OUP user menu

Risk Assessment of Moderate to Severe Alcohol Withdrawal—Predictors for Seizures and Delirium Tremens in the Course of Withdrawal

Florian Eyer , Tibor Schuster , Norbert Felgenhauer , Rudi Pfab , Tim Strubel , Bernd Saugel , Thomas Zilker
DOI: http://dx.doi.org/10.1093/alcalc/agr053 427-433 First published online: 18 May 2011

Abstract

Aims: To develop a prediction model for withdrawal seizures (WS) and delirium tremens (DT) during moderate to severe alcohol withdrawal syndrome (AWS) in a large cohort of inpatients treated for AWS (= 827). Methods: Re-analysis of a cohort study population treated between 2000 and 2009. All patients received a score-guided and symptom-triggered therapy for AWS. Multivariable binary logistic regression models with stepwise variable selection procedures were conducted providing odds ratio (OR) estimates. Results: In the multivariable regression, significant predictors of WS during AWS therapy were a delayed climax of withdrawal severity since admission [OR/10 h: 1.23; 95% confidence interval (CI): 1.1–1.4; P < 0.001)], prevalence of structural brain lesions in the patient's history (OR 6.5; 95% CI: 3.0–14.1; P < 0.001) and WS as the cause of admittance (OR 2.6; 95% CI: 1.4–4.8; P = 0.002). Significant predictors at admission for the occurrence of DT were lower serum potassium (OR/1 mmol/l 0.33; 95% CI: 0.17–0.65; P = 0.001), a lower platelet count (OR/100.000 0.42; 95% CI: 0.26–0.69; P = 0.001) and prevalence of structural brain lesions (OR 5.8; 95% CI: 2.6–12.9; P < 0.001). Conclusion: In this large retrospective cohort, some easily determinable parameters at admission may be useful to predict a complicated course of alcohol withdrawal regarding the occurrence of WS or DT. Using the provided nomograms, clinicians can estimate the percentage likelihood of patients to develop either WS or DT during their course of withdrawal. Prevalence of structural brain lesions in the patient's history does strongly warrant a careful observation of patients.

INTRODUCTION

Complications related to alcohol withdrawal account for a significant demand in healthcare resources and are associated with an increase of morbidity and mortality (Lerner and Fallon, 1985; Ferguson et al., 1996; Palmstierna, 2001). The reported numbers of patients who undergo a complicated course of alcohol withdrawal vary widely between 5 and 20% and are dependent on several factors such as the clinical setting of withdrawal, the applied therapeutic approach for alcohol withdrawal syndrome (AWS) and individual characteristics of patients, but it is generally accepted that withdrawal seizure (WS) or delirium tremens (DT) do occur in the range of 6–15 and 4–15%, respectively (Saitz and O'malley, 1997; Mennecier et al., 2008). Most studies indicate that WS is reported in the lifetime histories of between only 5 and 10% of alcohol-dependent individuals (Brown et al., 1988; Schuckit et al., 1993). Slightly higher rates might be observed among alcohol-dependent men and women with excessively high levels of alcohol intake over relatively longer periods of time (Victor and Adams, 1953). Regarding DT as a major complication of AWS, clinical reviews indicate that the lifetime risk that an alcohol-dependent individual will ever have a full-blown DT condition is between 5 and 10% (Hemmingsen et al., 1979; Schuckit et al., 1993, 1995).

The treatment strategy could be optimized if patients with a higher risk of developing WS or DT could be identified. Even when treatment is initiated at the onset of DT or WS, the course is often volatile and unpredictable (Findley et al., 2010). These complications may be prevented through prompt and intensive treatment, including clomethiazole or benzodiazepines (Saitz et al., 1994; Lee et al., 2005). Thus, prediction and prevention become the cornerstones of avoiding DT- or WS-related morbidity. However, existing literature data suggesting associations between patient factors and the risk for DT or WS either lack empirical data or are limited by an inadequate sample size or lack of multivariable analyses (Cushman, 1987; Milne et al., 1991). The relative rarity of WS or DT and the associated need for large clinical samples makes it difficult to determine which, if any, easily determinable parameters are the most relevant to predict WS or DT. The purpose of this study was therefore to identify, among patients hospitalized in our toxicological unit for alcohol withdrawal, the most parsimonious collection of risk factors present at the time of hospital admission that were predictive for the development of WS or DT.

METHODS

Patient selection and withdrawal schedule

This was a secondary analysis of a cohort of adult patients admitted to our unit seeking for alcohol detoxification (both elective and emergency admissions) between 2000 and 2009 that fulfilled the International Classification of Diseases (ICD-10) criteria of alcohol dependence and completed a protocol-driven course of alcohol detoxification. Details of patient selection are described elsewhere (Eyer et al., 2011). Severity of AWS was determined in analogy to the revised Clinical Institute Withdrawal Assessment for Alcohol (CIWA-Ar) scale (Sullivan et al., 1989) by a validated and standardized 11-item withdrawal score (AWS score) (Wetterling et al., 1997). Of a total number of 2691 charts that were primarily reviewed, 827 patients were finally included in this analysis. Given the retrospective nature of this study, it was exempt from the assessment by the institutional ethics committee.

Patients received a score-guided pharmacological treatment with orally administered clomethiazole (CMZ) to achieve a minimum of withdrawal symptoms as evaluated with the AWS score (usually every 2–4 h). Clonidine in the presence of noradrenergic hyperactivity and haloperidol in the case of productive hallucinations were administered as needed. Additional treatment with antiepileptic drugs (AED) using either valproic acid (VPA; n = 453) or carbamazepine (CBZ; n = 374) started simultaneously with CMZ. All patients received 100 mg oral thiamin thrice daily. The decision to start AWS treatment was score-guided and independent of the breath alcohol concentration at this time.

Data collection

The charts were reviewed to obtain data on demographics, withdrawal history and alcohol or drug use, laboratory data on admission, presence of cirrhosis (child classification), blood pressure and heart rate, both on admission and during the course of AWS, as well as medical and psychiatric comorbidity. Charts were also reviewed for known or reported structural brain lesions at admission (e.g. cerebral trauma or hemorrhage in the past, benign or malign brain tumor, neurosurgical interventions in the past, epilepsy). The amount of daily alcohol consumption reported by the patient (if recorded) was calculated as: grams ethanol = volume of the drink [ml] × 0.8 × alcohol content [%]/100. Complications (WS, DT), length and quantity of medical treatment as well as AWS score upon admission and during AWS treatment were collected. We compared the patient's withdrawal history, clinical and some analytical parameters of patients who developed DT (DT+) and/or WS (WS+) with those who did not (WS−, DT−).

Statistical analysis

Data were entered and analyzed using PASW Statistics version 18.0 (SPSS Inc., Chicago, IL, USA). Mean and standard deviation (±SD) were used to describe quantitative data meeting normal distribution according to the Kolmogorov–Smirnov test. Variables that were not normally distributed were expressed as median and interquartile range. As appropriate, the chi-square or Fisher's exact test was used to compare categorical data between independent groups. The Student's t-test or Mann–Whitney U-test was used to compare continuous data distribution between two independent samples.

Accounting for a potential bias due to different additional treatment with AED in our study population (CBZ in n = 374 patients or VPA in n = 453 patients) as well as for a potential imbalance of baseline parameters between these subgroups, a stepwise multiple regression analysis of binary frequency data was performed. Specifically, we aimed to detect risk differences independent of the used treatment regimen. In this term, confounder-adjusted estimates of odds ratios (ORs) have been reported with 95% confidence intervals (CIs). For the purpose of illustration and clinical applicability, nomograms were created on the basis of the final regression model (Harrell et al., 1996). In these nomograms, model-based score points are displayed for each predictor variable category, which have to be summarized for any individual patient data. For the resulting total number of points, the corresponding predicted probability can be deduced from the nomograms. All statistical tests were conducted at a two-sided 0.05 level of significance.

RESULTS

We reviewed a total of 827 charts of inpatients who developed an AWS severe enough to be treated with a protocol-driven withdrawal schedule. Of these, 46 patients (5.6%) developed DT and in 61 patients (7.4%) the AWS coursed with WS.

Comparing patients with WS + and WS–, we found significant differences regarding a higher incidence of reported WS in previous withdrawal episodes (59 vs. 41%; P = 0.007), a higher number of known structural brain lesions (21 vs. 4%; P < 0.001) and a higher incidence of WS as the cause of admittance (34 vs. 14%; P < 0.001). Patients with WS + also developed significantly more often DT in the course of AWS compared with patients with WS− (15 vs. 5%; = 0.004). Details are depicted in Table 1. On admission, patients with WS + showed significantly lower serum potassium (4.0 ± 0.4 vs. 4.2 ± 0.5 mmol/l; P = 0.029) and lower platelets (150 ± 81 vs. 189 ± 94 g/l; P = 0.001). Patients with WS + showed also a significant later climax of withdrawal severity measured with the AWS-score compared with patients with WS− (34 ± 36 vs. 18 ± 19 h; P < 0.001). For details, see Table 2. As expected, patients with WS + needed a significantly longer medical treatment (127 ± 87 vs. 86 ± 49 h; P < 0.001), had a longer in-hospital stay (12 ± 7 vs. 8 ± 5 day; P < 0.001), needed higher cumulative doses of clomethiazole to resolve AWS (8.1 ± 5.9 vs. 5.7 ± 4.4 g; P < 0.001) and were treated more frequently in intensive care unit (16 vs. 3.5%; P < 0.001; data not shown).

View this table:
Table 1.

Baseline characteristics of patients that experienced seizures

Seizures n = 61No Seizures n = 766P-value
Age (years)44 (10)45 (10)0.630
Male gender44 (72%)562 (73%)0.834
Preexisting comorbid conditions
 Chronic pancreatitis1 (2%)20 (3%)1.000
 Liver cirrhosis8 (13%)91 (12%)0.775
 Duration of addiction years16 (9)15 (10)0.519
 Approximated daily ethanol intakea(g)215 (141)233 (170)0.527
Addiction-related historyb
 Previous seizures36 (59%)314 (41%)0.007
 Delirium7 (12%)105 (14%)0.614
 Previous withdrawal45 (74%)544 (71%)0.681
 Structural cerebral lesions13 (21%)29 (4%)<0.001
Cause of admittance
 Manifest withdrawal8 (13%)114 (15%)0.725
 Elective for withdrawal15 (25%)242 (32%)0.253
 Ethanol intoxicationc17 (28%)301 (39%)0.076
 Seizures21 (34%)109 (14%)<0.001
Occurrence of delirium9 (15%)37 (5%)0.004
  • Data are expressed as mean (SD) for continuous variables and as frequency (n,%) for categorical variables. Significant P-values (P < 0.05) are indicated in italics.

  • aThe amount of daily alcohol consumption reported by the patient (if recorded) was calculated according to the formula: grams ethanol = volume of the drink [ml] × 0.8 × alcohol content [%]/100.

  • bAddiction-related history was derived from the patients themselves either at admission or during inpatient course.

  • cPatients developed subsequently AWS and were treated according to our detoxification protocol.

View this table:
300 Multiple Choices

This is a pdf-only article and there is no markup to show you.

    Comparing patients with DT + and DT−, we found significant differences regarding higher age (49 ± 10 vs. 45 ± 10 a; P = 0.023), lower approximated daily ethanol intake (164 ± 139 vs. 236 ± 169 g; P = 0.007), less frequent prior withdrawal episodes (52 vs. 72%; P = 0.006), but a significant higher rate of reported structural brain lesions (26 vs. 4%; P < 0.001). Patients with DT + were also admitted less frequently electively for alcohol withdrawal (11 vs. 32%; P = 0.002) or due to ethanol intoxication (20 vs. 40%; P = 0.007), but were admitted significantly more often with WS as the cause of admission (46 vs. 14%; P < 0.001). Those patients with DT + suffered significantly more often from WS in the course of AWS compared with patients with DT− (20 vs. 7%; P = 0.004). For details, see Table 3.

    View this table:
    Table 3.

    Baseline characteristics of patients that experienced delirium

    Delirium n = 46No delirium n = 781P-value
    Age (years)49 (10)45 (10)0.023I
    Male gender33 (72%)573 (73%)0.808
    Preexisting comorbid conditions
     Chronic pancreatitis021 (3%)0.624
     Liver cirrhosis7 (15%)92 (12%)0.485
     Duration of addiction years19 (9)14 (10)0.565
     Approximated daily ethanol intakea(g)164 (139)236 (169)0.007
    Addiction-related historyb
     Previous seizures22 (48%)328 (42%)0.450
     Delirium9 (20%)103 (13%)0.198
     Previous withdrawal24 (52%)565 (72%)0.006
     Structural cerebral lesions12 (26%)30 (4%)<0.001
    Cause of admittance
     Manifest withdrawal11 (24%)111 (14%)0.067
     Elective for withdrawal5 (11%)252 (32%)0.002
     Ethanol intoxicationc9 (20%)309 (40%)0.007
     Seizures21 (46%)109 (14%)<0.001
    Occurrence of WS9 (20%)52 (7%)0.004
    • Data are expressed as mean (SD) for continuous variables and as frequency (n,%) for categorical variables. Significant P-values (P < 0.05) are indicated in italics.

    • aThe amount of daily alcohol consumption reported by the patient (if recorded) was calculated according to the formula: grams ethanol = volume of the drink [ml] × 0.8 × alcohol content [%]/100.

    • bAddiction-related history was derived from the patients themselves either at admission or during inpatient course.

    • cPatients developed subsequently AWS and were treated according to our detoxification protocol.

    Patients with DT + showed significantly higher γGT-values (753 ± 953 vs. 399 ± 599 U/l; P < 0.001), lower serum sodium (138 ± 6 vs. 140 ± 5 mmol/l; P = 0.023) and potassium (3.8 ± 05 vs. 4.2 ± 0.5 mmol/l; P < 0.001) as well as a lower platelet count (119 ± 62 vs. 190 ± 93 g/l; P < 0.001). Of note, patients with DT + had a lower serum ethanol concentration (1.4 ± 1.7 vs. 2.4 ± 1.6 g/l; P < 0.001) at admission compared with patients with DT–. As demonstrated for WS + , patients with DT + also showed a significantly later climax of withdrawal severity measured with the AWS-score compared with patients with DT− (36 ± 28 vs. 18 ± 20 h; P < 0.001). For details, see Table 4. Not surprisingly, patients with DT + needed also significantly longer healthcare resources in regard to the length of stay (13 ± 11 vs. 8 ± 5 day; P < 0.001), medical treatment (135 ± 103 vs. 86 ± 49 h; P = 0.002) and need for intensive care treatment (59  1.3%; P < 0.001) (data not shown).

    View this table:
    300 Multiple Choices

    This is a pdf-only article and there is no markup to show you.

      In the multivariable regression model, significant predictors for WS during AWS independent of the administered therapy were a delayed climax of withdrawal severity since admission (OR for every increase of 10 h: 1.23; 95% CI: 1.1–1.4; P < 0.001), prevalence of structural brain lesions in the patient's history (OR 6.5; 95% CI: 3.0–14.1; P < 0.001) and WS as the cause of admittance (OR 2.6; 95% CI: 1.4–4.8; P = 0.002). The c-index of this prediction model was 0.73 (95% CI: 0.66–0.88). Significant predictors at admission for occurrence of DT independent of the administered therapy were lower serum potassium (OR per an increase of 1 mmol/l: 0.33; 95% CI: 0.17–0.65; P = 0.001), a lower platelet count (OR per an increase of 100.000: 0.42; 95% CI: 0.26–0.69; P = 0.001) and prevalence of structural brain lesions (OR 5.8; 95% CI: 2.6–12.9; P < 0.001). The c-index of this prediction model was 0.81 (95% CI: 0.74–0.87).

      Figures 1 and 2 provide nomograms for risk calculation for WS or DT, respectively, using each of the underlying parameters that retained significant in the stepwise logistic regression model. Each parameter (numeric and categorical) is related to a given score resulting in a net sum score allowing judgment of a corresponding predicted probability to develop each complication, WS or DT. According to the underlying data, the span of likelihood to develop DT ranges from 0.1 to 70%, whereas the span of likelihood to develop WS ranges from 5 to 80%.

      Fig. 1.

      Nomogram to predict the probability for developing WS during alcohol withdrawal. Each variable of the multivariable logistic regression model is combined with a given score (upper scale: ‘Points’). Summarizing these scores (total points) and connecting this total point score rectangular to the nomogram, one can easily depict the risk for a given patient to suffer from WS in the course of AWS.

      Fig. 2.

      Nomogram to predict the probability for developing DT during alcohol withdrawal. Each variable of the multivariable logistic regression model is combined with a given score (upper scale: ‘Points’). Summarizing these scores (total points) and connecting this total point score rectangular to the nomogram, one can easily depict the risk for a given patient to suffer from DT in the course of AWS.

      DISCUSSION

      The present study corroborates the importance of the addiction-related history of an individual patient that adds significantly to our appraisal of which patients should deserve our special attention during their course of withdrawal. The multivariable logistic regression model revealed that three factors significantly predicted each, WS or DT.

      For WS, a delayed climax of withdrawal severity measured with the AWS-score, WS as the cause of admittance as well as the prevalence of structural brain lesions in the history, predicted significantly for WS. Patients with WS + developed also DT three times as often during their course of AWS compared with the WS− patients. Whereas the first predictor for WS is not yet established, there is more or less consensus about the remaining predictors. One set of theories especially relevant to WS involves kindling, a condition where the brain is likely to demonstrate progressively more severe motor overactivity following repeated less intense (subthreshold) electrical stimulations in the past (Goddard et al., 1969). Both animal and human studies indicate a greater likelihood of severe AWS in individuals who have had a greater number of withdrawal episodes in the past (Ballenger Post, 1978; Clemmesen and Hemmingsen, 1984). It is more difficult to understand why a delayed climax of withdrawal severity, measured with the AWS-score, is a positive predictor for WS. This could in part be due to the possibility that the highest attention toward the withdrawing patient is certainly in the early phase when aggressive treatment of AWS is warranted. In those cases where severe AWS is absent in the early phase, it is conceivable that a less intensive treatment leads to a higher number of break-through WS. Even though the treatment of AWS was protocol-driven in our study, final dosing of GABAergic agents (e.g. clomethiazole) was on the discretion of the responsible healthcare provider at this time. However, there is no current evidence to support this theory, and other factors are likely to play a role as well.

      In the literature, known predictors for WS include a previous history of such events, previous detoxification admissions and a CIWA-A score higher than 15 (Lechtenberg and Worner, 1990; Morton et al., 1994; Rathlev et al., 2000). However, we could neither find relevant differences regarding previous detoxification episodes nor differences in the AWS score that was fairly identical between WS +  and WS− in our investigation.

      For DT, lower serum potassium at admission, lower platelet count and the prevalence of structural brain lesions predicted significantly for DT in the further course. In fact, a physiologically related hypothesis of severe AWS involves changes in electrolytes, especially potassium or magnesium (Tonnesen, 1982). Of these, hypokalemia is frequently reported to be associated with DT or WS. Serum potassium was normal in chronic alcoholics on admission to hospital, but in patients in whom DT developed it fell steadily, leading to hypokalemia when DT started and returning to normal when DT resolved (Wadstein and Skude, 1978; Manhem et al., 1984). It is speculated that an increased shift of potassium across the cell membrane could be due to an activation of Na+/K+ ATPase in the cells mediated by catecholamines as well as a direct effect of catecholamines through β2 stimulation (Flatman and Clausen, 1979; Clausen and Flatman, 1980). This effect seems to be mainly associated with elevated adrenaline levels, but independent of the renin activity or aldosterone levels (Manhem et al., 1984).

      Another study investigated risk factors to develop DT. They demonstrated that patients with (a) elevated systolic blood pressure, (b) medical comorbidity and (c) prior complicated AWSs (DT and/or WS) who showed these features (alone or in combination) were associated with an increased risk of DT (Fiellin et al., 2002). Patients with a combination of a clinical feature (one or two) together with a prior complicated course of withdrawal were at highest risk to develop DT. While some studies also found an increased risk of DT in patients with a prior history of this disorder (Whitfield et al., 1978; Schuckit et al., 1995), others did not (Essardas Daryanani et al., 1994; Ferguson et al., 1996). In line with our results of a stepwise multiple regression analysis, the study of Palmstierna found that a history of DT and WS contributed only 6 and 6.8%, respectively, to the risk of developing DT, despite their high correlations in the single regression analysis, showing that the practical importance of this correlation may be small. Instead, these authors found that the prevalence of a current medical disease, tachycardia (>120 bpm) and AWS—despite an accompanying blood alcohol level higher than 1 g/l—were the strongest predictors for DT (Palmstierna, 2001).

      A comparison of a total of 211 alcohol-dependent subjects who reported about WS and/or DT in previous withdrawal episodes with 1437 controls showed that those with histories of a complicated course of AWS (WS + and/or DT+) reported a higher age, a greater maximum number of drinks in any 24 h period, more years of heavy drinking and more withdrawal episodes, more non-medicinal use of sedative-hypnotics and a greater comorbidity. These authors did also show that a history of head injury was significantly more frequently associated with a complicated course of AWS. The multivariate regression analysis revealed that the most powerful differences between WS + /DT + and WS−/DT− are related to the maximum daily ethanol intake and the total number of withdrawal episodes (Schuckit et al., 1995). While some of their data are in accordance with our own findings, others are not. Differences may be related to the fact that the authors grouped both complications together and they compared the reported histories of complications that ensued during AWS, whereas our attempt was to differentiate between patients who actually suffered from WS + /DT + with those who did not. In a study in 1856 trauma patients, the authors found after stepwise logistic regression higher age (>40) and white race predictive for DT (Lukan et al., 2002). While higher age was significantly more frequent in the DT + group in our own study sample, this difference did not retain significant after stepwise logistic regression. Increased age as a significant risk factor seems intuitive for two reasons. First, it may be associated with medical comorbidities and overall worse physical condition, being reported as significant risk factors in some studies (Soderstrom et al., 1992; Yost, 1996). Second, advanced age may be an indicator of tolerance, prolonged duration and increased amount of alcohol intake, factors known to be important, at least in animal models. However, unlike animal models, only one-third of the variance in the development of DT in humans can be attributed to the quantity, duration and frequency of alcohol ingestion, making these variables in general inapplicably for prediction models (Gorelick and Wilkins, 1986).

      In the study of Lukan et al., the level of blood alcohol at admission was not predictive for DT in the stepwise logistic regression, in congruence to our own data (Lukan et al., 2002). Instead, in our model the blood alcohol concentration at admission was significantly lower in the DT + group. In fact, Ferguson et al. proposed that a low blood alcohol concentration suggests a longer period of time since the last drink and therefore a higher risk of DT (Ferguson et al., 1996).

      Even though elevated mean corpuscular volume (Blondell et al., 2004) and aspartate aminotransferase levels have been recently demonstrated to be predictive for DT in trauma patients (Findley et al., 2010), this could not been established by our own data. Notwithstanding, we found significantly higher values of γGT in DT + vs. DT− patients (not significant after adjusting for potential confounders). We did not detect relevant differences in the aspartate aminotransferase levels or in the mean corpuscular volume, neither in the DT + /DT− nor in the WS + /WS− comparison (data not shown). It is important that abnormalities of any laboratory findings are neither isolated to alcoholism nor are they abnormal in all alcoholics. This is supported by the poor sensitivities, specificities and positive and negative predictive values of these tests (Story and Hoffman, 2010).

      While prospective validation of our data is needed, there are a number of important implications of this study. First, clinicians can stratify patients with respect to risk for WS or DT based upon easily accessible parameters at admission using the applied nomograms. Second, patients classified at risk to develop WS or DT should carefully be observed (e.g. in an intermediate or high-dependency unit) and should receive both, early and aggressive treatment (e.g. with benzodiazepines or clomethiazole). Third, clinicians should remain cautious about offering outpatient detoxification to alcohol-dependent patients known to have risk factors for DT or WS.

      Despite the possible valuable implications of this study, however, we have to account for some important limitations. First, it was a retrospective re-analysis of a data set of patients that was initially intended to compare two groups of patients that received two different AEDs (either CBZ or VPA) additionally to a standardized protocol-driven withdrawal schedule. Therefore, we have to account for a treatment-dependent bias on our results. Even though in this study we analyzed parameters of patients at admission (e.g. withdrawal history, laboratory parameters) being certainly independent of the applied AED, we have yet to consider differences in the baseline parameters of these two groups that were not necessarily balanced in all respects. Actually, those two treatment groups showed some differences regarding a younger age, a longer duration of addiction, a higher daily ethanol intake and a higher incidence of previous WS or WS as the cause of admittance in the CBZ group compared with the VPA group (Eyer et al., 2011). There were no significant differences, neither in the clinical nor in the analytical findings on admission between those two groups—in fact those parameters that were investigated as predictors in this current study. However, to account for these unbalanced baseline parameters, we calculated differences independent of the underlying treatment using a stepwise multiple regression model.

      Second, the information about prior DT and WS was according to patient's report and we were unable to verify the accuracy of these reports. For instance, patients might be likely to report a prior history of DT or seizure if they had previously experienced tremors or perceptual illusions during an AWS. Nonetheless, the information in the medical record was obtained during routine clinical care and therefore represents the level of detail and accuracy routinely available at the time patients were present for AWS.

      Third, increment of false-positive (false significant) results with increased number of hypothesis formally tested is a dilemma (the so-called multiple test problem), alike in this study, when comparing multiple endpoints and more than one potential predictive factor. However, to avoid finickiness and therefore to retain a maximum of power, no adjustment of overall alpha error level was conducted within the statistical assessments performed in this study, as suggested as a practical solution by Saville (1990).

      Despite these limitations, the current study provides an important starting point for further efforts to identify those at risk for WS or DT, whether prophylaxis is effective in preventing WS or DT and, if so, which type is the best. Potential risks of prophylaxis will need to be identified and, finally, the effect on outcome will need to be evaluated.

      REFERENCES

      View Abstract