Biomarkers of alcohol abuse potentially predict delirium, delirium duration and mortality in critically ill patients

Summary Carbohydrate-deficient transferrin (CDT) and the γ-glutamyltransferase-CDT derived Anttila-Index are established biomarkers for sustained heavy alcohol consumption and their potential role to predict delirium and mortality in critically ill patients is not clear. In our prospective observational study, we included 343 consecutive patients admitted to our ICU, assessed the occurrence of delirium and investigated its association with biomarkers of alcohol abuse measured on the day of ICU admission. 35% of patients developed delirium during ICU stay. We found significantly higher CDT levels (p = 0.011) and Anttila-Index (p = 0.001) in patients with delirium. CDT above 1.7% (OR 2.06), CDT per percent increase (OR 1.26, AUROC 0.75), and Anttila-Index per unit increase (OR 1.28, AUROC 0.74) were associated with delirium development in adjusted regression models. Anttila-Index and CDT also correlated with delirium duration exceeding 5 days. Additionally, Anttila-Index above 4, Anttila-Index per unit increase, and CDT per percent increase were independently associated with hospital mortality.


INTRODUCTION
Delirium comprises acute brain dysfunction and represents a frequently encountered clinical entity as estimations concerning incidence of delirium in the intensive care unit (ICU) range between 20% and 80% across different institutions and study populations. 1Delirium is the result of a continuum of multiple predisposing and precipitating factors and poses significant risks for negative patient outcomes. 2To facilitate prevention of ICU-delirium, preemptive identification of patients at high risk is crucial. 3lcoholism constitutes a relevant predisposing factor for development of ICU-delirium, 1 and alcoholism related admissions to ICU are reported to be as high as 30%. 4,5Differentiation between alcohol-withdrawal subtype of delirium and delirium of different origin is paramount, as therapeutic consequences are directly implied. 2,6Benzodiazepines have been shown to be effective in the treatment of alcohol-withdrawal associated delirium, whereas their application for other indications is considered as a trigger for delirium. 7,8In the ICU-setting, traditional methods of diagnosing alcoholism based on self-reported surveys and questionnaires can be challenged by the acute illness, making diligent history taking often not feasible.The unreliable diagnosis of alcoholism by medical history taking can in turn obscure the opportunity to identify it as a contributing factor for ICU-delirium and may hinder the implementation of preventive measures targeting this specific risk factor.
Carbohydrate deficient transferrin (CDT) is a liver-produced variant of serum transferrin that indicates chronic heavy alcohol abuse with adequate accuracy. 9,10Anttila et al. proposed using CDT and g-glutamyltransferase (gGT) measurements in a mathematical equation to enhance the diagnostic accuracy of excessive alcohol abuse disorder. 11No study has evaluated the relationship between these biomarkers and ICU-delirium.This study aimed to investigate the potential predictive role of these alcohol associated biomarkers in ICU-delirium.

Baseline characteristics of study population
As shown in the patient flow diagram (Figure 1), we screened 412 critically ill patients after admission to our medical ICU and excluded 69 patients due to absence of informed consent, dementia, missing CDT values, or hospital stay longer than 7 days before ICU admission.
Comatose patients, who died with a RASS score of À4 or À5 before CAM-ICU assessment was possible, were counted as ''absence of informed consent.''Thus, we ultimately included 343 patients.Thereof, 35% patients (n = 121) developed delirium during their stay in the ICU.The admission diagnosis is depicted in Table 1 and demographic and clinical characteristics of our study population are detailed in Table 2.
Only 35% (119/343) of included patients were female with a similar sex distribution in the two subgroups with and without delirium (p = 0.193).Critically ill patients with delirium had significantly higher SAPS3, TISS28, and SOFA scores at the time of ICU admission (p < 0.001, respectively); they were more frequently in need of mechanical ventilation and deep sedation (p < 0.001 for both).Accordingly, ICU-length of stay and hospital length of stay was significantly longer in delirious patients (p < 0.001, respectively).ICU mortality was approximately three times higher in patients with ICU delirium (18% vs. 5%, p < 0.001), while the ICU-mortality rate for the general study population was roughly 10%.
The laboratory parameters at time point of ICU admission are shown in Table S1.Critically ill patients with ICU-delirium had significantly lower platelet counts (p < 0.001), lower oxygenation index (p = 0.007), and lower serum-albumin (p < 0.001).The inflammatory biomarkers C-reactive protein and procalcitonin (p = 0.005 and 0.006) as well as serum creatinine (p = 0.014) and urea (p = 0.021) were significantly elevated in delirious patients when compared to those without delirium.

Biomarkers of alcohol abuse are associated with development of ICU-delirium
In line with our hypothesis, CDT levels were significantly higher in critically ill patients with ICU-delirium when compared to patients without delirium (p = 0.011).Indeed, 52% (33/63) of critically ill patients with CDT above 1.7% developed delirium during their stay in the ICU, whereas only 31% (88/280) of patients with a CDT level below 1.7% became delirious (p = 0.002).Furthermore, the Anttila-Index was significantly elevated in patients with delirium with a median of 3.69 in delirious patients compared to 3.22 in CAM-ICU negative patients (p = 0.001).Patients with an Anttila-Index exceeding 4 demonstrated a significantly higher incidence of delirium (49%, 45/91) in comparison to those with an Anttila-Index below 4 (30%, 76/249) (p = 0.002).(For 3 patients the Anttila-Index could not be calculated due to missing g-glutamyltransferase values).
In our study cohort, we observed a documented history of alcohol abuse in only 16% (55/343) of individuals.However, we found that 27% of patients (91/340) had an Anttila-Index value above the cutoff, indicating the presence of excessive alcohol consumption.Cramer's V as a measure of effect size for the correlation between documented alcohol abuse and the Anttila-Index above 4 was 0.23, indicating only moderate correlation.
This finding supports our hypothesis that alcohol abuse may be challenging to accurately diagnose in acute care settings, given the strong diagnostic performance demonstrated by the Anttila Index for detecting excessive alcohol consumption. 11n unadjusted binary logistic regression analysis, we found that both increase in CDT per percent point (OR of 1.34 (95% CI 1.10-1.69)),and CDT levels above the cutoff of 1.7%, (OR of 2.40 (95% CI 1.38-4.20))were significantly associated with the development of ICU-delirium (p values of 0.008 and 0.002, respectively).Similarly, Anttila-Index per unit increase (OR of 1.47 (95% CI 1.21-1.79)),and Anttila-Index above the cutoff of 4 (OR of 2.23 (95% CI 1.36-3.65))significantly correlated with the development of delirium in unadjusted analysis (p values of < 0.001 and 0.001, respectively).The results of unadjusted binary logistic regression analysis are depicted in Table S2.
Figure 2 illustrates the correlation between predicted probabilities of delirium occurrence and biomarkers of alcohol-abuse, SOFA-score, as well as serum albumin levels derived from unadjusted logistic regression analysis.Most importantly, a CDT value above 1.7%, CDT per percent increase, and Anttila-Index per unit increase remained significantly associated with delirium development in our multivariable models including the covariates SOFA-score, age, mechanical ventilation, and serum albumin (Figure 3).

Predictive performance of CDT and Anttila-Index
The predictive performance of the CDT and Anttila-Index was evaluated using Receiver operator characteristic (ROC) analysis and the area under the receiver operator characteristic curve (AUROC).The AUROC for CDT was found to be 0.75 (95% CI: 0.69-0.81).Similarly, the AUROC for the Anttila Index was calculated to be 0.74 (95% CI: 0.68-0.80).
Optimal cutoff values were established for both the CDT and Anttila Index using the Youden Index.The CDT demonstrated an optimal threshold of 1.79%, with associated sensitivity of 0.65 and specificity of 0.75.Likewise, the Anttila Index yielded an optimal cutoff of 3.61, accompanied by sensitivity of 0.66, and specificity of 0.74 for development of ICU-delirium.The respective receiver operating characteristic (ROC) curves are outlined in Figure 4.

Higher Anttila-Index and CDT are associated with longer delirium duration
The median duration of delirium in our cohort was 3 days (2-5 days).Based on recently published data concerning critically ill patients, which showed a range of delirium duration of 4-11 days with a median of 5 days, 12 and the 75 th percentile for delirium duration of our study population at 5 days, we defined 5 days of delirium in the ICU as ''prolonged ICU-delirium.''Thirty-four of 121 patients who developed delirium stayed CAM-ICU positive for 5 days or longer.Covariate adjusted logistic regression analysis in the subgroup of only CAM-ICU positive patients showed that an increase of one unit in Anttila-Index was associated with a SOFA, mechanical ventilation, age and albumin adjusted OR of 1.70 (95% CI 1.21-2.51,p = 0.004) for developing delirium lasting 5 days or more.An increase of CDT by 1% was associated with an adjusted OR of 1.34 (95% CI 1.04-1.84,p = 0.042) for developing prolonged ICU-delirium.
Anttila-Index and CDT are linked to mortality in critically ill patients Our study found that 14% of patients admitted to our ICU died during their stay at the hospital.Patients who developed delirium had a significantly higher mortality rate compared to patients who did not develop delirium, as demonstrated by a 23% death rate in the delirium group compared to a 9% death rate in the non-delirium group (p < 0.001).We found that Anttila-Index above 4 was associated with doubled risk of death during the hospital stay when compared to Anttila-Index below 4 in adjusted Cox proportional hazard analysis (HR 2.20, 95% CI 1.21-4.00,p = 0.010).Moreover, CDT per percent increase and Anttila-Index as continuous variables showed to be significantly associated with death during the stay at the hospital (Table 3).
As depicted in the Kaplan-Meier-curve (Figure 5), the 30-day survival rate for patients with an Anttila-Index above 4 was 77%, whereas patients with an Anttila-Index below 4 had a survival rate of 69% (p = 0.036 by Log rank test).

DISCUSSION
The results of this prospective observational study clearly demonstrate that elevated CDT and particularly the derived Anttila-Index at time point of ICU-admission were associated with increased risk of delirium, longer delirium duration, and higher hospital mortality.We could show that these biomarkers may have a potential role in prediction of ICU-delirium and other relevant outcomes.In addition and in line with previous studies, 13 we show that delirium causes a significant burden of disease with long ICU and hospital length of stay and most importantly high ICU and hospital mortality.
Bowman and colleagues propose specific clinical and biological subphenotyping of delirium to elucidate causal relationships between symptoms, risk factors, and biological mechanisms. 145][16] Additionally, assessing alcohol abuse disorder in a structured manner in critical care settings can be impractical and difficult.The complexity and multidimensional nature of alcohol abuse disorder assessments may also contribute to a lack of documentation.Our study highlights this issue, as there was a discrepancy between the proportion of patients with elevated biomarkers associated with alcohol abuse and the percentage of patients with documented histories of alcohol abuse.This discrepancy is further supported by an only moderate correlation (Cramer's V = 0.23) between documented history of alcohol abuse and an Anttila Index above 4.This finding underscores the need for a biomarker-based approach to assess the risk of delirium in critically ill patients.While there is evidence pointing toward a higher risk for delirium development in association with use of benzodiazepines, 7,8 a potential subphenotype of delirium identified by elevated CDT and Anttila-Index at time point of ICU-admission could potentially benefit from early use of benzodiazepines as preventive measure.However, this assumption needs to be verified in future controlled trials.
Our study harbors several strengths and limitations.RASS and CAM-ICU were performed by trained medical staff blinded to biochemical analyses of CDT to preclude influence of delirium assessment and performance of CAM-ICU.Moreover, we used prospectively collected data to assess potential association between alcohol consumption related markers and delirium occurrence.Most importantly, our findings are clinically important and relevant to intensivists, as CDT and Anttila-Index are readily available biomarkers able to identify patients at high risk for subsequent delirium development facilitating initiation of early preventive measures.Although model-based prediction of ICUdelirium has been found to have moderate to good discriminative ability, 4,[17][18][19] output from these models has been judged impractical for consequent real time action by intensivists, 2 which further advocates a more linear approach for delirium risk assessment, as proposed in our study.

Limitations of the study
Limitations of our study include the following aspects: (a) ICU-delirium depicts a clinical syndrome of acute brain dysfunction resulting from a complex interplay of different and partly obscure pathophysiological mechanisms, making it challenging to predict.This poses methodological problems in two dimensions.Firstly, single biomarkers have reduced discriminative performance in predicting delirium in the ICU due to the multifactorial nature of its pathogenesis and numerous confounding factors. 20Our findings align with these challenges, as demonstrated by the modest discriminative performance with AUROC values of 0.75 for CDT and 0.74 for the Anttila-Index.The AUROC values, however, while not achieving high levels of discrimination, still signify a capacity to differentiate patients with a heightened risk of developing ICUdelirium from those at lower risk.So, despite the intricacies inherent in the pathogenesis of delirium, our findings hold clinical relevance in identifying patients at risk.Secondly, statistical modeling to assess delirium in the ICU is challenging due to the involvement of multiple confounding factors.Therefore, we prioritized a priori covariate selection based on clinical judgment and evidence synthesis.To account for illness severity, which is a strong predictor of delirium development, we used the SOFA-score, a highly validated illness-severity score, as a covariate instead of relying on admission diagnosis. 14,21,223][24] We aimed to balance complexity, collinearity, and redundancy to aid interpretability of our results, hence the selection these four covariates.(b) Due to the single center, single ICU-design, our study cohort only involved critically ill medical patients which lead to lack of generalizability of our findings especially with regards to surgical patients.On the other hand, surgery and especially emergency surgery have been reported as precipitating risk factors for development of delirium per se, so risk assessment in critically ill subjects not exposed to these predisposing factors limits potential bias. 19,25(c) Our study was not designed to consider different delirium-psychomotor subtypes using for instance the Delirium Motor Subtype Scale (DMSS). 26It has, however, been proposed previously that heading away from using psychomotor subtypes for categorization of delirium might enable identification of more specific subphenotypes.Thereby, linking of precipitants with the syndrome and identification of clusters may be facilitated. 14This was conclusively the underlying ambition of our study.

Conclusion
CDT and particularly the derived Anttila-Index are independently associated with development of delirium, longer delirium duration, and higher mortality in critically ill patients.Therefore, elevated CDT and Anttila-Index are not only a specific biomarker for sustained heavy alcohol consumption but may also have a role in the identification of patients at risk for delirium in the ICU setting.These findings provide a clear rationale for future interventional trials in critically ill patients to identify patients at high risk and prevent ICU-delirium.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:   At the time of ICU-admission, we recorded demographic data and laboratory characteristics of these critically ill patients.All personal data were pseudonymized.The study related laboratory parameter CDT was blinded to avoid any potential influence of prior knowledge of CDT on the assessment of delirium and measurements were un-blinded after discharge of the last patient at the end of the study.All patients were of Caucasian ethnicity, and 34.7% of patients were female.Female gender had no impact on the occurrence of ICU-delirium in our study population (Table S2).

Figure 4 .
Figure 4. Receiver operating characteristic curves for CDT [%] and the Anttila-Index Receiver operating characteristic (ROC) curves to evaluate the predictive and discriminative ability of CDT (black line) and the Anttila-Index (gray line) for development of ICU-delirium are shown.Respective areas under the receiver operating characteristic curves (AUROC) are depicted as well.AUROC, area under the receiver operating characteristic curve; ROC, receiver operating characteristic; CDT, Carbohydrate deficient transferrin; ICU, Intensive care unit.

Figure 5 .
Figure 5. Kaplan-Meier survival curve 30-day Kaplan-Meier survival estimates for critically ill patients with Anttila-Index below 4 (black line) and Anttila-Index above 4 (gray line).Three patients were not considered because of missing values for g-glutamyltransferase.

Table 2 .
Demographic and clinical characteristics of the study population Median values are shown with 25 th and 75 th percentile in brackets.Statistically significant p-values are given in bold.
w/o, without; SAPS, simplified acute physiology score; TISS, therapeutic intervention scoring system; SOFA, sequential organ failure assessment; RASS, Richmond agitation-sedation scale.a Chi-Square-Test or rank-sum test as appropriate.

TABLE d
B Study population and study design d METHOD DETAILS B Outcomes B Laboratory testing B Quantification and statistical analysis B Covariates B Additional resources