Research Article

Characterising adrenal function using directly measured plasmafree cortisol in stable severe liver disease

Terrence Tan1,*, Linus Chang1, Aidan Woodward1, Brett McWhinney2, John Galligan2,Graeme A. Macdonald1,3, Jeremy Cohen4, Bala Venkatesh4,5

1Department of Gastroenterology and Hepatology, Princess Alexandra Hospital, Australia; 2Queensland Pathology, Australia; 3The DiamantinaInstitute of Cancer, Immunology and Metabolic Medicine, The University of Queensland, Brisbane, Australia; 4Department of Intensive Care,Princess Alexandra Hospital and Royal Brisbane and Women’s Hospitaland, Australia; 5Critical Care Endocrinology and Metabolism Research

Unit, Australia

Background & Aims: Adrenal insufficiency (AI) has been Introduction

reported in patients with advanced liver disease. Diagnosing AIis problematic owing to controversies in using total serum corti-sol as a measure of adrenal function. No published data exist ondirectly measured plasma free cortisol (PFC) in patients with liverdisease.Methods: This prospective study compared serum total and mea-sured plasma free cortisol to evaluate adrenal function in clini-cally stable cirrhotic patients and healthy controls. Cortisollevels were measured at baseline and following 250 lg cortico-trophin. AI was defined by total cortisol increments (delta corti-sol) of less than 250 nmol/L, or a peak total cortisol under500 nmol/L after cosyntropin. We used a peak plasma free corti-sol concentration of 33 nmol/L as the threshold for AI.Results: Forty-three consecutive patients and 10 healthy controlswere studied. Cirrhotic patients had significantly lower peak (526vs. 649 nmol/L, p = 0.004) and delta total cortisol (264 vs.397 nmol/L, p = 0.002) responses compared to healthy controls.However, basal plasma free cortisol was higher in patients(10.9 vs. 6.4 nmol/L, p = 0.03), and there were no differences inpeak plasma free cortisol (p = 0.69) between the two groups.The prevalence of AI using total cortisol criteria was 58% com-pared to 12% using free cortisol (p <0.001).Conclusion: In patients with stable severe liver disease, a signif-icant discrepancy exists between the rates of diagnosis of AIusing the total and free cortisol criteria. We would advise cautionin the interpretation of adrenal function testing using total corti-sol measurements in this group.Crown copyright � 2010 Published by Elsevier B.V. on behalf ofthe European Association for the Study of the Liver. All rightsreserved.

Journal of Hepatology 20

Keywords: Liver cirrhosis; Advanced liver disease; Adrenal function; Cortisol.Received 29 December 2009; received in revised form 17 May 2010; accepted 17 May2010; available online 17 July 2010*Corresponding author. Address: Department of Gastroenterology and Hepatol-ogy, Princess Alexandra Hospital, Ipswich Road, Woolloongabba, 4102 Queens-land, Australia. Fax: +61 732405111.E-mail address: perthfracp@gmail.com (T. Tan).

Cortisol, the primary glucocorticoid secreted by the adrenal cor-tex [1], is a crucial component of the host’s response to severestress through the activation of the hypothalamo-pituitary–adre-nal axis (HPA). It is vital for the maintenance of the vascular tone[1,2], cardiac output [1,2], immunomodulation [3] metabolic bal-ance and various other key intracellular functions. Functioning ofthe HPA axis is usually assessed in clinical practice by measure-ments of total serum cortisol, either at baseline or following stim-ulation by synthetic ACTH (the short Synacthen test – SST). Usingthese criteria, adrenal insufficiency is usually defined as an earlymorning cortisol below 100 nmol/L or peak cortisol in response tothe SST of <500 nmol/L [4]. Alternative criteria have been pro-posed in the setting of critical illness; a random cortisol of lessthan 275 nmol/L, or an increment in response to the SST of<250 nmol/L have been suggested to be diagnostic of Critical Ill-ness Related Corticosteroid Insufficiency (CIRCI) [5]. Recent stud-ies have suggested that adrenal insufficiency (AI) may beprevalent in patients with liver disease [6,7], particularly cirrho-sis [8]. The term ‘‘hepatoadrenal syndrome” has been coined todescribe apparent adrenal insufficiency in patients withadvanced liver disease in the setting of sepsis and/or other stress-ors [9,10]. The issue is relevant because cirrhotic patients withsepsis [11] and other forms of critical illnesses are at an increasedrisk of hemodynamic compromise [7,12], cardiovascular collapseand death which may be contributed by adrenal dysfunction [13].The presence of AI in advanced liver disease appears to be relatedto functional liver reserve and disease severity, and is associatedwith hemodynamic instability, renal dysfunction and reducedsurvival [14].

Estimating adrenal function in liver disease remains difficultand controversial, due to the use of total serum cortisol to defineadequate adrenal function [14–17]. More than 90% of cortisol inblood is bound to corticosteroid binding globulin (CBG), albuminand erythrocytes [18,19], whilst the remainder exists as plasmafree cortisol (PFC). PFC is the biologically active hormone, ableto pass across cellular membranes to exert its effects on the intra-cellular glucocorticoid receptors [18]. However prediction of PFCfrom total cortisol is difficult as the relationship between the twois non-linear. Additionally in liver disease, both CBG and albumin

10 vol. 53 j 841–848

Research Article

production are impaired, thus potentially resulting in increasedPFC concentrations making total cortisol a potentially unreliableindex of PFC [20,21]. A recent study utilising the Free CortisolIndex (FCI), a surrogate marker for plasma free cortisol, suggestedthat there is a discrepancy between PFC and TC in the diagnosis ofadrenal insufficiency in patients with severe liver disease [22].Other studies have used salivary cortisol as an index of PFC[23]. However, surrogate markers may not be completely reliableas indicators of free cortisol activity, and there are currently nopublished data on measured free cortisol levels in patients withsignificant hepatic impairment.

We elected to perform a prospective observational study tocharacterise adrenal function using measured PFC and comparethese data with simultaneous total plasma cortisol levels at base-line and following ACTH stimulation in a group of patients withclinically stable liver cirrhosis. We also aimed to compare theprevalence of adrenal insufficiency using the existing diagnosticcriteria with that determined using plasma free cortisol. In addi-tion we wished to examine the relationship between measuredfree cortisol and calculated indices of free cortisol in this cohortof patients.

Materials and methods

Patients

This prospective study was conducted between June 2008 and June 2009 in a sin-gle centre teaching hospital which serves as a tertiary referral centre for adultliver transplantation in the state of Queensland, Australia. The study protocolwas approved by the Princess Alexandra Hospital Human Research Ethics Com-mittee. All study participants gave informed consent prior to enrolment. Theinclusion criteria were patients with liver cirrhosis referred for evaluation fororthotopic liver transplant (OLT). The diagnosis of cirrhosis was made on clinical,haematological and biochemical grounds with all patients having presented withprior episodes of decompensation or deteriorating liver function. Histologic con-firmation on liver biopsies and/or the presence of radiological features of cirrhosis(CT, MRI or ultrasound) with portal hypertension served as adjuncts to the diag-nosis of cirrhosis in these patients. Controls were recruited from groups of nurs-ing, administrative and medical staff with no documented history of liver, knownendocrine or autoimmune disease and biochemically normal serum albumin andliver enzymes. All participants were clinically stable and patients were outpa-tients at the time of evaluation. Exclusion criteria included acutely decompen-sated liver disease, active bacterial infection, corticosteroid therapy within thepreceding 6 months, pregnancy, age <18 years, active malignancy other thanhepatocellular carcinoma, patients with known endocrine or autoimmune dis-eases other than diabetes mellitus and patients referred for OLT evaluation fornon-cirrhotic indications (e.g. symptomatic polycystic liver disease). All studyparticipants were followed up on a monthly basis in the pre-transplant liverclinic. Outcomes were recorded for composite endpoints of death and/or adverseevents (defined as either hepatorenal syndrome, sepsis, spontaneous bacterialperitonitis or variceal bleeding), as well as liver transplantation.

Laboratory investigations

All participants underwent an outpatient SST [24] performed by a single author(TT). The SSTs were undertaken between 0800 and 0900 h in an outpatient clinicsetting. Following an overnight fast, blood was collected for initial early morningbaseline measurement of TC, PFC and ACTH. Synacthen stimulation was per-formed by administering 250 lg cosyntropin (Synacthen, Novartis, North Ryde,NSW, Australia) through a peripheral intravenous cannula. Serial TC and PFC weremeasured at 30 and 60 min intervals after cosyntropin administration. We alsomeasured baseline serum dehydroepiandrostenedione sulfate (DHEAS), andCBG. Serum albumin, INR, liver enzyme and renal function were measured onthe day of SST. Fasting lipid profiles were assessed within 2 weeks of the SST inthe patient cohort.

CBG was measured by radioimmunoassay (IBL International GMBH, Ham-burg, Germany). The determined intra-assay coefficient of variation for thisassay was 2.8% at 54 mg/L. Serum total and plasma free cortisol were measured

842 Journal of Hepatology 201

in whole plasma using an accurate, reproducible and automated liquid chroma-tography–mass spectrometry (LC-MS/MS) multiple reactant monitoring tech-nique, modified from the method previously described by Huang et al. [25].In brief, 250 ll of sample plasma was equilibrated and centrifuged (tempera-ture controlled 37 �C) in a Amicon Ultra 30,000 MWCO centrifugal filter device(Millipore, MA, USA). The ultrafiltrate was then extracted with a C18 SEP PAKVAC solid phase cartridge (Waters Corp., Milford, MA, USA). The eluant wasdried and reconstituted in mobile phase. Chromatography was performed ona Waters Acquity Ultra Performance Liquid Chromatography (UPLC) systemcoupled to a Waters Quattro Premier Tandem Mass Spectrometer. Multiplereactant monitoring was carried out on each individual analyte and results cal-culated. The assay was linear to at least 2000 nmol/L for all analytes(R2 >0.995).

The peak cortisol concentration was defined as the highest cortisol value afterSynacthen administration at 60 min post-injection. The serum/plasma cortisolincrement (or delta cortisol) was defined as the difference between the peakand basal cortisol concentrations. AI was diagnosed using three sets of criteria;standard endocrinological, CIRCI, and free cortisol. Patients were considered tofulfil standard criteria if the peak total cortisol was below 500 nmol/L, CIRCI cri-teria if the delta total cortisol was below 250 nmol/L, and finally free cortisol cri-teria if the peak PFC was under 33 nmol/L [26–28]. Patients who fulfilled any setof criteria were designated the ‘‘AI” cohort and those who did not the ‘‘non-AI”cohort.

Calculated PFC was derived using the Coolens method with the followingequation: U2 � K(1 + N) + U[1 + N + K(G � T)] � T = 0, where T is the molar con-centration of total cortisol, G is CBG, U is unbound cortisol, and K is the affinityof CBG for cortisol at 37 �C. N is the ratio of albumin bound to free cortisol, forwhich 1.74 is the value conventionally used [21,29]. The Free Cortisol Index(FCI) was calculated by dividing the total cortisol by the patient’s CBG [30]. AnFCI of >12 is considered to represent adequate adrenal function [31].

Statistical analysis

All statistical analyses were performed using the GraphPad Prism 4.00 (San Diego,CA) software. Categorical non-continuous data were compared using the Chi-squared test. The Mann–Whitney U non-parametric rank sum test was used tocompare the medians of continuous variables within the non-Gaussian distribu-tion data. A p-value of less than 0.05 was considered to be statistically significant.A z-test for difference between proportions was used to compare the prevalencesof AI using total and free cortisol criteria. When more than two groups were com-pared, the Kruskal–Wallis one-way analysis of variance (ANOVA) test was used todetect statistical significance. The Spearman correlation coefficient was used toanalyse the relationship between continuous variables such as the results ofthe SST and parameters such as liver disease severity, CBG and serum albumin.The strength of agreement between the direct measurements of PFC and thoseobtained by Coolens equation was analysed by using the Bland–Altman method.All statistical values were reported as the mean ± standard error of the mean(SEM) unless otherwise stated.

Results

Demographic data

Forty-three patients with clinically stable liver cirrhosis and 10healthy controls were enrolled into the study. The demographicdata of the participants are presented in Table 1. The aetiologiesof liver disease were alcohol in 10 patients, hepatitis C (HCV) in10, a combination of alcohol and HCV in 8, primary biliary cirrho-sis (PBC) in 3, primary sclerosing cholangitis (PSC) in 4 patients,non-alcoholic steatohepatitis (NASH), in 4, and 1 patient eachhad hepatitis B (HBV), haemochromatosis, cystic fibrosis and bil-iary atresia.

The median Child-Pugh score was 9 (range 5–13). Twenty-twopatients had Child-Pugh A or B cirrhosis and 21 had class C cir-rhosis. The median MELD score was 13 (range 6–27). Nine(22%) patients had a history of hepatocellular carcinoma. Thecontrol group were significantly younger than the patient group(44 ± 2.5 vs. 51 ± 1.3 years, p = 0.01),

0 vol. 53 j 841–848

Table 1. Demographics in patient and control groups.

All values are presented as mean ± SEM unless otherwise stated. A p-value of <0.05is considered statistically significant. Total, LDL and HDL-cholesterol as well asserum triglycerides all expressed in mmol/L. *n/a denotes results not available ornot applicable. Normal ranges for the following clinical and biochemistry variablesare: Body Mass Index (BMI 20–25), serum Na (135–145 mmol/L), serum K (3.5–4.5 mmol/L), serum bilirubin (<20 lmol/L), INR (0.9–1.1), serum albumin (35–50 g/L), serum creatinine (70–108 lmol/L), Total cholesterol (<5 mmol/L), LDL-choles-terol (2–3.4 mmol/L), HDL-cholesterol (0.9–1.6 mmol/L), triglycerides (<2 mmol/L), serum CBG reference range (male 22–55 mg/L, female 40–154 mg/L).

Table 2. Hormone profiles at baseline and post short Synacthen test.

Values are presented as mean ± SEM. A p-value of <0.05 is considered statisticallysignificant. Normal range for the following biochemical parameters are: baselinetotal cortisol (>150 nmol/L), serum ACTH (10–50 ng/L), DHEAS (1–11 lmol/L),peak total cortisol >500 nmol/L, delta total cortisol >250 nmol/L and peak FreeCortisol Index >12.

JOURNAL OF HEPATOLOGY

Plasma hormone profiles

BaselineThe baseline total cortisol (TC) values were similar between thepatients and controls (Table 2). The PFC was significantly higherin the study cohort compared to controls (10.9 ± 1.3 vs.6.4 ± 1 nmol/L, p = 0.03, Fig. 1A). Baseline plasma ACTH and DHEASlevels were significantly lower in the patient group compared tocontrols (19.7 ± 1.2 vs. 31.5 ± 3.7 ng/L for ACTH, p = 0.004 and0.8 ± 0.1 vs. 4.4 ± 0.9 lmol/L for DHEAS, p <0.0001). For ACTH,these values were all still within the normal reference range(10–50 ng/L), while for DHEAS, most patients had a serum concen-tration of <1.1 lmol/L, values of less than a quarter of that seen inhealthy controls. The DHEAS values did not differ significantlybetween the different Child-Pugh groups (1.3 ± 0.43 for Childs Avs. 0.73 ± 0.07 for Childs B vs. 0.71 ± 0.06 for Childs C, p = 0.44).Serum CBG was significantly lower in the patient group comparedto healthy controls (37.6 ± 2.1 vs. 47.1 ± 1.8 mg/L, p = 0.001).

Post corticotropinThere were significantly lower peak TC and delta TC responsesfollowing ACTH in patients as compared to controls; peak TC(526 ± 21 vs. 649 ± 25 nmol/L, p = 0.004) (Fig. 1B) and delta TC(264 ± 18 vs. 397 ± 32, p = 0.002, Fig. 1C), Eighteen (42%) patientshad post-ACTH peak TC of <500 nmol/L. Peak PFC following ACTHstimulation was similar for patients and controls (p = 0.48).

Prevalence of AIA total of 25 patients (58%) were diagnosed with AI by any set ofcriteria. Eighteen patients (39%) fulfilled standard criteria for AI.Twenty (47%) patients fulfilled the criteria for CIRCI. Thirteenpatients (30%) had discordant results from the two methods.Using CIRCI criteria was associated with a 7% higher probability

Journal of Hepatology 201

of a diagnosis of AI than standard criteria (95% CI, �1.0 to 23.0,p = 0.4). Utilising the PFC criteria five patients (12%) were diag-nosed with AI. All of these 5 patients also fulfilled the standardcriteria and 4 out of 5 fulfilled the CIRCI criteria. Using CIRCI cri-teria was associated with a 33% higher likelihood of a diagnosis ofAI compared to the free cortisol criteria (95% CI 14 to 51%p <0.001). No controls met the standard or free cortisol criteriafor diagnosis of AI. One healthy control met the CIRCI criteriabut had normal peak total cortisol.

Comparison of AI cohort and controlsAs expected, there was a significant difference between the peakand delta total cortisol values in the AI and control groups (peaktotal cortisol 434 ± 23 vs. 670 ± 25 nmol/L, p <0.001, delta cortisol201 ± 13 vs. 397 ± 32 nmol/L, p <0.001). However there were nosignificant differences in peak free cortisol values between thegroups (peak free cortisol 42 ± 3.6 nmol/L vs. 58 ± 4.6 nmol/L,p = 0.09).

Comparison of AI and non-AI cohortsDemographic factors, hormonal profiles and disease severity werecompared between the AI (n = 25) and non-AI (n = 18) cohorts(Table 3). The AI cohort of patients had a higher proportion ofHCC and ascites but these differences were not statistically signif-icant. The severity of liver disease was comparable between thetwo cohorts. The AI cohort had lower peak PFC values (46.2 ± 3.6vs. 66.5 ± 4.1 nmol/L, p = 0.001) and CBG concentrations (33 ± 1.8vs. 44 ± 3.5 mg/L, p = 0.005) than the non-AI cohort. No significantdifferences were observed between the two cohorts in relation toserum albumin (p = 0.41), ACTH (p = 0.06), plasma DHEAS(p = 0.65), lipid profiles or liver disease severity. There was no sig-nificant correlation observed between liver disease severity (MELDscore) and total cortisol (r = 0.29), or free cortisol (r = 0.12).

Comparisons were also made between AI (n = 5) and non-AI(n = 38) patients as determined by the PFC criteria (Table 4).Using this criteria, subjects in the AI group had significantlyhigher median MELD scores (19 ± 3.2 vs. 13 ± 0.8, p = 0.03) anda significantly higher mortality rate (60% vs. 5%, p = 0.0007. Ofthe 3 patients who died in the AI group, two died of sepsis relatedcomplications and the remainder from early post OLT related

0 vol. 53 j 841–848 843

A

B

TOTAL CORTISOL

Patie

nts

Con

trols

Patie

nts

Con

trols

Patie

nts

Con

trols

0

250

500

750

1000

Baseline Peak Delta

p = 0.004

p = 0.002

SUBJECT GROUPS

SUBJECT GROUPS

nmol

/L

Patie

nts

Con

trols

Peak

Patie

nts

Con

trols

Baseline

PLASMA FREE CORTISOL

0

20

40

60

80

100

120

p = 0.001

**p = NS

nmol

/L

Fig. 1. Box-and-Whisker plots comparing the total and plasma free cortisolvalues between patient and control groups. (A) The total cortisol for patientsand controls at baseline, peak and delta values. (B) Plasma free cortisol profiles forboth groups at baseline and maximal levels post-ACTH.

Table 3. Clinical, hormonal and biochemical profiles of AI and non-AI groups.

*NS denotes non-significant p-values. Units for cholesterol and triglycerides inmmol/L.

Table 4. Subgroup analysis of AI and non-AI patients according to the plasmafree cortisol (PFC) criteria. A p-value of <0.05 is considered statisticallysignificant.

Research Article

complications. The overall adverse event outcomes other thandeath did not differ significantly between these two groups.

Comparisons between total and measured plasma free cortisol inpatientsSerum total and measured plasma free cortisol correlated well inpatients who had albumin values of greater than 30 g/L (n = 21)with correlation coefficients of r = 0.71 and 0.69 for baselineand peak cortisol, respectively. However, the strength of correla-tion remained strong for patients who had serum albumin of lessthan 30 g/L (n = 22) (r = 0.78 and 0.61 for baseline and peak cor-tisol, respectively, Fig. 2). Subgroup analysis of the AI and non-AIgroups revealed modest positive correlation between TC and PFCin the AI cohort (r = 0.54) but a poor negative correlation(r = 0.14) in the non-AI patients (Fig. 2).

Comparisons between measured and calculated markers of plasmafree cortisol

We performed comparisons of measured PFC with the valuesobtained by two other measures; the FCI and the Coolens

844 Journal of Hepatology 201

method. The FCI is a surrogate marker for serum free cortisol thatis calculated from the ratio of total cortisol (nmol/L) to serum

0 vol. 53 j 841–848

BASELINE CORTISOL FOR ALBUMIN < 30g/L

A PEAK CORTISOL FOR ALBUMIN < 30g/L

C

0 100 200 300 400 500 6000

10

20

30

40

50

Total Cortisol (nmol/L)Spearman Coefficient r = 0.78

p <0.0001

Total Cortisol (nmol/L)Spearman Coefficient r = 0.61

p = 0.003

Total Cortisol (nmol/L)Spearman Coefficient r = 0.69

p = 0.0005

Plas

ma

Free

Cor

tisol

(nm

ol/L

)

Plas

ma

Free

Cor

tisol

(nm

ol/L

)

Plas

ma

Free

Cor

tisol

(nm

ol/L

)

BASELINE CORTISOL FOR ALBUMIN > 30g/L

B

0 100 200 300 400 5000

5

10

15

20

Total Cortisol (nmol/L)Spearman Coefficient r = 0.71

p = 0.0003Pl

asm

a Fr

ee C

ortis

ol (n

mol

/L)

0 100 200 300 400 500 600 700 8000

2040

6080

100120

PEAK CORTISOL FOR ALBUMIN > 30g/L

DPEAK CORTISOL IN

AI GROUP

E

Peak Total Cortisol (nmol/L)Spearman Coefficient r = 0.54

p = 0.005

Plas

ma

Free

Cor

tisol

(nm

ol/L

)

0 100 200 300 400 500 600 700 8000

2040

6080

100120

0 200 400 600 800 10000

20

40

60

80

100

PEAK CORTISOL NON AI GROUP

F

PeakTotal Cortisol (nmol/L)Spearman Coefficient r = 0.14

p = 0.58Pl

asm

a Fr

ee C

ortis

ol (n

mol

/L)

400 500 600 700 800 900 10000

2040

6080

100120

–

Fig. 2. Correlation analysis between total and measured plasma free cortisol for patients with albumin values above or below 30 g/L. Data are presented for baseline(A and B) and peak (C and D) cortisol measurements and showed a strong degree of correlation between total and plasma free cortisol despite having lower albumin levels.However poor correlations existed between peak total and plasma free cortisol in the AI and non-AI groups (E and F).

JOURNAL OF HEPATOLOGY

CBG (mg/L). The Coolens equation provides a method of calculat-ing the actual serum PFC indirectly based on measured CBG andtotal cortisol with the provision of certain assumptions on theconditions in which the TC was measured [29].

The PFC correlated significantly with the FCI at baseline(r = 0.73) but less strongly with the peak free cortisol response(r = 0.47, see Fig. 3). Only 10 (24%) patients had a peak FCI of<12 to suggest inadequate adrenal function. Out of these 5(12%) subjects had an inadequate peak TC of <500 nmol/L. Thecorrelation coefficient between calculated (using Coolens equa-tion) and measured PFC was 0.66. The Bland–Altman analysisrevealed a poor level of agreement between the 2 sets of values(bias of 5.53, limits of agreement from �4.8 to 15.9, see Fig. 4).The calculated PFC values were significantly higher than the mea-sured in both the control (11.3 vs. 6.4 nmol/L, p = 0.008) andpatient cohorts (16.4 vs. 10.9 nmol/L, p = 0.003) with variationsin values by up to 160%.

Clinical outcomes and survival analysis

Clinical outcomes assessed included death or OLT, survival at 6and 12 months, as well as adverse events including varicealhaemorrhage, spontaneous bacterial peritonitis, hepatorenal syn-drome or sepsis. The presence of AI was not associated withworse outcomes in terms of death without OLT (3 patients in AIvs. 2 in the non-AI cohort) or combined adverse events (5 eventsin AI vs. 8 in non-AI). The number of patients in both cohorts who

Journal of Hepatology 201

underwent OLT was also similar (10 in non-AI vs. 13 in AI).Cumulative overall survival at 6 and 12 months were similar inthe two cohorts.

Discussion

Our results suggest that in patients with stable severe liver dis-ease using PFC measurements results in a significantly lowerprevalence of AI compared to using criteria based upon total cor-tisol values. We used three separate diagnostic criteria to deter-mine the prevalence of AI in our population. The standardendocrinological criteria are well established [4] and are usedin the investigation of AI in stable patients. In patients with crit-ical illness, alternative criteria based upon the delta cortisol havebeen proposed by a recent consensus conference [5]. In contrastto total cortisol estimations, there are scant published data onthe normal PFC response to Synacthen stimulation. Vogeseret al. studied the peak PFC response to a standard ACTH test in10 normal subjects and documented a range of 33–138 nmol/L[28]. A similar range of 33.7–168.5 nmol/L was reported by Lewiset al. [27] Higher ranges from 52 nmol/L have also been observed[20] Additionally, a comparison of salivary cortisol levels follow-ing ACTH testing in healthy controls and patients with docu-mented adrenal insufficiency suggested a normal lower limitfor salivary levels of 23.5 nmol/L [26]. As salivary cortisol concen-trations are usually estimated to be roughly 60% of the corre-sponding plasma free levels [32], this finding is consistent with

0 vol. 53 j 841–848 845

A

Measured Plasma Free Cortisol (nmol/L)Spearman Coefficient r = 0.73

p <0.0001

BASELINE FREE CORTISOL

0 10 20 30 40 500.0

2.5

5.0

7.5

10.0

12.5

15.0

17.5

Free

Cor

tisol

Inde

x

B

Measured Plasma Free Cortisol (nmol/L)Spearman Coefficient r = 0.47

p = 0.001

PEAK FREE CORTISOL

0 25 50 75 100 125

Free

Cor

tisol

Inde

x

0

10

20

30

Fig. 3. Spearman Correlation analysis for (A) baseline and (B) peak freecortisol between measured plasma values and calculated Free Cortisol Index(FCI) in all patients (n = 43). Measured plasma free cortisol correlated well withFCI at baseline (r = 0.73) but not with peak FCI (r = 0.47).

Calculated Free Cortisol (nmol/L)

Mea

sure

d fre

e C

ortis

ol (n

mol

/L)

0 5 10 15 20 30 35 40 450

20

40

10

30

50

Measured vs Coolens Calculated Baseline Free

Cortisol in Patients

R = 0.66532

r = 0.66

Mean of Measured/Calculated

Diff

eren

ce C

aclu

late

d-M

easu

red

0 10 20 30 40 50

-505

15

-25

-15

25

Bland Altman Analysis of Measured and Coolens Calculated

Free Cortisol in Patients

ig. 4. Correlation (A) and Bland Altman (B) analysis of measured andalculated Coolens free cortisol in patients demonstrating a poor strength ofgreement despite a high degree of correlation assessed by Spearman method

= 0.66).

Research Article

the previous data. We elected to use 33 nmol/L as the thresholdfor adrenal insufficiency as this was the lowest published limitsof PFC response to corticotropin in healthy subjects.

Using total cortisol criteria, we noted an overall prevalence of AIof 58% in our cohort. However, using only the PFC criteria ourobserved incidence of AI was just 12%. These results are in contrastto the prevailing opinion that AI is a significantly more commoncomplication of severe liver disease. Marik et al., using total cortisolcriteria reported a prevalence of AI of 66% in patients described ashaving chronic stable liver disease [9], which is consistent with ourresults. McDonald et al. reported on 27 patients with severe liverdisease who underwent corticotropin testing [13]. Peak and deltatotal cortisol concentrations were significantly lower in the patientcohort than in controls. Calculated free cortisol concentrationsfrom Coolens method were unchanged at baseline between thetwo groups. An incidence of AI of 62% was observed in 45 patientswith a broad range of liver dysfunction [6]. The authors used a com-bination of baseline, peak and delta total cortisol concentrations asdiagnostic criteria. Other studies using criteria based upon totalcortisol estimations have reported similar results [14,33].

Interpreting these findings is problematic due to their reli-ance upon total cortisol, which may be misleading in this patient

846 Journal of Hepatology 201

A

B

Fca(r

group owing to alterations in serum concentrations of albuminand CBG. In settings where CBG and plasma protein levels arealtered, there may be clinically significant changes in the activefree cortisol fraction which would not be appreciated by mea-surement of the total plasma concentration. Evidence to supportthis hypothesis has come from studies in critically ill patientswhich have demonstrated a poor correlation between total andfree cortisol estimations, and suggested that free cortisol fractionis more closely related to disease severity than is the total con-centration [21,20]. A recent study examining 26 patients withliver impairment compared the Free Cortisol Index (FCI) withstandard total cortisol diagnostic criteria and demonstrated thatin 46% of patients there was a discrepancy between the twomethods [22]. However, this method cannot account for thenon-linear binding properties of CBG and thus the higher pro-portional increase in free cortisol following ACTH stimulation[28]. To our knowledge our investigation represents the firstreport of measured plasma free cortisol concentrations in thisgroup.

The most appropriate method of diagnosing AI in the settingof chronic liver failure is not clear. The standard endocrinological

0 vol. 53 j 841–848

JOURNAL OF HEPATOLOGY

criteria for investigating AI in stable patients using total cortisolestimations are well validated in the general population outpa-tient setting [4,34], but their applicability in the setting of chronicsevere liver dysfunction is unknown. Likewise, it has been sug-gested that there are similarities between some pathophysiolog-ical features of liver disease and severe critical illness [9,35,36].These might suggest that diagnostic criteria described in patientswith critical illness may be more appropriate. However, these cri-teria have been derived primarily from observations of patientswith severe sepsis requiring vasopressor support [5], a very dif-ferent group to our study cohort.

In recognition of this uncertainty we elected to apply both ofthese well described criteria to our patient population. We foundthat significant abnormalities were present using both sets of crite-ria, but there was a moderate level of agreement between the two.In 30 patients (70%) the two tests correlated. The CIRCI criteriawere not significantly more likely to return a positive result for AI.

A mechanism that may further complicate the assessment ofadrenal function in this patient group is that of tissue glucocorti-coid resistance. This is a plausible mechanism that has beendescribed in several chronic conditions, including asthma, rheu-matoid arthritis and inflammatory bowel disease [37] and hasmore recently been proposed to be relevant in critical illness[5]. Alterations in glucocorticoid metabolism including changesin glucocorticoid receptor number and binding capacity havebeen described but the effect of these changes upon clinical out-comes is currently unclear [38].

We were not able to demonstrate an association between adiagnosis of AI and the severity of liver disease. The presence ofAI, however, was associated with significantly lower CBG levelsthan in the non-AI cohort. This observation coupled with the lackof significant difference between PFC levels in the AI group andcontrols supports the contention that PFC levels may be a betterguide to adrenal function in this group. The liver is the major siteof cholesterol biosynthesis (the precursor of cortisol). However,the lack of significant differences in the lipid profiles betweenthe AI and non-AI groups suggests that impaired cholesterol pro-duction is not an adequate explanation for the observed adrenalinsufficiency.

The techniques required to measure plasma free cortisol are notwidely available. In lieu of direct measurements, surrogate mark-ers have been advocated, of which the Coolens equation and theFree Cortisol Index (FCI) are the most frequently described. TheCoolens method has been found to provide values that correlatedhighly with measured PFC [21] and the FCI has been argued to bea better determinant of HPA-axis function than total cortisol [22].In our study we found no significant correlation between the peakPFC and FCI. This is not an unexpected finding given the relativeover-production of cortisol compared to CBG following ACTH stim-ulation, resulting in a relatively greater increase in PFC [28]. Theunderlying assumptions in the Coolens equation may not applyin severe liver disease, as shown by the high bias and imprecisionnoted in the Bland Altman analysis.

We also observed a significant difference in the serum DHEASlevels between patients and controls. This observed differencemay be due to two factors: either reductions in DHEA synthesisin the adrenals or impaired DHEA sulfation in the liver and adre-nal glands. Because cortisol and DHEAS synthesis are linked bytheir intermediate metabolites of pregnenolone and progesteroneconversion, it is possible that a decline in DHEAS levels mayreflect underlying adrenal dysfunction or liver disease.

Journal of Hepatology 201

In this study the presence of AI by currently accepted criteriawas not associated with a worse outcome in terms of adverseclinical events or mortality rates. However, there was a statisti-cally significant increase in observed mortality in those patientswho fulfilled free cortisol criteria for AI compared to the remain-der of the cohort. The study, however, was not powered to per-form survival analysis, so this result should be interpreted withcaution.

In summary, in patients with stable severe liver disease wehave demonstrated a significant discrepancy between the ratesof diagnosis of adrenal insufficiency using total and free cortisolcriteria. Further studies are warranted in a wider clinical rangeof patients, including decompensated cirrhosis, in order to inves-tigate the predictive value of free cortisol measurements in thissetting. In addition, we would advise caution in the interpreta-tion of adrenal function testing using either total cortisol or sur-rogate methods for estimating free cortisol measurements inthis group.

Conflict of interest

The authors declare no financial conflict of interest during thewriting and production of this article and throughout the periodof the research study.

References

[1] Arlt W, Stewart PM. Adrenal corticosteroid biosynthesis, metabolism, andaction. Endocrinol Metab Clin North Am 2005;34:293–313, viii.

[2] Handoko K, Yang K, Strutt B, Khalil W, Killinger D. Insulin attenuates thestimulatory effects of tumor necrosis factor alpha on 11beta-hydroxysteroiddehydrogenase 1 in human adipose stromal cells. J Steroid Biochem Mol Biol2000;72:163–168.

[3] Cronstein BN, Kimmel SC, Levin RI, Martiniuk F, Weissmann G. A mechanismfor the antiinflammatory effects of corticosteroids: the glucocorticoidreceptor regulates leukocyte adhesion to endothelial cells and expressionof endothelial-leukocyte adhesion molecule 1 and intercellular adhesionmolecule 1. Proc Natl Acad Sci USA 1992;89:9991–9995.

[4] Arlt W, Allolio B. Adrenal insufficiency. Lancet 2003;361:1881–1893.[5] Marik PE, Pastores SM, Annane D, et al. Recommendations for the diagnosis

and management of corticosteroid insufficiency in critically ill adultpatients: consensus statements from an international task force by theAmerican College of Critical Care Medicine. Crit Care Med 2008;36:1937–1949.

[6] Harry R, Auzinger G, Wendon J. The clinical importance of adrenalinsufficiency in acute hepatic dysfunction. Hepatology 2002;36:395–402.

[7] Harry R, Auzinger G, Wendon J. The effects of supraphysiological doses ofcorticosteroids in hypotensive liver failure. Liver Int 2003;23:71–77.

[8] Manglik S, Flores E, Lubarsky L, et al. Glucocorticoid insufficiency in patientswho present to the hospital with severe sepsis: a prospective clinical trial.Crit Care Med 2003;31:1668–1675.

[9] Marik PE, Gayowski T, Starzl TE. The hepatoadrenal syndrome: a commonyet unrecognized clinical condition. Crit Care Med 2005;33:1254–1259.

[10] Marik PE. Adrenal-exhaustion syndrome in patients with liver disease.Intensive Care Med 2006;32:275–280.

[11] Borzio M, Salerno F, Piantoni L, et al. Bacterial infection in patients withadvanced cirrhosis: a multicentre prospective study. Dig Liver Dis2001;33:41–48.

[12] Iwasaki Y, Aoki Y, Katahira M, Oiso Y, Saito H. Non-genomic mechanisms ofglucocorticoid inhibition of adrenocorticotropin secretion: possible involve-ment of GTP-binding protein. Biochem Biophys Res Commun1997;235:295–299.

[13] McDonald JA, Handelsman DJ, Dilworth P, Conway AJ, McCaughan GW.Hypothalamic–pituitary–adrenal function in end-stage non-alcoholic liverdisease. J Gastroenterol Hepatol 1993;8:247–253.

[14] Tsai MH, Peng YS, Chen YC, et al. Adrenal insufficiency in patients withcirrhosis, severe sepsis and septic shock. Hepatology 2006;43:673–681.

0 vol. 53 j 841–848 847

Research Article

[15] Annane D, Bellissant E, Sebille V, et al. Impaired pressor sensitivity to

noradrenaline in septic shock patients with and without impaired adrenalfunction reserve. Br J Clin Pharmacol 1998;46:589–597.

[16] Bollaert PE, Charpentier C, Levy B, et al. Reversal of late septic shock withsupraphysiologic doses of hydrocortisone. Crit Care Med 1998;26:645–650.

[17] Marik PE. The diagnosis of adrenal insufficiency in the critically ill patient:does it really matter? Crit Care 2006;10:176.

[18] Lentjes EG, Romijn FH. Temperature-dependent cortisol distribution amongthe blood compartments in man. J Clin Endocrinol Metab 1999;84:682–687.

[19] Lentjes EG, Romijn FP, Moolenaar AJ. Isotope effect in the binding of tritiumand 14C-labelled cortisol to corticosteroid-binding-globulin in serum. JSteroid Biochem Mol Biol 1997;60:255–260.

[20] Hamrahian AH, Oseni TS, Arafah BM. Measurements of serum free cortisol incritically ill patients. N Engl J Med 2004;350:1629–1638.

[21] Ho JT, Al-Musalhi H, Chapman MJ, et al. Septic shock and sepsis: acomparison of total and free plasma cortisol levels. J Clin Endocrinol Metab2006;91:105–114.

[22] Vincent RP, Etogo-Asse FE, Dew T, et al. Serum total cortisol and free cortisolindex give different information regarding the hypothalamus–pituitary–adrenal axis reserve in patients with liver impairment. Ann Clin Biochem2009;46:505–507.

[23] Galbois A, Rudler M, Massard J. Assessment of adrenal function incirrhotic patients: salivary cortisol should be preferred. J Hepatol 2010;52:839–845.

[24] Hurel SJ, Thompson CJ, Watson MJ, et al. The short Synacthen and insulinstress tests in the assessment of the hypothalamic–pituitary–adrenal axis.Clin Endocrinol 1996;44:141–146.

[25] Huang W, Kalhorn TF, Baillie M, Shen DD, Thummel KE. Determination offree and total cortisol in plasma and urine by liquid chromatography–tandem mass spectrometry. Ther Drug Monit 2007;29:215–224.

[26] Deutschbein T, Unger N, Mann K, Petersenn S. Diagnosis of secondaryadrenal insufficiency in patients with hypothalamic–pituitary disease:comparison between serum and salivary cortisol during the high-dose shortSynacthen test. Eur J Endocrinol 2009;160:9–16.

848 Journal of Hepatology 201

[27] Lewis JG, Bagley CJ, Elder PA, Bachmann AW, Torpy DJ. Plasma free cortisolfraction reflects levels of functioning corticosteroid-binding globulin. ClinChim Acta 2005;359:189–194.

[28] Vogeser M, Briegel J, Zachoval R. Dialyzable free cortisol after stimulationwith Synacthen. Clin Biochem 2002;35:539–543.

[29] Coolens JL, Van Baelen H, Heyns W. Clinical use of unbound plasma cortisolas calculated from total cortisol and corticosteroid-binding globulin. JSteroid Biochem 1987;26:197–202.

[30] le Roux CW, Sivakumaran S, Alaghband-Zadeh J, et al. Free cortisol index as asurrogate marker for serum free cortisol. Ann Clin Biochem2002;39:406–408.

[31] Bonte HA, van den Hoven RJ, van der Sluijs Veer G, Vermes I. The use of freecortisol index for laboratory assessment of pituitary–adrenal function. ClinChem Lab Med 1999;37:127–132.

[32] Wood P. Salivary steroid assays – research or routine? Ann Clin Biochem2009;46:183–196.

[33] Zietz B, Lock G, Plach B, et al. Dysfunction of the hypothalamic–pituitary–glandular axes and relation to Child-Pugh classification in male patientswith alcoholic and virus-related cirrhosis. Eur J Gastroenterol Hepatol2003;15:495–501.

[34] Clark PM, Neylon I, Raggatt PR, Sheppard MC, Stewart PM. Defining thenormal cortisol response to the short Synacthen test: implications for theinvestigation of hypothalamic–pituitary disorders. Clin Endocrinol (Oxf)1998;49:287–292.

[35] Albillos A, de la Hera A, Gonzalez M, et al. Increased lipopolysaccharidebinding protein in cirrhotic patients with marked immune and hemody-namic derangement. Hepatology 2003;37:208–217.

[36] Wiest R, Cadelina G, Milstien S, et al. Bacterial translocation up-regulatesGTP-cyclohydrolase I in mesenteric vasculature of cirrhotic rats. Hepatology2003;38:1508–1515.

[37] Barnes PJ, Adcock IM. Glucocorticoid resistance in inflammatory diseases.Lancet 2009;373:1905–1917.

[38] Cohen J, Venkatesh B. Assessment of tissue cortisol activity. Crit Care Resusc2009;11:287–289.

0 vol. 53 j 841–848