Several previous investigations in non-ICU patients established an association between inflammation and delirium, as correlations between proinflammatory cytokine levels and delirium have been found [6,8-10]. Furthermore, in elderly delirious patients with hip fractures, increased concentrations of IL-6, IL-8, and cortisol were correlated with elevated levels of the brain-specific protein (BSP) S100-β (a marker for astrocyte damage) [11]. Interestingly, sepsis is also associated with elevated levels of BSP [12,13]. Furthermore, it has been hypothesized that serious illness such as sepsis, as well as the use of sedatives and analgesic, could result in apoptosis and long-term cognitive impairment [14]. In mice, tumor-necrosis factor (TNF)-α is a mediator of apoptotic cellular death in the brain [15] and may therefore be causally associated with the development of delirium in patients with severe inflammation.

In the long term, delirium is associated with a more than 12-fold increased risk for developing dementia [16], resulting in permanent impairment of cognitive function that is associated with altered levels of amyloid-β [17,18]. The association between biomarkers in delirious patients and long-term cognitive function is unknown.

This is the first study investigating plasma amyloidβ (Aβ) levels and human Tau in critically ill patients in relation to the presence of delirium. In view of the reported increased incidence of dementia after ICU/hospital admission [16], our findings could provide a possible mechanistic link, because noninflamed delirium is associated with Aβ, but this must be confirmed in a longitudinal study focusing on these biomarkers combined with more-extensive cognitive testing. Furthermore, Aβ is associated with sustained long-term subjective cognitive dysfunction in ICU patients. Studies comparing plasma levels of Aβ between Alzheimer (AD) and non-Alzheimer dementia patients and controls [17,18,32] have yielded conflicting results with respect to levels of different forms of Aβ. Increased levels of Aβ_1-42 [17] as well as increased levels of Aβ_1-40 [18] were found in dementia patients [32]. In addition, increased levels of the Tau/Aβ_1-42 ratio have been found in cerebrospinal fluid (CSF) of patients with cerebral amyloid deposition [33], but this has not yet been investigated in plasma. In the present study, the difference in levels of total Tau and the Tau/Aβ_1-42 ratio between noninflamed delirious patients and noninflamed nondelirious patients approached statistical significance. It is known that plasma levels of Aβ are age dependent [34]; however, this could not have confounded our results because no differences in age existed between delirious and nondelirious patients in our study. Additionally, the patients investigated in this study were not recognized with a history of cognitive impairment by patients' medical history and information from their relatives, which could explain differential Aβ levels. The lower Aβ_1-42/40 ratio, probably due to an increase of Aβ_1-40 at a constant Aβ_1-42 level, in combination with a significant correlation with long-term cognitive failure on several domains of the CFQ are in accordance with findings that elevated levels of Aβ_1-40 increase the risk of developing dementia [16,18]. Importantly, this finding of early lower Aβ_1-42/40 ratio and ratio Aβ_1-40/N-40 in delirious patients without signs of serious previous cognitive impairment makes it tempting to speculate that this represents the first sign of an imbalance in the Aβ metabolism. To our knowledge, these early findings of imbalance in Aβ metabolism have not been reported. Our findings might therefore shed new light on the important question whether delirium plays a causative role in the development of dementia in later life, or if delirium is the first sign of dementia. Because deposition of Aβ in the brain is generally considered to be a long-term process, and samples in our study were drawn shortly after the onset of delirium, it is more plausible that delirium may be the first sign of an early dementia process. However, a cause-effect relation cannot be determined in a cross-sectional observational study like ours. This hypothesis of early imbalance in Aβ metabolism in delirious patients must be confirmed in future studies.

Previously, it was demonstrated that delirium is associated with elevated levels of IL-6, IL-8, and S100-β in non-ICU patients [9,10] and with IL-6 and S100-β in ICU patients with sepsis [13]. IL-8 levels were not measured in these sepsis patients. We showed, by using a multivariate logistic regression analysis, that levels of IL-8 in inflamed patients were associated with delirium, but IL-6 was not. A possible reason for this discrepancy might be that we determined biomarkers directly after the first positive delirium screening, whereas it has been shown that the highest levels of IL-6 occur in the later phase of delirium [10].

Several limitations of our study should be addressed. First, we used the CAM-ICU to diagnose delirium in ICU patients instead of the gold standard: the DSM-IV criteria [35]. It is recognized that it is not feasible to use this gold standard in ICU patients, and therefore the CAM-ICU is an accepted alternative to diagnose delirium in the ICU. The CAM-ICU has the highest sensitivity and specificity rate of all delirium-assessment tools [36,37] and is well implemented in the daily practice of our nurses with a high interrater reliability [26]. In addition, to strengthen the delirium diagnosis, all medical and nurse files of the patients were analyzed, and patients were not included when in doubt of the delirium diagnosis.

Second, we did not use a validated cognitive-assessment tool such as the informant questionnaire on cognitive decline short form (IQCODE-sf), which is a surrogate evaluation to determine whether the patient had serious cognitive impairment before ICU admission. Instead of this, we used information from medical records and the next of kin of the patients to identify whether the patient had a history of cognitive impairment. In case of any reference to or sign of cognitive impairment, patients were not included in our study. Furthermore, as a measure of patients' cognitive function 18 months after ICU discharge, we used the validated CFQ, which is a self-evaluated questionnaire to detect cognitive-based failures and not dementia and is also not a specific psychometric test, which may result in more-objective data. Although this can be considered a limitation of our study, our findings are the results of patient's own perception of cognitive functioning and are therefore informative and relevant.

Third, in this study, we measured biomarkers only at one point in time. In a longitudinal biomarker study [9], a difference in cytokine levels before and during delirium was found. In the absence of biomarker data in critically ill patients with delirium, we chose to perform an exploratory study to investigate which biomarkers were most strongly associated with delirium immediately after the onset of delirium. This was an exploratory hypothesis-generating study, of which the results may facilitate hypotheses for future research.

Fourth, potential covariates must be considered as a potential limitation of the study, in contrast to a randomized trial in which possible covariates are likely to be equally divided between the groups. Although baseline patient characteristics were comparable between the delirium and nondelirium groups, unbalanced influence of covariates cannot be ruled out in such an observational study as we performed.

• Delirium in inflamed patients is independently associated with the proinflammatory cytokine IL-8, and in noninflamed patients, delirium is independently associated with the antiinflammatory cytokine IL-10 and the ratio amyloidβ-_42/40.