Correspondence
Limitation in paediatric
logistic organ
dysfunction score
In their letter about the paediatric
logistic organ dysfunction (PELOD)
score (March 18, p 897),1 the authors
of the original report2 describe a
limitation in their previous calibration
of the score against mortality. That
is, the points given to each criterion
of the score (0, 1, 10, or 20) cannot
add up to a sum with predicted risk of
mortality between 5% and 15% when
using the equation: probability of
death = 1/(1 + exp[7·64–0·30*PELOD
score]). Despite this problem, the
authors still consider the PELOD score
a good measure of severity of illness.1
There seems to be another anomaly
in the PELOD score: the score is not a
continuous variable. Between zero
and 71 (the minimum and maximum
scores) the only possible scores are
0–4, 10–14, 20–24, 30–34, 40–43,
50–53, 60–62, 70 and 71. When
incorporated into the probability-of-
death equation, there is no mortality
prediction between 3·1% and 16·2%
because it is not possible to have
scores between 14 and 20. However,
there is another gap in the score
between 24 and 30, which means
that there is also no risk-of-mortality
prediction between 40% and 80%
(figure).
The PELOD score was developed as
a measurement of severity of illness
to be used in clinical trials. We believe
that many factors now invalidate its
use: the score is discontinuous and
provides no data for patients with
moderate (5–15%) or severe (40–80%)
risk of mortality, its calibration is poor,
and the relation with mortality is non-
linear. So, a 2-point increase in PELOD
score could refer to a 0·1% or a 10%
change in risk of mortality (score 1–3
or 20–22, respectively). Alternatively,
if we are interested in a change in
risk of mortality, then the gaps in
the curve (5–15% and 40–80%) are
problematic.
www.thelancet.com Vol 368 September 30, 2006
We declare that we have no conflict of interest.
Pedro Celiny R Garcia,
Pablo Eulmesekian, Ana Sffogia,
Augusto Perez, *Ricardo Garcia Branco,
Jefferson P Piva, Robert C Tasker
rgb35@cam.ac.uk
Hospital São Lucas da PUCRS, Porto Alegre, Brazil
(PCRG, AS, JPP); Hospital Italiano, Buenos Aires,
Argentina (PE, AP); and University of Cambridge,
Cambridge CB2 2QQ, UK (RGB, RCT)
1 Leteurtre S, Duhamel A, Grandbastien B,
Lacroix J, Leclerc F. Paediatric logistic organ
dysfunction (PELOD) score. Lancet 2006;
367: 897.
2 Leteurtre S, Martinot A, Duhamel A, et al.
Validation of the paediatric logistic organ
dysfunction (PELOD) score: prospective,
observational, multicentre study. Lancet 2003;
362: 192–97.
Authors’ reply
That the PELOD score is discontinuous is
neither a mystery nor a limitation: this is
due to the attribution of discontinuous
points to each organ dysfunction. In
fact, it was planned in the research
protocol that we would try to find clusters
of data for the different variables under
evaluation, and that we would attribute
more points to clusters associated with
more severe dysfunction. That we have
found the clusters and determined the
weights (the points) attributed to each
organ by using real data are strengths of
the PELOD score.1
The gaps in the risk of mortality
cannot be regarded as “problematic”
since they reflect a given population.
With the customised equation
applied to our population,2 there
were 216 patients (12% of 1806)
with a probability of death between
5% and 15%, and 54 (3% of 1806)
with a probability of death between
40% and 80% (the figure presented
by Pedro Celiny Garcia and colleagues
uses the development step predictive
equation3). Also, the fact that
customisation was needed to obtain
a good calibration2 is true for nearly all
scores. Finally, we do not understand
why the relation between a score and
the outcome must be linear:1 this
relation is logistic when scoring sys-
tems, such as many of those proposed
for intensive care units, are developed
by using logistic or Cox regressions.
100
Risk of mortality (%)
80
60
40
20
0
0 5 10 15 20 25 30 35 40 45 50
PELOD score
Figure: Actual possible PELOD scores and the calculated risk of mortality
We still believe that the PELOD score
is the best score presently available
when the purpose is to describe
the severity of illness of critically
ill children. Its excellent area under
the receiver-operating-characteristic
curve proves its good discriminatory
capacity. Evidently, the entire stay
PELOD should never be a mortality
prediction tool since it is calculated at
discharge (remember that the use of
scores to predict death in individuals is
not recommended).
The usefulness of the PELOD score
is already supported by data showing
a good relation between the PELOD
score and other descriptors of severity.
A good relation between the PELOD
score and septic states (sepsis, severe
sepsis, and septic shock) was shown.4
Also, the average admission PELOD
score in children who received a red
blood cell transfusion was significantly
higher than in those who did not
receive a transfusion.5
The PELOD score is a quantitative
score, even though it is not a continuous
variable. It is mainly a descriptive score.
It is a useful measure of the severity of
illness of critically ill children, which is
the purpose for which it was validated.
We declare that we have no conflict of interest.
Stéphane Leteurtre, Alain Duhamel,
Bruno Grandbastien, Jacques Lacroix,
*Francis Leclerc
fleclerc@chru-lille.fr
Réanimation Pédiatrique, Hôpital Jeanne de Flandre,
59037 Lille Cedex, France
1 Dominguez TE, Huh JW. Do we need another
pediatric severity of illness score? Crit Care Med
2005; 33: 1643–45.
1151
 Correspondence
2 Leteurtre S, Martinot A, Duhamel A, et al.
Validation of the paediatric logistic organ
dysfunction (PELOD) score: prospective,
observational, multicentre study. Lancet 2003;
362: 192–97.
3 Leteurtre S, Duhamel A, Grandbastien B,
We do not have the Lacroix J, Leclerc F. Paediatric logistic organ
rights to reproduce dysfunction (PELOD) score. Lancet 2006; 367:
897.
this image on the
4 Leclerc F, Leteurtre S, Duhamel A, et al.
web. Cumulative influence of organ dysfunctions and
septic state on mortality of critically ill children.
Science Photo Library
Am J Respir Crit Care Med 2005; 171: 348–53.
5 Armano R, Gauvin F, Ducruet T, Lacroix J.
Determinants of red blood cell transfusions in a
pediatric critical care unit: a prospective,
descriptive epidemiological study. Crit Care Med
2005; 33: 2637–44.
Malaria and deaths in the
English marshes
I read with dismay the article by
Robert Hutchinson and Steven Lindsay
(June 10, p 1947).1 Their thesis is
that Mary J Dobson2 was wrong to
attribute high mortality rates in the
Kentish marshes in the 16th and 17th
centuries to malaria. Hutchinson and
Lindsay suggest that these high rates
were caused by “poor hygiene and
sanitation”.
Hutchinson and Lindsay back up these
claims by frequent reference to the
publications of S Rickard Christophers
and S P James on the malaria situation
in the Canal Colonies and elsewhere
in the Punjab in and after 1903, on
quinine as prophylactic, and so on.
I have shown3 that everything
Christophers (after 1903) and James
wrote about malaria was deliberately
designed to support UK investors’
claims that malaria was caused by
environmental factors that had nothing
to do with British-financed irrigation
projects (ie, water, waterlogging, and
man-made marshes). The Christophers–
James investment-friendly view was
criticised as medically unsound by
Ronald Ross,4 and by Walter King (after
whom the King research laboratory at
Madras was named) at the All-India
Malaria Conference held at Simla in
1908, and by the League of Nations
Malaria Commission in their report on
their study tour of India in 1929.5
1152
I declare that I have no conflict of interest.
Sheldon Watts
sjwatts@aucegypt.edu
c/o Social Research Center, American University in
Cairo, PO Box 2511, 11511 Cairo, Egypt
1 Hutchinson RA, Lindsay SW. Malaria and deaths
in the English marshes. Lancet 2006; 367:
1947–51.
2 Dobson MJ. Contours of death and disease in
early modern England. Cambridge: Cambridge
University Press, 1997.
3 Watts S. British development policies and
malaria in India 1897–c.1929. Past Present 1999;
165: 141–81.
4 Ross R. A memorandum on the present position
of malaria prevention in India. J Comm Dis 1999;
29: 187–200 [extracts from original 1911
typescript].
5 League of Nations. Report of the Malaria
Commission on its study tour of India 23 August
to 28 December 1929. Geneva: League of
Nations, 1930.
Authors’ reply
In our paper we make one reference
to the classic work of Sir Rickard
Christophers in the Punjab.1 This
report showed that mortality rates
had strikingly increased in an area
of the Punjab where there was an
epidemic of falciparum malaria, but
that several hundred kilometres
away, where there was an epidemic
of vivax malaria, the death rate had
not changed. Even if one was to
reject this evidence, there are still
overwhelming data to show that
vivax malaria rarely kills. To choose
just one example, in Sri Lanka
in 1969 there was an epidemic
of vivax malaria that resulted in
500 000 cases, but no deaths.2 We
continue to maintain that vivax
malaria was not the killer in the
English marshes.
We declare that we have no conflict of interest.
*S W Lindsay, R A Hutchinson
s.w.lindsay@durham.ac.uk
Institute of Ecosystems Science, School of Biological
and Biomedical Sciences, Durham University,
Durham DH1 3LE, UK
1 Christophers SR. Malaria in the Punjab.
Scientific Memoirs by Officers of the Medical
and Sanitary Departments of the
Governments of India, Vol 46. Calcutta:
Superintendent Government Printing,
1911.
2 Gillies H, Warrell D. Bruce-Chwat’s
essential malariology. London: Edward
Arnold, 1993.
www.thelancet.com Vol 368 September 30, 2006
Overview of randomised
trials of ACE inhibitors
We fully agree with Giuseppe Remuzzi
and Piero Ruggenenti (Aug 12, p
555)1 that there is a direct relation
between the global cardiovascular
risk and the cardiovascular benefit
given by angiotensin-converting-
enzyme (ACE) inhibitors in placebo-
controlled trials, and that it is not
a pure blood-pressure-dependent
effect.
When compared with other active
drugs (that decrease blood pressure
slightly more), their benefit is quite
heterogeneous and organ-specific:2
when compared with calcium-
antagonists, ACE inhibitors have a
benefit in heart failure (relative risk
0·82, 95% CI 0·73–0·92) but it is the
reverse for stroke prevention (1·12,
1·01–1·25). When compared with
thiazide, they have no advantage for
heart failure and a lower protective
effect for stroke (1·09, 1·00–1·18).
For end-stage renal disease
(ESRD), when all studies in the Casas
meta-analysis3 are considered, ACE
inhibitors and angiotensin-receptor
blockers significantly decrease its
risk. However this decrease is not
significantly greater in diabetic
than in non-diabetic patients,
whereas patients with diabetes are
presumably at higher cardiovascular
risk. This finding could be caused
by the ALLHAT study in which the
large (n=5944) subgroup of diabetic
patients with baseline glomerular
filtration rate of 60–89 mL/min had
a 74% increased risk of ESRD with
lisinopril compared with thiazide.4 The
lower rate of ESRD in these patients
with unknown proteinuria compared
with macroproteinuric diabetic
patients5 (0·4 vs 4·7% per year) could
account for this observation.
Considering the usual need for at
least two drugs for adequate control
of blood pressure (especially in renal
patients), the claim that ACE inhibitors
have the broadest cardiovascular
protective effect1 would be better