Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
OF HEPATOLOGY
Introduction 1
Gastroenterology Unit, Ospedale
Cirrhosis is typically classified as compensated or not capture clinical state transitions following V. Cervello, Via Trabucco 180,
Palermo, Italy
decompensated, based on the absence or presence the occurrence of a competing event, before the 2
Medical Statistics Unit,
(or previous history) of variceal bleeding, ascites, event of interest. For example, decompensation Università di Milano, Milano, Italy
jaundice or encephalopathy.1–3 The significantly
3
before death is not captured by a survival curve Radiology Department,
Università di Palermo, Palermo,
longer survival, usually symptomless, and better of compensated cirrhosis. In this situation, the
Italy
quality of life experienced by patients with com- cumulative incidence function (CIF), based on the 4
Dipartimento di Medicina e
pensated cirrhosis compared to those with decom- Aalen-Johansen estimator allows a more realistic Chirurgia Università di
pensated cirrhosis, has brought about the concept description of the disease course, by providing Milano-Bicocca, Milano, Italy
that compensated and decompensated cirrhosis the incidence of the competing events.22–24
are distinct clinical states of the disease.4,5 Further Herein, the clinical course, competing risks and Key point
disease states have been identified according to clinical states of cirrhosis will be reviewed and
The clinical course of cir-
the presence of oesophageal varices and to the examples of possible multistate models of the dis-
rhosis encompasses several
presence of only one or more disease complica- ease will be provided. disease states which
tions.4,6–10 require multistate models
Development of gastro-oesophageal varices and competing risks anal-
and decompensation usually do not occur below The clinical course of cirrhosis ysis for proper assessment.
the portal pressure threshold of clinically signifi- Cirrhosis may result from chronic liver inflamma-
cant portal hypertension (CSPH) defined by hep- tion from any cause, following parenchymal
atic venous pressure gradient (HVPG) ≥10 necrosis, activated fibrogenesis, angiogenesis and
mmHg.11–13 Increasing portal pressure, bacterial profound vascular changes. Increased hepatic
translocation, inflammation and hyperdynamic resistance to blood flow, derived from both
circulation are likely mechanisms of decompensa- mechanical obstacle and vasoconstriction result-
tion,14 which progresses to a late decompensation ing from endothelial dysfunction and hepatic stel-
state characterised by further worsening of liver late cells contraction, gradually leads to portal
function associated with other organ dysfunction, hypertension. The ensuing adaptive splanchnic
while acute-on-chronic liver failure (ACLF) may vasodilation further contributes to aggravate por-
occur in any disease state. Histological stages15 tal hypertension and progressively results in
⇑ Corresponding author.
of cirrhosis also parallel the progression of portal hyperdynamic circulation.25–27
Address: Gastroenterology Unit,
hypertension16,17 and clinical states of the When established, cirrhosis remains compen- Ospedale V Cervello, Via
disease.18,19 sated for a variably long time, depending on cur- Trabucco 180, 90146 Palermo,
This body of evidence supports a multistate ability of the underlying disease. In this state, Italy. Tel.: +39 091 6802780 or
+39 091 6802730; fax: +39 091
approach to the clinical course of cirrhosis, which persistence of liver damage results in increasing
6802739.
implies a specific statistical methodology, since fibrosis and portal hypertension.16,17,28 Decom- E-mail address: gedamico@
the Kaplan-Meier (KM) survival curves20,21 may pensation and oesophageal varices may occur libero.it (G. D’Amico).
above the CSPH threshold (HVPG ≥10 mmHg)13,29 mortality of 23%.61 Overall mortality within one
and hence decompensation is more frequent in year of the infectious episode is 60%.63 However,
patients with oesophageal varices.6,7,10 In com- infections are relatively frequent, as well as being
pensated cirrhosis, the rate of development of significant events also in compensated cirrhosis,
oesophageal varices and of decompensation is where they are associated with increased risks of
about 7–8% and 5% per year, respectively.11,30,31 long-term decompensation and mortality.64
Following their appearance, varices grow in cali- Hepatocellular carcinoma (HCC) occurs in 2–8%
bre at a similar rate30,32 and may rupture in 5% patients per year.65–70 The risk is higher in
to 15% of patients per year, with a higher risk in patients with CSPH, higher body mass index, oeso-
patients with large oesophageal varices and red phageal varices and decompensated cirrhosis.70,71
signs or Child B-C class.33 The ensuing variceal Median survival after HCC detection is nine
bleeding is one of the most critical emergencies months in untreated patients72 and approximately
in medicine, with a mortality rate of 10–20%34,35 two years in treated patients, ranging from >10
Key point (it was 50% in the early eighties).36 In untreated years in Barcelona Clinic Liver Cancer (BCLC) stage
patients, rebleeding and death occur in approxi- 0,73 to <6 months in stage D.74
The earliest consequence mately 60% and 30% of patients, respectively, one Liver failure, bleeding, HCC, infections, hepa-
of cirrhosis is a progressive to two years after the index bleeding,37,38 and torenal syndrome, and ACLF are the most frequent
increase in portal pressure
are significantly reduced by non-selective beta- causes of death in patients with cirrhosis.
up to the CSPH threshold of
≥10 mmHg. Bleeding, blockers plus endoscopic variceal ligation or, in Acute-on-chronic liver failure (ACLF) is charac-
ascites, encephalopathy selected patients, by early transjugular intrahep- terised by acute decompensation, organ failure(s)
and jaundice indicate atic portosystemic shunt (TIPS).39–41 and high short-term mortality75–78 and may occur
decompensated cirrhosis. Ascites is a hallmark of decompensation and is either in compensated or in decompensated cir-
Renal function impair- associated with a five-year mortality of about rhosis. Organ failures (liver, renal, coagulation,
ment, refractory ascites,
50%.9,42 The progression of the hyperdynamic cir- cerebral, respiratory and circulatory) are defined
infections and circulatory
dysfunction indicate more
culation and the activation of pro-inflammatory according to the Chronic Liver Failure-Sequential
advanced decompensation mechanisms14 result in deterioration of renal Organ Failure (CLIF-SOFA) score or by its simplified
and are associated with function, and ascites may become refractory. The version Chronic Liver Failure-Organ Failure Assess-
very poor survival. cumulative risk of refractory ascites is in the order ment (CLIF-OF) score.75,76 Severity of ACLF is
of 20% within five years of the development of graded according to the number of organ failures
ascites.42 Two-year mortality following refractory as ACLF-1, ACLF-2, and ACLF-3. There is a large
ascites is approximately 65% and may be reduced body of evidence indicating that ACLF is triggered
by TIPS in selected patients.43,44 by a systemic inflammatory response78–82 to sev-
Overt hepatic encephalopathy and/or jaundice eral critical events, which may act as intra- or
typically occur in advanced cirrhosis, rarely as a extrahepatic factors.83,84 Bacterial by-products ter-
first sign of decompensation and are associated med pathogen associated molecular patterns
with a five-year survival of about 20%.8,10 The (PAMPS) and damage associated molecular pat-
prognostic weight of covert hepatic encephalopa- terns (DAMP) released after liver tissue injury play
thy45 remains to be defined, and is possibly more a key role in triggering inflammation.14 Infections
dependent on the grade I encephalopathy compo- are not only a triggering factor in up to 37% of
nent than on minimal encephalopathy.46–48 patients, but they also occur during follow-up in
Renal function is often impaired in patients 46% of patients free of infection at ACLF diagnosis,
with advanced cirrhosis, as a consequence of and are associated with a significantly increased
either progressive haemodynamic derangement mortality risk in ACLF grades 1–2.85 No triggering
or acute events such as bleeding, infections or factors may be identified in up to 40% of
nephrotoxicity. The usual clinical presentation is patients,75,84 in whom PAMPs resulting from sub-
that of acute kidney injury (AKI), including hepa- clinical bacterial translocation55 or DAMPs result-
torenal syndrome type 1 or 2.49–51 AKI is usually ing from liver tissue injury86,87 may be key-factors.
followed by progressive worsening of renal func- ACLF is a dynamic syndrome and 28-day mor-
tion, and progressive reduction of mean arterial tality ranges from 10% to 87% depending on the
pressure, indicating progressive cardiac and number of failing organs between day three and
haemodynamic impairment.52 One-year mortality, seven.78,77,88 The CLIF-C ACLF score enables pre-
following renal failure in this advanced disease diction of mortality risk more accurately than the
state, is in the order of 60%,53,54 although it varies MELD, MELD sodium and Child-Pugh scores,89
according to the definition used.49–51 while another specific prognostic score has been
Because of bacterial translocation driven by the developed for patients with acute decompensation
progression of portal hypertension and liver dys- without ACLF, the CLIF-C AD score.90
function,55 infections are common in advanced In patients with ACLF, 23% of those with mostly
cirrhosis,56–62 particularly in patients with ascites. alcohol and/or hepatitis C virus (HCV)-related cir-
The inflammatory response is frequently severe rhosis,75 and 45–52% of those with mostly HBV-
and associated with organ failures.55,60 Mortality related cirrhosis83,91 may not have history of pre-
may reach 38%,58 while discharged patients have vious decompensation. However, in a recently
a 30-day readmission rate of 35% and six-month reported study of the incidence of ACLF in outpa-
Increasing fibrosis
METAVIR 4A METAVIR 4B METAVIR 4C
Aetiological cure
Fig. 1. Schematic representation of the progression of pathophysiological mechanisms, haemodynamic and clinical manifestations of cirrhosis
according to increasing severity along the course of the disease. METAVIR 4A, 4B, 4C: substaging of METAVIR fibrosis stage 4 according to increasing
thickness of septa and decreasing nodules size. ACLF, acute-on-chronic liver failure; CO, cardiac output; CSPH, clinically significant portal hypertension; HRS,
hepatorenal syndrome; HVPG, hepatic venous pressure gradient; MPH, mild portal hypertension; PSE, portosystemic encephalopathy; SAV, splanchnic
arteriolar vasodilatation.
Key point and systemic inflammation result in a more Although timolol (an NSBB) has been proven inef-
advanced, late decompensation state where fective in preventing development of oesophageal
LSM>15 identify cACLD multi-organ dysfunction becomes clinically evi- varices,11 since prevention of decompensation is
which is equivalent to
dent.14,49,53,59–63,127–131 Refractory ascites, infec- now recognised as a major management objective
compensated cirrhosis and
values ≥20–25 KPa denote
tions, persistent encephalopathy and/or jaundice, in these patients (states 0–2),104,107 other preven-
CSPH with 0.90 specificity. renal, circulatory and respiratory dysfunctions tative approaches, aiming at reducing intrahepatic
are typical presentations of this disease state. resistance and splanchnic flow, should undergo
MPH (HVPG >5mmHg and One-year mortality for these conditions ranges appropriate clinical trials. Patients in states 3–5
<10mmHg), CSPH and
from 60 to 80%. should be appropriately treated for the prevention
gastroesophageal varices
are associated with ACLF72–75 may occur in any disease state and is of bleeding, rebleeding104,107 and further decom-
increasing disease severity associated with six-month mortality from 38% to pensation. Patients in state 6 and those with ACLF
in compensated cirrhosis. 96%.77 Patients resolving ACLF, may remain in a require a multidisciplinary approach to their man-
decompensated state with or without organ dys- agement, with access to specific treatments and
Bleeding alone, any first
function or even in a compensated state, although early selection of candidates for liver transplanta-
decompensating event and
≥2 decompensating events the proportions of these state transitions are not tion.132–134
are associated with clearly defined.
worsening outcome of
decompensated cirrhosis.
Competing risks in cirrhosis
Management implications of clinical states A competing risk is the risk of an event whose
of cirrhosis occurrence either precludes the occurrence of
Key point Recognising different clinical states may have another event or modifies the probability that it
Refractory ascites, HRS, important implications on the most likely clinical will occur. Along the clinical course of cirrhosis,
infections, circulatory dys- outcomes. Hence, clinical states may be used to many clinical conditions, or states, may be charac-
function and ACLF are inform treatment interventions to prevent disease terised by competing outcomes which require
markers of late advanced progression. In fact, there is evidence that in com- competing risks analysis to correctly assess the
decompensation and very
pensated patients without varices (state 1) the relevant risks. Several such conditions are shown
poor survival, although
ACLF may also occur in
most appropriate approach is aetiological treat- (Table 1), where examples of outcomes of interest
compensated cirrhosis. ment, particularly in those with MPH (state 0) in are reported together with potentially relevant
whom NSBBs do not appreciably reduce portal competing events, according to the baseline condi-
hypertension.13,97 In these patients, non-invasive tion and the research aim requiring competing
monitoring based on LSM and platelet count may risks analysis.135
identify patients at risk of CSPH and guide endo- It is important to note that in decompensated
scopic screening for oesophageal varices.104,109,110 cirrhosis the underlying risk of death is high, with
Table 1. Examples of clinically relevant outcomes and potentially relevant competing events according to several baseline conditions and specific
research aims.
Clinical condition Event of interest Potentially relevant Aims of competing risks analysis
competing events
Compensated cirrhosis
Decompensation Death Decompensation; death before decompensation
Specific Death; other decompensating First event to occur in compensated cirrhosis
decompensating event events
ACLF Death; decompensation ACLF before overt decompensation; death before ACLF
HCC Death; decompensation HCC in compensated cirrhosis; death before HCC
MPH (HVPG >5 and Oesophago-gastric Death; decompensation Development of varices before decompensation; death before
<10 mmHg) varices varices and decompensation
CSPH (HVPG ≥10 Oesophago-gastric Death; decompensation Development of varices before decompensation; death before
mmHg) varices varices and decompensation
Oesophageal varices Bleeding Death, decompensation Assessment of bleeding before decompensation
Decompensated cirrhosis
Variceal bleeding Rebleeding Death; decompensating Incidence of rebleeding before other decompensating events
events
Variceal bleeding Rebleeding Death Incidence of rebleeding; death before rebleeding
Ascites Refractory ascites Death Incidence of refractory ascites; death before refractory ascites
Ascites Other decompensating Death; other decompensating First new clinically relevant event; death before other events
events events
≥1 decompensating Other decompensating Death; other decompensating First new clinically relevant event; death before other events
events events events
Any type of ACLF; HCC; HRS Death ACLF, death before ACLF; HCC, death before HCC; HRS, death before
decompensation HRS
Late decompensation
Refractory ascites Death OLT Death before OLT; OLT
Major infections Death OLT Death before OLT; OLT
Renal failure Death OLT Death before OLT; OLT
Acute on chronic liver Death OLT Death before OLT; OLT
failure
ACLF, acute-on-chronic liver failure; CSPH, clinically significant portal hypertension; HCC, hepatocellular carcinoma; HRS, hepatorenal syndrome; HVPG, hepatic venous
pressure gradient; OLT, orthotopic liver transplantation; MPH, mild portal hypertension.
a median survival of approximately two years.4 developing varices is clearly a competing event
This implies that whenever assessing the inci- for the occurrence of oesophageal varices. When
dence of any further decompensating event, like using the KM estimator (Fig. 2A) the 20-year 1-
refractory ascites, hepatorenal syndrome, infec- KM estimate of the probability of developing
tions and ACLF, it is important to consider death varices was 0.61. The 20-year 1-KM cumulative
as a competing event to achieve reliable risk probability of death without varices was 0.46.
estimates.135 Since the two risks are mutually exclusive, they
Key point
are clearly upward biased, summing up to 1.07
instead of ≤1. Notably, when assessing the risk of Competing risks are fre-
Competing risks analysis varices the KM estimator censors death, and when quent in cirrhosis. Death
Recognising competing risks is important because, should always be consid-
estimating the risk of death without varices it cen-
ered a competing risk for
in the presence of competing risks, the KM estima- sors the occurrence of varices: for this reason the
any clinically relevant
tor21 invariably results in upward biased esti- number of patients at risk per each observation event in the course of the
mates.22–24,135 Competing risks analysis is based period is identical for the two KM estimates. To disease, particularly for
on the CIF which, by using the Aalen-Johansen show how competing risks analysis works we decompensated cirrhosis.
estimator,23 partitions the probability of any event assessed the risk (1-KM estimate) of the composite
into the probabilities that each event occurs first, outcome (death or varices, whichever occurs first):
resulting in an overall event probability (or the in this way neither event is censored. By doing
sum of the probabilities of each event) correctly this, we have reduced the three-state model (no
ranging from zero to one. The essential difference varices, varices, death) into a two-state model
between the two methods is that the KM estima- ([no varices] ? [varices or death]), the situation
tor counts only the events of interest and censors where the KM method may be safely applied.20,21
or ignores the competing events, while the This estimate in our cohort is plotted (Fig. 2B [1-
Aalen-Johansen estimator correctly counts both KM curve]) together with the CIF of each of the
the event of interest and the competing events. two competing events obtained by the competing
As a practical example consider assessing the risks analysis. It may be seen that the 1-KM esti-
incidence of oesophageal varices, and the proba- mate for the risk of the composite outcome death
bility of survival without developing varices, in a or varices (0.79) corresponds exactly to the sum of
cohort of patients included in a prospective study each of the two risks (0.53 varices + 0.26 death =
of the natural history of cirrhosis.10 Death before 0.79 composite) computed by the competing risks
analysis. The example shows that the competing 0.51, variceal bleeding 0.35 and HCC 0.28 (Fig. 3A),
risks analysis partitions the risk of any of the com- yielding a total probability of 1.77, which is of
peting events (whichever occurs first) into the risk course impossible for mutually exclusive risks. In
that each of the competing events occurs first. fact, this analysis does not provide us with the
Therefore, the real risk of developing varices in probability of the next event to occur, but instead
the included cohort of patients was 0.53 as esti- gives us the 20-year cumulative risk of each event,
mated by the CIF and not 0.61 as estimated by independently of the occurrence of other events,
the KM estimator. Moreover, censoring death like in a hypothetical world where death (or the other
it was not a clinically relevant event does not make events) do not exist. In particular, the KM curve on
sense and is technically wrong.135 The correspond- death is the correct estimate of the overall mortal-
ing real probability of dying before developing ity, as defined before. Note that, as already men-
varices was 0.26 and not the 0.46 1-KM estimate. tioned, death is censored when estimating the
By considering any death (before or after varices) risk of each of the other events, while for any other
we could have correctly estimated the overall mor- event of interest, patients experiencing the com-
tality by the KM estimator, however this estimate peting events are still at risk of the event of inter-
would not consider the possible change of mortal- est. Competing risks analysis showed that the 20-
Key point ity risk after the development of varices. year probability of being the next event to occur
Another example of a situation where the com- was 0.29 for ascites, 0.16 bleeding, 0.15 HCC,
Competing risks analysis is peting risks analysis should be applied is the 0.12 death summing up to 0.72 (Fig. 3B). In a sim-
based on the Aalen- assessment of the probability that a clinically rel- ilar study117 it was erroneously concluded that
Johansen estimator which
evant event will be the first to occur among sev- HCC was the first event to occur based on a 17-
partitions the risk of any
competing event in the eral events. In such a situation all the assessed year 1-KM cumulative risk of 0.55 compared with
probabilities that each events are mutually exclusive, because only one 0.35 ascites, 0.27 jaundice, 0.10 bleeding, 0.03
competing event occurs of them may be the next to occur, and the sum encephalopathy, summing up to a total probability
first. of all the probabilities shall range between zero of 1.3. An inference from this study would be that
and one. However, by the KM model the sum of HCC is the most likely first event in an ideal world
the probabilities obtained may be higher than where the other events do not exist.
one, because per each assessed event, the occur- More insight into the reason why the use of the
rence of the competing events is ignored or cen- KM estimator is correct only in two-state settings
sored. For example, looking for the next event to is provided by the relationship between rate (or
occur in a cohort of 377 patients with compen- incidence rate) and risk. This has been thoroughly
sated cirrhosis,10 the 1-KM estimates showed that explained elsewhere.22,135 Here we only recall that
20-year cumulative risk of death was 0.63, ascites while the incidence rate is the ratio D/Y (number of
A 1.0
1-KM estimate
B 1.0
0.79
0.8 0.8 Varices or death
estimated by 1-KM
Probability
0.61
0.6
Probability
0.6 0.53
Varices
Varices (CIF)
0.4 0.46
0.4
Death 0.26
Death (CIF)
0.2 0.2
0.0 0.0
0 60 120 180 240 0 60 120 180 240
Months Months
Patients
at risk 243 153 98 74 52 243 153 98 74 52
Fig. 2. Development of oesophageal varices and mortality in 243 patients free of varices at diagnosis of compensated
cirrhosis.10 (A) 1-KM estimates of the cumulative risks of developing oesophageal varices (death censored) and of dying before
developing varices (varices censored). The sum of the two risks is higher than 1 (0.46 + 0.61 = 1.07). (B) Cumulative incidence
of varices and of death assessed by the competing risks analysis CIF, estimated by the Aalen-Johansen estimator. The 1-KM
estimate of the composite endpoint (death or development of varices) is also plotted to show how the CIF partitions the risk of
any event in the risks of each event (0.53 + 0.26 = 0.79). (A, B) The abscissa denotes the number of months of observation and
the numbers below the abscissa are the number of patients at risk per each observation period. Note that the number of
patients at risk per each observation period for the two KM plots in (A) is the same because when estimating the risk of death
without developing varices, the KM estimator censors the occurrence of varices and when estimating the risk of varices it
censors deaths. Therefore, per each observation period the number of patients at risk is the number at risk at the beginning of
the previous observation period – (n developing varices + n death + n with truncated observation) for both curves. For the
same reason, the number of patients at risk in the competing risks analysis and in the 1-KM plot in the right side of the figure
are the same as those in the plots of the left side. CIF, cumulative incidence function; KM, Kaplan-Meier.
A 1.0
Cumulative risks
by the KM estimator
B 1.0 Cumulative incidence
by the Aalen-Johansen estimator
0.8 0.8
Death
Death
Ascites
Bleeding
0.63 Ascites
0.6 0.6
Probability
Bleeding
Probability
HCC
0.51 HCC
0.12
0.0 0.0
0 60 120 180 240 0 60 120 180 240
Months Months
Patients 377 309 251 207 141 377 276 201 142 106
at risk
377 288 219 160 120
377 292 231 164 125
377 304 244 178 131
Fig. 3. Occurrence of death, ascites, bleeding and HCC in a cohort of 377 patients with compensated cirrhosis at
diagnosis.10 (A) Cumulative risk of each event by the KM estimator plotted as 1-KM estimates. (B) Competing risks analysis
showing the cumulative incidence function for the same events, computed by the Aalen-Johansen estimator. Each curve in (B)
plots the probability for each event to be the first to occur along the disease course. (A, B) The abscissa denotes the number of
months of observation and the numbers below the abscissa are the number of patients at risk per each observation period. It is
to note that in (A), the number of patients at risk is the same for all the assessed events at the first observation but it differs
thereafter because the Kaplan-Meier estimator considers only one event of interest and does not account for the others.
Therefore, it is the occurrence of only the event of interest to reduce the number of subjects at risk per each relevant curve and
hence the number of subjects at risk may be different for different events. By contrast in the competing risks analysis by the
Aalen-Johansen estimator shown in (B), all the patients are at risk of all the considered events and at any time one of the
considered events occurs, the number of subjects at risk is correspondingly reduced for all the considered events. Therefore, the
number of subjects at risk is the same for any event at each observation time. HCC, hepatocellular carcinoma; KM, Kaplan-Meier.
subjects developing the disease/the total amount estimates the risk that each of several events of
of person-time at risk), the risk is the ratio D/N, interest occurs first, by the CIF.
(number of subjects who develop the disease/ These concepts also imply that the association
number of subjects disease-free at the beginning of covariates with the rate of one event of interest
of the observation time).136 Therefore, while the may be different from their association with the
incidence rate is essentially an average measure risk of that event, in the presence of competing
of the speed of occurrence of the event of interest, risks. For this reason, the Cox model137 may not
assuming a constant hazard over time, and is be appropriate to estimate the risk of events in dif-
expressed as the number of events occurring per ferent groups135 in a competing risks setting,
unit of time, the risk is a measure of the cumula- because it assumes a one to one relationship
tive probability of the event in a given observation between rate and risk.
time. Consequently, while the cumulative risk may
not decrease over time (because D either increases
or is constant with time), the incidence rate may Building multistate models in cirrhosis
increase, remain near constant or decrease. Clinical states should be conceived as a flexible
Key point
It may be noted that while in two-state set- concept and specific clinical states should be
tings, there is a direct one to one relationship defined whenever a specific prognostic question In two-state settings there
between rate and risk, because the number of sub- makes it appropriate (Table 1). As an example, in is a direct one to one
jects at risk and of events taken as a basis for cal- the same study of the clinical course of cirrhosis relation between rate and
risk, this is lost in
culations is the same for rate and risk, in a referred to earlier,10 the 20-year cumulative prob-
multi-state settings, where
multistate setting (competing risks) this unique ability of death (1-KM) was 0.62, in 377 patients
the number of subjects at
relationship is lost, because the number of sub- with compensated cirrhosis, and 0.93 after decom- risk decreases at any time
jects at risk decreases whenever one of the com- pensation in 224 patients who developed decom- one of the competing
peting events occurs. In fact, here the interest is pensation during the follow-up (Fig. 4A). events occurs.
in the next event to occur, and the risk depends However, in clinical practice, it might also be of
on more than one rate of events. interest to know the probability of decompensa-
In other words, the KM assumes that there is tion and of death before decompensation. These
only one event of interest and that only this will probabilities are appropriately assessed by the
contribute to reduce the number of patients at competing risks analysis (Fig. 4B), which shows a
risk. When more than one event causes the reduc- 20-year cumulative probability of 0.58 and 0.14
tion of patients at risk, like in the competing risks for decompensation and death before decompen-
situation, the Aalen-Johansen estimator correctly sation, respectively. The probability of death after
Decompensated
Event probability
0.6
n = 224
0.62
pensation, and 34% of being alive in a compen-
0.6 Decompensation 0.58
sated state.
0.4 0.4 A schematic representation of a hypothetical
Compensated multistate model encompassing the whole course
0.2 n = 377 0.2 Death before decompensation 0.14
of cirrhosis is shown (Fig. 5).
0.0 0.0
0 60 120 180 240 0 60 120 180 240
Months Months Competing risks and clinical research
Patients 377 309 251 186 141 Patients 377 274 203 146 109
at risk 224 62 35 23 10 at risk As shown in previous sections in this article, the
Alive, compensated Dead after decompensation most obvious field of application for competing
C D100 Alive, decompensated Dead without decompensation
risks analysis is the study of the clinical course of
Compensated diseases in every situation where relevant compet-
Prediction probabilities %
34
80 46 ing risks are identified.
59
60
78 8 A second relevant field where competing risks
0.58 0.14 100 10 should be appropriately accounted for is prognosis
40
13
41 research. A major aim of this type of research is to
20 10 18
31
find accurate predictors of a disease outcome in a
7
13 17 given time. The hypothesis underlying this
Decompensated Death 5 10
0.93 0
0 60 120 180 240 research is that one or more among several patient
Months or disease characteristics may be linked to the dis-
ease outcome through a causal mechanism (causal
Fig. 4. A three-state model for cirrhosis is shown, including compensated cirrhosis,
decompensated cirrhosis and death. The model is derived from a cohort of 377 patients with factors) or through indirect association (predictive
compensated cirrhosis at diagnosis.10 (A) 20-year Cumulative risk of death for 377 patients factors). In studies of causal factors, the hypothesis
with compensated cirrhosis at diagnosis and of death after decompensation in the 224 of them is that some biological mechanism links the candi-
who developed decompensation during the observation period. The risks are estimated by the date predictor to the event under study. In this
KM estimator and plotted as 1-KM estimates. (B) Cumulative incidence of decompensation or
death by competing risks analysis of for the 377 patients with compensated cirrhosis shown in
type of study the interest is in assessing whether
(A) The CIF by the Aalen-Johansen estimator is plotted for the two competing events. (C) the event occurs earlier in patients presenting
Three-state model showing 20-year transition probabilities from compensated cirrhosis the candidate causal factor than in those without,
towards decompensation and death and from decompensation to death in the same cohort as thus focussing on the rate, not the risk. In this sit-
in previous panels. (D) Five, ten, fifteen and twenty-year state occupation probabilities for the
uation, the cause specific Cox model138 is appro-
three-state model shown in (C). The disease states considered are compensated, decompen-
sated and death. The probability of being dead after decompensation or before decompen- priate, since it looks at the instantaneous rate of
sation are exploited to allow a more complete prognostic assessment. (A, B, D) the abscissa only the event of interest and may disclose a cau-
denotes the number of months of observation; (A, B) the numbers below the abscissa are the sal effect when all the potential confounders are
numbers of patients at risk per each observation period. CIF, cumulative incidence function; accounted for.
KM, Kaplan-Meier.
When the interest is in identifying factors asso-
ciated with the occurrence of the event of interest
decompensation is already known by the KM anal- (risk predictors), competing events, which may
ysis and the full three-state model is shown modify the risk of the event of interest must be
(Fig. 4C). accounted for. In this type of research, the stan-
In a multistate model it is also possible to dard Cox model is not appropriate because it
assess the probability of patients in the earliest ignores the competing events and may result in
state occupying one of the other possible states upwards biased estimates similar to the KM
after a certain observation time.135 In the three- estimator. A specific multiple regression propor-
state model above, the possible outcomes along tional hazards model has been developed for com-
the disease course are: alive still compensated, peting risks, by Fine and Gray.139 This model
alive decompensated, dead before decompensa- provides the sub(distribution)-hazard ratio (sHR),
tion, dead after decompensation. The relevant which measures the relationship between the
state occupation probabilities may be estimated candidate predictive factor and the outcome,
by the Aalen-Johansen estimator, which is an accounting for both the association between the
extension of the KM estimator for multistate mod- factor and the outcome and for the modifying
els. Statistical software for this kind of analysis are effect of the competing events on this association.
available in the R statistical package (msSurv or Although it does not have a meaningful interpreta-
mstate routins).138 State occupation probabilities tion, it allows for risk predictions in individual
for the above model are shown (Fig. 4D) at differ- patients.140
ent observation times over the whole disease Therefore, it is important when planning
course. The probability of being alive in a decom- research on prognostic indicators, to ensure that
pensated state is low, and this is coherent with the statistical method used is appropriate for the
the current knowledge that patients who transi- specific research aims. The Cox model should be
0.8
Death probability
0.6 dictable sequence, because of the variable inter-
0.6
Rebleeding 0.41 play between pathophysiological mechanisms.
0.4 0.4 Rebleeding 0.28
Therefore, a multistate approach provides a more
0.2 Death 0.08 0.2 Death 0.10
realistic description of the disease course. Mul-
0.0
tistate models require competing risks analysis
0.0
0 7 14 21 28 35 42 0 7 14 21 28 35 42 to assess the probabilities of transition across
Days Days states and appropriate multiple regression models
Patients 101
at risk
82 77 71 67 65 61 Patients
at risk
97 82 80 71 74 71 66 to investigate prognostic indicators, because the
101 99 99 97 94 94 93 97 95 94 90 89 89 87
traditional KM estimator and the Cox model may
Placebo Octreotide provide biased results in the presence of compet-
C 1.0 Competing risks analysis D 1.0 Competing risks analysis ing risks. These concepts should be thoroughly
(Aalen-Johansen estimator) (Aalen-Johansen estimator)
accounted for when planning clinical research
0.8 0.8
either of prognosis or of treatment efficacy in
Death probability
Death probability
Fig. 7. Estimates of rebleeding and death risks in a double blind, placebo controlled
randomised clinical trial of octreotide.141 (A, B) 1-Kaplan-Meier estimates in (A) the placebo
Conflict of interest
group and in (B) the octreotide group, respectively. (C) Placebo group and (D) octreotide The authors declare no conflicts of interest that
group: rebleeding and death cumulative incidence functions, estimated by the Aalen-Johansen pertain to this work.
estimator in the same trial. The fig. shows that deaths occurred mostly after rebleeding in the Please refer to the accompanying ICMJE disclo-
placebo group, while half of the total deaths in the octreotide group occurred before sure forms for further details.
rebleeding. (A–D) The abscissa denotes the number of months of observation and the numbers
below the abscissa are the numbers of patients at risk per each observation period.
Authors’ contribution
size) and the total number of patients included Gennaro D’Amico: review concepts and project,
would need to be 452. Instead, by using the 0.53 data collection, elaboration of examples, text and
risk derived by the competing risks analysis, figures. Alberto Morabito: statistical concepts
accounting for death before the formation of and data analysis for examples. Mario D’Amico:
varices (Fig. 2B), the total number of patients data collection; cohort study for the examples
needed would be 520 under the assumption of a included in the review; text revision. Linda Pasta:
hazard ratio of 0.7 as before.142 Therefore, it is responsible for the cohort study used for the
likely that even if the correct hypothesis has been examples included in the review. Giuseppe Mal-
made, the trial would be underpowered if the izia: review project and text revision. Paola Reb-
competing risk of death has not been appropri- ora: statistical concepts, examples and text
Key point ately accounted for.143 revision. Maria Grazia Valsecchi: statistical con-
cepts, examples, text revision.
Accounting for competing
risks may enable a more
adequate sample size esti-
Conclusions
mation when planning The clinical spectrum of cirrhosis encompasses Supplementary data
clinical research, particu- several clinical states, from a very early state of Supplementary data associated with this article
larly when analysing compensated cirrhosis with subclinical portal can be found, in the online version, at
treatment efficacy. hypertension, to the late decompensation state https://doi.org/10.1016/j.jhep.2017.10.020.
[46] Thomsen KL, Macnaughtan J, Tritto G, Mookerjee RP, Jalan R. Clinical in hepatitis C-related advanced liver disease. Gastroenterology
and pathophysiological characteristics of cirrhotic patients with grade 2009;136:138–148.
1 and minimal hepatic encephalopathy. PLoS One 2016;11:e0146076. [69] Giannini EG, Risso D, Testa R, Trevisani F, Di Nolfo MA, Del Poggio P,
[47] Bajaj JS, Cordoba J, Mullen KD, Amodio P, Shawcross DL, Butterworth et al. Prevalence and prognostic significance of the presence of
RF, et al. Review article: the design of clinical trials in hepatic esophageal varices in patients with hepatocellular carcinoma. Clin
encephalopathy–-an International Society for Hepatic Encephalopathy Gastro Hepatol 2006;4:1378–1384.
and Nitrogen Metabolism (ISHEN) consensus statement. Aliment [70] Ripoll C, Groszmann RJ, Garcia-Tsao G, Bosch J, Grace N, Burroughs A,
Pharmacol Ther 2011;33:739–747. et al. Hepatic venous pressure gradient predicts development of
[48] Patidar KR, Thacker LR, Wade JB, Sterling RK, Sanyal AJ, Siddiqu MS, hepatocellular carcinoma independently of severity of cirrhosis. J
et al. Covert hepatic encephalopathy is independently associated with Hepatol 2009;50:923–928.
poor survival and increased risk of hospitalization. Am J Gastroenterol [71] Shim JJ, Oh CH, Kim JW, Lee CH, Kim BH. Liver cirrhosis stages and the
2014;109:1757–1763. incidence of hepatocellular carcinoma in chronic hepatitis B patients
[49] Piano S, Rosi S, Maresio G, Fasolato S, Cavallin M, Romano A, et al. receiving antiviral therapy. Scand J Gastroenterol 2017. https://doi.org/
Evaluation of the Acute Kidney Injury Network criteria in hospitalized 10.1080/00365521.2017.1335773.
patients with cirrhosis and ascites. J Hepatol 2013;59:482–489. [72] Giannini EG, Farinati F, Ciccarese F, Pecorelli A, Rapaccini GL, Di Marco
[50] Fagundes C, Barreto R, Guevara M, Garcia E, Solà E, Rodríguez E, et al. A M, et al. Prognosis of untreated hepatocellular carcinoma. Hepatology
modified acute kidney injury classification for diagnosis and risk 2015;61:184–190.
stratification of impairment of kidney function in cirrhosis. J Hepatol [73] Llovet JM, Bru C, Bruix J. Prognosis of hepatocellular carcinoma: the
2013;59:474–481. BCLC staging classification. Semin Liver Dis 1999;19:329–338.
[51] Angeli P, Ginès P, Wong F, Bernardi M, Boyer TD, Gerbes A, et al. [74] Liu PH, Hsu CY, Hsia CY, Lee YH, Su CW, Huang SC, et al. Prognosis of
Diagnosis and management of acute kidney injury in patients with hepatocellular carcinoma: Assessment of eleven staging systems. J
cirrhosis: Revised consensus recommendations of the International Hepatol 2016;64:601–608.
Club of Ascites. J Hepatol 2015;62:968–974. [75] Moreau R, Jalan R, Gines P, Pavesi M, Angeli P, Cordoba J. Acute-on-
[52] Tsien CD, Rabie R, Wong F. Acute kidney injury in decompensated chronic liver failure is a distinct syndrome that develops in patients
cirrhosis. Gut 2013;62:131–137. with acute decompensation of cirrhosis. Gastroenterology
[53] Fede G, D’Amico G, Arvaniti V, Tsochatzis E, Germani G, Georgiadis D, 2013;144:1426–1437.
et al. Renal Failure and cirrhosis: a Systematic review of mortality and [76] Arroyo V, Jalan R. Acute-on-chronic liver failure: definition, diagnosis
prognosis. J Hepatol 2012;56:810–818. and clinical characteristics. Semin Liver Dis 2016;36:109–116.
[54] Salerno F, Cazzaniga M, Merli M, Spinzi G, Saibeni S, Salmi S, et al. [77] Bajaj JS, O’Leary JG, Reddy R, Wong F, Biggins SW, Patton H, et al.
Diagnosis, treatment and survival of patients with hepatorenal Survival in infection-related acute-on-chronic liver failure is defined by
syndrome: A survey on daily medical practice. J Hepatol extrahepatic organ failures. Hepatology 2014;60:250–256.
2011;55:1241–1248. [78] Hernaez R, Solà E, Moreau R, Ginès P. Acute-on-chronic liver failure: an
[55] Wiest R, Lawson M, Geuking M. Pathological bacterial translocation in update. Gut 2017;66:541–553.
liver cirrhosis. J Hepatol 2014;60:197–209. [79] Claria J, Stauber ER, Coenraad MJ, Moreau R, Jalan R, Pavesi M.
[56] Dionigi E, Garcovich M, Borzio M, Leandro G, Majumdar A, Tsami A Systemic inflammation in decompensated cirrhosis. Characterization
et al. Bacterial Infections Change Natural History of Cirrhosis Irrespec- and role in acute-on-chronic liver failure. Hepatology 2016;64:
tive of Liver Disease Severity. AJG 2017 e-published ahed of print. doi: 1249–1264.
10.1038/ajg.2017.19. [80] Solé C, Solà E, Morales-Ruiz M, Fernandez G, Huelin P, Graupera I, et al.
[57] Borzio M, Salerno F, Piantoni L, Cazzaniga M, Angeli P, Bissoli F, et al. Characterization of inflammatory response in acute-on-chronic liver
Bacterial infection in patients with advanced cirrhosis: a multicentre failure and relationship with prognosis. Sci Rep 2016;6:32341.
prospective study. Dig Liver Dis 2001;33:41–48. [81] Moreau R. The pathogenesis of ACLF. The inflammatory response and
[58] Fernandez J, Navasa M, Gomez J, Colmenero J, Vila J, Arroyo V, et al. immune function. Semin Liver Dis 2016;36:133–140.
Bacterial infections in cirrhosis: epidemiological changes with invasive [82] Claria J, Arroyo V, Moreau R. The acute-on-chronic liver failure
procedures and norfloxacin prophylaxis. Hepatology 2002;35:140–148. syndrome, or when the innate immune system goes astray. J Immunol
[59] Fernandez J, Acevedo J, Castro M, Garcia O, Rodrıguez de Lope C, Roca D, 2016;197:3755–3761.
et al. Prevalence and risk factors of infections by multiresistant bacteria [83] Shi Y, Yang Y, Hu Y, Wu W, Yang Q, Zheng M, et al. Acute-on-
in cirrhosis: a prospective study. Hepatology 2012;55:1551–1561. chronic liver failure precipitated by hepatic injury is distinct from
[60] Bajaj JS, O’Leary JG, Wong F, Reddy KR, Kamath PS. Bacterial infections that precipitated by extrahepatic insults. Hepatology 2015;62:
in end-stage liver disease: current challenges and future directions. Gut 232–242.
2012;61:1219–1225. [84] Solà E, Fernandez J, Gines P. Acute-on chronic liver failure: the role of
[61] Piano S, Morando F, Carretta G, Tonon N, Vettore E, Rosi S, et al. precipitating illness. Semin Liver Dis 2016;36:117–122.
Predictors of early readmission in patients with cirrhosis after the [85] Fernandez J, Acevedo J, Weist R, Gustot T, Amoros A, Deulofeu C, et al.
resolution of bacterial infections. Am J Gastroenterol 2017. https://doi. Bacterial and fungal infections in acute-on-chronic liver failure:
org/10.1038/ajg.2017.253. prevalence, characteristics and impact on prognosis. Gut 2017.
[62] Jalan R, Fernandez J, Wiest R, Schnabl B, Moreau R, Angeli P. Bacterial https://doi.org/10.1136/gutjnl-2017-314240.
infections in cirrhosis: A position statement based on the EASL Special [86] Albillos A, Lario M, Alvarez-Mon M. Cirrhosis associated imuune
Conference 2013. J Hepatol 2014;60:1310–1324. dysfunction: distinctive features and clinical relevance. J Hepatol
[63] Arvaniti V, D’Amico G, Fede G, Manosou P, Tsochatzis E, Pleguezuelo M, 2014;61:1385–1396.
et al. Infections in patients with cirrhosis increases mortality four-fold [87] Kubes P, Mehal WZ. Sterile inflammation in the liver. Gastroenterology
and should be used in determining prognosis. Gastroenterology 2012;143:1158–1172.
2010;139:1246–1256. [88] Gustot T, Fernandez J, Garcia E, Morando F, Caraceni P, Alessandria C,
[64] Nahon P, Lescat M, Layese R, Bourcier V, Talmat N, Allam S, et al. Bacterial et al. Clinical Course of acute-on-chronic liver failure syndrome and
infection in compensated viral cirrhosis impairs 5-year survival (ANRS effects on prognosis. Hepatology 2015;62:243–252.
CO12 CirVir prospective cohort). Gut 2017;66:330–341. [89] Jalan R, Pavesi M, Saliba F, Amoros A, Moreau R, Jines P, et al.
[65] Zhao C, Nguyen MH. Hepatocellular carcinoma screening and surveil- Development and validation of a prognostic score to predict mortality
lance. practice guidelines and real-life practice. J Clin Gastroenterol in patients with acute-on-chronic liver failure. J Hepatol
2016;50:120–133. 2014;61:1038–1047.
[66] Fattovich G, Stroffolini T, Zagni I, Donato F. Hepatocellular carcinoma in [90] Jalan R, Pavesi M, Saliba F, Amoros A, Fernandez J, Holland-Fisher P,
cirrhosis: incidence and risk factors. Gastroenterology 2004;127: et al. The CLIF consortium acute decompensation score (CLIF-C Ads) for
s35–s50. prognosis of hospitalized cirrhotic patients without acute-on-cronich
[67] Ioannou G, Splan M, Weiss N, McDonald G, Beretta L, Lee S. Incidence liver failure. J Hepatol 2015;62:831–840.
and predictors of hepatocellular carcinoma in patients with cirrhosis. [91] Li H, Chen LY, Zhang NN, Li ST, Zeng B, Pavesi M, et al. Characteristics,
Clin Gastroenterol Hepatol 2007;5:938–945. diagnosis and prognosis of acute-on-chronic liver failure in cirrhosis
[68] Lok AS, Seeff LB, Morgan TR, Di Bisceglie AM, Sterling RK, Curto TM, associated to hepatitis B. Sci Rep 2016;6:25487. https://doi.org/
et al. Incidence of hepatocellular carcinoma and associated risk factors 10.1038/srep25487.
[139] Fine JP, Gray RJ. A proportional hazards model for the subdistribution [142] Schoenfeld DA. Sample-size formula for the proportional-hazards
of competing risk. J Am Stat Ass 1999;94:496–509. regression model. Biometrics 1983;39:499–503.
[140] Wolbers M, Koller MT, Wittenam JCM, Steyerberg EW. Prognostic [143] BC Tai, ZJ Chen, D Machin. Estimating sample size in the presence of
models with competing risks. Methods and application to coronary risk competing risks – Cause-specific hazard or cumulative incidence
prediction. Epidemiol 2009;20:555–561. approach? Stat Methods in Med Res First published date: December-
[141] D’Amico G, Politi F, Morabito A, D’Antoni A, Guerrera D, Giannuoli G, 27-2015 DOI 10.1177/0962280215623107.
et al. Octreotide compared with placebo in a treatment strategy for
early rebleeding in cirrhosis. A double blind, Randomized pragmatic
trial. Hepatology 1998;28:1206–1214.