Leucocyte ratios are biomarkers of mortality in patients with acute decompensation of cirrhosis and acute-on-chronic liver failure

BACKGROUND
In patients with cirrhosis, progression to acute decompensation (AD) and acute-on-chronic liver failure (ACLF) has been associated with poor prognosis. Differential leucocyte ratios might predict mortality in systemic inflammatory conditions.


AIM
To evaluate differential leucocyte ratios as prognostic biomarkers in patients with cirrhosis.


METHODS
Patients with AD and ACLF were recruited from four centres in three countries. Peripheral blood differential leucocytes were measured (three centres using flow cytometry) on hospital admission and at 48 hours. Ratios were correlated to model for end-stage liver disease (MELD), chronic liver failure-sequential organ failure (CLIF-SOFA), suspected/culture-positive bacterial infection and survival.


RESULTS
Nine hundred twenty-six patients (562 (61%) male, median age 55 (25-94) years) were studied. Overall, 350 (37%) did not survive to hospital discharge. Neutrophil-lymphocyte ratio (NLR) and monocyte-lymphocyte ratio (MLR) were elevated in patients with AD and ACLF who died during their hospital stay. On multivariate analysis NLR retained statistical significance independently of CLIF-SOFA or MELD. NLR >30 was associated with an 80% 90-day mortality risk in patients with ACLF but not AD. On sensitivity analysis for subgroups (alcohol-related liver disease and suspected sepsis), NLR and MLR retained statistically robust accuracy for the prediction of mortality. Significant predictive accuracy was only observed in centres using flow cytometry.


CONCLUSION
Leucocyte ratios are simple and robust biomarkers of outcome in ACLF, which are comparable to CLIF-SOFA score but dependent on leucocyte quantification method. NLR and MLR may be used as screening tools for mortality prediction in patients with acutely deteriorating cirrhosis.


| INTRODUC TI ON
Liver cirrhosis is a pressing public health concern due to its rising global incidence and increasing attributable mortality. 1 Acute decompensation (AD) 2 of cirrhosis and subsequent progression to acute-on-chronic liver failure (ACLF) 3 are conditions with a substantial morbidity and mortality and associated healthcare costs. 4 The cause of acute deterioration is often bacterial infection, and the consecutive systemic inflammatory response syndrome (SIRS) and organ failure are associated with poor prognosis. [5][6][7] Recently, scoring systems have been developed that classify acutely decompensated cirrhotic patients by reference to disease severity and predict 28-day survival. 6,8 In analogy to the long-established sequential organ failure assessment (SOFA) score for critical illness, 9 the chronic liver failure (CLIF-SOFA) score is calculated at the patients' admission to hospital. These scores stratify patients with AD and with organ failure (ACLF), hereby predicting 28-day-mortality rates between the CLIF subgroups. 8 The score has been further optimised for patients without and with organ failure, respectively, and named CLIF-C AD score and CLIF-C ACLF score. 2,10 Despite their accuracy in predicting mortality, these composite scores are, however, dependent on complex calculation and online-tools for their assessment. For primary patient contact in both out-patient-and in-patient scenarios, rapid and simpler predictors of poor clinical course and mortality would be preferable for immediate and accurate triage of the patient.
A simultaneous increase in neutrophil-and decline in lymphocyte counts is a physiological response of circulating leucocytes to stress including sepsis. 11 Therefore, a number of simple leucocyte ratio indices have been developed in inflammatory syndromes to capture this response including the neutrophil-lymphocyte ratio (NLR), monocyte-lymphocyte ratio (MLR) and systemic immune-inflammation index (SII = P × N/L; P=platelet count). They have been evaluated in various acute and chronic inflammatory conditions and diverse malignancies. Their use as prognostic marker for patient stratification at hospital admission have been proposed in patients with extrahepatic disorders such as critical illness in general, 12,13 blood stream infection [12][13][14] or sepsis 15 in particular. NLR has also been found prognostic for the prediction of acute kidney injury in patients with sepsis. 13,15,16 Bacterial infection is a common primary event precipitating AD and ACLF. It lengthens hospital stay and is an independent risk factor for early mortality. 5,17 Predicting those at risk of secondary infection (occurring once the initial insult has passed but preventing hospital discharge) would have important clinical benefits in clinical management and discharge planning. 5 NLR has been assessed as a potential marker of hospital acquired infection but not as a predictor of later sepsis. 18 Furthermore the aetiology-specific behaviour of NLR has not been systematically assessed. 19 NLR has also extensively been studied in patients with hepatocellular carcinoma, as a prognostic marker and marker of tumour recurrence after hepatectomy or transplantation, [20][21][22][23][24][25] and may complement alpha-fetoprotein. 26 Recently the prognostic use of the NLR has been evaluated in studies of patients with chronic liver disease without hepatocellular carcinoma. The prognostic use of the NLR has been evaluated prior to 18,[27][28][29] or after transplantation 30 and linked to poor survival. 31 Recent further evidence suggests that using NLR > 9 in patients who are hospitalised with cirrhosis may add to prognostic accuracy at 1 year. However, in patients with ACLF NLR > 9 is ubiquitous 32 and so the question remains as to which levels of NLR or other leucocyte indices are more useful in the different scenarios of AD and ACLF.
It is important to assess whether leucocyte ratios are valid not only in different scenarios of hepatic decompensation and organ failure, but also in different centres where the methodology for leucocyte counting may differ. Most modern clinical haematology laboratories will use flow cytometry methods although others will continue to use automated counters where sensitivity in cytopenic patients is impaired and the risk of mis-assigning immature cell types persists. 33 Differential leucocyte ratios are, therefore, candidates for novel, easy to use biomarkers of inflammation, infection and mortality in patients with AD and ACLF. In this multi-centre study, we sought to assess the admission and sequentially measured NLR, MLR and other leucocyte differentiation ratios in different cohorts of cirrhotic patients with predominantly AD or ACLF, and compare their accuracy in predicting disease trajectory, future infection and mortality to established composite scores.

| Patients-primary derivation cohort
Between January 2009 and December 2016 consecutive admissions with AD or ACLF to the liver intensive therapy unit at King's College Hospital had prospective predefined capture of baseline demographic and clinical data by dedicated auditors as previously described. 34 The worst result/score in the 24-hour period was recorded. These data were collected daily for the total critical care admission period. Detailed intensive care unit (ICU) discharge documents were produced by senior medical staff and were also utilised as a data source.
Patients presenting with acute liver failure, hepatocellular carcinoma, chronic liver disease not consistent with cirrhosis and malignancy were excluded. The presence of cirrhosis was determined from clinical, biochemical, radiological or histopathological results. Patients post-liver transplant or transplanted during their ICU or hospital stay were also excluded due to bias of this outcome.
Readmissions to ICU were also excluded from this analysis; their first ICU admission being the only one analysed regarding physiological parameters and eventual outcome. Elective admissions were also excluded such as planned elective surgery or paracentesis.
ACLF, defined as AD with organ "dysfunction/failure" for the purposes of this study cohort were defined as per CLIF-SOFA definitions. 8 For those cohorts with a higher rate of AD compared with ACLF, then, the CLIF-AD score was used preferentially. Data captured on day 1, 3 and 7 of ICU stay were used for this analysis (see Supplementary methods for data captured and data captured in validation cohorts).
Demographics, full blood count, international normalised ratio, liver and renal function tests, lactate, ammonia, hospital mortality and infection status and clinical variables were collected prospectively by trained audit nurses. The following disease severity scores were calculated: Child-Pugh, 34

| Statistical methods
The primary outcome was hospital mortality (or 90 days whichever came sooner). All continuous data were tested for normality using the Kolmogorov-Smirnov test and expressed as mean (standard deviation) or median (inter-quartile range [IQR]) as appropriate.
Comparison between continuous variables was performed using the Student's t-test or Mann-Whitney U test for two variable compari-

| Derivation cohort-Liver Intensive Therapy Unit, King's college hospital, London
Six hundred and seventeen patients (65% male, median age 55  years) were studied in this cohort recruited consecutively between January 2009 and December 2016. Median MELD was 17 , median APACHE II 21(7-50) and median CLIF-SOFA 10(0-20). All patients were managed in an ICU and 258 (42%) died in hospital. One hundred and eighty-eight patients had AD (31%) and the remainder ACLF. In AD patients the rate of sepsis was 30% on admission and this was higher (50%) in those with ACLF on admission.
Cohort characteristics are summarised in Table 1. NLR (P < 0.001), MLR (P < 0.001) and neutrophil-white blood cell ratio (NWR) (P = 0.020) were higher in nonsurvivors compared to survivors ( Figure 1A). There was no difference in monocyte-white blood cell ratio (MWR) between survivors and nonsurvivors (P > 0.05). Following correction for multiple comparisons NLR MWR did not predict morality in this cohort. In patients with ACLF NLR, NWR and MLR were significantly higher in those who did not survive (P < 0.001 for all comparisons). When comparing NLR with CLIF-SOFA and MELD scores, NLR did not predict mortality better in all patients and in patients with ACLF, respectively. The P value for comparison of the AUROC was significant (P = 0.012).
The pattern of NLR and MLR with outcome was also seen in the subgroups of patients with and without sepsis and with and without alcohol-related liver disease (ARLD; Figure 2).
On Kaplan-Meier survival analysis mortality to 90 days increased with rising admission NLR. An NLR >30 was associated with an 80% mortality (see Figure 3).

| Multivariate analysis
For the purposes of assessing prognostic performance against accepted modalities the CLIF-SOFA score was taken as the gold standard. In the Kings cohort, NLR were independently associated with mortality in both a multivariate analysis using physiological variables (model 1) and a CLIF-SOFA controlled analysis (model 2-see

| D ISCUSS I ON
This is the first multi-centre study to confirm the utility of differential leucocyte count to predict mortality in patients with AD and ACLF. We propose its use primarily to guide stratification as it can be calculated simply in clinic or at the bedside. It can be used as a simple and effective proxy for ACLF in predicting the risk of hospital mortality in the early phases of admission. Our data suggests that NLR is an important screening tool for early mortality due to its ease of calculation. While less accurate than the main clinical scoring systems it also modestly adds prognostic accuracy when used in combination to these, indicating that it may be capturing an immune-based adverse process that SOFA or MELD scores do not reflect.
Our data are important in that we systematically look at AD and ACLF cohorts in different hospitals, with different measurement techniques and where we are just only looking at where the patients received treatment (hospital ward or critical care area). Furthermore, we assess the risk of secondary infection and make important stratifications using novel indices.
In this study, we focussed on the prognostic value of acute deterioration of liver disease (AD/ACLF) while other papers had evaluated NLR in the setting pre-and post-transplantation. 27    as its performance is less open to degradation on sensitivity analysis.
Nevertheless, in two cohorts MLR did not provide further prognostic benefit.
One finding is that the use of these ratios is dependent to some degree on the methodology used to automatically generate cell counts. The Edmonton cohort where reduced prognostic accuracy is reported used alternative methods for cell counting before upgrading to flow cytometry-based methods as in the derivation cohort. In this cohort, the estimation of leucocyte counts was also reported to two significant figures, which may further reduce the accuracy of the subsequent ratios. In addition to this, patients in the Edmonton cohort presented with more advanced disease compared to the other cohorts. Edmonton has a number of other geographical differences in that some patients will come to the centre after stays in other ICUs and so the potential for lead time bias is informative. At present, we would suggest that early, nonhigh grade ACLF, in centres with flow-based cell counters are most likely to benefit from the use of these ratios and that further allowances may be required for their use outside these settings.
The limitations of this study include its retrospective nature.
However, most studies in this area are retrospective in nature. As most other studies are also single centre, they have limited opportunity to assess for measurement bias. The strengths include its multi-centre nature and appreciation of the importance of assessing leucocyte ratios in sub-cohorts with ACLF, ARLD and whether sepsis was a primary cause of decompensation. The result that in one cohort the leucocyte ratios were not of prognostic benefit is important for the liver community to appreciate. The causes of this are likely to be multifactorial but may be due to lead time bias (Edmonton/University of Alberta has a geographically dispersed referral area), antibiotic usage, reason for decompensation and possibly due to technical difference in white cell count automation.
MLR and NLR are useful adjunctive markers of prognosis in patients ACLF. They are simple to calculate and can be used for early identification of high-risk mortality groups.