Beruflich Dokumente
Kultur Dokumente
Early report
Summary
Background Current risk-stratification systems for patients
with acute upper-gastrointestinal bleeding discriminate
between patients at high or low risks of dying or rebleeding.
We therefore developed and prospectively validated a risk
score to identify a patients need for treatment.
Methods Our first study used data from 1748 patients
admitted for upper-gastrointestinal haemorrhage. By logistic
regression, we derived a risk score that predicts patients
risks of needing blood transfusion or intervention to control
bleeding, rebleeding, or dying. From this score, we developed
a simplified fast-track screen for use at initial presentation.
In a second study, we prospectively validated this score
using receiver operating characteristic (ROC) curvesa
measure of the validity of a scoring systemand 2
goodness-of-fit testing with data from 197 patients. We also
validated the quicker screening tool.
Findings We calculated risk scores from patients admission
haemoglobin, blood urea, pulse, and systolic blood pressure,
as well as presentation with syncope or melaena, and
evidence of hepatic disease or cardiac failure. The score
discriminated well with a ROC curve area of 092 (95% CI
088095). The score was well calibrated for patients
needing treatment (p=084).
Interpretation Our score identified patients at low or high
risk of needing treatment to manage their bleeding. This
score should assist the clinical management of patients
presenting with upper-gastrointestinal haemorrhage, but
requires external validation.
Lancet 2000; 356: 131821
See Commentary page 1289
1318
Introduction
Acute upper-gastrointestinal haemorrhage is a common
reason for emergency hospital admission in the UK, with
up to 172 admissions per 100 000 adults yearly.1 Patients
who have a history of acute upper-gastrointestinal
bleeding present with a wide range of clinical severity,
ranging from insignificant bleeds to catastrophic
exsanguination.2 Several systems have been designed to
identify patients with high risks of adverse outcomes and
to differentiate them from patients with lower risks.27 The
use of such scores is not unequivocally supported,8,9 and
none have been widely adopted. These measures have
been developed from mathematical models of patients
risks of death or rebleeding. However, since clinical
treatment aims to prevent patients from dying, we believe
that it is more logical to identify which patients will
require clinical intervention rather than which might die.
We therefore developed and validated a risk score to
assess whether patients presenting with acute uppergastrointestinal bleeding will require admission for
treatment to manage their bleeding.
Methods
In our first study, we developed the score with data
collected in an audit of admissions for acute uppergastrointestinal haemorrhage in all 19 hospitals in west
Scotland.1 Data from 1748 patients were used to build a
logistic regression model with the need for treatment as a
response variable. Patients were defined as needing
treatment if they had had a blood transfusion or any
operative or endoscopic intervention to control their
haemorrhage, or if they had undergone no intervention
but had died, rebled, or had a substantial fall in
haemoglobin concentration after admission. The
regression model was built by stepwise selection of
explanatory variablesclinical and laboratory data
obtained at the time of admission. The coefficients
obtained from the logistic regression were multiplied by a
scaling factor to produce a scoring system that required
the addition of integer values, which were associated with
specific risk factors identified at patients initial
assessments.
In our second study, we prospectively validated this risk
scoring system in a group of 197 consecutive adult
patients
admitted
with
upper-gastrointestinal
haemorrhage during a subsequent 3-month period in
three hospitals in west Scotland. One was a teaching
hospital, the other two were district general hospitals. We
excluded patients whose records were incomplete (two) or
whose final outcome could not be ascertained (six). A
single researcher was responsible for identification and
data collection for the patients in the initial audit and
those included in the subsequent score validation.
We assessed the validity of the scoring system by
plotting a receiver operating characteristic (ROC) curve
for the second (validation) set of patients. This curve was
For personal use only. Not to be reproduced without permission of The Lancet.
EARLY REPORT
2
3
4
6
1
3
6
1
6
1
2
3
Other markers
Pulse 100 (per min)
Presentation with melaena
Presentation with syncope
Hepatic disease
Cardiac failure
1
1
2
2
2
Patients (n)
40
35
30
25
20
15
10
5
0
70
60
50
40
30
20
10
0
9 10 11 12 13 14
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Total
No
intervention
needed
Intervention
needed
Predicted
need for
intervention
Intervention
needed
276 (158%)
185 (106%)
115 (66%)
101 (58%)
97 (55%)
72 (41%)
61 (35%)
32 (18%)
14 (08%)
15 (09%)
3 (02%)
5 (03%)
1 (01%)
3 (02%)
0 (0%)
980 (561%)
5 (03%)
11 (06%)
15 (09%)
10 (06%)
30 (17%)
44 (25%)
62 (35%)
85 (49%)
58 (33%)
53 (30%)
77 (44%)
113 (65%)
74 (42%)
55 (31%)
76 (43%)
768 (439%)
06 (03%)
18 (09%)
14 (07%)
12 (06%)
21 (11%)
42 (21%)
71 (36%)
94 (48%)
105 (53%)
31 (16%)
58 (29%)
124 (63%)
89 (45%)
57 (29%)
60 (30%)
802 (407%)
1 (05%)
3 (15%)
1 (05%)
3 (15%)
4 (20%)
4 (20%)
11 (56%)
10 (51%)
10 (51%)
4 (20%)
5 (25%)
12 (61%)
9 (46%)
6 (30%)
6 (30%)
89 (452%)
30
25
20
15
10
5
0
0
4
Score
Results
Table 1 shows the risk markers from the score
development study of 1748 patients who were included in
the final logistic regression model, alongside their
associated score component values. Patients risk scores
were calculated by adding the score components
associated with each clinical risk marker at the initial
presentation. Table 2 shows the score values for the
development and validation groups, and the numbers of
patients needing treatment. The stepwise logistic
regression confirmed that, after the factors in table 1 were
taken into account, patients age, diastolic blood pressure,
creatinine, and urea/creatinine ratio were not significant
predictors of the need for clinical intervention. Fewer
patients with low scores on our risk assessment needed
clinical interventions than those who had low Rockall
admission and postendoscopy scores (figure 1).
Our score was well calibrated for patients who needed
clinical intervention (p=084). The area under the ROC
curve of our score was 092 (95% CI 088095), which
was significantly higher than that for Rockall admission
(071 [064078]) and the full postendoscopy scores
(075 [067083], figure 2). Our score was also more
highly correlated with patients lengths of stay and the
1319
For personal use only. Not to be reproduced without permission of The Lancet.
EARLY REPORT
10
Predicted need of
treatment
09
No
Yes
Total
08
True positives
07
Yes
Total
43
65
108
1
88
89
44
153
197
06
05
04
03
New score
Rockall admission score
Rockall postendoscopy score
02
01
0
0
01 02 03 04 05 06 07 08 09 10
False positives
Discussion
Our score discriminated well between patients who
needed clinical intervention to control uppergastrointestinal haemorrhage and those who did not.
We modelled the score on the process rather than on
the outcome of clinical treatment, thus it identifies
which patients may need clinical treatment, rather
than identifying those at risk of death. Since the aim
of treatment for acute upper-gastrointestinal haemorrhage
is to prevent death, calculation of patients risk of
death is analogous to identifying the risk of treatment
failure.
Rockall scores were developed to predict patients risks
of death or rebleeding.8 Our score better predicted the
need for treatment in this group of patients and correlated
highly with two proxy markers of severity of uppergastrointestinal bleeding. Because blood transfusion
formed part of the combined endpoint used to build our
score, it might be expected that this score would have a
Scoring system
New score
Rockall admission score
Rockall postendoscopy score
0569
0453
0380
0739
0320
0408
*All p<00001.
1320
For personal use only. Not to be reproduced without permission of The Lancet.
EARLY REPORT
Acknowledgments
We thank Lindsay Davidson and members of the upper-gastrointestinal
haemorrhage audit steering group of the Royal College of Physicians and
Surgeons of Glasgow for initiating and supporting the first phase of this
study. We thank also the hospital clinicians for allowing us to access
anonymous patients data and Harper Gilmour and Catriona Smith for
helpful comments.
The first study was supported by the Scottish Office Clinical Resource
and Audit Group (CRAG grant CA91/24) and the second by the Chief
Scientists Office (CSO grant K/OPR/15/7/11).
13
14
15
16
References
1
17
18
19
20
21
22
23
24
25
26
27
28
1321
For personal use only. Not to be reproduced without permission of The Lancet.