Professional Documents
Culture Documents
Abstract
Liver Transplantation (LT) is the only proven treatment for paediatric acute liver failure (PALF).
However, over a period of time, spontaneous native liver survival is increasingly reported, making
us wander if we are over-transplanting children with acute liver failure (ALF). An effective
prognostic model for PALF would help direct appropriate organ allocation. Only patients that
would die would undergo liver transplantation (LT), and those who would spontaneously recover,
would avoid unnecessary LT. Deriving and validating such a model for PALF, however,
encompasses numerous challenges. In particular, the heterogeneity of age and aetiology in PALF,
as well as lack of understanding of the natural history of the disease, contributed by the
laboratory variables have been identified, and incorporation into scoring systems have been
attempted. A reliable targeted prognostic model for ALF in Wilson’s Disease, has been established
and, externally validated. The role of physiological, immunological and metabolomic parameters
in prognosis are being investigated. This review discusses the challenges with prognostic modelling
in PALF and describes predictive methods that are currently available and in development for the
future.
Abbreviations; AIH, autoimmune hepatitis; ALFED, ALF early dynamic model; ALF, acute liver failure;
ALT, alanine transaminase; APACHE II, The Acute Physiology and Chronic Health Evaluation II; AST,
aspartate transaminase; AUC, area under the curve; CK 18, total cytokeratin 18; ELSS, extracorporeal
This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process which may lead to
differences between this version and the Version of Record. Please cite this article as an
‘Accepted Article’, doi: 10.1002/lt.24501
liver support system; GALD, Gestational Alloimmune Liver Disease INR, International Normalised
Ratio; KCH, Kings College Hospital; HE, hepatic encephalopathy; HLH, haemophagocytic
lymphohistiocytosis; HSV, herpes simplex virus; LIU, Liver Injury Units; LT, liver transplantation;
MELD, Model for End-stage Liver Disease; NH, Neonatal Haemochromatosis; NIH, The National
Institute of Health; ; NPV, negative predictive value; PALF, paediatric acute liver failure; PELD,
pediatric end stage liver disease; PIM , Paediatric Index of Mortality; PPV, positive predictive value;
PRISM, Paediatric Risk of Mortality; PT, Pro-Thrombin Time; PCM, paracetamol; ROC, receiver
operator curve; SOFA , Sequential Organ Failure Assessment; SBR, serum bilirubin; SNL, survival
native liver; UK, United Kingdom;, US, United States, WD, wilson’s disease; WCC, leukocyte count.
Paediatric acute liver failure (PALF), although rare, is a condition associated with high mortality 1.
Earlier literature on the subject has used adult definitions of acute liver failure (ALF) as well as adult
prognostic criteria to consider liver transplantation (LT). However, with increased experience, it
became clear that PALF differs from adult ALF, in terms of type and diversity of aetiologies and the
later appearance of hepatic encephalopathy (HE). A consensus definition of PALF is “ the acute
onset of liver disease with hepatic-based coagulopathy (Pro-Thrombin Time, PT ≥ 20s; International
Normalised Ratio, INR ≥ 2) not corrected by parenteral vitamin K with or without HE or hepatic-
based coagulopathy (PT 15-19.9s; INR 1.5-1.9) with HE 2. Emergency liver transplantation (LT)
remains the only definitive treatment, with PALF accounting for 10-15% of all paediatric LT1, 3.
Aetiology in PALF varies significantly according to age and worldwide location1-4; identifiable causes
immune dys-regulation. In the developed world, the commonest cause amongst all age groups
remains indeterminate. Clinical management of PALF entails (i) identifying conditions that have
specific medical treatments (ii) ensuring supportive therapy and (iii) determining the likelihood of
either spontaneous native liver recovery or death. Accurate prognostication in PALF would ensure
that only patients that were likely to die would be listed for emergency LT. Prognostication is also
key in appropriately allocating extra-corporeal liver support systems (ELSS) and auxiliary LT to
PALF outcomes
The creation of multi-centre registries of PALF data have led to a much better understanding of PALF
management in the developed world. The National Institute of Health (NIH) sponsored PALF Study
database 2, initiated in 1999, includes data from 24 participating centres from United States (US),
Canada and the United Kingdom (UK). In the pre-transplantation era, Kings College Hospital (KCH),
5
London reported spontaneous recovery only in 28% PALF patients, albeit with limitation of the
data, whilst the majority (72%) died. Post-LT, the PALF Study database demonstrated a much higher
spontaneous recovery of 56% (1999-2004). More recently, the Paediatric Health Information System
(PHIS), from the US 6, has reported an increasing spontaneous survival of 73.4% (2008-2012) (Figure
1). Despite these encouraging results, optimal prognostic criteria to decide who will die without LT is
lacking 7. The PALF Study database 2 highlighted 10% of all patients listed for emergency LT were de-
listed due to death or spontaneous recovery in 20 and 80% of cases respectively. The increased
availability of living donor LT could further complicate the natural history of PALF by unnecessary
transplantation where spontaneous liver recovery could have occurred. Auxiliary LT, with native liver
regeneration shown in up to 70% of select patient groups, provides insight into the potential for
Figure 1 Comparison of survival native liver (SNL), LT and death rates (%) in PALF in the pre-LT and
post-LT eras.
A prognostic model in medicine is designed to produce indices to enable the estimation of the risk of
future events in individual patients/groups and to risk stratify these patients 9. Figure 2 highlights the
principles behind ideal derivation, assessment and validation of a prognostic model. Prior to deriving
a predictive model, its clinical relevance should be clear, aimed at aiding clinicians in decision-
making. Prognostication in PALF is essential for clinical practice, to accurately differentiate patients
that are likely to survive or die, in order to allocate emergency LT appropriately. The ideal model-
derivation population should be large, representative of the diseased cohort and entail a reasonable
proportion of the outcome measure. The outcome measure(s) should be clearly defined. Ideally,
PALF prognosis should comprise two outcomes; survival and death. However, artificially intervening
with LT skews the data, as the true outcome for patients undergoing LT, is unknown. Variables in a
model should be easy to measure and utilise, ideally objective and not too extensive. Variables for
adult ALF and PALF models are usually derived from uni- and/or multi-variate analyses on historical
patient cohorts. To test if the variables/model display acceptable predictive reliability, the actual
observed and predicted outcomes need to be plotted and compared, in a process called ‘calibration’.
between patients who do or do not experience the event. In order to determine discriminatory
power, the sensitivity and specificity for the model at different threshold settings must be calculated
10
and then plotted on a receiver-operator curve (ROC) . The area under the curve (AUC) or ‘c’-
statistic summarises how good the model is at discriminating between outcomes. AUC or ‘c’-statistic
of ‘1.0’ would be ideal, representing 100% discrimination, however, in practice, AUC/c-statistic > 0.8
essential, as reduced sensitivity (low positive predictive value; PPV) could lead to the failure to list a
patient for LT who would have subsequently died, but reduced specificity (low negative predictive
value; NPV) carries a risk of unnecessary LT in a patient who was likely to recover spontaneously.
Once a model demonstrates good calibration and discrimination, it should be tested outside of the
11
derivation dataset, in a process called ‘validation’ . Validation can take place internally, on a
validation (i.e different centre) is preferred, in order to test transportability of the model.
Figure 2 Diagram to illustrate the principles of the ideal prognostic model development and
validation. The accuracy of the model developed is summarised by (a) calibration and (b)
discrimination. A well calibrated model shows good correlation between observed and predicted
mortality with a reasonable ‘goodness to fit’. Sensitivity and specificity at different threshold
settings are plotted on a receiver-operator curve (ROC) and discrimination is summarised by the
area under the curve (AUC). AUC as close to 1.0 shows optimal sensitivity and specificity (0.8 is
Adult ALF, given the higher patient population, fewer aetiologies and strict definition, consists of
12, 13
more literature on prognostic modelling as compared to children . The two most recognised
models (Figure 3) were derived from logistic regression analysis of historical patient cohorts in the
pre-transplantation era 14, 15. The Kings College Hospital (KCH) criteria, first described in 1989, is the
standard upon which other models have been judged, and is currently the basis for the UK NHS
Blood and Transplant emergency LT selection. It incorporates paracetamol (PCM) and non-PCM
aetiologies as well as age, INR, HE, serum bilirubin (SBR), renal function and acidosis. Although a
small number of paediatric patients (n=29) were included in this cohort, details are limited, and
16
infants were excluded. Numerous studies, including a recent meta-analysis , have consistently
demonstrated good specificity (PCM 95%; non-PCM 81%) but poor sensitivity (PCM 58%; non-PCM
68%), hence, not adequately discriminating the patients that would die without LT. The Clichy-
14
Villejuif criteria , was similarly devised in the late 1980’s from an adult cohort of viral hepatitis
patients, and incorporates age, HE and factor V concentration. It is the criteria by which patients are
listed for emergency LT in Northern Europe. However, similar to KCH criteria, sensitivity is low.
Assessment of other variables to supplement existing criteria, e.g. lactate, continues 17. Despite its
initial development for cirrhotic patients, Model for End-stage Liver Disease (MELD) has been
18, 19
evaluated in ALF . McPhail et al 20, in their meta-analysis, revealed higher sensitivity of MELD
21
over KCH criteria, but at the cost of reduced specificity. Kumar et al recently developed, and
validated, the ALF early dynamic (ALFED) model. This model was based on early changes of four
variables (arterial ammonia, INR, HE and serum bilirubin; SBR), over 3 days, as a prediction of
mortality. Good calibration and discrimination (AUC 0.91) were revealed. Although we can learn
from the principles of adult ALF modelling, extrapolation to PALF is challenging, due to fundamental
Figure 3 Adult ALF prognostic models; King’s College Hospital Criteria and Clichy-Villejuif criteria.
The pursuit for the ‘ideal’ prognostic model in PALF is undoubtedly crucial, but complex. Figure 4
Presentation and definition. Definitions for adult ALF and PALF incorporate different concepts,
adult ALF, hence, incorporated into both KCH and Clichy models. However, in PALF, HE is difficult to
children and rapidly progressive in some cases1, 22, 23. Another important discrepancy between adult
and paediatric definitions, is that the ‘absence of pre-existing liver disease’ is required in adult ALF,
but in PALF, children can present with an acute de-compensation of a previously unknown
Age and Aetiology. PALF comprises widespread aetiologies and age groups, which carry different
prognostic outcomes. Figure 5a shows overall PALF aetiology, as well as categorisation by age in the
1, 2, 24
developed world. Several paediatric studies have highlighted the contribution of aetiology to
prognosis (Figure 5b), with an increased likelihood of spontaneous recovery in conditions such as,
PCM-toxicity, hepatitis A and ischemic hepatitis and a reduced likelihood of spontaneous recovery in
indeterminate, Wilson’s Disease (WD), idiosyncratic drug reactions, autoimmune hepatitis (AIH), and
and herpes simplex virus; HSV). The PALF Study database comprises a wide range of aetiologies, but
subgroup analysis demonstrates a 94% spontaneous recovery of PCM- induced PALF. PCM toxicity
accounts for over 50% of adult ALF in the UK 25; O’Grady et al15. (1989) incorporated PCM toxicity
separately within the KCH criteria. This raises the question as to whether aetiology specific scores
would be more appropriate than an overall PALF score. A successful prognostic model for WD,
which is discussed in a later section, further supports this view. Aetiology specific scores would need
to incorporate the variability of a single condition, the variable impact of specific directed therapy in
some disease processes (e.g. AIH, HSV, Neonatal Haemochromatosis; NH) as well as the potential for
score modification if new treatments become available (e.g possible immunosuppression in cases of
indeterminate PALF)26
Young age is associated with poorer PALF outcomes, partly due to comprising higher risk aetiologies
with multi-systemic involvement, but also due to the specific technical challenges posed by younger
infants27. Figure 5a highlights age-dependent aetiologies; NH, HLH and HSV most common in
neonates, metabolic in infancy/early childhood and drugs, WD and AIH in older children 1, 2, 4, 22. The
term, ‘NH’, is being expressed rather than ‘GALD” (Gestational Allo-immune Liver Disease), as only
the phenotypic description of this disease can be inferred and not the allo-immune aetiological
component. The largest aetiological group for PALF across all ages, indeterminate, accounting for
1, 2, 4, 6, 24
over 50% cases in UK . The PALF Study database revealed an increased risk of death/LT
amongst <3 versus ≥ 3 years old groups. Durand et al (2001), in their single-centre retrospective
22
analysis of infants with ALF (n=80, 1986-2000) demonstrated death/LT in 47% cases, which was
28
higher than in older children. Sundaram et al , in particular, highlighted the challenges in very
young infants (< 3 months) from the PALF Study database. They identified 148 young infants (median
age 18 days, range 11-37 days) and compared outcomes to the older group (≥ 3 months).
Spontaneous survival and mortality without LT were 59.5 and 24% respectively in young infants
which was significantly worse than figures for older children in the PALF cohort (79 and 10.5%
respectively). Indeterminate (38%), NH (13.6%) and HSV (12.8%) were the commonest categories in
this young cohort. The same group showed that only 40% of young infants that were listed for LT,
actually received an organ, compared to 66% in the older group, emphasizing potential technical
issues, such as, lack of suitable size-matched organs. Hence, similar to aetiology, age plays an
essential role in prognostication. The incorporation of age parameters, in particular young age in
prognostic modelling, is vital for accurate scoring. The ‘ideal prognostic model’ would need to be
Figure 5 (a) Overall and age categorised PALF aetiology in the developed world (b) Aetiology
dependent PALF prognosis. Data for overall aetiology adapted from Dhawan et al 4
PALF natural history. A significant limitation in developing the ‘ideal’ model for both adult ALF and
PALF, is the restricted insight into the natural history of the disease, in light of artificially intervening
with LT. Data from the pre-LT era 5 and from countries where LT is not available 29, allows glimpses
into PALF natural history, however, differences in medical care and aetiologies within both these
cohorts make extrapolation to current practice in developed countries, challenging. PALF prognostic
models often incorporate LT and death as a combined measure of poor outcome, but as some
patients undergoing LT would have spontaneously recovered, models comparing native liver survival
to death without LT, are more appropriate. Ideally, parameters of liver function which can determine
the number of remaining parenchymal cells, the level of cell dysfunction and their capacity of
Additional model limitations. Decisions regarding optimal threshold cut-offs for a PALF model are
challenging, as sensitivity and specificity need to be balanced appropriately. Most prognostic models
are derived and validated retrospectively, hence, missing information and different time-points for
data collection (admission versus study enrolment data; early versus late transfer to tertiary
centre30), can be confounders. Intensive care procedures, such as ELSS and use of blood products
can affect laboratory and physiological parameters used in prognostic models, however, integrating
these measures into models is difficult. Worldwide, there are differences in organ availability (lack of
donors, less experience with living related techniques). In countries with reduced organ availability
and increased waiting times, although PALF modelling would still guide management, there may be
more a role for proactive listing for LT as compared to countries with greater organ availability. In
addition, PALF aetiologies and management vary across the world, hence, external validation of PALF
models across different geographic locations need to be interpreted cautiously. Lastly, the
improvement of PALF outcomes over time, with enhanced medical care, suggests the need to
Despite the above challenges, numerous groups have investigated the role of various prognostic
variables in PALF (Table 1). A retrospective analysis of 31 children with PALF (predominantly viral
aetiology) 5 at KCH in the pre-LT era demonstrated a significantly higher chance of death if the pro-
thrombin time exceeded 90 seconds. Subsequent studies in the post-LT era demonstrated that PT
and INR were significant predictors of death/LT or death alone 24, 27, 31. At KCH, an INR > 4 was shown
to have the best sensitivity (86%) and specificity (73%.) and AUC-ROC 0.79. The PALF study database
revealed an INR ≥ 2.5 to be a predictor of death. Hence, different thresholds may be applicable to
different populations. UK organ allocation is based on INR > 4. Other laboratory variables, namely
SBR, alanine transaminase (ALT), aspartate transaminase (AST), leukocyte count and factor V
concentration, have been studied, in order to supplement INR and improve its discriminative power.
SBR was significantly associated with death/LT and death alone in the PALF study database on
multivariate analysis, along with several other studies. Dhawan et al 27, in a multi-variate followed by
uni-variate analysis, and using ROC methodology, identified a cut-off value of total SBR 235
µmmol/l, with a a sensitivity and specificity of 85 and 65% (AUC 0.76). Leukocyte count (WCC) ≥ 9
x109/L and age < 2 years were identified as additional risk factors. Figure 6 shows the increasing
discriminatory value of combining all 4 parameters (INR, SBR, WCC, age), predicting near 100 %
mortality without LT. Low transaminases 23, 24 and factor V 22, 32, 33 at admission have been associated
with poor outcome. Factor V is used in few European countries, for PALF organ allocation. The
John Wiley10
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Page 11 of 27 Liver Transplantation
severity of HE, if present, is a poor prognostic marker, with grade IV HE nearly always associated
2, 24
with fatality . Admission electroencephalogram (EEG) abnormalities in conjunction with clinical
HE grading could add further prognostic advantage34. Currently, INR/PT together with other
laboratory variables, aetiology and young age is the optimal way to assess PALF outcomes. There are
no paediatric studies, which have evaluated the role of type of PALF presentation (e.g acute versus
Figure 6 Four PALF prognostic variables associated with cumulative risk of mortality. Adapted
from Dhawan et al 27
Incorporating PALF prognostic variables into a scoring system would aid clinical staff in making
decisions regarding LT. However, this is proving to be a highly complex task. Several groups have
attempted to apply adult criteria and chronic liver disease scoring systems in PALF. Newer PALF
models have been devised, with the disease specific score, the revised Wilson’s Index, being in
widespread use.
KCH Criteria. The parameters incorporated into KCH criteria are known to be prognostic markers for
death in PALF. KCH criteria was applied to a large cohort of patients (n=522) with non-PCM induced
PALF 35, who did not undergo LT, from the PALF Study database. This cohort comprised 36.8% infants
and 43.1% overall indeterminate aetiology. Of the patients that fulfilled KCH criteria, 66.9% survived,
and of those that did not, 11.8% died. Hence, although specificity for mortality was similar to adult
ALF, sensitivity was lower, at 61%. Missing data, in particular for HE, was a significant limitation to
this study. Several other studies have applied KCH criteria to PALF groups, however, results are
33
difficult to interpret due to small or non-heterogenous cohorts. Ciocca et al demonstrated
John Wiley11
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Liver Transplantation Page 12 of 27
sensitivity and specificity of 72% and 96% respectively, for mortality without LT in their large cohort
(n=210), but infants were excluded and aetiology was dominated by Hepatitis A.
MELD/PELD. The Pediatric End-stage Liver Disease (PELD) score, was derived in 2000, from a
multicentre North American database, in order to aid allocation and prioritisation of LT to children
36
(up to 12 years old) and infants listed with end stage liver disease . It incorporates SBR, INR,
albumin and age and growth failure. It has been validated and approved for use in organ allocation
in many countries outside of USA. After 12 years of age, the MELD scoring system has been
validated. PELD-MELD score at different time-points (admission, PALF diagnosis, peak) was applied
to a small cohort of PALF patients 37 at a single centre in Australia (n=54, median age 17 months, 1
day-15.6 years). Scores were significantly different between spontaneous survivors and non-
transplant death at diagnosis and peak values, but not at admission. Construction of AUC-ROC curves
demonstrated that PELD-MELD score cut-offs >27 and >42 at PALF diagnosis led to sensitivity of 76
and 66% and specificity of 60 and 92% respectively. Another recent study 38, in a small PALF cohort,
demonstrated good sensitivity (86%) and specificity (81%) for poor outcome with a cut-off value of
33. However, missing data, increased prevalence of hepatitis A (43%) and AIH (17.5%) and combined
Liver Injury Units (LIU). Single centre retrospective analysis in a cohort of PALF patients (n=81)
revealed that peak bilirubin, INR (or PT) and ammonia were significantly associated with death/LT 39.
A score called ‘Liver Injury Units’ (LIU) was developed by weighting the peak values of these
parameters by coefficients obtained from the Cox regression. The score derived from this formula
was stratified into risk categories for likelihood of death or LT (low risk ≤ 85; moderate risk 85-138;
high risk > 139). High specificity and sensitivity of the LIU score for predicting death/LT was
40
demonstrated by an AUC-ROC 90.5%. Lu et al subsequently validated this score internally on
retrospective analysis of a subsequent group of children (n=53). The transportability of this score to
a larger, more heterogenous population (n=461), was assessed by application to the PALF Study
John Wiley12
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Page 13 of 27 Liver Transplantation
41
database . In contrast to the preceding studies, LT and death alone were assessed as separate
outcome measure, in addition to the combined death/LT parameter. AUC-ROC values for LIU score
prediction of death/LT, LT alone or death alone was 0.81, 0.84 and 0.76 respectively. Although this
scoring model contains measurable, objective parameters and demonstrates good specificity and
sensitivity for predicting death/LT in the derivation, internal and external validation cohorts, several
limitations exist. The model is only accurate at predicting death or LT, not death alone, which would
be a more reliable outcome measure. Also the model was derived for peak laboratory values,
irrespective of day (i.e. no specific time-point), and score values were divided into low/moderate
and high-risk categories, which would be challenging to apply on a practical, day-to-day level.
41
Admission LIU scores were not predictive of death/LT. Lu et al sought to use the score in a
dynamic way, by calculating the highest daily LIU score within the 21 days of enrolment. Good
discrimination for death/LT (AUR-ROC 0.81) was revealed on using the highest values. Although a
dynamic role for LIU score is suggested, the limitations of using enrolment rather than admission
data, must be considered. In addition, neither age, nor aetiology is incorporated into the model, and
in fact, stratification by age revealed that the LIU score was not reliable in early infancy. To our
KCH revised Wilsons index. PALF, is a heterogenous condition, with aetiology-specific therapies and
outcomes. Hence, aetiology- specific prognostic models may be more appropriate than an overall
PALF model. Wilson’s Disease (WD) is autosomal recessive, predominantly presenting in childhood
metabolism is a result of a mutation in the ATP7B gene located on the long arm of chromosome 13
42
. In children, WD usually manifests after 3 years of age, with either an incidental finding of
abnormal liver function, chronic liver disease or rarely, as PALF. The rarity of WD associated PALF is
illustrated in the PALF Study database, where only 3% of PALF was secondary to WD 2. WD,
2, 42
presenting with acute decompensation, and associated with HE, has a near-100% mortality .
John Wiley13
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Liver Transplantation Page 14 of 27
Likewise, some patients with acute decompensation, but without encephalopathy, succumb to LT.
The challenge over the past 3 decades, had been to understand who will respond to copper
chelation therapy, and to calculate the risk of mortality in WD PALF not associated with HE, in order
to optimize organ allocation. Nazer et al43, identified SBR, AST and PT as prognostic parameters in
WD ALF, in a small cohort of both adult and paediatric patients at KCH. Dhawan et al 42, at the same
centre, improved the prognostic power of this score, by performing multivariate and ROC curve
Leukocyte count (WCC) was identified as an additional prognostic marker. Table 2. demonstrates the
incorporation of these parameters into a ‘revised Wilson’s Index’. At a cut-off value ≥ 11, sensitivity
and specificity of 93 and 97% for mortality prediction was achieved, respectively. The revised
Wilson’s Index has been prospectively validated and is widely used in children, including parts of the
It is well established, in both adults and children, that ALF is a complex multi-organ illness, and that
mortality increases with severe sepsis and multi-organ failure 45. In an attempt to improve current
prognostic scores, several groups have studied the efficacy of established intensive care unit (ICU)
predictive models, as a means to grade adult ALF and PALF severity. The Acute Physiology and
Chronic Health Evaluation II (APACHE II) and the Sequential Organ Failure Assessment (SOFA) are
46, 47
validated adult ICU-specific prognostic scores developed from general ICU populations . Both
scores have been validated in patients with cirrhosis 48. Mixed results are available when applying
49
these models to ALF. Cholangitas et al applied APACHE II and SOFA to a single-centre large cohort
(n=125) of PCM-induced ALF. Discrimination between survival and LT/death was better in SOFA
(AUC-ROC 0.79) than KCH criteria, although KCH criteria had higher specificity. Similarly, in
paediatrics, a wide range of general ICU-scoring systems exist; Paediatric Risk of Mortality (PRISM)
and Paediatric Index of Mortality (PIM) have been widely established and validated in conditions
John Wiley14
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Page 15 of 27 Liver Transplantation
such as sepsis, meningococcal disease, cardio- and respiratory failure 50. PRISM, first published by
Pollack et al 50 in 1987, is a 14-point score, based on clinical and laboratory parameters, including 23
ranges, and accounting for infants and children separately. Several attempts at applying this score to
51
PALF, have yielded varied results. Admission PRISM scores at a French centre, for a large
heterogenous group of PALF (n=109), did discriminate between survivors and death/LT groups, but
observed and PRISM score based expected mortality demonstrated poor calibration. Matthews et al
52
applied a revised version of the PIM score, PIM2, to assess surival outcomes in PALF, utilising the
PALF Study Database (n=103) cohort. The AUC-ROC for mortality in the non-transplanted patients
was only 0.712, which would not be sensitive or specific enough to minimize the risk of unnecessary
transplantation. Currently, ICU-scoring systems, do not confer extra advantage to PALF prognostic
models, but in conjunction with established PALF criteria, physiological and inflammatory
PALF comprises immune activation and inactivation, apoptosis and liver regeneration, which can all
variably affect its outcome. Biomarkers incorporating these processes are being researched for their
role in prognosis. In adult ALF, IL6 and IL853 and soluble CD154 have been associated with
poorer prognosis of PALF. Biomarkers of regeneration, detectable at the time of presentation, might
help determine the likelihood of spontaneous recovery. Specific amino acids and elevated levels of
alpha-fetoprotein, have been associated with native liver survival in PALF, and, studies suggest there
56-58
may be a role for growth factors in prognosis . Conversely, cell death markers have been
investigated as potential prognostic variables for poor outcome. The US ALF Study group59 reported
that the addition of the apoptosis marker M30 of the CK18 molecule improved sensitivity of the KCH
criteria to 86% but specificity was reduced to 69%. In paediatrics, an increased level of M30 epitope
John Wiley15
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Liver Transplantation Page 16 of 27
Conclusions
Early identification and diagnosis in PALF is paramount, with rapid initiation of disease specific, and
overall supportive therapy. Dynamic prognostic scoring systems for PALF are important for
emergency LT allocation, but must incorporate optimal accuracy before use. To date, no optimal
PALF prognostic model exists, therefore, organ allocation should be based on clinical and
biochemical assessment, adjusted for age and aetiology, as well as taking into consideration any
contraindications for LT. Neither adult nor chronic liver disease-derived scores are suitable for PALF.
The revised Wilson’s Index can be reliably utilised in WD PALF. We suggest that there is an urgent
need for further aetiology and age specific scoring systems, with potential inclusion of physiological,
References
1. Devictor D, Tissieres P, Afanetti M, Debray D. Acute liver failure in children. Clin Res Hepatol
Gastroenterol. 2011;35(6-7):430-7.
2. Squires RH, Jr., Shneider BL, Bucuvalas J, Alonso E, Sokol RJ, Narkewicz MR, et al. Acute liver
failure in children: the first 348 patients in the pediatric acute liver failure study group. J Pediatr.
2006;148(5):652-8.
3. Dhawan A. Acute liver failure in children and adolescents. Clin Res Hepatol Gastroenterol.
2012;36(3):278-83.
4. Dhawan A. Etiology and prognosis of acute liver failure in children. Liver Transpl. 2008;14
Suppl 2:S80-4.
5. Psacharopoulos HT, Mowat AP, Davies M, Portmann B, Silk DB, Williams R. Fulminant
hepatic failure in childhood: an analysis of 31 cases. Arch Dis Child. 1980;55(4):252-8.
6. Kulkarni S, Perez C, Pichardo C, Castillo L, Gagnon M, Beck-Sague C, et al. Use of Pediatric
Health Information System database to study the trends in the incidence, management, etiology,
and outcomes due to pediatric acute liver failure in the United States from 2008 to 2013. Pediatr
Transplant. 2015;19(8):888-95.
7. Shanmugam NP, Dhawan A. Selection criteria for liver transplantation in paediatric acute
liver failure: the saga continues. Pediatr Transplant. 2011;15(1):5-6.
8. Faraj W, Dar F, Bartlett A, Melendez HV, Marangoni G, Mukherji D, et al. Auxiliary liver
transplantation for acute liver failure in children. Ann Surg. 2010;251(2):351-6.
9. Moons KG, Royston P, Vergouwe Y, Grobbee DE, Altman DG. Prognosis and prognostic
research: what, why, and how? BMJ. 2009;338:b375.
10. Zweig MH, Campbell G. Receiver-operating characteristic (ROC) plots: a fundamental
evaluation tool in clinical medicine. Clin Chem. 1993;39(4):561-77.
John Wiley16
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Page 17 of 27 Liver Transplantation
11. Altman DG, Vergouwe Y, Royston P, Moons KG. Prognosis and prognostic research:
validating a prognostic model. BMJ. 2009;338:b605.
12. Bernal W, Auzinger G, Dhawan A, Wendon J. Acute liver failure. Lancet. 2010;376(9736):190-
201.
13. Bernal W, Lee WM, Wendon J, Larsen FS, Williams R. Acute liver failure: A curable disease by
2024? J Hepatol. 2015;62(1 Suppl):S112-20.
14. Bernuau J, Goudeau A, Poynard T, Dubois F, Lesage G, Yvonnet B, et al. Multivariate analysis
of prognostic factors in fulminant hepatitis B. Hepatology. 1986;6(4):648-51.
15. O'Grady JG, Alexander GJ, Hayllar KM, Williams R. Early indicators of prognosis in fulminant
hepatic failure. Gastroenterology. 1989;97(2):439-45.
16. McPhail MJ, Wendon JA, Bernal W. Meta-analysis of performance of Kings's College Hospital
Criteria in prediction of outcome in non-paracetamol-induced acute liver failure. J Hepatol.
2010;53(3):492-9.
17. Bernal W. Lactate is important in determining prognosis in acute liver failure. J Hepatol.
2010;53(1):209-10.
18. Schmidt LE, Larsen FS. MELD score as a predictor of liver failure and death in patients with
acetaminophen-induced liver injury. Hepatology. 2007;45(3):789-96.
19. Kremers WK, van IM, Kim WR, Freeman RB, Harper AM, Kamath PS, et al. MELD score as a
predictor of pretransplant and posttransplant survival in OPTN/UNOS status 1 patients. Hepatology.
2004;39(3):764-9.
20. McPhail MJ, Farne H, Senvar N, Wendon JA, Bernal W. Ability of King's College Criteria and
Model for End-Stage Liver Disease Scores to Predict Mortality of Patients With Acute Liver Failure: A
Meta-Analysis. Clin Gastroenterol Hepatol. 2015.
21. Kumar R, Shalimar, Sharma H, Goyal R, Kumar A, Khanal S, et al. Prospective derivation and
validation of early dynamic model for predicting outcome in patients with acute liver failure. Gut.
2012;61(7):1068-75.
22. Durand P, Debray D, Mandel R, Baujard C, Branchereau S, Gauthier F, et al. Acute liver failure
in infancy: a 14-year experience of a pediatric liver transplantation center. J Pediatr.
2001;139(6):871-6.
23. Rivera-Penera T, Moreno J, Skaff C, McDiarmid S, Vargas J, Ament ME. Delayed
encephalopathy in fulminant hepatic failure in the pediatric population and the role of liver
transplantation. J Pediatr Gastroenterol Nutr. 1997;24(2):128-34.
24. Lee WS, McKiernan P, Kelly DA. Etiology, outcome and prognostic indicators of childhood
fulminant hepatic failure in the United kingdom. J Pediatr Gastroenterol Nutr. 2005;40(5):575-81.
25. Bernal W, Wendon J. Acute liver failure. N Engl J Med. 2014;370(12):1170-1.
26. Maggiore G, Socie G, Sciveres M, Roque-Afonso AM, Nastasio S, Johanet C, et al.
Seronegative autoimmune hepatitis in children: Spectrum of disorders. Dig Liver Dis. 2016.
27. Dhawan A, Cheeseman P, Mieli-Vergani G. Approaches to acute liver failure in children.
Pediatr Transplant. 2004;8(6):584-8.
28. Sundaram SS, Alonso EM, Narkewicz MR, Zhang S, Squires RH, Pediatric Acute Liver Failure
Study G. Characterization and outcomes of young infants with acute liver failure. J Pediatr.
2011;159(5):813-8 e1.
29. Poddar B, Saigal S, Kumar A, Singh RK, Azim A, Gurjar M, et al. Factors associated with
outcome in acute liver failure in an intensive care unit. Indian J Gastroenterol. 2013;32(3):172-8.
30. Sturm E, Lexmond WS, Verkade HJ. Pediatric acute liver failure: variations in referral timing
are associated with disease subtypes. Eur J Pediatr. 2015;174(2):169-75.
31. Bhaduri BR, Mieli-Vergani G. Fulminant hepatic failure: pediatric aspects. Semin Liver Dis.
1996;16(4):349-55.
32. Devictor D, Desplanques L, Debray D, Ozier Y, Dubousset AM, Valayer J, et al. Emergency
liver transplantation for fulminant liver failure in infants and children. Hepatology. 1992;16(5):1156-
62.
John Wiley17
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Liver Transplantation Page 18 of 27
John Wiley18
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Page 19 of 27 Liver Transplantation
53. Ilonen I, Koivusalo AM, Repo H, Hockerstedt K, Isoniemi H. Cytokine profiles in acute liver
failure treated with albumin dialysis. Artif Organs. 2008;32(1):52-60.
54. Zheng YB, Gao ZL, Zhong F, Huang YS, Peng L, Lin BL, et al. Predictive value of serum-soluble
CD154 in fulminant hepatic failure. J Int Med Res. 2008;36(4):728-33.
55. Bucuvalas J, Filipovich L, Yazigi N, Narkewicz MR, Ng V, Belle SH, et al. Immunophenotype
predicts outcome in pediatric acute liver failure. J Pediatr Gastroenterol Nutr. 2013;56(3):311-5.
56. Rudnick DA, Dietzen DJ, Turmelle YP, Shepherd R, Zhang S, Belle SH, et al. Serum alpha-NH-
butyric acid may predict spontaneous survival in pediatric acute liver failure. Pediatr Transplant.
2009;13(2):223-30.
57. Aw MM, Mitry RR, Hughes RD, Dhawan A. Serum hepatocyte growth factor and vascular
endothelial growth factor in children with acute liver failure. J Pediatr Gastroenterol Nutr.
2007;44(2):224-7.
58. Schmidt LE, Dalhoff K. Alpha-fetoprotein is a predictor of outcome in acetaminophen-
induced liver injury. Hepatology. 2005;41(1):26-31.
59. Larson AM, Polson J, Fontana RJ, Davern TJ, Lalani E, Hynan LS, et al. Acetaminophen-
induced acute liver failure: results of a United States multicenter, prospective study. Hepatology.
2005;42(6):1364-72.
60. Mustafa AM, R.; Hussain, M.; Dhawan, A. Mechanisms of hepatocellular death in acute liver
failure of indeterminate aetiology in children. [Abstract]
. Journal of Paediatric Gastroenterology and Nutrition: King's College London, UK; 2009.
John Wiley19
& Sons, Inc.
This article is protected by copyright. All rights reserved.
Liver Transplantation Page 20 of 27
Pre-LT Post-LT
80
72 73.4
70
60 58 56
%
50 SNL
40 Death
31
30 28 28 LT
22
20 15 13
10 4.6
0
1969-1977 1985-1993 1999-2004 2008-2012
Western Hemisphere
John Wiley & Sons, Inc.
This article is protected by copyright. All rights reserved.
Page 21 of 27 Liver Transplantation
Model Development
Relevant
will model aid medical decisions?
Validation
Population
Internal
The ideal large, represent diseased cohort
40
1
35
observed mortality %
30
ROC
True
25 +ve
20 rate
15 AUC 0.8
10 Goodness to fit
5
0 0
0 10 20 30 40 50 John Wiley & Sons, Inc.
Predicted mortality % This article is protected by copyright. All rights
0 reserved.
False +ve rate 1
Liver Transplantation Page 22 of 27
Clichy-Villejuif Criteria
Presence of HE
AND
Factor V level;
< 20% of normal in patients < 30 years old
OR
< 30% of normal in patients > 30 years old
John Wiley & Sons, Inc.
This article is protected by copyright. All rights reserved.
Page 23 of 27 Liver Transplantation
Age
Presentation
Retrospective validation Aetiology Incorporation bias
Missing data
Incorporation of
Calibration and ICU/ELSS, blood
discrimination cut-offs product interventions
Challenges
Outcome?
In Timepoint?(i)Admission
(i) LT/Death PALF (ii)Peak
(ii) Death alone (iii) multi-centre study
prognosis enrollment
NH Infection
Drugs
8% Indeterminate
Likely to recover
PCM
(b) Ischemic
spontaneously
Hepatitis A
Indeterminate
Idiosyncratic drug reaction Unlikely to survive
Autoimmune Hepatitis without LT
Wilson’s Disease
Mitochondrial diseases
Haemophagocytic Lymphohistiocytosis
Herpes Simplex Virus John Wiley & Sons, Inc.
Unlikely to survive or
benefit from LT
This article is protected by copyright. All rights reserved.
Page 25 of 27 Liver Transplantation
Bhadhuri et al abstract Y INR > 4 peak • 16.6% LT-free survival in INR > 4
1996, UK (31)
Dhawan et al. 44 Mixed Y INR ≥ 4 peak Age < 2 66 34 - • All independent prognostic
2004, UK (27) SBR ≥ 235 μmol/l HE or no HE parameters for death or survival
WCC ≥ 9 x 109/l HE I/II vs III/IV • Presence of HE not significant
• HE I-II and III-IV mortality risk 44%
and 78% respectively (p=0.002)
Squires et al. 348 Mixed Y INR ≥ 2.55 Admission Gender 56 13 31 • INR, SBR and HE independent
2006, 49% SBR ≥ 5mg/dL and peak HE or no HE prognostic factors for death/LT
US/Canada/UK indeterminate HE III and IV • HE III 33% and HE IV 22% survival
(2) 14% PCM • F>M
10% metabolic
Ciocca et al 210 Mixed Y INR > 4, PT < 22% peak HE III/IV 28 29 43 • Univariate analysis; FV, PT/SBR/HE
2008, Argentina 61% HAV SBR > 17 mg/dL III-IV significantly different for all 3
(2 centres) (33) 32% FV < 30 groups
indeterminate • Multivariate analysis; SBR/INR/HE
III-IV independent predictors for
death/LT and SBR/PT/HE III-IV for
death
Table 1 Paediatric studies identifying prognostic variables for PALF. Outcomes 1. Survival with native liver (SNL) 2. Death without LT 3.
LT