Serum Osmolality, Cerebrospinal Fluid Specific Gravity, and Overt Hepatic Encephalopathy Severity in Patients with Liver Failure

Abstract

Background and Aims:

Hepatic encephalopathy (HE) is a leading contributor to morbidity in liver disease. While hyperammonemia plays a key role, the mechanisms of cerebral toxicity are unclear. We hypothesized that serum hyperosmolality contributes to HE during acute (ALF) and acute-on-chronic liver failure (ACLF) through mechanisms that affect the water and solute composition of the cerebral environment.

Methods:

We performed a retrospective analysis of serum osmolality, cerebral spinal fluid (CSF) solute density (specific gravity, determined from computed tomography attenuation), and clinical HE severity (Glasgow Coma Score [GCS]) at the time of intensive care admission in a prospectively identified cohort of liver failure patients with overt HE.

Results:

Seventy-three patients (39 ALF, 34 ACLF) were included, of whom 28 (38%) were comatose. Serum osmolality (303.9±15.4 mOsm/kg) was elevated despite normal serum sodium (136.6±6.3 mEq/L). Increased osmolality was independently associated with more severe encephalopathy (ordinal adjusted OR 0.26 [95% CI 0.22, 0.31] for higher GCS per standard deviation increase in osmolality) and lower CSF specific gravity (linear adjusted β= −0.039 [95% CI −0.069, −0.009] Hounsfield unit per 1 mOsm/kg).

Conclusions:

In the context of related research, these data suggest that hyperosmolality increases brain exposure to metabolic toxins by blood-brain barrier alteration and may be a unique therapeutic target.

Lay Summary:

In patients with acute and acute-on-chronic liver failure, increasing serum solute concentration (osmolality) is associated with more severe brain dysfunction (encephalopathy) and lower solute density (specific gravity) in the cerebral spinal fluid. In the context of related research, these data suggest that elevated serum osmolality may contribute to encephalopathy severity by increasing brain exposure to metabolic toxins through blood-brain barrier alteration.

INTRODUCTION

Hepatic encephalopathy (HE)—liver disease-related brain dysfunction—is a leading reason for hospitalization and contributes to morbidity and mortality in liver disease.(1–4) Historically, HE has primarily been attributed to astrocyte dysfunction as a result of impaired ammonia metabolism. However, the inconsistent association between serum ammonia levels and encephalopathy severity during both acute liver failure (ALF) and acute-on-chronic liver failure (ACLF) suggests that other molecular and cellular pathways are involved.(2, 5, 6) Systemic processes, such as the inflammatory response, have been hypothesized to function synergistically with ammonia and other toxins to produce HE by disrupting both the blood brain barrier (BBB) and blood cerebrospinal fluid barrier (BCSFB), thereby increasing cerebral exposure to these toxins.(6–9)

In a previous study, we described acute shifts in serum osmolality as a mechanism of neurologic deterioration and cerebral edema formation during hospitalization for severe HE in the context of ALF and ACLF.(10) We also observed that the majority of both ALF and ACLF patients demonstrated elevated serum osmolality at the time of hospital admission, despite normal or low serum sodium. Animal studies have demonstrated that induced hyperosmolality increases the permeability of the BBB and increases penetration of bilirubin and chemotherapeutic agents into the brain.(11–15) Given that context, we hypothesized that increased baseline serum osmolality and alterations in the solute density (specific gravity) of the cerebrospinal fluid independently contribute to HE during severe liver failure.

MATERIALS AND METHODS

We retrospectively abstracted electronic record data for patients greater than or equal to 18 years old admitted to an intensive care unit (ICU) at Northwestern Memorial Hospital between 2012 and 2017 with ALF or ACLF and overt HE. Patients with liver failure and overt HE, as diagnosed by a board-certified neurologist, were prospectively identified from our neurologic critical care consultation census. Neurologic critical care consultation is standardized care for patients with liver failure and encephalopathy at our institution. Liver failure was identified from attending intensivist or hepatologist documentation and confirmed according to published definitions.(16, 17) Given similar phenotypic presentations and our intention to explore pathophysiologic mechanisms, we included patients with either ALF or ACLF, an approach consistent with recommendations from the International Society for Hepatic Encephalopathy and Nitrogen Metabolism.(7, 10, 18, 19) Study inclusion required 1) measurement of serum osmolality within 24 hours of ICU admission for liver failure and 2) serially measured Glasgow Coma Scale (GCS) scores. The GCS is a neurologic examination scale ranging from 3 (deepest coma) to 15 (alert and oriented) and composed of motor, eye, and verbal response components that has good inter-rater reliability and is recommended as a measure of encephalopathy severity in HE.(18, 20–22) We excluded patients with focal brain lesions (e.g., stroke) and those who received hyperosmolar therapy (mannitol or hypertonic saline) prior to initial serum osmolality measurement or GCS assessment. No patient received hyperosmolar therapy prior to computed tomography (CT) scan acquisition. Additionally, no patient received albumin infusion or had renal replacement therapy performed prior to osmolality measurement, GCS assessment, or CT scan acquisition. Patients who received sedative or paralytic medications (such as for endotracheal intubation) prior to ICU admission were excluded unless a documented sedative medication hold of at least four hours duration occurred during the first 24 hours of ICU admission; in those cases, the highest GCS assessment corresponding to the sedation hold was used as the ICU admission GCS score.

Consistent with U.S. Acute Liver Failure Study Group practices, we infrequently use invasive intracranial pressure monitors to manage HE during liver failure.(23, 24) In this population, we use hourly neurologic examinations and repeat head CT scans as previously described.(10) All HE patients admitted to our ICUs receive hourly neurologic assessments, including GCS, performed and electronically documented by trained ICU nurses.(25, 26)

We collected demographic and clinical data including GCS, serum osmolality measured by freezing point depression (normal range 275–295 mOsm/kg), and serum chemistry and laboratory data from the electronic medical record for each patient at the time of ICU admission.(27) Serum osmolar gap was calculated as previously published.(27) We calculated Model for End-Stage Liver Disease—Sodium (MELD-Na) and Acute Physiology and Chronic Health Evaluation—II (APACHE II) scores as measures of liver failure and overall illness severity, respectively. The MELD-Na score uses creatinine and dialysis utilization, bilirubin, international normalized ratio (INR), and sodium level to estimate liver disease severity.(28) The APACHE II score uses age, presence of chronic organ insufficiency or immunocompromise, and 12 routine physiologic variables measured during the first 24 hours of ICU admission (temperature, mean arterial blood pressure, serum pH, heart rate, respiratory rate, sodium, potassium, creatinine and presence of acute renal failure, hematocrit, white blood cell count, arterial partial pressure of oxygen and fraction of inhaled oxygen, and GCS) to quantify multisystem disease severity on a scale from 0 to 71.(29) Since GCS was used as our measure of HE severity, we subtracted the points associated with the GCS from the APACHE II score when APACHE II was included as a covariate in statistical models. Demographic, neurologic examination, laboratory, sedative or paralytic medications, and neuroimaging data were collected as separate blinded queries.

Determination of CSF and Brain Attenuation and Specific Gravity:

Since both acquiring magnetic resonance imaging and cerebrospinal fluid (CSF) samples during severe liver failure represent high risk procedures, we interrogated the solute density of the cerebral environment using a technique that measures specific gravity from CT scans.(9, 30–32) CT attenuation measured in Hounsfield units (HU) can be used to calculate the specific gravity of regions of interest on a CT scan according to the formula: specific gravity = (1 + HU/1000) where HU is the average attenuation of the voxels in the region of interest.(9, 31, 32). Since CT scans are calibrated to water (specific gravity 1 g/mL) as 0 HU, specific gravity calculated from CT attenuation can be expressed as a physical density (g/mL).

CT scans (GE Medical Systems) of the brain acquired within 24 hours of ICU admission and as 5-mm thick contiguous slices were analyzed as Digital Imaging and Communications in Medicine formatted images using commercial semiautomated computer software (Analyze Direct 11.0, Overland Park, KS). Our group has considerable experience in the use of Analyze Direct for quantitative neuroimaging analysis.(10, 25, 33, 34) Regions of interest corresponding to the CSF-filled lateral, third, and fourth ventricles were defined with attention to exclude soft tissue structures. The average CT attenuation of the CSF was calculated from the attenuation of each voxel and the total number of voxels in the region. The average CT attenuation of the brain was calculated by defining a region of interest that encompassed the entire intracranial space and then truncating all voxels with attenuation ≥80 HU (corresponding to calcium and bone) and all voxels with attenuation below the 95^th percentile of attenuation for the voxels in the patient’s CSF measurement.

Statistical Analysis and Approvals

Continuous variables are presented as mean±SD for normally distributed variables and median (interquartile range [IQR]) for non-normally distributed variables. Univariate associations were assessed using Spearman rank correlation tests (r) and Wilcoxon rank-sum tests. Two-sided P≤0.05 was considered significant and all analyses were performed in R version 3.5.0 (R Foundation for Statistical Computing, Vienna, Austria). Our institutional care protocols require hourly documentation of GCS by trained ICU nurses. Because GCS documentation is inconsistent in the emergency department, admission GCS was defined as the initial ICU assessment. We defined admission serum osmolality and CT scan as the measurement nearest in time to the admission GCS.

Admission Encephalopathy Severity and Serum Osmolality:

Proportional odds ordinal logistic regression was used to identify independent associations between ICU admission GCS and admission serum osmolality. Our first model (A Priori Model) selected covariates for model inclusion a priori based on the published literature and biological plausibility as mechanistic contributors to HE severity and included: age, admission osmolality, admission serum ammonia level, APACHE II score (GCS points subtracted), and ALF versus ACLF. Ammonia and APACHE II score (as a marker of systemic inflammation(35)) were included in a priori models because they are widely available in clinical practice and the literature frequently invokes hyperammonemia and inflammation as mechanistic contributors to HE, cerebral metabolic dysfunction, and BBB/BCSFB dysfunction(2, 6, 7). The same a priori variables were selected for models of encephalopathy severity and CSF attenuation because CSF composition is expected to reflect cerebral pathophysiology that manifests clinically as HE.(8) To account for the possibility of unrecognized contributors to encephalopathy, we then developed a Fully Adjusted Model that included the a priori variables and all variables with univariate association with GCS at p≤0.2; the variables considered for inclusion were gender, MELD-Na, and the individual components of the MELD-Na and APACHE II scores. We then insured that models weren’t overfit to the data by using a backward selection algorithm based on Akaike Information Criteria optimization to develop a Parsimoniously Adjusted Model. In addition, a binary logistic regression model for admission in coma (GCS≤8) was developed using those variables included in the fully adjusted ordinal GCS model and this model was used for subsequent sensitivity analysis.

Admission Serum Osmolality and CSF Attenuation (Specific Gravity):

Linear regression was used to identify independent associations between admission osmolality and CSF attenuation. We used the same approach detailed above for modeling the association between admission GCS and osmolality to develop A Priori, Fully Adjusted, and Parsimoniously Adjusted Models of CSF attenuation.

To ensure that there was not meaningful confounding from endotracheal intubation performed for pulmonary indications rather than primarily for encephalopathy, we repeated fully adjusted models using only the sum of the motor and eye GCS components. In addition, to ensure that there was not an effect from multicollinearity between APACHE II, age, and osmolality or MELD-Na and osmolality, we repeated each model that included APACHE II or MELD-Na scores after subtracting any contribution of age, sodium, or potassium to the calculation of the APACHE II or MELD-Na score. In each case, repeating the models after subtracting the contribution of age, sodium, and potassium had no meaningful impact on the covariate effect estimates.

A primary concern for assessing the likelihood of causal relationships from observational data is that unmeasured factors related to both the exposure and the outcome might explain the association. To help address this concern we performed a sensitivity analysis with the fully adjusted model for coma and osmolality using the E-value. The E-value is a sensitivity analysis that makes minimal assumptions and represents the minimum strength of association, on the risk ratio scale, that an unmeasured confounder would need with both the exposure and the outcome, above and beyond the measured confounders, to completely negate a specific exposure-outcome association.(36) In biomedical research, an E-value ≥2 or 3-fold is considered supportive of causality because effects of this size are uncommon and an E-value of this size requires two such effects to be present (i.e. unmeasured confounder to exposure and unmeasured confounder to outcome).(36)

The study was approved by the Northwestern University Institutional Review Board with waiver of consent for retrospective data analysis. Data and statistical code will be shared upon request from a qualified investigator for the purpose of analysis replication.

RESULTS

There were 140 patients with liver failure and overt HE. Reasons for exclusion were: structural brain lesion (32), first osmolality measured after 24 hours (20), prolonged ICU admission at an outside institution (7), no sedation free period for GCS assessment (7), and mannitol therapy prior to ICU admission (1), yielding 73 (52%) patients for analysis. Sixty-seven (92%) analyzed patients had a CT brain scan available within 24 hours of ICU admission; all patients without a CT scan had mild encephalopathy (GCS ≥13). Eight (11%) patients received sedative medications prior to ICU admission and their GCS was assessed after a sedation hold of ≥4 hours; exclusion of these patients from the multivariate models did not meaningfully change the results.

Demographic and clinical data for the cohort are summarized in Table 1. The median admission GCS was 11 (6, 14) and 28 (38%) patients were comatose at admission. Admission serum osmolality was elevated at 303.9±15.4 mOsm/kg, whereas admission serum sodium was normal at 136.6±6.3 mEq/L. The median time from admission GCS assessment to osmolality measurement was 2.9 [1.0, 7.6] hours. Most osmolality measurements preceded CT brain scans (median time from osmolality measurement to CT scan, 2.9 [−1.7, 6.4] hours). Seventeen (23%) patients had either serum osmolality measurement or CT scan obtained in the emergency department while awaiting ICU bed assignment. Table 2 shows Spearman correlation coefficients between admission GCS, CSF attenuation and other important variables. Admission GCS was correlated with serum osmolality (r = −0.72, p<0.001) and disease severity measured by APACHE II (r = −0.49, p<0.001). Admission GCS was also lower in those with ACLF than those with ALF (8 [6, 12] versus 13 [9, 14], p = 0.004). CSF attenuation was correlated with admission osmolality (r = −0.37, p = 0.002) and admission GCS (r = 0.27, p = 0.027) but did not differ between ALF and ACLF (p = 0.45). In addition, CSF attenuation was correlated with brain tissue attenuation (r = 0.56, p<0.001) although neither osmolality nor GCS were significantly correlated (p>0.2) with brain tissue attenuation.

Table 1.

Cohort demographics and initial clinical and radiographic characteristics

Demographic and clinical characteristics, N= 73 patients	Result
Age (years, mean (sd))	49.0 (15.8)
Male	28 (38.4%)
Acute Liver Failure	39 (53.4%)
Acetaminophen and other intoxications	26 (66.7%)
Autoimmune	1 (2.6%)
Cryptogenic	3 (7.7%)
Hepatitis B	4 (10.3%)
Other	3 (7.7%)
Wilson’s Disease	2 (5.1%)
Acute-on-chronic Liver Failure	34 (46.6%)
Alcoholic cirrhosis	7 (20.6%)
Alpha-1 antitrypsin deficiency	1 (2.9%)
Autoimmune	3 (8.8%)
Cryptogenic	2 (5.9%)
Hepatitis B	1 (2.9%)
Hepatitis C	7 (20.6%)
Nonalcoholic steatohepatitis cirrhosis	9 (26.5%)
Other	4 (11.8%)
MELD-Na Score (mean (sd))	32.6 (8.4)
APACHE II Score (mean (sd))	19.3 (9.0)
Renal Replacement Therapy in First 24 Hours of Admission†	31 (42.5%)
Admission Glasgow Coma Scale Score Coma (GCS≤8)	11 [6, 14] 28 (38.4%)
Admission Serum Osmolality (mean (sd), reference 275–295 mOsm/kg)	303.9 (15.4)
Admission Serum Sodium (mean (sd), reference 135–145 mEq/L)	136.6 (6.3)
Admission Serum Blood Urea Nitrogen (median [IQR], reference 6–20 mg/dL)	28 [17, 52]
Admission Serum Glucose (mean (sd), reference 70–100 mg/dL)	136.9 (54.4)
Admission Serum Osmolar Gap (mean (sd), reference −10 to 10 mOsm/kg)	9.1 (8.2)
Admission Serum Ammonia (median [IQR], reference 15–45 μg/dL,)	182.5 [108.8, 305.0]
Admission International Normalized Ratio (median [IQR], reference 0.8–1.1)	2.5 [2.0, 4.1]
Admission Total Bilirubin (median [IQR], reference 0.1–1.2 mg/dL)	4.6 [2.2, 10.9]
Admission Creatinine (median [IQR], reference 0.7–1.3 mg/dL)	2.17 [1.07, 3.45]
Admission Temperature (Celsius, mean (sd))	36.7 (0.8)
Admission Mean Arterial Blood Pressure (mmHg, mean (sd))	87.3 (14.2)
Admission Heart Rate (beats/minute, mean (sd))	97 (18)
Admission Respiratory Rate (breaths/minute, median [IQR])	17 [16, 20]
Admission Arterial Partial Pressure to Fractional Inhalation of Oxygen (mean (sd), reference >350 mmHg/%)	328.6 (133.1)
Admission Arterial pH (median [IQR], reference 7.35–7.45)	7.39 [7.34, 7.44]
Admission Potassium (mean (sd), reference 3.5–5.0 mEq/L)	4.2 (0.9)
Hematocrit (mean (sd), reference male 40.750.3%, female 36.1–44.3%)	31.3 (6.9)
Admission White Blood Cell Count (mean (sd), reference 4.5–11 thousand/μL)	12.2 (8.5)
Admission Platelet Count (mean (sd), reference 150–400 thousand/μL)	123 (71)
CSF CT Attenuation (HU, mean (sd))	9.69 (1.69)
Brain Tissue CT Attenuation (HU, mean (sd))	34.17 (2.74)
Liver Transplantation During Hospitalization	12 (16.4%)
Discharge Disposition
Transfer to Acute Hospital	2 (2.7%)
Acute Rehab	6 (8.2%)
Death	25 (34.2%)
Home	30 (41.1%)
Hospice	1 (1.4%)
Long-term Acute Care Facility	5 (6.8%)
Skilled Nursing Facility	4 (5.5%)

^†Renal replacement therapy was performed after osmolality measurement, GCS assessment, and CT acquisition in each case.

Reference ranges represent the normal ranges reported for our hospital laboratory.

Table 2.

Spearman correlation coefficients for associations between osmolality, Glasgow Coma Scale, disease severity scores, and CSF attenuation

Variable	Sodium	Blood Urea Nitrogen	Glucose	Gap Osmoles	Ammonia	APACHE II	MELD-Na	GCS	CSF Attenuation
Osmolality	0.36 P=0.002	0.45 P<0.001	0.41 P<0.001	0.36 P=0.002	−0.006 P=0.96	0.46 P<0.001	0.003 P=0.98	−0.72 P<0.001	−0.37 P=0.002
Sodium	1	−0.28 P=0.017	0.14 P=0.24	−0.45 P<0.001	−0.17 P=0.14	−0.19 P=0.10	−0.32 P=0.006	−0.25 P=0.036	−0.027 P=0.83
Blood Urea Nitrogen		1	−0.03 P=0.77	0.19 P=0.11	−0.11 P=0.37	0.51 P<0.001	0.10 P=0.38	−0.36 P=0.002	−0.33 P=0.006
Glucose			1	0.12 P=0.32	0.18 P=0.12	0.14 P=0.24	0.021 P=0.86	−0.33 P=0.004	0.066 P=0.60
Gap Osmoles				1	0.23 P=0.055	0.45 P<0.001	0.34 P=0.003	−0.22 P=0.057	−0.11 P=0.37
Ammonia					1	0.093 P=0.44	0.21 P=0.072	−0.14 P=0.25	0.13 P=0.29
APACHE II						1	0.23 P=0.050	−0.49 P<0.001	−0.24 P=0.052
MELD-Na							1	−0.029 P=0.81	−0.20 P=0.10
GCS								1	0.27 P=0.027

CSF = Cerebrospinal fluid, GCS = Glasgow Coma Scale, MELD-Na = Model for End-Stage Liver Disease—Sodium, APACHE II = Acute Physiology and Chronic Health Evaluation—II

Two-sided P≤0.05 was considered significant for all analyses.

Table 3 summarizes multivariate models for admission GCS and CSF attenuation. In addition to those chosen a priori, variables associated with admission GCS at p≤0.2 and included in the fully adjusted ordinal model of GCS were: admission INR, total bilirubin, creatinine, hematocrit, white blood cell count, platelet count, and ratio of arterial partial pressure of oxygen to inhaled fraction of oxygen (PaO2/FiO2). Increasing serum osmolality was associated with lower GCS (OR 0.92, 95% CI [0.91, 0.93], p<0.001 for higher GCS per 1 mOsm/kg increase in osmolality; this is equivalent to OR 0.26, 95% CI [0.22, 0.31] for higher GCS per standard deviation increase in osmolality) in the fully adjusted model. The association between serum osmolality and admission GCS was similar in a fully adjusted model using only those 67 patients who also had CT scans available (OR 0.92, 95% CI [0.91, 0.93], p<0.001). Similar, significant effects were found if total GCS was replaced by the sum of motor and eye GCS components, or by evaluating the ALF and ACLF groups with separate models (Table 4). Consistent with these models, the odds of being comatose at time of ICU admission increased with increasing serum osmolality (fully adjusted OR 3.84 95% CI [1.47, 9.99] for being comatose per standard deviation increase in osmolality).

Table 3:

Models for Intensive Care Admission Glasgow Coma Scale and Cerebrospinal Fluid Attenuation

Variables	A Priori			Fully Adjusted			Parsimonious
Ordinal Regression Model for Intensive Care Admission GCS†
	OR	95% CI	p	OR	95% CI	p	OR	95% CI	p
Admission Osmolality (per mOsm/kg)	0.902	[0.895, 0.908]	<0.001	0.916	[0.905, 0.927]	<0.001	0.909	[0.903, 0.915]	<0.001
Admission Ammonia (per μg/dL)	0.997	[0.994, 1.001]	0.10	0.997	[0.993, 1.001]	0.19	--	--	--
Acute Liver Failure (reference Acute-on-Chronic Liver Failure)	3.09	[1.08, 8.82]	0.035	2.53	[0.716, 8.94]	0.15	2.53	[0.897, 7.11]	0.079
APACHE II (per point)	0.924	[0.847, 1.008]	0.074	0.951	[0.835, 1.08]	0.46	--	--	--
Age (per year)	1.03	[0.998, 1.06]	0.064	1.03	[0.987, 1.07]	0.20	1.03	[0.996, 1.06]	0.088
Admission INR	--	--	--	1.16	[0.818, 1.65]	0.40	--	--	--
Admission Total Bilirubin (per mg/dL)	--	--	--	1.007	[0.957, 1.07]	0.79	--	--	--
Admission Creatinine (per mg/dL)	--	--	--	1.08	[0.753, 1.54]	0.69	--	--	--
PaO2/FiO2 (per mmHg)	--	--	--	1.003	[0.998, 1.007]	0.22	1.004	[1.000, 1.008]	0.035
Admission Hematocrit (per %)	--	--	--	1.02	[0.908, 1.04]	0.45	--	--	--
Admission White Blood Cell Count (per thousand/μL)	--	--	--	0.974	[0.908, 1.04]	0.45	--	--	--
Admission Platelet Count (per thousand/μL)	--	--	--	1.00	[0.996, 1.01]	0.36	--	--	--
Linear Regression Model for CSF Attenuation on Computed Tomography
	β	95% CI	p	β	95% CI	p	β	95% CI	p
Admission Osmolality (per mOsm/kg)	−0.041	[−0.071, −0.010]	0.012	−0.039	[−0.069, −0.009]	0.015	−0.033	[−0.057, −0.009]	0.009
Admission Ammonia (per μg/dL)	0.002	[−0.001, 0.005]	0.18	0.006	[0.000, 0.005]	0.062	0.003	[0.000, 0.005]	0.064
Acute Liver Failure (reference Acute-on-Chronic Liver Failure)	−0.400	[−1.29, 0.489]	0.38	−0.157	[−1.05, 0.737]	0.73	--	--	--
APACHE II (per point)	0.015	[−0.072, 0.101]	0.74	0.071	[−0.031, 0.172]	0.18	--	--	--
Age (per year)	−0.025	[−0.052, 0.002]	0.074	−0.039	[−0.088, −0.005]	0.022	−0.021	[−0.044, 0.002]	0.082
MELD-Na (per point)	--	--	--	−0.039	[−0.088, 0.010]	0.13	−0.052	[−0.095, −0.009]	0.020
Admission Creatinine (per mg/dL)	--	--	--	−0.132	[−0.415, 0.151]	0.36	--	--	--
Heart Rate (per beat/minute)	--	--	--	−0.017	[−0.041, 0.007]	0.18	--	--	--

GCS=Glasgow coma scale, OR = Odds Ratio, CI = Confidence Interval, APACHE II = Acute Physiology and Chronic Health Evaluation—II, INR = International normalized ratio, PaO2/FiO2 = partial pressure of oxygen to inhaled fraction of oxygen, MELD-Na = Model for End-Stage Liver Disease—Sodium

^†The proportional odds ordinal regression model generates one covariate adjusted odds ratio for each predictor variable in the model. This odds ratio represents the effect estimate corresponding to a one-unit change in the predictor variable. For each predictor variable, the ordinal regression odds ratio represents the odds of a higher GCS versus a lower GCS being observed and this value is the same for each step on the GCS scale where the GCS may be dichotomized as higher versus lower observed GCS.

Two-sided P≤0.05 was considered significant for all analyses.

Table 4:

ALF, ACLF, and Motor-eye Subscore Parsimonious Ordinal Regression Models for Admission GCS

Variables	Total GCS in ALF			Total GCS in ACLF			Motor and Eye Subscores, ALF and ACLF
	OR	95% CI	p	OR	95% CI	p	OR	95% CI	p
Admission Osmolality (per mOsm/kg)	0.880	[0.872, 0.887]	<0.001	0.926	[0.916, 0.936]	<0.001	0.909	[0.904, 0.914]	<0.001
Age (per year)	1.036	[0.993, 1.081]	0.10	1.012	[0.958, 1.069]	0.67	1.018	[0.989, 1.048]	0.22
PaO2/FiO2 (per mmHg)	1.005	[0.999, 1.011]	0.09	1.002	[0.997, 1.007]	0.45	1.004	[1.000, 1.008]	0.04

ALF = Acute liver failure, ACLF = Acute-on-chronic liver failure, GCS=Glasgow coma scale, OR = Odds Ratio, CI = Confidence Interval, PaO2/FiO2 = partial pressure of oxygen to inhaled fraction of oxygen

Two-sided P≤0.05 was considered significant for all analyses

In addition to those chosen a priori, variables associated with CSF CT attenuation, our biomarker of CSF solute density, at p≤0.2 and included in the fully adjusted linear model were: MELD-Na, creatinine, and heart rate. Increasing serum osmolality was associated with lower CSF attenuation and had similar effect size in each model (fully adjusted β= −0.039, 95% CI [−0.069, −0.009], p=0.015 HU per 1 mOsm/kg osmolality). In addition, the association between serum osmolality and CSF attenuation was similar in separate models for patients with ALF (β= −0.049, 95% CI [−0.091, −0.008], p=0.026) and for those with ACLF (β= −0.056, 95% CI [−0.097, −0.015], p=0.012).

The sensitivity analysis using the fully adjusted model for coma and the odds ratio associated with a standard deviation change in serum osmolality yielded an E-value of 3.33 on the risk ratio scale (risk ratio is approximately the square root of the odds ratio; OR≈11).(36) Therefore, an unmeasured confounder would need to be associated with both a standard deviation change in osmolality and presentation in coma at a risk ratio of 3.33, above and beyond the measured confounders, to completely negate the association between osmolality and comatose presentation.

DISCUSSION

In our cohort of patients with overt HE and liver failure, serum osmolality was elevated at the time of ICU admission despite normal serum sodium, and higher admission serum osmolality was independently associated with more severe encephalopathy. The association between serum osmolality and encephalopathy severity was observed in both ALF and ACLF patients, suggesting a common mechanism. In addition, we found that both higher serum osmolality and greater encephalopathy severity were associated with lower CSF CT attenuation, consistent with lower CSF specific gravity. Lower CSF specific gravity could result from dilution of high-density solutes by a greater proportion of water relative to solute and/or entry of low-density solutes, such as ammonia, into the CSF.(37) This observation suggests that serum hyperosmolality might affect more severe encephalopathy through a mechanism that also alters the composition of CSF and therefore the cerebral environment. It should be noted that our study does not address the effect on CSF osmolality as this cannot be measured from CT attenuation. Decreased CSF osmolality might be expected in the case of a greater proportion of water relative to solute in CSF, and increased CSF osmolality might be expected in the case of entry of low-density solutes in to the CSF.

In addition, we observed that CSF CT attenuation was significantly associated with brain tissue CT attenuation; however, despite this association and the significant association between serum osmolality and CSF attenuation, we did not demonstrate an association between serum osmolality and brain tissue attenuation. Our data do not allow us to determine if serum osmolality in fact has no relationship to brain tissue attenuation or if our study is under-powered to detect an effect without accounting for patients’ premorbid brain tissue attenuation. Brain tissue attenuation has been used in ischemic stroke research as a quantitative measure of cerebral edema but, because of variability in premorbid brain tissue attenuation between patients, the approach requires attenuation measurements of both the edematous brain infarct and unaffected tissue (in ischemic stroke the un-infarcted hemisphere can provide an estimate of the premorbid attenuation).(38) The unavailability of premorbid CT brain imaging in our cohort prevented us from quantifying cerebral edema at the time of admission and investigating cerebral edema as a contributing factor to the association between serum osmolality and encephalopathy severity. However, quantifying cerebral edema at a given timepoint in critically ill patients who are without premorbid reference neuroimaging is not a challenge unique to our study. The availability of MRI techniques are limited by logistical challenges and safety concerns in critically ill patients, intracranial pressure (ICP) recording is a poor quantitative measure of cerebral edema because ICP compensatory mechanisms allow edema to become advanced before it is reflected in ICP elevation, and qualitative neuroimaging approaches are insensitive to cerebral edema due to factors such as premorbid cerebral atrophy.(19, 30, 39) As a future direction, a prospective study with longitudinal neuroimaging data, including when patients are free of encephalopathy, may be able to help address this challenge and better inform the role of cerebral edema in critically ill patients presenting with HE.

Although our observational study was not designed to determine the specific mechanisms by which hyperosmolality might contribute to overt HE, the existing literature suggests multiple biologically plausible and potentially additive mechanisms. First, hyperosmolality is known to suppress synaptic transmission and reduce neuronal excitability in the neocortex.(40) Second, hyperosmolality increases the permeability of the BBB and BCSFB through mechanisms that likely involve tight junction separation, cytoskeletal alterations, and nitric oxide mediated effects.(11–14) Inflammation-mediated BBB disruption is proposed to increase cerebral exposure to ammonia, and hyperosmolality may function analogously.(5, 7, 41) Similarly, hyperosmolality increases the permeability of the BBB for bilirubin, and bile acids appear to be involved in microglial activation and neurologic decline in models of ALF induced HE.(13, 42, 43) Thirdly, hyperosmolality increases the expression of aquaporin-4 (AQP-4) water channels on the plasma membranes of astrocytes, and AQP-4 expression is known to increase during both ALF and uremia, which can result from liver failure and contribute to a hyperosmolar state.(44–47) Given the pathologic role of AQP-4 suggested by murine studies in which AQP-4 deletion conferred protection against cerebral edema development and reduced HE severity, increased expression of AQP-4 resulting from hyperosmolality could contribute to HE pathophysiology.(48) Of these mechanisms, BCSFB disruption and alteration of AQP-4 expression might be expected to affect CSF specific gravity through changes in the quantity of water and solute that cross the BCSFB and through changes in the AQP-4 facilitated clearance of solutes from the brain parenchyma that occurs with the exchange of CSF and interstitial fluid.(49) These mechanisms would not necessarily require the concurrent development of cerebral edema because, for instance, altering the rate of exchange between CSF and interstitial fluid could affect solute clearance from the brain parenchyma without a net change in brain tissue fluid content.

Data from this study and previous work from our group suggests that serum osmolality might impact HE through multiple, potentially opposing mechanisms related to both absolute osmolality and change in osmolality over time. In our prior studies of liver failure patients with severe HE (West Haven HE grade 3 [stupor] and 4 [coma]), we used serial neuroimaging and serum osmolality measurements to demonstrate that acute increases in osmolality with hypertonic saline were associated with cerebral edema reduction and improved encephalopathy while large, acute reductions in osmolality exacerbated cerebral edema and contributed to neurologic deterioration, likely through generation of osmotic gradients between serum and brain tissue.(10, 25) However, in patients receiving renal replacement therapy, an intervention expected to reduce serum osmolality, blunting the rate of serum osmolality reduction with concurrent hypertonic saline administration (titrated to avoid osmolality reductions greater than 10 [West Haven grade 4] to 20 [West Haven grade 3] mOsm/kg per day) was associated with improved cerebral edema and encephalopathy severity compared to patients who received renal replacement therapy without attempts to manage osmolality reduction.(10) These data suggest that reducing serum osmolality in a controlled manner may remove the insult of hyperosmolality without separately exacerbating cerebral edema through large osmotically driven fluid shifts. Furthermore, this hypothesis would not preclude a benefit to hyperosmolar therapy targeting acute serum osmolality elevation in those cases where cerebral edema has progressed to critical levels and has become the dominant factor contributing to encephalopathy severity. Regardless, a therapeutic strategy of controlled serum osmolality reduction to manage HE will require further, targeted investigation.

There are limitations to our study. Unmeasured confounding, rather than causation, is a concern for observational data. We attempted to minimize this concern by performing a priori and fully adjusted models, separate stratified models for ALF and ACLF, models using only motor and eye GCS components, and models with coma as the outcome. Furthermore, our E-value sensitivity analysis provides evidence in favor of a causal relationship between serum osmolality and HE severity since unmeasured confounding of risk ratio ≥3.33-fold, above and beyond measured confounders, is unlikely.(36) While patients were prospectively identified, laboratory and imaging studies were collected as clinically indicated and analyzed retrospectively; a prospective investigation with contemporaneous imaging and biochemical sample acquisition would be required to eliminate the potential bias of timing delays. Since only clinically indicated laboratory studies were available, biomarkers of systemic inflammation such as tumor necrosis factor alpha (TNF-α) and interleukin-6 (IL-6) were not available as potential model covariates. However, we considered the APACHE II score as a measure of multisystem disease severity and APACHE II is strongly correlated with both TNF-α and IL-6.(35) In addition, we did not have access to a control cohort for comparison since we do not routinely monitor osmolality in another patient population without structural brain injury. As previously mentioned, our single center design may limit generalizability. However, Weiss et al observed changes in the CSF specific gravity of cirrhotic patients consistent with our findings.(9) In that study, CSF specific gravity was reduced in patients with ACLF compared to cirrhotic patients without ACLF.(8) Furthermore, CSF specific gravity trended lower (p=0.18) in cirrhotic patients with HE compared to patients without HE, an association that may not have reached significance due to small sample size of liver failure patients and dichotomous rather than ordinal measurement of HE severity.(9) In addition, our data suggest a potential influence of patient age on CSF specific gravity during liver failure (Table 3) that is compatible with existing literature. With increasing age, the choroid plexus experiences reduced enzymatic function in pathways including glycolysis, cellular respiration, and ion transport that contribute to decreased CSF production.(50) Impaired choroid plexus function with aging may represent an increased vulnerability to the effects of liver failure on CSF composition suggested by our study. While our approach of studying both ALF and ACLF is consistent with recommendations from the International Society for Hepatic Encephalopathy and Nitrogen Metabolism and we performed separate stratified analyses for ACLF and ALF, the vulnerability of these groups to the effects of hyperosmolality may differ in magnitude. Clinical trials based on mechanistic research studies in HE should investigate therapeutic effects in ALF and ACLF separately.

CONCLUSIONS

Serum osmolality at the time of ICU admission is independently associated with severity of HE in patients with liver failure. These findings, in the context of existing literature, suggest possible mechanisms involving BBB disturbance with exposure of the brain to metabolic toxins and alterations in water and solute composition within the cerebral environment. Whether interventions to address hyperosmolality in a controlled fashion may be of benefit as a therapeutic option for HE in patients with liver failure will require further investigation.

Abbreviations:

HE

hepatic encephalopathy

ALF

acute liver failure

ACLF

acute-on-chronic liver failure

GCS

Glasgow Coma Scale

CSF

cerebrospinal fluid

mOsm

milliosmoles

kg

kilogram

mEq

milliequivalent

L

liter

OR

odds ratio

CI

confidence interval

BBB

blood brain barrier

BCSFB

blood cerebrospinal fluid barrier

ICU

intensive care unit

U.S.

United States

CT

computed tomography

MELD-Na

Model for End-Stage Liver Disease—Sodium

APACHE II

Acute Physiology and Chronic Health Evaluation—II

ICP

intracranial pressure

INR

international normalized ratio

HU

Hounsfield units

g

gram

mL

milliliter

mm

millimeter

SD

standard deviation

IQR

interquartile range

PaO2/FiO2

Arterial partial pressure of oxygen to inhaled fraction of oxygen ratio

AQP-4

aquaporin-4

TNF-α

tumor necrosis factor alpha

IL-6

interleukin-6