Sanchez-Leira J, Perez-Becerra
E, Gonzalez-Quintela A (2004) Liver disease in heavy drinkers with
and without alcohol withdrawal syndrome. Alcohol Clin Exp Res
28:131–136.
Berggren U, Fahlke C, Berglund KJ, Blennow K, Zetterberg H, Balldin J
(2009) Thrombocytopenia in early alcohol withdrawal is associated with
development of delirium tremens or seizures. Alcohol Alcohol 44:382–386.
Bleich S, Bayerlein K, Hillemacher T, Degner D, Kornhuber J, Friel-
ing H (2006) An assessment of the potential value of elevated hom-
ocysteine in predicting alcohol-withdrawal seizures. Epilepsia 47:934–
938.
Borenstein M, Higgins JP (2013) Meta-analysis and subgroups. Prev Sci
14:134–143.
Brousse G, Arnaud B, Vorspan F, Richard D, Dissard A, Dubois M, Pic D,
Geneste J, Xavier L, Authier N, Sapin V, Llorca PM, De Chazeron I,
Minet-Quinard R, Schmidt J (2012) Alteration of glutamate/GABA bal-
ance during acute alcohol withdrawal in emergency department: a prospec-
tive analysis. Alcohol Alcohol 47:501–508.
Ceccanti M, Sasso GF, Nocente R, Balducci G, Prastaro A, Ticchi C, Ber-
tazzoni G, Santini P, Attilia ML (2006) Hypertension in early alcohol
withdrawal in chronic alcoholics. Alcohol Alcohol 41:5–10.
Chan GM, Hoffman RS, Gold JA, Whiteman PJ, Goldfrank LR, Nelson LS
(2009) Racial variations in the incidence of severe alcohol withdrawal. J
Med Toxicol 5:8–14.
Charlson ME, Pompei P, Ales KL, MacKenzie CR (1987) A new method of
classifying prognostic comorbidity in longitudinal studies: development
and validation. J Chronic Dis 40:373–383.
Coffey SF, Schumacher JA, Brady KT, Cotton BD (2007) Changes in PTSD
symptomatology during acute and protracted alcohol and cocaine absti-
nence. Drug Alcohol Depend 87:241–248.
Collins GB, Brosnihan KB, Zuti RA, Messina M, Gupta MK (1992) Neuro-
endocrine, fluid balance, and thirst responses to alcohol in alcoholics.
Alcohol Clin Exp Res 16:228–233.
GOODSON ET AL.
Cushman P Jr (1987) Delirium tremens. Update on an old disorder. Postgrad
Med 82:117–122.
Dolman JM, Hawkes ND (2005) Combining the audit questionnaire and
biochemical markers to assess alcohol use and risk of alcohol withdrawal
in medical inpatients. Alcohol Alcohol 40:515–519.
Driessen M, Lange W, Junghanns K, Wetterling T (2005) Proposal of a com-
prehensive clinical typology of alcohol withdrawal—a cluster analysis
approach. Alcohol Alcohol 40:308–313.
Dufour MC, Adamson MD (2003) The epidemiology of alcohol-induced
pancreatitis. Pancreas 27:286–290.
Eyer F, Schuster T, Felgenhauer N, Pfab R, Strubel T, Saugel B, Zilker T
(2011) Risk assessment of moderate to severe alcohol withdrawal–predic-
tors for seizures and delirium tremens in the course of withdrawal. Alcohol
Alcohol 46:427–433.
Ferguson JA, Suelzer CJ, Eckert GJ, Zhou XH, Dittus RS (1996) Risk
factors for delirium tremens development. J Gen Intern Med 11:410–
414.
Fiellin DA, O’Connor PG, Holmboe ES, Horwitz RI (2002) Risk for delir-
ium tremens in patients with alcohol withdrawal syndrome. Subst Abuse
23:83–94.
Findley JK, Park LT, Siefert CJ, Chiou GJ, Lancaster RT, Demoya M, Ger-
vasini A, Velmahos GC (2010) Two routine blood tests-mean corpuscular
volume and aspartate aminotransferase - as predictors of delirium tremens
in trauma patients. J Trauma 69:199–201.
Frieling H, Leitmeier V, Haschemi-Nassab M, Kornhuber J, Rhein M,
Bleich S, Hillemacher T (2012) Reduced plasma levels of asymmetric
di-methylarginine (ADMA) in patients with alcohol dependence normalize
during withdrawal. Eur Neuropsychopharmacol 22:836–840.
Hall W, Zador D (1997) The alcohol withdrawal syndrome. Lancet
349:1897–1900.
Higgins J, Green S, eds (2011) Cochrane Handbook for Systematic Reviews
of Interventions. Available at: www.cochrane-handbook.org. Accessed
May 9, 2014.
Hillemacher T, Bayerlein K, Wilhelm J, Poleo D, Frieling H, Ziegenbein M,
Sperling W, Kornhuber J, Bleich S (2006) Volume intake and craving in
alcohol withdrawal. Alcohol Alcohol 41:61–65.
Huang MC, Chen CH, Liu HC, Chen CC, Ho CC, Leu SJ (2011) Differential
patterns of serum brain-derived neurotrophic factor levels in alcoholic
patients with and without delirium tremens during acute withdrawal. Alco-
hol Clin Exp Res 35:126–131.
Jacques D, Zdanowicz N, Reynaert C, Janne P, Timary P (2011) Intensity of
symptoms from alcohol withdrawal in alcohol-dependent patients: com-
parison between smokers and non-smokers. Psychiatr Danub 23(Suppl 1):
S123–S125.
Jamal MM, Morgan TR (2003) Liver disease in alcohol and hepatitis C. Best
Pract Res Clin Gastroenterol 17:649–662.
Kraemer KL, Mayo-Smith MF, Calkins DR (2003) Independent clinical cor-
relates of severe alcohol withdrawal. Subst Abuse 24:197–209.
Lee JH, Jang MK, Lee JY, Kim SM, Kim KH, Park JY, Lee JH, Kim HY,
Yoo JY (2005) Clinical predictors for delirium tremens in alcohol depen-
dence. J Gastroenterol Hepatol 20:1833–1837.
Limosin F (2002) Clinical and biological specificities of female alcoholism.
Encephale 28:503–509.
Lukan JK, Reed DN Jr, Looney SW, Spain DA, Blondell RD (2002)
Risk factors for delirium tremens in trauma patients. J Trauma 53:901–
906.
Marik P, Mohedin B (1996) Alcohol-related admissions to an inner city hos-
pital intensive care unit. Alcohol Alcohol 31:393–396.
Mennecier D, Thomas M, Arvers P, Corberand D, Sinayoko L, Bonnefoy S,
Harnois F, Thiolet C (2008) Factors predictive of complicated or severe
alcohol withdrawal in alcohol. Gastroenterol Clin Biol 32:792–797.
Moher D, Liberati A, Tetzlaff J, Altman DG (2009) Preferred reporting
items for systematic reviews and meta-analyses: the PRISMA statement.
PLoS Med 6:e1000097.
Monte R, Rabu~ nal R, Casariego E, Bal M, Pertega S (2009) Risk factors for
delirium tremens in patients with alcohol withdrawal syndrome in a hospi-
tal setting. Eur J Intern Med 20:690–694.

Substance abuse variables
History of DT
Evaluated by 9 studies. 3 reported data
inadequately for aggregation (Kraemer
et al., 2003; Mennecier et al., 2008;
Wetterling et al., 2001)
History of alcohol withdrawal seizure
Evaluated by 7 studies. 3 reported
data inadequately for aggregation
(Kraemer et al., 2003; Mennecier et al.,
2008; Wetterling et al., 2001)
Prior inpatient alcohol detoxifications
Evaluated by 10 studies. 6 reported
data inadequately for aggregation
(Fiellin et al., 2002; Kraemer et al.,
2003; Mennecier et al., 2008; Monte
et al., 2009; Shaw et al., 1998;
Wetterling et al., 2001)
History of prior AWS
Evaluated by 4 studies. 1 reported
data inadequately for aggregation
(Shaw et al., 1998)
Duration of alcohol abuse
Evaluated by 8 studies. 3 reported
data inadequately for aggregation
(Kraemer et al., 2003; Shaw et al.,
1998; Wetterling et al., 2001)
Daily intake of alcohol
Evaluated by 9 studies. 4 reported
data inadequately for aggregation
(Kraemer et al., 2003; Mennecier
et al., 2008; Shaw et al., 1998;
Wetterling et al., 2001)
Age at onset of alcohol abuse
Evaluated by 3 studies. 1 reported
data inadequately for aggregation
(Mennecier et al., 2008)
Time since last drink
Evaluated by 6 studies. 5 reported
data inadequately for aggregation
(Ferguson et al., 1996; Fiellin et al.,
2002; Kraemer et al., 2003; Lee et al.,
2005; Shaw et al., 1998)
DT, delirium tremens; SAWS, severe alcohol withdrawal syndrome; OR, odds ratio; CI, confidence interval; MD, mean difference; AWS, alcohol withdrawal syndrome; Z, test statistic of test for
overall effect.
Table 2. Meta-Analysis of Substance Abuse Variables
Strength to predict DT versus no-DT
OR significant based on 6 studies (Berggren et al.,
2009; Eyer et al., 2011; Ferguson et al., 1996;
Fiellin et al., 2002; Lee et al., 2005; Wright et al.,
2006) with a total of 1,637 patients. 206 patients
had DT. OR 2.58 (95% CI 1.41 to 4.7). Z = 3.09,
p = 0.002. Heterogeneity I2 = 59%
OR not significant based on 4 studies (Berggren
et al., 2009; Eyer et al., 2011; Fiellin et al.,
2002; Huang et al., 2011) with a total of 1,269
patients. 96 patients had DT. OR 1.78 (95%
CI 0.87 to 3.64). Z = 1.57, p = 0.12.
Heterogeneity I2 = 33%
MD not significant based on 3 studies (Wetterling
et al., 1994; Wojnar et al., 1999; Wright et al.,
2006) with a total of 1,296 patients. 1,170 patients
had DT. MD 0.82 episodes (95% CI 2.17, 0.53).
Z = 1.19; p = 0.24. Heterogeneity I2 = 37%
OR not significant based on 2 studies (Eyer et al.,
2011; Ferguson et al., 1996) with a total of 1,027
patients. 94 patients had DT. OR 0.6 (95% CI 0.29
to 1.24). Z = 1.38, p = 0.17. Heterogeneity
I2 = 63%
MD not significant based on 4 studies (Eyer et al.,
2011; Huang et al., 2011; Wetterling et al., 1994;
Wojnar et al., 1999) with a total of 2,127 patients.
1,210 patients had DT. MD 2.06 years (95% CI
0.7, 4.81). Z = 1.46, p = 0.14. Heterogeneity
I2 = 70%
MD not significant based on 4 studies (Eyer et al.,
2011; Huang et al., 2011; Monte et al., 2009;
Wright et al., 2006) with a total of 1,251 patients.
246 patients had DT. MD 6.55 g (95% CI
34.35, 21.26). Z = 0.46, p = 0.64. Heterogeneity
I2 = 79%
Only evaluated by 1 study (Huang et al., 2011)
Only evaluated by 1 study (Monte et al., 2009)
Strength to predict SAWS versus no-SAWS
Same as analysis for DT versus no-DT
Same as analysis for DT versus no-DT
MD not significant based on 4 studies (Bleich
et al., 2006; Wetterling et al., 1994; Wojnar
et al., 1999; Wright et al., 2006) with a total
of 1,384 patients. 1,188 patients had SAWS.
MD 0.41 episodes (95% CI 1.59, 0.78).
Z = 0.67, p = 0.5. Heterogeneity I2 = 78%
OR not significant based on 3 studies
(Brousse et al., 2012; Eyer et al., 2011;
Ferguson et al., 1996) with a total of
1,106 patients. 117 patients had SAWS.
OR 1.2 (95% CI 0.33 to 4.35). Z = 0.28,
p = 0.78. Heterogeneity I2 = 89%
MD not significant based on 5 studies with
a total of 2,215 patients (Bleich et al., 2006;
Eyer et al., 2011; Huang et al., 2011;
Wetterling et al., 1994; Wojnar et al., 1999).
1,228 patients had SAWS. MD 1.33 years
(95% CI 1.11, 3.77). Z = 1.07, p = 0.29.
Heterogeneity I2 = 72%
MD not significant based on 5 studies (Eyer
et al., 2011; Huang et al., 2011; Monte et al.,
2009; Wojnar et al., 1999; Wright et al., 2006)
with a total of 3,184 patients. 298 patients had
SAWS. MD 21.85 g (95% CI 55.2, 11.51).
Z = 1.28, p = 0.2. Heterogeneity I2 = 88%
MD not significant based on 2 studies (Huang
et al., 2011; Wojnar et al., 1999) with a total of
1,993 patients. 74 patients had SAWS. MD
0.05 years (95% CI 2.23, 2.14). Z = 0.04,
p = 0.97. Heterogeneity I2 = 0%
Only evaluated by 1 study (Monte et al., 2009)
2668
Strength to predict seizure versus no-seizure
OR not significant based on 2 studies
(Berggren et al., 2009; Eyer et al.,
2011) with a total of 1,143 patients.
69 patients had a seizure. OR
0.8 (95% CI 0.36 to 1.74). Z = 0.57, p = 0.57.
Heterogeneity I2 = 0%
OR significant based on 2 studies (Berggren
et al., 2009; Eyer et al., 2011) with a total of
1,143 patients. 69 patients had a seizure. OR
2.8 (95% CI 1.09 to 7.19). Z = 3.85,
p = 0.0001. Heterogeneity I2 = 44%
MD not significant based on 2
studies with a total of 2,021 patients
(Bleich et al., 2006; Wojnar et al., 1999).
70 patients had a seizure. MD 0.01 episodes
(95% CI 0.35, 0.32). Z = 0.08, p = 0.94.
Heterogeneity I2 = 0%
Only evaluated by 1 study (Eyer et al., 2011)
MD not significant based on 3
studies (Bleich et al., 2006; Eyer et al.,
2011; Wojnar et al., 1999) with a total of
2,848 patients. 131 patients had a seizure.
MD 0.11 years (95% CI 1.27, 1.48).
Z = 0.15, p = 0.88. Heterogeneity I2 = 0%
MD not significant based on 2 studies (Eyer
et al., 2011; Wojnar et al., 1999) with a total
of 2,760 patients. 113 patients had a seizure.
MD 48.23 g (95% CI 106.53, 10.07).
Z = 1.62, p = 0.1. Heterogeneity I2 = 81%
Only evaluated by 1 study (Wojnar et al., 1999)
Not evaluated
GOODSON ET AL.

Initial diagnostic results variables
Platelet count
Evaluated by 5 studies. 1 reported
data inadequately for aggregation
(Kraemer et al., 2003)
Potassium
Evaluated by 7 studies. 3 reported
data inadequately for aggregation
(Ferguson et al., 1996; Kraemer
et al., 2003; Lee et al., 2005)
ALT
Evaluated by 7 studies. 2 reported
data inadequately for aggregation
(Lee et al., 2005; Wojnar et al.,
1999)
GGT
Evaluated by 7 studies. 2 reported
data inadequately for aggregation
(Lee et al., 2005; Mennecier et al.,
2008)
Sodium
Evaluated by 6 studies. 3 reported
data inadequately for aggregation
(Ferguson et al., 1996; Kraemer
et al., 2003; Lee et al., 2005)
BAL
Evaluated by 5 studies. 3 reported
data inadequately for aggregation
(Brousse et al., 2012; Fiellin et al.,
2002; Kraemer et al., 2003)
AST
Evaluated by 9 studies. 5 reported
data inadequately for aggregation
(Brousse et al., 2012; Ferguson
et al., 1996; Kraemer et al., 2003;
Lee et al., 2005; Mennecier et al.,
2008)
Systolic blood pressure
Evaluated by 7 studies. 4 reported
data inadequately for aggregation
(Ferguson et al., 1996; Fiellin et al.,
2002; Kraemer et al., 2003; Lee et
al., 2005)
Table 3. Meta-Analysis of Initial Diagnostic Results Variables
Strength to predict DT versus no-DT
MD significant based on 4 studies (Berggren
et al., 2009; Eyer et al., 2011; Huang et al.,
2011; Monte et al., 2009) with a total of 1,527
patients. 227 patients had DT. MD 45.64/mm3
(95% CI 75.95, 15.33). Z = 2.95, p = 0.003.
Heterogeneity I2 = 76%
MD significant based on 4 studies (Berggren
et al., 2009; Eyer et al., 2011; Monte et al.,
2009; Wetterling et al., 1994) with a total of
1,525 patients. 222 patients had DT.
MD 0.26 mEq/l (95% CI 0.45, 0.08).
Z = 2.75, p = 0.006. Heterogeneity
I2 = 71%
MD not significant based on 4 studies
(Berggren et al., 2009; Huang et al., 2011;
Monte et al., 2009; Wetterling et al., 1994)
with a total of 757 patients. 200 patients had
DT. MD 6.09 U/l (95% CI 5.51, 17.69).
Z = 1.03, p = 0.3. Heterogeneity I2 = 0%
MD not significant based on 5 studies
(Berggren et al., 2009; Eyer et al., 2011;
Huang et al., 2011; Monte et al., 2009;
Wetterling et al., 1994) with a total of 1,586
patients. 245 patients had DT. MD 86.09 U/l
(95% CI 36.1, 208.29). Z = 1.38, p = 0.17.
Heterogeneity I2 = 43%
MD not significant based on 3 studies (Eyer
et al., 2011; Monte et al., 2009; Wetterling
et al., 1994) with a total of 1,191 patients.
212 patients had DT. MD 1.69 mEq/l
(95% CI 3.73, 0.36). Z = 1.61, p = 0.11.
Heterogeneity I2 = 79%
Only evaluated by 1 study (Eyer et al., 2011)
MD not significant based on 4 studies
(Berggren et al., 2009; Huang et al.,
2011; Monte et al., 2009; Wetterling
et al., 1994) with a total of 760 patients.
200 patients had DT. MD 16.41 U/l (95%
CI 2.92, 35.74). Z = 1.66, p = 0.1.
Heterogeneity I2 = 7%
MD not significant based on 3 studies
(Berggren et al., 2009; Eyer et al.,
2011; Monte et al., 2009) with a total
of 1,464 patients. 203 patients had DT.
MD 1.73 mmHg (95% CI 11.72, 8.26).
Z = 0.34, p = 0.73. Heterogeneity I2 = 86%
Strength to predict SAWS versus no-SAWS
Same analysis for DT versus no-DT
Same analysis for DT versus no-DT
MD significant based on 5 studies (Berggren
et al., 2009; Huang et al., 2011; Mennecier
et al., 2008; Monte et al., 2009; Wetterling
et al., 1994) with a total of 939 patients. 220
patients had SAWS. MD 20.97 U/l (95%
CI 0.89, 41.05). Z = 2.05, p = 0.04.
Heterogeneity I2 = 54%
Same analysis as for DT versus no-DT
Same analysis for DT versus no-DT
MD not significant based on 2 studies with
a total of 915 patients (Bleich et al., 2006;
Eyer et al., 2011). 64 patients had SAWS.
MD 45.28 mg/dl (95% CI 152.1, 61.53).
Z = 0.83, p = 0.41. Heterogeneity I2 = 89%
Same analysis as for DT versus no-DT
Same analysis for DT versus no-DT
Strength to predict seizure versus no-seizure
MD significant based on 2 studies (Berggren et al.,
2009; Eyer et al., 2011) with a total of 1,160 patients.
69 patients had a seizure. MD 59.91/mm3 (95%
CI 105.69, 14.13). Z = 2.57, p = 0.01.
Heterogeneity I2 = 78%
MD significant based on 2 studies (Berggren et al.,
2009; Eyer et al., 2011) with a total of 1,160 patients.
69 patients had a seizure. MD 0.2 mEq/l (95% CI
0.3, 0.1). Z = 3.85, p = 0.0001. Heterogeneity
I2 = 0%
SEVERE ALCOHOL WITHDRAWAL PREDICTORS
Only evaluated by 1 study (Berggren et al., 2009)
MD significant based on 2 studies (Berggren et al.,
2009; Eyer et al., 2011) with a total of 1,160 patients.
69 patients had a seizure. MD 202.56 U/l (95% CI
3.62, 401.5). Z = 2, p = 0.05. Heterogeneity I2 = 0%
Only evaluated by 1 study (Eyer et al., 2011)
MD not significant based on 2 studies with a total of
915 patients (Bleich et al., 2006; Eyer et al., 2011).
79 patients had a seizure. MD 4.33 mg/dl (95% CI
29.35, 38.01). Z = 0.25, p = 0.8. Heterogeneity
I2 = 0%
Only evaluated by 1 study (Berggren et al., 2009)
MD not significant based on 2 studies (Berggren et al.,
2009; Eyer et al., 2011) with a total of 1,160 patients.
69 patients had a seizure. MD 1.03 mmHg (95%
CI 7.14, 5.08). Z = 0.33, p = 0.74. Heterogeneity
I2 = 6%
Continued.
2669

Initial diagnostic results variables
Diastolic blood pressure
Evaluated by 3 studies
Heart rate
Evaluated by 5 studies. 2 reported
data inadequately for aggregation
(Fiellin et al., 2002; Lee et al., 2005)
White blood cell count
Evaluated by 3 studies. 1 reported
data inadequately for aggregation
(Berggren et al., 2009)
Mean corpuscular volume
Evaluated by 4 studies. 2 reported
data inadequately for aggregation
(Berggren et al., 2009; Mennecier
et al., 2008)
Creatinine
Evaluated by 5 studies. 1 reported
data inadequately for aggregation
(Kraemer et al., 2003)
Chloride
Evaluated by 3 studies. 2 reported
data inadequately for aggregation
(Kraemer et al., 2003; Lee et al.,
2005)
Albumin
Evaluated by 4 studies. 3 reported
data inadequately for aggregation
(Ferguson et al., 1996; Kraemer
et al., 2003; Lee et al., 2005)
DT, delirium tremens; SAWS, severe alcohol withdrawal syndrome; OR, odds ratio; CI, confidence interval; MD, mean difference; ALT, alanine aminotransferase; GGT, gamma-glutamyl transpep-
tidase; BAL, blood alcohol level; AST, aspartate aminotransferase; Z, test statistic of test for overall effect.
2670
Table 3. (Continued)
Strength to predict DT versus no-DT Strength to predict SAWS versus no-SAWS Strength to predict seizure versus no-seizure
MD not significant based on 3 studies Same analysis for DT versus no-DT MD not significant based on 2 studies (Berggren et al.,
(Berggren et al., 2009; Eyer et al., 2011; 2009; Eyer et al., 2011) with a total of 1,160 patients.
Monte et al., 2009) with a total of 1,464 69 patients had a seizure. MD 0.86 mmHg (95%
patients. 203 patients had DT. MD 0.92 CI 4.38, 2.66). Z = 0.48, p = 0.63.
mmHg (95% CI 7.99, 6.15). Z = 0.25, Heterogeneity I2 = 0%
p = 0.8. Heterogeneity I2 = 77%
MD not significant based on 3 studies Same analysis for DT versus no-DT MD not significant based on 2 studies (Berggren et al.,
(Berggren et al., 2009; Eyer et al., 2011; 2009; Eyer et al., 2011) with a total of 1,160 patients.
Monte et al., 2009) with a total of 1,464 69 patients had a seizure. MD 0.44 bpm (95% CI
patients. 203 patients had DT. MD 0.08 4.08, 4.96). Z = 0.19, p = 0.85. Heterogeneity
bpm (95% CI 2.8, 2.64). Z = 0.06, I2 = 0%
p = 0.95. Heterogeneity I2 = 0%
MD not significant based on 2 studies Same analysis for DT versus no-DT Not evaluated
(Huang et al., 2011; Monte et al., 2009)
with a total of 366 patients. 171 patients
had DT. MD 0.2/mm3 (95% CI 0.43, 0.83).
Z = 0.62, p = 0.54. Heterogeneity I2 = 0%
MD not significant based on 2 studies Same analysis for DT versus no-DT Not evaluated
(Monte et al., 2009; Wetterling et al., 1994)
with a total of 600 patients. 48 patients had
DT. MD 0.3 fl (95% CI 1.63, 2.22). Z = 0.3,
p = 0.76. Heterogeneity I2 = 0%
MD not significant based on 4 studies (Eyer Same analysis for DT versus no-DT Only evaluated by 1 study (Eyer et al., 2011)
et al., 2011; Monte et al., 2009; Wetterling
et al., 1994; Wright et al., 2006) with a total
of 1,601 patients. 240 patients had DT. MD 0
mg/dl (95% CI 0.05, 0.05). Z = 0.19, p = 0.85.
Heterogeneity I2 = 41%
Only evaluated by 1 study (Wetterling et al., Only evaluated by 1 study (Wetterling et al., Not evaluated
1994) 1994)
Only evaluated by 1 study (Monte et al., Only evaluated by 1 study (Monte et al., 2009) Not evaluated
2009)
GOODSON ET AL.
SEVERE ALCOHOL WITHDRAWAL PREDICTORS
1994), or fixed doses of benzodiazepines (Fiellin et al.,
2002; Huang et al., 2011) were the common strategies. The
findings are summarized in Tables 1–3.
Of the demographic and comorbidity variables including
age, male gender, chronic liver, or pancreatic disease
(Table 1), none were found to be different between
patients with and without SAWS. However, the presence
of an acute medical illness was more likely in patients who
developed DT versus no-DT using a fixed-effect model.
The details of the analyses are presented in the online
Supplementary data.
Among variables associated with substance abuse
(Table 2), a history of DT was predictive of an incident diag-
nosis of DT (OR 2.58 vs. no-DT; 95% CI 1.41, 4.7;
I2 = 59%) and a history of alcohol withdrawal seizures was
predictive of incident seizures (OR 2.8 vs. no-seizures; 95%
CI 1.09, 7.19; I2 = 44%). In fixed-effect but not random-
effects modeling, the number of prior inpatient admissions
for alcohol detoxification, a history of prior AWS, and the
duration of alcohol abuse was associated with developing
DT. Similarly, the amount of daily intake of alcohol was
associated with developing seizures in fixed-effect but not
random-effects modeling. Further details of these analyses
are presented in Data S1.
Fig. 2. Forest plot of categorical variables significantly associated with delirium tremens (DT) versus no-DT. I2 = test statistic of test for heterogeneity,
Z = test statistic of test for overall effect.
Fig. 3. Forest plot of categorical variables significantly associated with seizure versus no-seizure. I2 = test statistic of test for heterogeneity, Z = test
statistic of test for overall effect.
2671
Nine studies evaluated a history of DT as a predictor of
SAWS (Berggren et al., 2009; Eyer et al., 2011; Ferguson
et al., 1996; Fiellin et al., 2002; Kraemer et al., 2003; Lee
et al., 2005; Mennecier et al., 2008; Wetterling et al., 2001;
Wright et al., 2006). Three of these studies could not be
included in the meta-analysis because they did not report
event rates for both milder AWS and more severe SAWS
cohorts (Kraemer et al., 2003; Mennecier et al., 2008; Wet-
terling et al., 2001). Six studies reporting a total of 1,637
patients, 206 of whom had DT, were included. Heterogeneity
was significant among these studies (I2 = 59%). In this
analysis, a history of DT correlated with an incident episode
of DT (OR 2.58 vs. no-DT; 95% CI 1.41, 4.7) as shown in
Fig. 2. In addition, 2 studies evaluated a history of DT as a
risk factor for seizures (Berggren et al., 2009; Eyer et al.,
2011). When aggregated, 69 of 1,143 patients in these 2 stud-
ies experienced a seizure (6.0%). Heterogeneity was low
(I2 = 0%). The analysis identified that a history of DT did
not correlate with a probability of seizure during the incident
admission (OR 0.8 vs. no-seizure; 95% CI 0.36, 1.74). Of
note, these studies used different methods to determine his-
tory of DT. Wright and colleagues (2006) reported prior DT
from the medical record, 2 studies used patient-reported data
(Eyer et al., 2011; Fiellin et al., 2002), and 3 studies did not
2672
specify their source (Berggren et al., 2009; Ferguson et al.,
1996; Lee et al., 2005).
Seven studies evaluated a patient-reported history of
seizures as a predictor of SAWS (Berggren et al., 2009;
Eyer et al., 2011; Fiellin et al., 2002; Huang et al., 2011;
Kraemer et al., 2003; Mennecier et al., 2008; Wetterling
et al., 2001). Three of these could not be included in the
meta-analysis because of inadequate reporting. Four stud-
ies with 1,269 patients were included in the meta-analysis;
96 patients had DT. Heterogeneity was moderate
(I2 = 33%). We found no association between history of
seizures and incident occurrence of DT (OR vs. no-DT
1.78; 95% CI 0.87, 3.64). In addition, the 2 studies report-
ing seizure as an outcome evaluated history of seizure as
a risk factor for an incident seizure (69 of 1,143 patients;
Berggren et al., 2009; Eyer et al., 2011). Heterogeneity
was moderate (I2 = 44%). The analysis identified that
prior seizures did correlate with incident seizure (OR 2.8
vs. no-seizure; 95% CI 1.09, 7.19) as shown in Fig. 3.
Initial laboratory diagnostic variables (Table 3) were
evaluated as potential predictors. Platelet count and potas-
sium level were both significantly lower in patients with
DT (platelet count MD 45.64/mm3 vs. no-DT, 95% CI
75.95, 15.33; potassium level MD 0.26 mEq/l vs. no-
DT, 95% CI 0.45, 0.08) and in patients who experi-
enced seizures (platelet count MD 59.91/mm3 vs. no-sei-
zure, 95% CI 105.69, 14.13; potassium level MD
0.2 mEq/l vs. no-seizures, 95% CI 0.3, 0.1). Alanine
aminotransferase (ALT) was higher in patients with
SAWS (MD 20.97 U/l vs. no-SAWS; 95% CI 0.89,
41.05), and gamma-glutamyl transpeptidase (GGT) was
higher in patients with seizures (MD 202.56 U/l vs. no-sei-
zure; 95% CI 3.62, 401.5). Random-effects modeling for
other diagnostic variables failed to identify significant dif-
ferences. However, fixed-effect models identified lower
Fig. 4. Forest plot of numeric variables significantly associated with delirium tremens (DT) versus no-DT. Please note scales differ. I2 = test statistic of
test for heterogeneity, MD = mean difference, Z = test statistic of test for overall effect.
GOODSON ET AL.
sodium level as predictive of incident DT. Additional
details of these analyses are reported in Data S1.
Five studies evaluated initial platelet count as predictive of
SAWS (Berggren et al., 2009; Eyer et al., 2011; Huang et al.,
2011; Kraemer et al., 2003; Monte et al., 2009). One study
could not be included because of inadequate reporting (Kra-
emer et al., 2003). Two hundred and twenty-seven of 1,527
aggregated patients from 4 studies had DT (14.9%). Hetero-
geneity was considerable (I2 = 76%). Platelets were signifi-
cantly lower in patients with DT (MD 45.64/mm3 vs.
no-DT; 95% CI 75.95, 15.33) as shown in Fig. 4. In addi-
tion, 2 studies evaluated initial platelet count as a predictor
of seizures (Berggren et al., 2009; Eyer et al., 2011). Sixty-
nine of 1,160 aggregated patients had seizures (5.9%).
Heterogeneity was considerable (I2 = 78%). The analysis
identified admission platelet level as significantly lower in
patients with seizures (MD 59.91/mm3 vs. no-seizure;
95% CI 105.69, 14.13) as shown in Fig. 6.
Seven studies evaluated initial potassium as a predictor of
SAWS (Berggren et al., 2009; Eyer et al., 2011; Ferguson
et al., 1996; Kraemer et al., 2003; Lee et al., 2005; Monte
et al., 2009; Wetterling et al., 1994). Three studies could not
be included in the meta-analysis because of inadequate
reporting (Ferguson et al., 1996; Kraemer et al., 2003; Lee
et al., 2005). Two hundred and twenty-two of 1,525 patients
aggregated from 4 studies had DT (14.6%). Heterogeneity
was substantial (I2 = 71%). The analysis identified that ini-
tial potassium level was lower in patients with DT (MD
0.26 mEq/l vs. no-DT; 95% CI 0.45, 0.08) as shown in
Fig. 4. In addition, 2 studies evaluated initial potassium as a
predictor of seizures (Berggren et al., 2009; Eyer et al.,
2011). Sixty-nine of 1,160 patients had seizures (5.9%). Het-
erogeneity was low (I2 = 0%). Initial potassium was also
lower in patients with incident seizures (MD 0.2 mEq/l vs.
no-seizures; 95% CI 0.3, 0.1) as shown in Fig. 6.
SEVERE ALCOHOL WITHDRAWAL PREDICTORS 2673

Fig. 5. Forest plot of numeric variables significantly associated with severe alcohol withdrawal syndrome (SAWS) versus no-SAWS. I2 = test statistic
of test for heterogeneity, MD = mean difference, Z = test statistic of test for overall effect.

Seven studies evaluated initial serum ALT in relation to
SAWS (Berggren et al., 2009; Huang et al., 2011; Lee et al.,
2005; Mennecier et al., 2008; Monte et al., 2009; Wetterling
et al., 1994; Wojnar et al., 1999). Two studies could not be
included in the meta-analysis because of inadequate report-
ing (Lee et al., 2005; Wojnar et al., 1999). Two hundred of
757 aggregated patients from 4 studies had DT (26.4%).
Heterogeneity was low (I2 = 0%). We found no association
between initial ALT and DT (MD 6.09 U/l vs. no-DT; 95%
CI 5.51, 17.69). Mennecier and colleagues (2008) evaluated
ALT in relation to SAWS defined as a Cushman score ≥8.
The Cushman score is a validated scoring system for AWS.
A score of 8 is considered severe, 12 is considered compli-
cated (includes hallucinations), and 15 indicates DT (Cush-
man, 1987). When combined with the 4 DT studies from our
initial analysis, we identified 220 of 939 aggregated patients
with SAWS (23.4%). Heterogeneity was substantial
(I2 = 54%). In this second analysis, higher ALT was signifi-
cantly associated with SAWS (MD 20.97 U/l vs. no-SAWS;
95% CI 0.89, 41.05) as shown in Fig. 5.
Seven studies evaluated GGT as a predictor of SAWS
(Berggren et al., 2009; Eyer et al., 2011; Huang et al., 2011;
Lee et al., 2005; Mennecier et al., 2008; Monte et al., 2009;
Wetterling et al., 1994). Two studies could not be included in
the meta-analysis because of inadequate reporting (Lee
et al., 2005; Mennecier et al., 2008). The 5 included studies
reported a total of 245 of 1,586 patients aggregated from 5
studies had DT (15.45%). Heterogeneity was moderate
(I2 = 43%). We found no association between GGT and DT
(MD 86.09 U/l vs. no-DT; 95% CI 36.1, 208.29). Addi-
tionally, 2 studies evaluated GGT as a predictor of seizures.
Sixty-nine of 1,160 in those 2 studies had seizures (5.9%).
Heterogeneity was low (I2 = 0%). The analysis identified
higher GGT as a predictor of incident seizures (MD
202.56 U/l vs. no-seizure; 95% CI 3.62, 401.5) as shown in
Fig. 6.

Fig. 6. Forest plot of numeric variables significantly associated with seizure versus no-seizure. Please note scales differ. I2 = test statistic of test for
heterogeneity, MD = mean difference, Z = test statistic of test for overall effect.
2674
Funnel plots were constructed for visual analysis (Higgins
and Green, 2011) that demonstrated asymmetry that was
subtle for the variables reported as proportional but marked
for interval data. These figures can be found in Data S1.
DISCUSSION
The primary finding of this analysis was that prediction of
SAWS is highly variable, and few demographic, clinical, or
biochemical parameters are consistently predictive of a
SAWS episode. The analysis did identify that an incident
occurrence of DT or alcohol withdrawal seizures was signifi-
cantly predicted by history of a similar event. Both a lower
initial platelet count and serum potassium level were predic-
tive of an incident occurrence of DT, seizures, and SAWS.
Higher initial ALT was commonly seen in patients with
SAWS. Higher initial serum GGT was also consistently asso-
ciated with incident alcohol withdrawal seizures. The pub-
lished reports included in the analyses were predominantly
retrospective cohort or case–control series and of intermedi-
ate to low quality. Data synthesis was limited by significant
heterogeneity in methods, outcomes, and reporting among
these studies.
Age as a determinant of severity of AWS, while highly het-
erogeneous, was not a significant factor. Our analysis verified
the individual study findings that gender is also not a signifi-
cant predictor of severity of AWS in this typically male-
dominated population. This finding is somewhat at odds
with prior observations that females are more prone to alco-
hol-related organ damage, less likely to seek care for their
substance disorder, develop alcoholism later in life, and
develop alcoholism with lower daily intake of alcohol (Barrio
et al., 2004; Limosin, 2002). While there is no clear mecha-
nistic reason for this discordance, reconciliation of these
observations will require analysis in SAWS populations that
include greater numbers of females.
The role of acute or chronic medical illness as a risk factor
for SAWS is unclear from the results of this meta-analysis.
The 2 studies that evaluated acute medical illness included in
this analysis were conducted in different countries, and so
the disparity in results suggests a potential unmeasured con-
founding variable. The limited data regarding the contribu-
tion of chronic medical conditions suggest either these
contribute little to the severity of AWS or are systematically
under reported in this patient population. Future, prospec-
tive investigations regarding the contribution of chronic
comorbid conditions would be significantly enhanced by uni-
form evaluation such as the Charlson score (Charlson et al.,
1987).
A surprising finding was that history of liver disease, and
pancreatitis was not statistically correlated with severity of
AWS in our analysis. The studies we evaluated, however,
include patient populations with a relatively lower prevalence
of these diseases than reported for patients with alcohol use
disorders. Dufour and Adamson (2003) report a 5 to 15%
lifetime risk for chronic pancreatitis in alcoholic patients,
GOODSON ET AL.
only slightly higher than the prevalence in our meta-analysis.
Lifetime risk for cirrhosis is estimated to be 15 to 20%
(Jamal and Morgan, 2003). In contrast, Barrio and col-
leagues (2004) reported a lower rate of cirrhosis in patients
with AWS than in all alcoholics. In that study, alcoholics
who developed cirrhosis had more continuous alcohol use
rather than episodic “binge” drinking. The episodes of bing-
ing were hypothesized to be the primary precipitant of
SAWS (Barrio et al., 2004). The authors determined that
alcoholic hepatitis was closely associated with SAWS in that
population. Appreciating the observation that females are
more prone to alcohol-related organ damage with lower
amounts of alcohol (Limosin, 2002), a potential explanation
for these data is the possibility that females are more likely to
be steady versus episodic drinkers.
The clinical spectrum of SAWS previously experienced by
a patient appears to be an important predictor of both the
severity and clinical manifestations of an index, recurrent
SAWS episode. In this respect, seizures and DT appear to be
discrete manifestations with little overlap. Of the studies that
evaluated patients for both outcomes, Berggren and col-
leagues (2009) reported no patients with both seizures and
DT. Eyer and colleagues (2011) reported that 15 to 20% of
the patients with either seizure or DT also had the other con-
dition, with a suggestion that having one condition increased
the likelihood for the other. Aside from potassium and plate-
lets, most of our findings were significantly predictive for DT
but not seizures or vice versa. This suggests that they may be
distinct pathologies and perhaps should not be aggregated
into 1 syndromic category.
Analysis of patients who had previously experienced
AWS yielded novel findings. There were trends toward
fewer prior AWS symptoms and fewer admissions for
detoxification in patients with DT. This seems to contra-
dict the kindling theory (Ballenger and Post, 1978) which
postulates that episodes of withdrawal are progressive. A
notable study by Wojnar and colleagues (1999) reported
that only a subset of patients with consecutive admissions
for AWS experienced progressively severe episodes of
AWS. Our analyses support the hypothesis that the
majority of patients with sequential episodes continue to
experience uncomplicated AWS. In connection with the
hypothesis of Barrio and colleagues (2004), it is possible
that the subset of patients with progressive AWS drinks
more episodically than steadily.
Measures of severity of alcoholism have not yet shown a
clear correlation with severity of AWS. Longer duration of
abuse correlates with DT and lower daily alcohol intake cor-
relates with seizure only in the less-stringent fixed-effect
models. Perhaps these results are skewed from significance
by the patients themselves. Indeed, some authors elected not
to record these data as they felt that patient-reported data on
this subject were too unreliable (Berggren et al., 2009; Lee
et al., 2005).
The contribution of polysubstance abuse to SAWS has
not been studied sufficiently. It is logical that benzodiazepine
SEVERE ALCOHOL WITHDRAWAL PREDICTORS
or barbiturate abuse would lead to more severe AWS given
the similarity of their neurochemical effect with that of
EtOH. Some (Kraemer et al., 2003; Wojnar et al., 1999) but
not all (Eyer et al., 2011) studies support this hypothesis,
and it deserves further study.
Among initial clinical vital signs our analysis identified no
predictors of SAWS; a result that correlates with most prior
studies. By contrast, among the laboratory values analyzed,
there were a few notable findings. Initial values of serum
platelet count and potassium level were lower in DT, SAWS,
and seizures. These results are consistent with those of the
individual studies.
Alcohol is toxic to the bone marrow resulting in decreased
megakaryocyte production (Berggren et al., 2009). This is an
acute and sustained response to alcohol in all patients. We
found no studies to evaluate whether this is a dose–response
relationship. Thrombocytopenia is also a well-known result
of cirrhosis. We found that platelet count was lower in
patients with SAWS. If driven by a higher cumulative alco-
hol intake in SAWS patients, this would indicate that sever-
ity of alcoholism is the true correlate of SAWS. This
hypothesis is not supported by our findings from patient-
provided variables. Alternately, this relationship may be dri-
ven by higher acute alcohol intake. With reference to Barrio
and colleagues’ (2004) hypothesis of episodic versus steady
alcohol intake, thrombocytopenia may indicate a more
severe recent binge. This would imply that the severity of epi-
sodic drinking is the true correlate of SAWS. The result
could be driven by a higher incidence of cirrhosis in the
SAWS group; the rest of our data does not support this.
Hypokalemia has many physiologic effects that may
explain its correlation with SAWS. Increased catecholamine
activity in AWS activates the sodium–potassium ATPase
pump causing hypokalemia (Eyer et al., 2011). Chronic alco-
hol abuse is associated with increased angiotensin II levels
yet suppressed vasopressin levels and unchanged renin and
aldosterone levels (Collins et al., 1992). DT is marked by
elevated vasopressin levels despite low-serum osmolality
(Trabert et al., 1992). Wetterling and colleagues (1994)
hypothesized that this inappropriately high vasopressin level
drives low osmolality and hypokalemia in DT. Elevated
vasopressin should cause more hyponatremia than hypokale-
mia. However, if the sodium–potassium ATPase is also acti-
vated, hyponatremia would be partly compensated and
hypokalemia worsened. This may explain why hyponatremia
was not significantly associated with SAWS in our meta-
analysis, only reaching significance in fixed-effect modeling
for DT and SAWS.
Higher ALT was found in patients with SAWS, but there
was no difference in aspartate aminotransferase (AST). Typi-
cally AST and ALT are increased in a 2-to-1 ratio in alcohol-
related diseases. It is probable that the ALT result in our
analysis is falsely positive as it is driven by 1 study. That
study used Cushman score ≥8 as the outcome, a definition,
that is, not as strict as DT (Mennecier et al., 2008). Notably,
a lack of difference in AST and ALT in patients with SAWS
2675
versus no-SAWS would not support Barrio and colleagues’
(2004) finding that alcoholic hepatitis is more common in
SAWS.
In our analysis, higher initial GGT level was significantly
associated with seizure. GGT is increased by moderate or
greater alcohol consumption and liver disease (Niemela,
2007). Our findings, therefore, are similar to our analysis for
thrombocytopenia. Degree of GGT elevation may corre-
spond either to severity of recent alcohol binge or cumulative
alcohol intake, and thus higher risk for SAWS.
With respect to blood alcohol level (BAL), our analysis
supports the conclusion that there is no significant relation-
ship to severity of AWS. The finding of a lower BAL in
patients with SAWS in fixed-effect modeling is likely spuri-
ous as the majority of the studies were not included in the
meta-analysis.
There are several important limitations and confounders
of this analysis. While we conducted a comprehensive
search and review of the published literature, we were not
able to include non-English publications. Heterogeneity of
study and reporting methods as well as definitions limited
our ability to aggregate data. Additionally, there was evi-
dence of systemic bias as indicated by the asymmetry of
the funnel plots. We did not consider treatment approaches
for DT or SAWS for the purposes of inclusion in the
analysis. Therefore, intervention bias is likely to impact our
conclusions. A spectrum of treatment methods was
employed including no standardization or benzodiazepine-
based protocols. It is reasonable to hypothesize that treat-
ment affects the incidence of SAWS and that this variable
contributes to the heterogeneity in our results and is the
focus of continued analysis. An additional important
potential confounder is publication bias including language
of publication. We did find that results sometimes varied
greatly with the dominant ethnicity of the study popula-
tion. Most studies were retrospective in design with impor-
tant and potentially significant inferences regarding
systematic biases in reported outcomes and analysis report-
ing bias. Finally, true heterogeneity between and within the
selected studies and potential SAWS predictors are likely
to have contributed to our findings. Specifically patterns of
drinking behavior are likely to be highly heterogeneous.
Our analysis is revealing with regard to the insufficien-
cies of previous studies designed to derive and validates a
reliable risk prediction score for SAWS. Lack of standard
definitions, insufficient statistical power, or a true lack of
predictive power in the evaluated variables may all have
contributed. In our analyses, we used standard definitions
where possible and leveraged the increased power through
meta-analysis. The majority of variables we included are
identifiable from routine admission data. We conclude that
the course of prior episodes of AWS is the most reliable
predictor of subsequent episodes. Thrombocytopenia and
hypokalemia should be viewed as potential warning signs
for SAWS although we were not able to identify cut off
points for these values. However, no single variable is
2676
sufficient to predict SAWS, DT, or seizures, and the avail-
able biochemical parameters are poorly calibrated to serve
as reliable biomarkers. It remains unclear if severity of
alcohol abuse disorder is a significant contributor to devel-
oping SAWS. The role of gender and medical comorbidi-
ties while incompletely evaluated thus far may yield
important insights into future prospective studies.
Taken together, retrospective analysis is insufficient to
derive a robust risk prediction model but provides strong
justification for future well-designed, prospective approaches
to derive and validate severity risk prediction and outcomes
models for SAWS and identify novel predictive biomarkers
for risk prediction, clinical stratification, and treatment
response measurement.
ACKNOWLEDGMENT
This research was funded by Denver Health Medical
Center Division of Pulmonary and Critical Care, Internal
Funding.
REFERENCES
American Psychiatric Association (1987) Diagnostic and Statistical Manual
of Mental Disorders. 3rd ed., text rev. Author, Washington, DC.
American Psychiatric Association (1994) Diagnostic and Statistical Manual
of Mental Disorders. 4th ed. Author, Washington, DC.
Ballenger JC, Post RM (1978) Kindling as a model for alcohol withdrawal
syndromes. Br J Psychiatry, 133:1–14.
Barrio E, Tome S, Rodriguez I, Gude F, Sanchez-Leira J, Perez-Becerra
E, Gonzalez-Quintela A (2004) Liver disease in heavy drinkers with
and without alcohol withdrawal syndrome. Alcohol Clin Exp Res
28:131–136.
Berggren U, Fahlke C, Berglund KJ, Blennow K, Zetterberg H, Balldin J
(2009) Thrombocytopenia in early alcohol withdrawal is associated with
development of delirium tremens or seizures. Alcohol Alcohol 44:382–386.
Bleich S, Bayerlein K, Hillemacher T, Degner D, Kornhuber J, Friel-
ing H (2006) An assessment of the potential value of elevated hom-
ocysteine in predicting alcohol-withdrawal seizures. Epilepsia 47:934–
938.
Borenstein M, Higgins JP (2013) Meta-analysis and subgroups. Prev Sci
14:134–143.
Brousse G, Arnaud B, Vorspan F, Richard D, Dissard A, Dubois M, Pic D,
Geneste J, Xavier L, Authier N, Sapin V, Llorca PM, De Chazeron I,
Minet-Quinard R, Schmidt J (2012) Alteration of glutamate/GABA bal-
ance during acute alcohol withdrawal in emergency department: a prospec-
tive analysis. Alcohol Alcohol 47:501–508.
Ceccanti M, Sasso GF, Nocente R, Balducci G, Prastaro A, Ticchi C, Ber-
tazzoni G, Santini P, Attilia ML (2006) Hypertension in early alcohol
withdrawal in chronic alcoholics. Alcohol Alcohol 41:5–10.
Chan GM, Hoffman RS, Gold JA, Whiteman PJ, Goldfrank LR, Nelson LS
(2009) Racial variations in the incidence of severe alcohol withdrawal. J
Med Toxicol 5:8–14.
Charlson ME, Pompei P, Ales KL, MacKenzie CR (1987) A new method of
classifying prognostic comorbidity in longitudinal studies: development
and validation. J Chronic Dis 40:373–383.
Coffey SF, Schumacher JA, Brady KT, Cotton BD (2007) Changes in PTSD
symptomatology during acute and protracted alcohol and cocaine absti-
nence. Drug Alcohol Depend 87:241–248.
Collins GB, Brosnihan KB, Zuti RA, Messina M, Gupta MK (1992) Neuro-
endocrine, fluid balance, and thirst responses to alcohol in alcoholics.
Alcohol Clin Exp Res 16:228–233.
GOODSON ET AL.
Cushman P Jr (1987) Delirium tremens. Update on an old disorder. Postgrad
Med 82:117–122.
Dolman JM, Hawkes ND (2005) Combining the audit questionnaire and
biochemical markers to assess alcohol use and risk of alcohol withdrawal
in medical inpatients. Alcohol Alcohol 40:515–519.
Driessen M, Lange W, Junghanns K, Wetterling T (2005) Proposal of a com-
prehensive clinical typology of alcohol withdrawal—a cluster analysis
approach. Alcohol Alcohol 40:308–313.
Dufour MC, Adamson MD (2003) The epidemiology of alcohol-induced
pancreatitis. Pancreas 27:286–290.
Eyer F, Schuster T, Felgenhauer N, Pfab R, Strubel T, Saugel B, Zilker T
(2011) Risk assessment of moderate to severe alcohol withdrawal–predic-
tors for seizures and delirium tremens in the course of withdrawal. Alcohol
Alcohol 46:427–433.
Ferguson JA, Suelzer CJ, Eckert GJ, Zhou XH, Dittus RS (1996) Risk
factors for delirium tremens development. J Gen Intern Med 11:410–
414.
Fiellin DA, O’Connor PG, Holmboe ES, Horwitz RI (2002) Risk for delir-
ium tremens in patients with alcohol withdrawal syndrome. Subst Abuse
23:83–94.
Findley JK, Park LT, Siefert CJ, Chiou GJ, Lancaster RT, Demoya M, Ger-
vasini A, Velmahos GC (2010) Two routine blood tests-mean corpuscular
volume and aspartate aminotransferase - as predictors of delirium tremens
in trauma patients. J Trauma 69:199–201.
Frieling H, Leitmeier V, Haschemi-Nassab M, Kornhuber J, Rhein M,
Bleich S, Hillemacher T (2012) Reduced plasma levels of asymmetric
di-methylarginine (ADMA) in patients with alcohol dependence normalize
during withdrawal. Eur Neuropsychopharmacol 22:836–840.
Hall W, Zador D (1997) The alcohol withdrawal syndrome. Lancet
349:1897–1900.
Higgins J, Green S, eds (2011) Cochrane Handbook for Systematic Reviews
of Interventions. Available at: www.cochrane-handbook.org. Accessed
May 9, 2014.
Hillemacher T, Bayerlein K, Wilhelm J, Poleo D, Frieling H, Ziegenbein M,
Sperling W, Kornhuber J, Bleich S (2006) Volume intake and craving in
alcohol withdrawal. Alcohol Alcohol 41:61–65.
Huang MC, Chen CH, Liu HC, Chen CC, Ho CC, Leu SJ (2011) Differential
patterns of serum brain-derived neurotrophic factor levels in alcoholic
patients with and without delirium tremens during acute withdrawal. Alco-
hol Clin Exp Res 35:126–131.
Jacques D, Zdanowicz N, Reynaert C, Janne P, Timary P (2011) Intensity of
symptoms from alcohol withdrawal in alcohol-dependent patients: com-
parison between smokers and non-smokers. Psychiatr Danub 23(Suppl 1):
S123–S125.
Jamal MM, Morgan TR (2003) Liver disease in alcohol and hepatitis C. Best
Pract Res Clin Gastroenterol 17:649–662.
Kraemer KL, Mayo-Smith MF, Calkins DR (2003) Independent clinical cor-
relates of severe alcohol withdrawal. Subst Abuse 24:197–209.
Lee JH, Jang MK, Lee JY, Kim SM, Kim KH, Park JY, Lee JH, Kim HY,
Yoo JY (2005) Clinical predictors for delirium tremens in alcohol depen-
dence. J Gastroenterol Hepatol 20:1833–1837.
Limosin F (2002) Clinical and biological specificities of female alcoholism.
Encephale 28:503–509.
Lukan JK, Reed DN Jr, Looney SW, Spain DA, Blondell RD (2002)
Risk factors for delirium tremens in trauma patients. J Trauma 53:901–
906.
Marik P, Mohedin B (1996) Alcohol-related admissions to an inner city hos-
pital intensive care unit. Alcohol Alcohol 31:393–396.
Mennecier D, Thomas M, Arvers P, Corberand D, Sinayoko L, Bonnefoy S,
Harnois F, Thiolet C (2008) Factors predictive of complicated or severe
alcohol withdrawal in alcohol. Gastroenterol Clin Biol 32:792–797.
Moher D, Liberati A, Tetzlaff J, Altman DG (2009) Preferred reporting
items for systematic reviews and meta-analyses: the PRISMA statement.
PLoS Med 6:e1000097.
Monte R, Rabu~ nal R, Casariego E, Bal M, Pertega S (2009) Risk factors for
delirium tremens in patients with alcohol withdrawal syndrome in a hospi-
tal setting. Eur J Intern Med 20:690–694.
SEVERE ALCOHOL WITHDRAWAL PREDICTORS 2677

Niemela O (2007) Biomarkers in alcoholism. Clin Chim Acta 377:39–49. Wright T, Myrick H, Henderson S, Peters H, Malcolm R (2006) Risk factors
Palmstierna T (2001) A model for predicting alcohol withdrawal delirium. for delirium tremens: a retrospective chart review. Am J Addict 15:213–
Psychiatr Serv 52:820–823. 219.
Secades RM, Vales EC, Diaz SP, Rey RR, Zemsch MP, Fernandez SP
(2008) Clinical course and features of the alcohol withdrawal syndrome in
a general hospital. Rev Clin Esp 208:506–512. SUPPORTING INFORMATION
Shaw GK, Waller S, Latham CJ, Dunn G, Thomson AD (1998) The detoxi-
fication experience of alcoholic in-patients and predictors of outcome. Additional Supporting Information may be found in the
Alcohol Alcohol 33:291–303.
online version of this article:
Tadic A, Dahmen N, Szegedi A, Rujescu D, Giegling I, Koller G, Anghe-
lescu I, Fehr C, Klawe C, Preuss UW, Sander T, Toliat MR, Singer P, Fig. S1. Forest plot of categorical variables and associa-
Bondy B, Soyka M (2005) Polymorphisms in the NMDA subunit 2B are tion with DT vs. no-DT.
not associated with alcohol dependence and alcohol withdrawal-induced Fig. S2. Forest plot of categorical variables and associa-
seizures and delirium tremens. Eur Arch Psychiatry Clin Neurosci tion with SAWS vs. no-SAWS.
255:129–135.
Fig. S3. Forest plot of categorical variables and associa-
de Timary P, Luts A, Hers D, Luminet O (2008) Absolute and relative
stability of alexithymia in alcoholic inpatients undergoing alcohol tion with seizure vs. no-seizure.
withdrawal: relationship to depression and anxiety. Psychiatry Res 157: Fig. S4. Forest plot of numerical variables and association
105–113. with DT vs. no-DT.
Trabert W, Caspari D, Bernhard P, Biro G (1992) Inappropriate vaso- Fig. S5. Forest plot of numerical variables and association
pressin secretion in severe alcohol withdrawal. Acta Psychiatr Scand
with SAWS vs. no-SAWS.
85:376–379.
Wetterling T, Driessen M, Kanitz RD, Junghanns K (2001) The severity of Fig. S6. Forest plot of numerical variables and association
alcohol withdrawal is not age dependent. Alcohol Alcohol 36:75–78. with seizure vs. no-seizure.
Wetterling T, Junghanns K (2000) Psychopathology of alcoholics during Fig. S7. Funnel plot of categorical variables and associa-
withdrawal and early abstinence. Eur Psychiatry 15:483–488. tion with DT vs. no-DT.
Wetterling T, Kanitz RD, Besters B, Fischer D, Zerfass B, John U, Spranger
Fig. S8. Funnel plot of categorical variables and associa-
H, Driessen M (1997) A new rating scale for the assessment of the alcohol-
withdrawal syndrome (AWS scale). Alcohol Alcohol 32:753–760. tion with SAWS vs. no-SAWS.
Wetterling T, Kanitz RD, Veltrup C, Driessen M (1994) Clinical predictors Fig. S9. Funnel plot of categorical variables and associa-
of alcohol withdrawal delirium. Alcohol Clin Exp Res 18:1100–1102. tion with seizure vs. no-seizure.
Wojnar M, Bizon Z, Wasilewski D (1999) Assessment of the role of kindling Fig. S10. Funnel plot of numerical variables and associa-
in the pathogenesis of alcohol withdrawal. Alcohol Clin Exp Res 23:
tion with DT vs. no-DT.
204–208.
Wojnar M, Wasilewski D, Matsumoto H, Cedro A (1997) Differences in the Fig. S11. Funnel plot of numerical variables and associa-
course of alcohol withdrawal in women and men: a Polish sample. Alcohol tion with SAWS vs. no-SAWS.
Clin Exp Res 21:1351–1355. Fig. S12. Funnel plot of numerical variables and associa-
Wojnar M, Wasilewski D, Zmigrodzka I, Grobel I (2001) Age-related differ- tion with seizure vs. no-seizure.
ences in the course of alcohol withdrawal in hospitalized patients. Alcohol
Data S1. Additional results.
Alcohol 36:577–583.