Prediction of outcome following
acute variceal haemorrhage
Br. J. Surg. 1985, Vol. 72, February,
91-95
0. J. Garden, H. Motyl,
W. H. Gilmour, R. J. Utley
and D. C. Carter
University Department of Surgery, Royal
Infirmary, Glasgow and Department of
Statistics, University of Glasgow, UK
Correspondence to: Mr 0 . J. Garden,
University Department of Surgery, Royal
Infirmary, Glasgow G31 2ER, UK
I n order to identify factors predicting survival following acute variceal
haemorrhage, data were collected prospectively from 100 admissions in
70 patients managed by a standard policy employing oesophageal
tamponade, injection sclerotherapy and, i f necessary, oesophageal
transection. Of the ten predictive factors identified by univariate analysis,
only prothrombin ratio, serum creatinine and the presence of
encephalopathy on admission were shown by stepwise logistic regression
to have independent significance. T h e derived regression equation
allowed clearer identification than conventional scoring systems of high
and low risk groups and successfully predicted outcome in 90 per cent of
admissions.
Keywords: Portal hypertension, oesophageal varices. injection sclerotherapy
Child’s classification’ and its modified scoring system2 are
widely accepted as the mainstay of assessment of patients with
portal hypertension, and while both classifications have been
used to select patients for particular treatment option^^,^ and to
assess the comparability of patients5-’, neither has been
subjected to critical evaluation other than in patients undergoing portacaval shunting. The present study is intended as an
assessment of the value of various risk factors and scoring
systems in defining risk as soon as possible after admission in
patients with acute variceal haemorrhage.
Patients and methods
During the period 1 September 1979 to 31 October 1982, data were
collected prospectively on 70 patients considered at endoscopy to have
bled from varices and who were admitted on a total of 100 occasions to
the University Department of Surgery, Glasgow Royal Infirmary. These
patients will be referred to subsequently as Group A. Only patients who
had bled within 72 h of admission to the Department were included for
analysis. All patients were managed by a policy of resuscitation with
oesophageal tamponade where necessary, followed by injection sclerotherapy using a modified Negus oesophagoscope under general anaesthesia4. Stapled oesophageal transection was reserved for haemorrhage
which persisted despite sclerotherapy. Liver histology was obtained on
all but four patients in whom a severe coagulopathy prevented biopsy or
in whom post-mortem examination was refused. Admission mortality,
defined as death in hospital within 30 days of admission, was 28 per cent
(Table I ) .
Between 1 September 1982 and 31 January 1984, 26 patients were
admitted on 35 occasions with variceal haemorrhage which was
managed according to the same protocol. There were five deaths in
hospital in this group of patients who will be referred to as Group B
(Table I ) .
Patients were assessed according to Pugh’s modification’ of Child‘s
classification’ (Table2). Age, sex, cause and duration of liver disease, and
the time since first variceal haemorrhage was recorded. The clinician
(O.J.G.) involved in collection of data assessed patients for the presence
of ascites and encephalopathy on admission, grading them as present or
absent. The first recorded value for serum levels of bilirubin (pmol/l),
alanine aminotransferase (units/]), alkaline phosphatase (units/l), urea
(mmol/l), creatinine (pmol/l), total protein (g/I). albumin (g/I), prothrombin ratio, kaolin cephalin clotting ratio, thrombin ratio, haemoglobin (g/dl), white cell and platelet count were recorded.
Student’s t and Mann-Whitney tests for independent samples of data
were used to determine the significanceof differences between the group
of patients who died and those who were discharged in terms of mean
value for each factor. x2 analysis was used to evaluate whether
categorical variables could predict outcome. Stepwise logistic regression
analysis was used to minimize the number of admission factors needed
for optimum separation of patients who survived from those who died.
The analysis was performed using the BMDP computer package
program P7M on an ICL 2988 computer’. The results are expressed
as mean +s.d. for normally distributed data and median and quartile
when data was not normally distributed.
Results
Analysis of mortality
Seventy consecutive patients admitted on 100 occasions (Group
A) had an admission mortality of 28 per cent. The relationship
between the cause of portal hypertension, Child’s classification
and mortality are shown in Tables I and 3. In Group A, 3 out of
29 (10.4 per cent) Grade A and B patients and 25 of 71
(35.2 per cent) Grade C patients died.
Table 4 lists the deaths in both groups. Twenty-two patients
in Group A and two patients in Group B died in liver failure but
this was associated with a significant terminal haemorrhage in
Table 1 Cause of portal hypertension and admission mortality
Group A
No. patients
Alcohol cirrhosis/hepatitis
Primary biliary cirrhosis
Chronic active hepatitis
Cryptogenic cirrhosis
Idiopathic
Portal vein thrombosis
Posthepatitic cirrhosis
Metastatic breast carcinoma
Wilson’s disease
Total
0007-1323/85/02009145$3.00
No. admissions
Group B
No. deaths
No. patients
No. admissions
No. deaths
22
1
1
I
18
0
5
2
0
0
0
0
1
26
22
0
9
3
0
0
0
0
1
35
5
0
49
4
4
4
4
3
72
6
5
5
I
4
0
1
1
2
1
0
100
1
1
0
70
5
@ 1985 Butterworth & Co. (Publishers) Ltd
0
28
0
0
0
0
0
0
0
5
91
Outcome following acute variceal haemorrhage: 0. J. Garden et al.
four patients in Group A and in both patients in Group B. Two
patients died from continued variceal haemorrhage after a
positive decision had been taken to withhold further treatment;
the first was a mentally retarded woman with cryptogenic
cirrhosis and the second an 83-year-old who had experienced
multiple admissions with haematemesis attributable to continued alcohol abuse (Table 4).
Table 2 Pugh's modifcation of Child's classifcation
Points
1
2
3
Nil
Nil
Slight
Slight
Mod.-severe
Mod.-severe
< 34
> 35
34-51
28-35
1.3-1.5
<28
> 1.5
~
Encephalopathy
Ascites
Bilirubin (pmol/l)
Albumin (g/l)
Prothrombin
< 1.3
>51
Potential predictive factors and admission mortality
Group A . Nine individual risk factors obtained from Group
A patients demonstrated a significant association with
admission mortality (Tables 5, 6 and 7). Stepwise logistic
regression analysis showed that only prothrombin ratio (PTR),
serum creatinine (CR) and the presence of encephalopathy
(ENC) (in decreasing order of significance) were independent
predictors of mortality. The derived regression equation allows
estimation of the predicted probability of discharge. The
relationship between the probability of discharge and the three
predictors is given by:
~
Risk: Grade A, 5 4 points; B, 7-9 points; C, 10-15 points
Table 3 Modified Child's classification and admission mortality
Group A
Child's
category
A
B
C
Total
No.
admissions
6
23
71
100
Group B
No.
deaths
No.
admissions
No.
deaths
0
3
25
28
6
7
22
0
0
5
5
35
log(P/l -P)= 10'0-4.3 PTR-0.03 CR-0.85 ENC
where the actual values of PTR and CR are entered and ENC is
coded as - 1 when absent aiid + 1 when present.
The relationship between individual values of P and death or
discharge in Group A patients is shown in Figure l a ( P not
calculated in six patients with unrecorded serum creatinine
values). Decreasing values of P were associated with increasing
risk of death; for example, there were five deaths in 70
Table 4 Characterization of patients dying afer admission with variceal haemorrhage
Patient
admission No.
Cause of portal
hypertension
Child's grade
(Pugh's mod.)
46
61
39
47
52
59
56
44
60
79
36
39
45
40
60
68
42
53
54
51
68
26
38
58
34
66
51
63
AC
AC
AC
AC
AC
PBC
AC
AC
AC
AC
AC
AC
AC
AC
AC
Idiopathic
AC
AC
CAH
AC
PHC
AC
Ca
83
65
46
57
57
Sex
Age
M
M
M
M
M
F
M
M
F
M
M
M
M
F
M
M
M
M
F
M
F
M
F
F
M
M
F
M
M
M
M
M
M
P*
Cause of death
0.09
0.04
AC
AC
AC
AC
C
C
B
C
C
C
C
C
B
B
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
Liver failure, terminal bleed
Liver failure
Terminal bleed during endoscopy
Liver failure
Oesophageal perforation
Liver failure
Liver failure, terminal bleed
LVF after oesophageal transection
Liver failure
Myocardial infarction
Liver failure, terminal bleed
Liver failure
Liver failure
Liver failure
Liver failure
Liver failure
Liver failure
Liver failure
LVF after oesophageal transection
Liver failure
Liver failure
Liver failure
Liver failure
Treatment withdrawn
Liver failure
Liver failure
Liver failure
Liver failure
AC
AC
AC
AC
AC
C
C
C
C
C
0.26
0.21
Group A
7
8
9
10
16
20
23
31
33
34
35
39
42
43
50
51
52
53
55
60
64
72
77
87
93
95
97
100
Group B
102
105
106
111
129
cc
-
0.66
0.95t
0.17
-
040t
-
0.95t
0.02
0.02
0.17
0.19
010
0.13
0.06
0.04
0.90t
0.02
0.45
0.17
0.05
0.75t
0.00
0.00
0.26
001
000
0.59
0.75t
Treatment withdrawn
Myocardial infarction
Liver failure, terminal bleed
Liver failure, terminal bleed
Myocardial infarction
AC, alcoholic cirrhosis; PBC, primary biliary cirrhosis; PHC, posthepatitic cirrhosis; CC, cryptogenic cirrhosis; CAH, chronic active hepatitis: Ca,
carcinoma
* P=probability of discharge; see text for calculation of probability values
t Death occurring in low risk group
92
Br. J. Surg.. Vol. 72, No. 2, February1985
Outcome following acute variceal haemorrhage: 0. J. Garden at al.
admissions (7 per cent admission mortality) when P>O.66 (low
risk), and there were 20 deaths in 24 admissions (83 per cent
admission mortality) when P < 0-66 (high risk). The five deaths in
the low risk group are among those listed in Table 4. In all five
cases, death was not obviously related to the degree of liver
failure. The outcome following acute variceal haemorrhage was
correctly predicted in 90 per cent of Group A admissions using
this P value of 0-66as an arbitrary discriminant between low and
high risk groups.
Group B. Using the regression equation derived from Group
A patients, the individual value of P for each of the 35 patient
admissions was determined (Figure Zb). There were four deaths
in eight admissions when P x 0 . 6 6 , although one of the four
survivors died in liver failure within a week of discharge from
hospital. All but one ofthe 24 patients with P > 0.66 survived and
the one death occurred in a patient who developed a cardiac
arrest 24 h following injection sclerotherapy.
Discussion
Mortality following acute variceal haemorrhage in the 100
patient admissions of Group A was 28 per cent and we have
shown that nine of the 21 potential risk factors in this group had
a significant association with admission mortality. Stepwise
logistic regression identified prothrombin ratio, serum
creatinine and the presence of encephalopathy (in decreasing
order of significance) as having independent significance in
predicting admission mortality. The lower admission mortality
of 14 per cent in the Group B series of patients cannot be
explained by any alteration in management but may relate in
part to the relatively smaller proportion of Grade C patients.
Analysis of outcome in this second series of patients using the
derived regression equation suggests that the prediction method
is soundly based although experience with more high risk
patients will be required to sustain this conclusion. The total
Table 5 Relationship between values of normally distributed individual
variables and outcome of admission
Discharge
Death
Factor
n Mean ks.d.
n
Mean ks.d.
PTR
KCCR
Haemoglobin (g/dl)
Age (years)
Albumin (g/l)
Protein (g/l)
72
70
72
72
72
72
28
26
28
28
28
28
1.8
1.6
11.6
51.6
304
62.5
1.3
1.2
105
51.4
32.4
63.3
0.2
0.2
2.1
14.7
53
9.7
0.4
0.4
2.7
12.0
6.7
7.4
P
<OW1
iO.001
<0.05
n.s.
n.s.
ns.
PTR, prothrombin ratio; KCCR, kaolin cephalin clotting ratio
number of admissions rather than the actual number of patients
has been analysed because outcome was not always the same on
each admission. Although patients readmitted with variceal
haemorrhage might be expected to have an improved outcome
over initial admissions, univariate analysis showed that the
number of previous bleeds did not have a significant effect on
outcome.
The present study highlights the predictive value of prothrombin ratios at admission and is in keeping with its arbitrary
inclusion in Pugh’s modification of Child’s original classification. There is debate as to whether prothrombin time can predict
outcomeafter electiveportacaval shunting”.’ I , but a prolonged
prothrombin time has been shown to have an independently
significant association with admission mortality in alcoholic
hepatitis”. It is of course probable that the value of particular
predictive factors is influenced in the same way that the results of
treatment of variceal haemorrhage are influenced by the method
of selection of patients and the timing of their entry to the
programme of treatment13.
The present study has shown that the inclusion of serum
albumin and bilirubin (or jaundice) as in Child‘s and Pugh’s
classifications does not enhance prediction in our population of
patients when the severity of liver disease is assessed in the
period immediately following acute variceal haemorrhage.
Although Simert and colleague^'^ have suggested that bilirubin
and albumin are of value in determining early and late survival
respectively following elective portacaval shunting, we have
shown that there was no association between serum albumin
and admission mortality and that serum bilirubin had no
Table I Relationship of categorical variables and outcome in Group A
patients
No. of
admissions
Died
Sex
Male
Female
69
31
20
8
0.1 1
Cause of portal hypertension
Alcohol
Other
72
28
22
6
0.83
Ascites
Absent
Present
19
81
1
27
6.02*
Encephalopath y
Absent
Present
60
40
9
19
12.57t
x2
* P <0.05
t P <0~001
Table 6 Relationship between values of individual variables not normally distributed and outcome of admission
Discharge
Creatinine (mmol/l)
Bilirubin (pmol/l)
Urea (mmol/l)
MFB (months)
MLD (months)
White cell count ( x i09/i)
Platelets ( x 109/1)
Thrombin ratio
Alanine aminotransferase (units/l)
Alkaline phosphatase (units/])
Number of bleeds
Death
n
Median
Quartile
n
Median
Quartile
P
69
72
69
72
72
71
72
70
72
72
72
75
55
7.4
4.0
15
8.1
105
1.o
32
240
1
60,90
20, 110
4.9, 10.0
0, 18
2, 60
5 3 , 10.7
70, 150
1.0, 1.1
20,46
170, 355
0, 3
25
28
25
28
28
27
28
26
28
28
28
105
165
9.5
0
6
12.7
118
1.o
50
232
0
90, 150
78, 324
60, 12.9
0.5
0, 30
6.6, 18.4
74, 190
1.0, 1.2
28, 64
157,428
0, 2
0~001
0.005
0.05
0.05
n.s.
ns.
ns.
n.s.
n.s.
n.s.
ns.
MFB, months since first variceal haernorrhage; MLD, duration of liver disease in months
Br. J. Surg., Vol. 72, No. 2, February1985
93
Outcome following acute variceal haemorrhage: 0. J. Garden et al.
.
MSCHARGED
0.
0
02
04
.
I
I
I
I
I
I
.I
...I
06
..
.......0
me.
me.
0..
0..
m
..
.mom.
.me.
08
10
P
b
DIED
DISCHARGED
.
1:
independent significance. Maddrey and othersL2have shown
that serum bilirubin is independently associated with admission
mortality in alcoholic hepatitis. Since changes in kaolin cephalin
clotting ratio closely match those of prothrombin ratio, serum
creatinine has been identified by our study as the second most
important predictive factor by regression analysis. Although
creatinine is of value in predicting outcome following surgery for
obstructive jaundiceL5, its predictive value in portal hypertension has not previously been noted.
Cello and colleagues16 have claimed that Child’s original
classification, which included a clinical assessment of nutrition,
ascites and encephalopathy, was the most important factor in
determining early mortality after portacaval shunt. In the
present study no attempt has been made to grade the severity of
ascites and encephalopathy but we have confirmed that the
presence of these two subjective clinical parameters is associated
with a poor outcome. However, only encephalopathy aids
prediction of outcome when prothrombin ratio and serum
creatinine have been taken into consideration. The usefulness of
Child’s grading in predicting admission mortality has been
questioned by Simert and colleague^'^. Campbell and coworkers’ have described a scoring system to grade the severity
of the five individual factors described by Child but allocation to
high and low risk groups proved correct in terms of admission
mortality in only 62 per cent of their total patient population.
The addition of other pre-operative parameters did not improve
the predictive value of this scoring system.
Attempts have been made by others to use invasive measurements as a means of predicting outcome in patients undergoing
surgical decompression of the portal venous system. It has been
suggested that appearances at splenoportography1s.19,portal
venous pressure2’, portohepatic pressure
and
wedged hepatic blood flow” may be of value in selection of
patients and determination of outcome. However, other^^^,^^
have questioned haemodynamic selection since it has failed to
improve operative mortality o r long-term survival in shunted
patients.
With regard to the predictive value of liver histology, Kanel
and colleagues24 found no association between histological
evidence of continuing alcoholic hepatitis and survival in
patients with cirrhosis. O n the other hand, Grendell and cow o r k e r ~using
~ ~ linear logistic regression analysis found the
presence of panlobular fat taken in association with haematocrit
could be used to predict outcome in 79 per cent of patients
about to undergo portasystemic shunting. This two-variable
combination was not improved by addition of prothrombin
time. We have not used histological features to predict outcome
since liver biopsy on admission may be contra-indicated in the
presence of a prolonged prothrombin time.
The logistic regression equation obtained from the 100
Group A admissions has clearly identified a high and low risk
group of patients based on three variables which can be readily
obtained within a few hours of admission to hospital. The value
of these variables has been verified in a second independent
group of patients and their use in assessing outcome following
variceal haemorrhage appears to have considerable advantage
over Child’s and Pugh’s classifications. The use of such
classifications is clearly of limited value in our own practice
when no less than 71 per cent of our patients with variceal
haemorrhage are consigned to Child’s C grade when first seen.
It is our intention to continue to evaluate this method of
analysis once the patient’s clinical condition has stabilized
following the initial bleed and to assess its predictive value
within specific populations of patients since certain predictive
factors may assume greater importance in patients such as those
with alcoholic cirrhosis. We intend to employ this analysis in a
continuing audit of patients admitted with variceal haemorrhage and it is conceivable that it may offer a useful means of
selection for entry to clinical trials. It is also arguable whether it
can be used to identify that group of patients whose probability
of survival is so low that they should be denied active treatment.
Acknowledgements
The authors wish to express their thanks to clinicians who referred
patients for management and to the medical and nursing staff involved in
the care of these patients. The authors acknowledge the secretarial
assistance of Miss A. McKellar.
References
1.
2.
3.
4.
5.
’
94
6.
7.
8.
9.
10.
Child CG 111, Turcotte JG. Surgery and portal hypertension. In:
Child CG, ed. The Liver and Portal Hypertension. Philadelphia:
W. B. Saunders, 1964: 50.
Pugh RNH, Murray-Lyon IM, Dawson JL et al. Transection of
the oesophagus for bleeding oesophageal varices. Br J Surg 1973;
60: 646-9.
Johnston GW. Bleeding oesophageal varices: the management of
shunt rejects. Ann R Coil Surg Engl 1981; 63: 3-8.
Garden OJ, Osborne DH, Blamey SL, Carter DC. The management of acute variceal haemorrhage. Aust N Z J Surg 1983; 53:
197-202.
MacDougall BRD, Westaby D, Theodossi A et al. Increased
longterm survival in variceal haernorrhage using injection sclerotherapy. Lancet 1982; i : 124-7.
Burroughs AK, Bass NM, Osborne D. Randomised controlled
study of transhepatic obliteration of varices and oesophageal
stapling transection in uncontrolled variceal haemorrhage. Liver
1983; 3 : 122-8.
Burroughs AK, Jenkins WJ, Sherlock S et al. Controlled trial of
propranolol for the prevention of recurrent variceal haemorrhage
in patients with cirrhosis. New Engl J Med 1983; 309: 153942.
Terblanche J, Bornman PC. Kahn D et al. Failure of repeated
injection sclerotherapy to improve long-term survival after
oesophageal variceal bleeding. Lancer 1983; i: 1328-32.
Dixon WJ, Brown MB, eds. BMDP Biomedical Cornpurer
Programs P-Series 1977. University of California Press: 71 1-33.
Malt RB. Malt RA. Tests and management affecting survival after
Br. J. Surg., Vol. 72, No. 2, February 1985
Outcome following acute variceal haemorrhage: 0. J. Garden et al.
11.
12.
13.
14.
15.
16.
17.
18.
portacaval and splenorenal shunts. Surg Gynecol Obstet 1979;
149: 220-4.
Orloff MJ, Bell RH, Hyde PV, Skivolocki WP. Long-term results
of emergency portacaval shunt for bleeding esophageal varices in
unselected patients with alcoholic cirrhosis. Ann Surg 1980; 192:
32540.
Maddrey WC, Boitnott JK, Bedine MS et al. Corticosteroid
therapy of alcoholic hepatitis. Gastroenrerology 1978; 75: 193-9.
Smith JL, Graham DY. Variceal haemorrhage-a critical
evaluation of survival analysis. Gastroenrerology 1982; 82:
968-7 3.
Simert G. Persson T, Vang J. Factors predicting survival after
portacaval shunt: a multiple linear regression analysis. Ann Surg
1978; 187: 174-8.
Blarney SL, Fearon KCH, Gilmour WH, Osborne DH, Carter
DC. Prediction of risk in biliary surgery. Br J Surg 1983; 70:
535-8.
Cello JP, Deveney KE, Trunkey D D et al. Factors influencing
survival after therapeutic shunts: results of a discriminant
function and linear logistic regressions analysis. A m J Surg 1981;
141 : 25765.
Campbell DP, Parker DE, Anagnostopoulos CE. Survival
prediction in portacaval shunts: a computerized statistical
analysis. A m J Surg 1973; 126: 748-51.
Viamonte M, Warren WD. Fomon JJ, Martiney LO. Angiographic investigations in portal hypertension. Surg Gynecol
19.
20.
21.
22.
23.
24.
25.
Obstet 1970; 130: 37-53.
Warren WD, Fomon JJ, Viamonte M, Zeppa R. Preoperative
assessment of portal hypertension. Ann Surg 1967; 165: 999-1012.
McDermott WV Jr. Evaluation of the haemodynamics of portal
hypertension in the selection of patients for shunt surgery. Ann
Surg 1972; 176: 449-56.
Vine1 JP, Cassigneul J, Louis A et al. Clinical and prognostic
significance of portohepatic gradient in patients with cirrhosis.
Surg Gynecol Obstet 1982; 155: 347-52.
Smith GW. Use of haemodynamic selection criteria in the
management of cirrhotic patients with portal hypertension. Ann
Surg 1974; 179: 782-90.
Burchell AR, Moreno AH, Panke WF, Nealon TF Jr. Hemodynamic variables and prognosis following portacaval shunts.
Surg Gynecol Obstet 1974; 138: 35949.
Kanel GC, Kaplan MM, Zawacki JK, Callow AD. Survival in
patients with postnecrotic cirrhosis and Laennec’s cirrhosis
undergoing therapeutic portacaval shunt. Gastroenterology 1977;
73: 679-83.
Grendell JH, Cello JP, Margaretten W, Heilbron DC. Impact of
preshunt liver histology on survival following portasystemic
shunt surgery for bleeding esophageal varices. Dig Dis Sci 1983;
28: 44-55.
Paper accepted 14 August 1984
Announcements
The Publishers are pleased to announce that as from the April 1985 issue the senior author of each Paper, Review, Short Note or Case Report will receive
25 free offprints.Because these offprints will be produced at the time ofprinting the journal, they may include other authors’material either before or alter
their own.
Authors who would like to order reprints of their articles inside a special full colour cover may do so by completing the order form enclosed with their
proofs, which should then be sent with the appropriate sterling remittance to the Reprints Department, Butterworth Scientific Limited, PO Box 63,
Westbury House, Bury Street. Guildford, Surrey GU2 5BH.
Br. J. Surg., Vol. 72, No. 2, February 1985
95