Confusion assessment method accurately screens for hepatic encephalopathy and predicts short-term mortality in hospitalized patients with cirrhosis

Hepatic encephalopathy (HE), a subtype of delirium, is common in cirrhosis and associated with poor outcomes. Yet, objective bedside screening tools for HE are lacking. We examined the relationship between an established screening tool for delirium, Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) and short-term outcomes while comparing its performance with previously established measures of cognitive function such as West Haven criteria (WHC). Prospectively enrolled adults with cirrhosis who completed the CAM-ICU from 6/2014–6/2018 were followed for 90 days. Blinded provider-assigned West Haven Criteria (WHC) and other measures of cognitive function were collected. Logistic regression was used to test associations between CAM-ICU status and outcomes. Mortality prediction by CAM-ICU status was assessed using Area under the Receiver Operating Characteristics curves (AUROC). Of 469 participants, 11% were CAM-ICU( +), 55% were male and 94% were White. Most patients were Childs-Pugh class C (59%). CAM-ICU had excellent agreement with WHC (Kappa = 0.79). CAM-ICU( +) participants had similar demographic features to those CAM-ICU(-), but had higher MELD (25 vs. 19, p < 0.0001), were more often admitted to the ICU (28% vs. 7%, p < 0.0001), and were more likely to be admitted for HE and infection. CAM-ICU( +) participants had higher mortality (inpatient:37% vs. 3%, 30-day:51% vs. 11%, 90-day:63% vs. 23%, p < 0.001). CAM-ICU status predicted mortality with AUROC of 0.85, 0.82 and 0.77 for inpatient, 30-day and 90-day mortality, respectively. CAM-ICU easily screens for delirium/HE, has excellent agreement with WHC, and identifies a hospitalized cirrhosis cohort with high short-term mortality.


Introduction
Delirium is a manifestation of acute brain dysfunction characterized by a wide range of neuropsychiatric abnormalities including confusion, disorientation, and hallucinations (Deksnytė et al. 2012). A subtype of delirium, hepatic encephalopathy (HE), is a well-studied complication of cirrhosis that also encompasses a range of neuropsychiatric impairment from subtle personality changes to coma (Ferenci et al. 2002;American Psychiatric Association 2013;Rosenberg et al. 2013;Vilstrup et al. 2014;Amodio 2018). Delirium is as an independent predictor of short term mortality and longer length of stay in hospitalized patients leading many hospitals to routinely screen for delirium (Ely et al. 2004). Like delirium, HE is often precipitated by infections, electrolyte disturbances, and volume imbalances and is associated with significant morbidity and mortality as well as healthcare expenditures (Stepanova et al. 2012;Patidar et al. 2014;Vilstrup et al. 2014).
The gold standard tool for diagnosing and quantifying HE at the bedside is the West Haven criteria (WHC) which classifies the severity of HE on a scale of 0-4 (Cordoba 2011;Vilstrup et al. 2014). Due to the overlapping features of delirium and HE, objective screening tools for delirium may be particularly valuable in patients with cirrhosis (Vilstrup et al. 2014;Tapper et al. 2016). Among the various delirium screening tools, the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) is a simple 4-item, objective, validated tool that aids healthcare workers in quickly and reliably diagnosing delirium (Ely et al. 2001). Due to its excellent performance characteristics, and reliability, CAM-ICU is recommended in more than 30 clinical practice guidelines, used in a variety of clinical settings and translated into 20 languages leading to worldwide use (Patel et al. 2009;Guenther et al. 2010;Wong et al. 2010;Gusmao-Flores et al. 2012). Importantly, CAM-ICU can be administered accurately by nursing staff in approximately 1 min (Guenther et al. 2010). Despite the wide-spread use of CAM-ICU, it has been understudied in patients with cirrhosis (Cordoba 2011;Orman et al. 2015).
In this prospective study, we aimed to examine the relationship between CAM-ICU status and short-term outcomes in hospitalized patients with cirrhosis. We also examined the relationship between CAM-ICU, WHC and previously established measures of cognitive function.

Study design
This analysis uses data from adults with cirrhosis (age ≥ 18) hospitalized at Indiana University Hospital between June 2014 and June 2018 who were prospectively enrolled during their hospitalization and followed for 90 days as previously described (Orman et al. 2021). Those in the study cohort with at least one assessment with the Richmond Agitation Sedation Scale (RASS) and/or the CAM-ICU during their hospitalization were included in this analysis (n = 469, Fig. 1). Cirrhosis diagnosis was based on the presence of liver histology or on characteristic clinical, laboratory, and radiologic findings. Patients were excluded if they had prior solid organ transplant. In patients meeting these criteria, research staff coordinated with clinical staff (caring physician or nursing staff) to determine their WH stage. Only those with WH stage 0-1 were approached for enrollment and informed consent; patients with WH ≥ 2 were followed daily until mental status improved and then approached for enrollment. Those who maintained a WH stage ≥ 2 were deemed unable to provide informed consent and excluded from the study. The study was approved by the Indiana University Institutional Review Board. Informed consent was obtained from all individual participants included in the study.

Measures of cognition
Participants were assessed daily, first for level of sedation using the Richmond Agitation-Sedation Scale (RASS, Supplementary Fig. 1) (Sessler et al. 2002). Participants with a RASS score of − 4 (no response to voice, but movement or eye opening to physical stimulation) or − 5 (no response to voice or physical stimulation) were deemed comatose and ineligible for further assessment with CAM-ICU. For those eligible, assessment using CAM-ICU was completed ( Supplementary Fig. 1). Assessments were made daily by research staff. Blinded daily provider-assigned WHC was also recorded with WHC ≥ 2 defined as HE (Vilstrup et al. 2014). On the day of enrollment, participants with WHC of 0 or 1 (n = 415) also completed additional, previously established, assessments of HE: Clinical Hepatic Encephalopathy Staging Scale (CHESS), Modified-orientation log (MO-log), and Number Connection Test A (NCT). These scales were scored or timed as previously described (Amodio et al. 1999;Ortiz et al. 2007;Salam et al. 2012).

Demographic and clinical variables
Demographic data, social history, cirrhosis etiology, presence of cirrhosis complications such as hepatocellular carcinoma, varices, ascites, prior HE, and Charlson Comorbidity Index (CCI) were retrieved from the electronic medical record. During the hospitalization, reason for admission, presence and type of infections, and laboratory studies completed at the time of admission were recorded. Child-Pugh Class and Model for End-stage liver disease (MELD) score were also calculated.

Outcomes
The primary outcomes were inpatient, 30-day, and 90-day mortality. 30-day readmission and length of stay were also assessed.

Statistical analysis
Participants who were RASS/CAM-ICU positive ( +) at any point during the admission were compared to those who remained RASS/CAM-ICU negative (-) throughout the admission. Baseline characteristics were compared between two groups using Pearson's chi-squared test or Fisher's exact test for categorical variables, t-test for continuous variables and Wilcoxon Rank-Sum test for continuous skewed data.
Concurrent validity of RASS/CAM-ICU was established through several methods. First, we compared RASS/CAM-ICU status to WHC using the agreement statistic as well as the sensitivity and specificity of HE by RASS/CAM-ICU to diagnose hepatic encephalopathy using WHC ≥ 2 as the gold standard for HE diagnosis. In addition, performance on the CHESS, MO-Log, and NCT were compared by RASS/ CAM-ICU status using Wilcoxon Rank-Sum test.
Logistic regression models were used to assess the association between clinical characteristics, RASS/CAM-ICU status or WHC and mortality outcomes. Variables that have a p-value < 0.1 in univariate models were considered potential confounders and added to the multivariate models. In addition, age and history of HE were included in the models due to clinical relevance. We report the unadjusted and multivariate models testing the association between RASS/ CAM-ICU status or WHC ≥ 2 and mortality outcomes. The Area under the Receiver Operating Characteristics (AUROC) curves was estimated to assess the predictability of the models.
All analyses were conducted using SAS v9.4.

Concurrent validity between west haven criteria and RASS/CAM-ICU
RASS/CAM-ICU demonstrated concurrent validity when compared to WHC ≥ 2 for hepatic encephalopathy (Table 2). There was strong agreement between WHC and RASS/CAM-ICU (κ = 0.79). When compared to WHC, the sensitivity of RASS/CAM-ICU for HE is 92.5% and the specificity is 96.7%. Notably, 14 participants screened positive by RASS/CAM-ICU despite being categorized as WHC < 2 by the clinical team. Of these, WHC = 1 in 12 participants and WHC = 0 in 2 participants. Both participants with WHC = 0 were noted to be RASS = 0 but screened positive for all three features assessed by CAM-ICU (acute change or fluctuating mental status, inattention, and disorganized thinking). Of those with WHC = 1, RASS = 0 in 6 participants who all screened positive for all three features assessed by CAM-ICU while RASS ≠ 0 in the remaining 6 leading to the positive screen ( Supplementary Fig. 2).

Concurrent validity between RASS/CAM-ICU and objective instruments of cognitive function
RASS/CAM-ICU demonstrated concurrent validity when compared to objective measures of cognitive function including CHESS, MO-Log, and NCT in those with West Haven score ≤ 1 (Table 2)
RASS/CAM-ICU( +) status remained independently associated with higher odds of each outcome after adjusting for adjusting for MELD, CCI and ICU at admission (Table 3). These models showed strong AUROC curves of 0.85, 0.82 and 0.77 for inpatient, 30-day and 90-day mortality, respectively ( Fig. 3A-C). Notably, these models performed similarly to a model using WHC ≥ 2 instead of those RASS/CAM-ICU( +) status to predict short-term mortality ( Fig. 3A-C). In addition to adjusting for MELD, we looked at AUROC curves in different MELD sub-groups and found that AUC of RASS/CAM-ICU status for predicting inpatient, 30-day and 90-day mortality tended to be highest in those with MELD > 30, however, they were not statistically   Fig. 3).

Discussion
HE is a subtype of delirium and defining feature of decompensated cirrhosis (Ferenci et al. 2002;American Psychiatric Association 2013;Rosenberg et al. 2013;Vilstrup et al. 2014;Amodio 2018). HE remains an important target for quality improvement in hospital-based care (Bajaj et al. 2019). Importantly, both delirium and HE have been associated with longer hospitalizations, higher health care costs, and increased mortality (Ely et al. 2004;Stepanova et al. 2012;Patidar et al. 2014;Vilstrup et al. 2014). Despite these similarities and impact on outcomes, few studies have examined the value of an objective screening tool in patients with cirrhosis (Vilstrup et al. 2014;Orman et al. 2015). In this large prospective cohort of hospitalized patients with cirrhosis, we found substantial agreement between the most commonly used bedside measures of delirium and HE: RASS/CAM-ICU and WHC. We established concurrent validity of RASS/ CAM-ICU by noting lower performance on previously established measures of cognitive function in cirrhosis. Finally, we show a strong association between RASS/CAM-ICU status and short-term outcomes allowing for accurate prediction of these outcomes using the RASS/CAM-ICU tool.
Although the WHC are easy to understand and intuitive to use, the RASS/CAM-ICU tools also offer advantages. They are already implemented in diverse settings, including centers lacking expertise in end-stage liver disease (Ely et al. 2001;Patel et al. 2009;Guenther et al. 2010;Wong et al. 2010;Gusmao-Flores et al. 2012). In many institutions, no additional training would be required of bedside nursing staff to implement the use of RASS/CAM-ICU, even when specifically assessing individuals with cirrhosis (Tapper et al. 2016). In addition, RASS/CAM-ICU are reliably reproducible when used at the bedside (Ferenci et al. 2002;Prakash and Mullen 2010). Other objective bedside measures of cognitive function have been developed such as MOlog and CHESS and we show that RASS/CAM-ICU shares concurrent validity with these tools, however, these tools have not been widely validated as has RASS/CAM-ICU.
Despite excellent agreement between WHC and RASS/ CAM-ICU, 14 participants without overt HE by WHC (stage 0 or 1) had a positive RASS/CAM-ICU. Although this amounts to only 3% of those without overt HE, positive CAM-ICU screens in this group may provide an opportunity for earlier intervention for patients who may not otherwise be identified as being at risk for cognitive impairment by WHC. In our study, these individuals demonstrated agitation or sedation and/or acutely changed/fluctuating mental status, inattention and disorganized thinking. In prior studies, RASS has been used to adjust lactulose dosing (Tapper et al. 2016). Furthermore, CAM-ICU had the added advantage of early identification of hypoactive delirium characterized by flat affect or apathy in calm and outwardly alert individuals (Truman and Ely 2003). This form of delirium was noted to be the most common subtype of delirium and may be associated with a worse prognosis (Camus et al. 2000;Ely et al. 2001). Future studies could follow such patients longitudinally to assess for downstream cognitive outcomes and to investigate whether early treatment for delirium and/or HE may mitigate such outcomes. Notably, only 3 patients with WH grade 2 HE did not screen positive with RASS/ CAM-ICU. Our study also establishes that screening for delirium and HE with RASS/CAM-ICU or WHC can identify those at high risk for poor outcomes. For example,those who screened positive with RASS/CAM-ICU in our study experienced dramatically high rates of death: 37%, 51% and 62% during hospitalization, 30 days and 90 days post-discharge, respectively. Both RASS/CAM-ICU and WCH were found to have very good predictive validity for short-and intermediate-term mortality with AUC ~ 0.8 after adjusting for important covariates. These findings are consistent with prior work demonstrating poor outcomes associated with delirium in a variety of disease states including those with cirrhosis (Stepanova et al. 2012;Orman et al. 2015;Maldonado 2017). Interestingly, despite the value of delirium and HE screening in predicting mortality, RASS/CAM-ICU and WHC did not distinguish risk of 30-day readmission. Delirium has been associated with early readmission in older adults in a general medical population (LaHue et al. 2019). However, readmission in the cirrhosis population is notoriously difficult to predict, with risk prediction models achieving only fair-to-poor discrimination despite employing a broad array of analytic techniques and considering multidimensional risk factors (Desai and Reau 2016;Koola et al. 2020;Hu et al. 2021;Orman et al. 2021). Nevertheless, readmission should be considered as a potential outcome when evaluating interventions targeting HE and/or delirium.
We noted that RASS/CAM-ICU status was associated with several measures of disease severity, including MELD and ICU admission, and, as expected, it was more common in those admitted for encephalopathy or infection. Demographic characteristics were otherwise largely similar between RASS/CAM-ICU( +) and (-) individuals, except for a trend toward lower education level in the RASS/ CAM-ICU( +) group. A similar association between RASS/ CAM-ICU( +) status and educational level has been found elsewhere, attributed to enhanced cognitive reserve in those with greater educational attainment (Jones et al. 2006). In cirrhosis populations, cognitive impairment has also been linked to lower socioeconomic status (Bajaj et al. 2013;Tapper et al. 2019). Our work further supports the use of RASS/ CAM-ICU as a valid, objective screening tool for cognitive impairment in cirrhosis whose performance is not greatly impacted by demographic features of the individual.
This prospective study builds on prior retrospective work by incorporating real-time physician assessments of HE using WHC as a gold standard comparator for RASS/CAM-ICU. Research assessors and physicians were blinded to the WHC and RASS/CAM-ICU ratings, respectively, to prevent biased assessments. The prospective design also allowed for assessment of RASS/CAM-ICU predictive validity using short-term mortality, length of stay, and readmissions as outcomes of interest. However, these results must be interpreted in the context of the study design. Although WHC and RASS/CAM-ICU assessments were performed on the same day for comparison, they may have occurred at different times during the day and the assessments were completed by different staff (clinician for WHC and research coordinator for RASS/CAM-ICU). While these may have led to disagreements between instruments, this was relatively rare. Additionally, the paired measurements were not always done on day of admission. While our study underscores the importance of a formal delirium/HE screening protocol, future studies are needed to identify the ideal implementation of RASS/CAM-ICU-based screening into routine care with attention to timing of screening, staff training using published instruction manuals, and development of follow up management pathways. Finally, RASS/CAM-ICU is largely validated in an ICU setting as opposed to CAM, which has more evidence for those admitted to general hospital ward beds. However, 10% of the study cohort was admitted to the ICU, and RASS/CAM-ICU can be performed for those on mechanical ventilation. Thus, use of RASS/CAM-ICU allowed a standard assessment and comparison for the entire cohort, including those in the ICU. As compared to the WHC, a limitation of RASS/CAM-ICU is that it does not grade severity of HE and delirium. Newer tools such as CAM-7, which can measure delirium severity, may have additional usefulness in this population, and are deserving of further study (Khan et al. 2017;Boehm et al. 2016).
In conclusion, CAM-ICU is a valuable well-validated screening tool for HE that is widely utilized in a variety of hospital settings and performs similarly to WHC. Our study extends the validity of CAM-ICU to screen for delirium and HE in the hospitalized cirrhotic. The use of CAM-ICU by bedside staff may allow for earlier recognition and interventions for this common complication. Additionally, CAM-ICU status predicts important outcomes of in-hospital mortality, 30-day mortality, and 90-day mortality independent of liver disease severity and comorbidity burden. Future studies exploring the incorporation of CAM-ICU scores to liverspecific mortality prognosticating scores such as MELD-Na are needed.