COMMENTARY

Linear growth retardation and cognition There is now consistent evidence of growth retarded
See page 564
children having cognitive deficits but it would be a mistake
to attribute the entire deficit to poor nutrition or infection.
About 33% of children aged under 5 years in developing A major problem with all observational studies is the
countries have moderate or severe linear growth difficulty in controlling for the multitude of adverse
retardation, and the prevalence is over 50% in some least- environmental factors associated with IQ and both growth
developed countries.1 The possible link between linear retardation and infections that also detrimentally affect
growth retardation and poor development is therefore of child development. Berkman and colleagues controlled
immense importance both to the individuals concerned and for several factors but there remain others, such as
to national development. stimulation in the home, which could account for some
Several studies from Guatemala,2 the Philippines,3 or all of the relation between growth retardation and
Jamaica,4,5 and Kenya6 have shown associations between infections. In a randomised controlled trial12 growth-
height in the first 3 years and a wide range of deficits in retarded childrens development benefited from 2 years of
school performance and/or cognition in later childhood or supplementation, which suggested that poor nutrition
adolescence. The school-performance measures include caused at least some of their developmental deficit. In the
literacy, numeracy, general knowledge, maximum school trial, only the children who received both stimulation
grade reached,2 school achievement in girls,5 and school and supplementation caught up with the non-growth-
drop-out, grade repetition, and absenteeism.3 Cognitive retarded comparison group. Furthermore, benefits from
deficits were found in IQ,4 reasoning,2,3,4,6 vocabulary,4,6 and supplementation were no longer apparent at 11 years of
verbal analogies, visual-spatial and auditory age, although benefits from stimulation remained.4
working memory, sustained attention, and information Linear growth retardation clearly represents an enormous
processing.4 waste of potential in millions of children in developing
The report by Douglas Berkman and colleagues in this countries, and urgent attention needs to be paid to both
issue of The Lancet adds to the accumulating evidence of their nutrition and stimulation. The difficulties in obtaining
the long-term effects of linear growth retardation. These sustained benefits from supplementation indicates that
investigators report that severe growth retardation (height- prevention must be the goal.
for-age 3 SD of the National Center for Health Statistics
Sally Grantham-McGregor
references7) in Peruvian children between 12 and
Centre for International Child Health, Institute of Child Health, London
24 months of age was associated with a ten-point deficit in WC1N 1EH, UK
IQ on the Wechsler Intelligence Scale for Children-Revised (e-mail: [email protected])
(WISC)8 at 9 years. The size of the deficit approaches 1 SD
and would be clinically important. This finding is
remarkably similar to the findings from a Jamaican study4,9 1 UNICEF. The state of the worlds children 2001. Oxford: Oxford
University Press, 2001.
in which 46 children aged 11 years, who were severely 2 Martorell R, Rivera J, Kaplowitz J, Pollitt E. Long term consequences of
growth retarded before the age of 2 years, had a deficit of growth retardation during early childhood. In: Hernandez M, Argenta J,
99 IQ points on the same test. Furthermore, moderately eds. Human growth: basic and clinical aspects. Amsterdam: Elsevier,
growth-retarded children (n=70) had a 6-point deficit, 1992: 14349.
which suggests a dose response. However, the WISC was 3 Mendez MA, Adair LS. Severity and timing of stunting in the first two
years of life affect performance on cognitive tests in late childhood.
standardised in the USA, and its robust psychometric J Nutr 1999; 129: 155562.
properties may not be the same in Peru and Jamaica. 4 Walker SP, Grantham-McGregor SM, Powell CA, Chang SM. Effects
Berkman and colleagues examined the effect of age of of growth restriction in early childhood on growth, IQ and cognition at
onset, prevalence, and persistence of growth retardation. age 11-12 years and the benefits of nutritional supplementation and
psychosocial stimulation. J Pediat 2000; 137: 36-41.
However, in too few children was growth retarded before
5 Richardson SA. Severity of malnutrition in infancy and its relation to
6 months and after 18 months for these effects to be later intelligence. In: Brozek J, ed. Behavioral effects of energy and
examined with any confidence. In a much larger study protein deficits. Washington DC: National Institute of Arthritis,
in the Philippines3 onset of growth retardation before Metabolism, and Digestive Diseases. US Department of Health,
6 months of age and persistence of the retardation to Education, and Welfare, 1979: 17284.
6 Sigman M, McDonald MA, Neumann C, Bwibo N. Prediction of
8 years were related to cognitive function at 8 and 11 years. cognitive competence in Kenyan children from toddler nutrition, family
However, their effect seemed to be mediated through characteristics and abilities. J Child Psychol Psychiatry 1991;
degree of the retardation at 24 months, and when severity 32: 30720.
was taken into account, early onset and persistence of 7 Hamill PV, Drizd TA, Johnson CL, Reed RB, Roche AF. NCHS
growth retardation were no longer significant. growth curves for children birth-18 years. United States. Vital Health
Stat 11 1977; 165: 174.
Growth-retarded children have an increased likelihood of 8 Wechsler D. Wechsler intelligence scale for children, revised. New York:
infections, particularly diarrhoea. An important finding of The Psychological Corporation; 1974.
the Peruvian study is that the effects of growth retardation 9 Walker SP, Chang SM, Powell CA. Psychosocial consequences of early
were independent of the effects of diarrhoea. It is surprising childhood growth retardation. In: Martorell R, Hashcke F, eds.
that the investigators also found that infections with Giardia Nutrition and growth. Nestle Nutrition Workshop series, Pediatric
Program vol 47, Philacelphia: Lippincott Williams & Wilkins, 2001:
lamblia had an adverse effect on cognitive scores, which was 24154.
independent of the effects of diarrhoea and growth 10 Grantham-McGregor SM, Fernald LC. The effects of health and
retardation. It is difficult to hypothesise a mechanism nutrition on cognitive and behavioural development in children
linking the two, and Berkman and colleagues suggest that in the first three years of life. Part 2: Infections and micronutrient
deficiencies: iodine, iron and zinc. Food Nutr Bull 1999; 20:
the infection could be a marker for other environmental 7699.
conditions, or be related to micronutrient deficiencies and 11 Neumann C, McDonald MA, Sigman M, Bwibo N, Marquardt M.
fat malabsorption. Another hypothesis is that the chronic Relationships between morbidity and development in mildly to
inflammatory response might affect cognition. Previous moderately malnourished Kenyan toddlers. Pediat 1991; 88:
studies have found concurrent associations between 93442.
12 Grantham McGregor SM, Powell CA, Walker SP, Himes JH.
diarrhoea and other infections with childrens Nutritional supplementation, psychosocial stimulation, and mental
development10 but few have examined longitudinal development of stunted children: the Jamaican Study. Lancet 1991;
associations.11 338: 15.

542 THE LANCET Vol 359 February 16, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES

Effects of stunting, diarrhoeal disease, and parasitic infection

during infancy on cognition in late childhood: a follow-up study

Douglas S Berkman, Andres G Lescano, Robert H Gilman, Sonia L Lopez, Maureen M Black

Summary Introduction
Cognitive function in children is affected by environ-
Background Chronic malnutrition during infancy, marked by mental and health-related factors.13 Risk factors that
stunting, has been associated with poor cognitive function. interfere with cognitive function are especially important
We assessed the effect of stunting, diarrhoeal disease, and during infancy because the first 2 years of life are an
parasitic infections during infancy on cognitive function in essential period of rapid growth and development.
late childhood. Chronic malnutrition during infancy, marked by
stunting, has been related to poor cognitive function in
Methods We followed up from birth to 2 years, a cohort of late childhood in the Philippines,4 and in Jamaica,5
239 Peruvian children for anthropometrics, stool samples, although the mechanisms are not fully understood.
and diarrhoeal status. At 9 years of age, we assessed Diarrhoea is both a cause and an effect of malnutrition,
cognitive function in 143 (69%) with the full-scale and can lead to linear growth retardation.68
intelligence quotient of the Wechsler intelligence scale for Cryptosporidium parvum and Giardia lamblia, two
children-revised (WISC-R). common enteric infections often associated with
diarrhoea, adversely affect growth and nutritional status,
Findings All findings were adjusted for socioeconomic status especially during infancy.914 Research on the relation
and schooling; in addition, findings related to diarrhoea between cognitive function in children and diarrhoeal
prevalence, Giardia lamblia, and Cryptosporidium parvum disease during infancy is limited. Guerrant and
were adjusted for severe stunting. During the first 2 years of colleagues15 noted that diarrhoea during infancy was
life, 46 (32%) of 143 children were stunted. Children with negatively correlated with three tests of cognitive function
severe stunting in the second year of life scored 10 points in Brazilian children. However, in this study, neither
lower on the WISC-R test (95% CI 24175) than children environmental nor health-related factors were
without severe stunting. Children with more than one episode considered, thus raising the possibility that the
of G lamblia per year scored 41 points (0280) lower than associations seen could be attributed to confounding.
children with one episode or fewer per year. Neither diar- The complex interrelations among malnutrition,
rhoea prevalence nor C parvum infection was associated with diarrhoeal disease, and environmental factors such as
WISC-R scores. socioeconomic status and education, make it difficult to
determine the unique contribution of either malnutrition
Interpretation Malnutrition in early childhood, indexed by or diarrhoeal disease to cognitive development. For
stunting, and potentially G lamblia, are associated with poor instance, because diarrhoeal disease can result in
cognitive function at age 9 years. If the observed retarded growth,6,10,16 stunting becomes a potential
associations are causal, then intervention programmes confounder in the relation between diarrhoea and
designed to prevent malnutrition and G lamblia early in life cognitive function.
could lead to significant improvement in cognitive function of We aimed to examine the associations between health-
children in similar lower-income communities throughout the related factors during infancy and cognitive function in
less-developed world. late childhood. First, we extended previous research that
has linked stunting during infancy with poor cognitive
Lancet 2002; 359: 56471 function in late childhood by investigating multiple
See Commentary page 542 classifications of stunting and using a comprehensive,
standardised, and widely accepted measure of cognitive
function. Second, we assessed the independent effects of
diarrhoeal disease, and C parvum and G lamblia
infections during infancy on cognitive test scores in late
childhood, after adjusting for stunting, schooling, and
socioeconomic factors. If diarrhoeal disease during
infancy is associated with cognitive function at 9-years-
old after controlling for nutritional and environmental
factors, our understanding of the long-term burden
Departments of Epidemiology (D S Berkman ScM) and International associated with diarrhoea and the health-related
Health (A G Lescano MHS, Prof R H Gilman MD), The Johns Hopkins determinants of cognitive function will be greatly
School of Public Health, Baltimore, MD, USA; Proyectos de improved.
Informtica, Salud, Medicina y Agricultura (A B PRISMA), Lima,
Per (D S Berkman, A G Lescano, R H Gilman, Patients and methods
S L Lopez RN); Department of Pediatrics, University of Maryland Study setting
School of Medicine, Baltimore (M M Black PhD); Universidad This study is a follow-up assessment of a cohort of
Peruana Cayetano Heredia, Lima, Per (A G Lescano, R H Gilman) children who were examined longitudinally from
Correspondence to: Dr Robert H Gilman, Department of International 198991. The study was done in Pampas de San Juan de
Health, John Hopkins School of Public Health, 615 North Wolfe Miraflores, a periurban shanty town in the outskirts of
Street, Room W3503, Baltimore, MD 21205, USA Lima, Peru. In 1989, most families in this community
(e-mail: [email protected]) lived in houses constructed from woven thatch supported

564 THE LANCET Vol 359 February 16, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet Publishing Group.
 ARTICLES

by wooden poles. Less than a half of families had latrines wooden platform and sliding footboard. Diarrhoea was
or sewage connections, and water was largely supplied by surveyed twice weekly. Field workers obtained
municipal trucks and stored in cisterns or barrels that information from the mother or primary caretaker of the
often became contaminated with faecal matter.17 Previous child for the day of the interview and the 3 previous days.
studies11,18,19 have described the community in more For each day they inquired about the number of liquid or
detail. Over the past decade, the population has grown to semi-liquid stools the child had passed.11 We collected
40 000 people and the emigration rate has decreased. stool samples weekly and examined for C parvum and
Today, most homes are constructed at least partly from G lamblia as previously described.11,2022 Helminths and
brick, and have piped water and basic plumbing. Families bacterial and viral enteropathogens were not determined.
in the community are similar in socioeconomic status, When the children were 9-years-old, cognitive function
housing quality, and access to public utilities.11 A survey was assessed with the Wechsler intelligence scale for
in 2000 reported a median income for heads of children-revised (WISC-R),23 translated to Spanish, and
households of US$2100 per year (range $850$4600; used routinely in local hospitals. The WISC-R,
unpublished data). standardised on 2200 children, has excellent
psychometric properties. The full scale intelligence
Enrolment quotient has a reliability coefficient of 096, an SE of
Between September, 1989, and March, 1991, 253 measurement of 319, and a test-retest stability
mothers in their final trimester of pregnancy were coefficient of 095.23 Validity studies with tests of
randomly selected from the censused community and intelligence, receptive vocabulary, and achievement, as
asked to participate in a study to describe the effect of well as school grades, yield concurrent validity
C parvum infection on nutritional status.12 13 mothers coefficients from the upper 030s to low 080s.23
declined, and one child died of pneumonia early in the Six psychology interns, trained and supervised by a
study.12 The remaining 239 children (94%) were followed licensed psychologist from the Department of Psychology
up prospectively from birth until November, 1991, for of the Hospital Nacional Hipolito Unanue, Lima, Peru,
anthropometry and diarrhoeal illness. Census administered the WISC-R. We assigned children to a test
information was obtained by A B PRISMA, a Peruvian administrator using a random number table;
non-governmental organisation working in the administrators were unaware of each childs previous
community. diarrhoeal status and anthropometric assessment. One
In June, 1999, we selected all children from the interviewer had significantly higher scores (56, p=0047)
original cohort who had complete longitudinal data for at and two interviewers had significantly lower scores (51,
least 6 months from birth (210 [88%] of 239). This p=0029) compared with the other three interviewers.
criterion reduced the risk of misclassifying the exposures We therefore created a variable with three categories to
of interest (stunting, diarrhoeal disease, and parasitic adjust for these differences. This variable remained
infections), while maintaining as large a sample as significant in all our models and explained about 85% of
possible. 145 (69%) of the 210 selected children were the total variance.
located in the community, all agreed to participate in a Weight in late childhood was measured with a Soehnle
cognitive assessment, which were completed in 143 scale (Murrhardt, Germany). Height was measured with
(99%) children. There were no significant differences a locally constructed wooden backboard and sliding
between the 145 study participants and the 65 children headboard. A socioeconomic survey was administered,
who were not located, when compared by stunting and information on schooling was obtained at the end of
prevalence (p=021), proportion of days with diarrhoea the academic year. The children were offered free clinical
(p=035), proportion of C parvum-positive stool samples examinations that included vision and hearing screening.
(p=045), or by socioeconomic factors, including total None of these tests showed important deficiencies in
weekly expenses (p=045), weekly food expenses either hearing or vision.
(p=044), water quality (p=087), or floor type (p=019).
The proportion of giardia-positive stool samples was Variables analysed
greater in the children enrolled in the study (141% vs We defined grade level in school as the grade of the child
84%, p<001). However, study children were followed when the cognitive test was administered. Additional
up longer (176 months vs 132 months, p<0001) and information on school type (public or private), class size
therefore observed through periods of higher risk than (small class defined as <30 students), kindergarten and
children who were not located. When the comparability daycare attendance, whether the child worked, age of
analysis was restricted to the first 6 months of life, for school entry, and grade retentions was obtained. Study
which both study participants (145) and losses to follow- habits were described by inquiring whether the child
up (65) had complete surveillance, the proportion of completed homework assignments on a regular and
giardia-positive stool samples between the two groups consistent basis.
was no longer significant (p=013). Housing conditions were defined according to floor
The longitudinal surveillance and follow-up type (cement vs soil), water source (direct pipe
assessment were separately approved by the Committee connection, a neighbour, or other), and stool disposal
of Human Research of the Johns Hopkins University method (bathroom with plumbing, hole in the ground
School of Hygiene and Public Health and by the ethics (silo), or other). Information was gathered about
committee of A B PRISMA. maternal birthplace (coast, highlands, or jungle), whether
either parent had completed some college education
Data collection (college or less), the language spoken (Spanish or
We followed up children from September, 1989, until Quechua), the mothers parity at the birth of this child,
November, 1991, for anthropometry and diarrhoeal and the profession of the main source of income
illness. Field workers measured each childs height and (technical employee, construction worker, skilled worker,
weight about every 30 days. Weight was measured with street seller, chauffeur, small business owner, or
Salter scales (Salter Housewares Ltd, Tonbridge, independent). Regarding the primary caregiver, we
England). Length was measured with a locally made questioned the mother or guardian about who cared for

THE LANCET Vol 359 February 16, 2002 www.thelancet.com 565

For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES

the child on a daily basis during infancy (mother, father, cognitive test scores; significance of the contribution of
sibling, grandparent, or other relative) and whether the each variable was assessed with the F test after
father lived in the home during this period. adjustment for the effects of variables already included in
Standardised height-for-age and weight-for-height Z the model. We used the Wald test for the comparison of
scores were calculated with the EPINUT program in WISC-R scores for categorical variables with more than
EpiInfo (version 6.04b). We defined stunting as an two categories. Regression coefficients were used to
height-for-age Z score of less than 2; wasting was estimate the change in WISC-R scores between levels for
defined as weight-for-height Z score of less than 2, on categorical variables and for a one unit increase in
the basis of WHO reference data. continuous variables. We constructed multivariate
We assessed age of the first stunting event by 6-month models to adjust for the effects of socioeconomic status
intervals (05, 611, 1217, 1824 months). Severity of and education on cognitive test scores. Adjustment
stunting was classified as moderate (3 height-for-age variables were selected by sequentially adding significant
<2) and severe (height-for-age <3). Persistence of predictors from the univariate analysis into the model.
stunting was defined as persistent (stunted at least once We manually rotated the order of the variables to identify
during infancy and also in late childhood), catch-up interactions and collinearity.
(stunted at least once during infancy but not in late Continuous variables were truncated into fixed-size
childhood), late incident (stunted for the first time in late intervals and analysed as categorical variables to identify
childhood), and never stunted. We calculated stunting non-linear associations. If a monotonic trend or U-
prevalence as the number of anthropometric shaped relation was recorded between WISC-R scores
measurements classified as stunted divided by the total and a particular covariate, an attempt was then made to
number of measurements. This prevalence rate can be combine categories that had similar WISC-R average
interpreted as the number of months stunted per year. scores.
Wasting was analysed as both a continuous and Several variables were highly correlated and
dichotomous variable. Age at first wasting and wasting represented similar domains relating to cognitive
prevalence were also calculated. These expressions were function (eg, maternal and paternal education; school
then analysed by age intervals, but no associations with type and class size). In these cases, we chose the variable
WISC-R scores were recorded. that explained the largest percentage of the variance and
A day of diarrhoea was defined as one in which a child enhanced the strength of associations of the other
passed three or more liquid or semi-liquid stools. variables in the model.
Diarrhoea episodes were defined as at least 1 day of The presence of interactions in the final model was
diarrhoea followed by 2 or more diarrhoea-free days.16 tested for by introducing products of the variables by
Diarrhoea incidence rates were calculated by dividing the pairs into the model. The internal validity of our results
number of diarrhoea episodes by the total number of was assessed by repeating the analyses on children who
diarrhoea-free days. We calculated a diarrhoea had more than 12 months of prospective surveillance
prevalence rate by dividing the number of diarrhoea- from birth (ie, children who were followed up into the
positive days by the total number of days under second year of life). The three exposure variables were
surveillance for diarrhoea. Episodes of C parvum and also analysed by age categories in 6-month intervals (05,
G lamblia were defined as having one positive stool 611, 1217, 1824 months). This analytical strategy
sample followed by at least three consecutive negative enabled us to identify effects that acted only during
samples.1617 Incidence rates for C parvum and G lamblia specific periods in infancy, and to further reduce the risk
were calculated as the number of episodes divided by the of misclassification that could arise when comparing
number of stool samples at risk of a new episode (ie, total children who were followed up prospectively for different
episode-free samples). We multiplied incidence by 52 to lengths of time.
express it in episodes per year at risk, given that the
number of stool samples assessed per year was 52 due to Role of the funding source
weekly collections. C parvum and G lamblia prevalence The study sponsors had no role in study design, data
rates were calculated by dividing the number of positive collection, data analysis, data interpretation, or writing of
weekly stool samples by the total number of stool the report.
samples. Episodes of C parvum and G lamblia were
judged symptomatic if the child had at least one Results
diarrhoea-positive day in the period 2 weeks before the Description of the sample
beginning and 2 weeks after the end of the episode. Since Table 1 shows the characteristics of the children. All
giardia infections are commonly symptomless and can children were enrolled in school from grade 2 to 5.
last for several weeks or even months,24 we also defined 25 (18%) of 141 children had been held back in school
giardia-positive stools as symptomatic if the child had once and required to repeat a grade; nine (6%) had been
diarrhoea on any one of the 3 days surrounding or held back at least twice (table 2).
including the day the sample was collected. Thus, 46 (32%) of 143 children were stunted at least once
giardia-positive stools were also defined as symptomless during infancy. Of these, ten (22%) were severely
in the absence of diarrhoea. We then calculated incidence stunted. Most stunted children became stunted between
and prevalence rates for symptomatic C parvum and both 6 and 17 months of age. Nearly half of children who were
definitions of G lamblia infection by the same methods stunted during infancy had persistent stunting into late
described previously. We estimated the median age at childhood; the remaining stunted children had catch-up
onset of first infection with the Kaplan-Meier statistic.25 growth and were not stunted at 9 years of age. Several
expressions of stunting were correlated with each other.
Statistical analysis 18 (95%) of 19 persistently stunted children were stunted
We used Stata (version 6.0) for all analyses, and one-way for the first time between 6 and 17 months of age
analysis of variance (ANOVA) for univariate analysis of whereas only 18 (68%) of 27 catch-up children were
the WISC-R scores. Sequential multivariate ANOVA was stunted for the first time during this period (p=002, 2
applied to identify factors independently associated with test). Catch-up children were nearly seven times as likely

566 THE LANCET Vol 359 February 16, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet Publishing Group.
 ARTICLES

Children (n=143) n  [95% CI) p

Demographic characteristics Grade
Age (years)* 936 (046) Fifth 31 .. <0001
Male sex 76 (53%) Second 7 235 (329 to 142) <0001
WISC-R test scores* 889 (1253) Third 29 117 (175 to 60) <0001
Fourth 76 39 (86 to 09) 0110
Stunting severity
Severe (HAZ <3) 10 (7%) School type
Moderate (3 HAZ <2) 36 (25%) Public 115 .. <0001
Never stunted 97 (68%) Private 26 103 (52 to 154) ..
Diarrhoea episodes (total) Class size
02 29 (20%)
30 119 .. 0033
34 35 (25%) <30 22 62 (05 to 118) ..
56 35 (25%)
Held back
78 23 (16%)
No 107 ..

9 21 (14%)
Yes 34 124 (168 to 80) <0001
Diarrhoea prevalence (days/month) by age (months)*
Homework habits
05 360 (370)
Always or almost always 124 .. 0001
611 206 (324)
Sometimes or never 17 102 (164 to 41) ..
1217 167 (190)
1824 170 (190) Maternal education
011 293 (273) No college-level education 135 .. 0026

12 167 (195) Some college schooling 6 115 (137 to 217) ..

Giardia episodes (total) Paternal education
0 20 (14%) No college-level education 133 .. <0001
1 37 (26%) Some college schooling 8 154 (68 to 241) ..
2 37 (26%)
Parity
3 28 (20%)

4 40 ..
4 12 (8%)
13 101 51 (06 to 97) 0027
57 9 (6%)
Income source
Cryptosporidium episodes (total)
Other 5 .. 0013
0 66 (46%)
Father or mother 136 140 (30 to 251) ..
1 47 (33%)
Career
2 20 (14%)
Other 125 .. 0036
3 10 (7%)
Technical employee 13 77 (05 to 148) ..
Cryptosporidium incidence (episodes per year at risk)* 056 (061)
Floor type
Follow-up time (months) Cement 119 .. ..
611 28 (20%) Soil 24 86 (140 to 32) 0002
1217 50 (35%)
*Socioeconomic and education variables except for grade level and floor type,
1823 43 (30%) were determined from survey, which was completed in 141 children. From

24 22 (15%) ANOVA F test. Wald test. Three patients did not report a main source of
HAZ=height-for-age z score. *Values are mean (SD). household income. Factors not significantly associated with WISC-R scores:
toilet and water connection, language spoken at home during the first 5 years
Table 1: Childrens characteristics of life, mothers origin, primary caregiver, father living in the household during
infancy, attended day-care or kindergarten before school, began school late,
and worked while in school.
(33%, nine of 27 vs 5%, one of 19) to be stunted for the Table 2: Association between education and socioeconomic
first time before age 6 months and after 17 months factors and WISC-R test scores*
compared with children with persistent stunting.
Median age at onset of first diarrhoea episode was
09 months. The total number of diarrhoea episodes scored higher than children of less educated parents.
during infancy ranged from 0 to 29, with a mean of 55. Children who were born before the fourth child, or who
122 (85%) of 143 children had eight or fewer episodes. had parents who were the primary source of income in
The estimated median age at onset of first G lamblia the home or who were technical employees had higher
infection was 72 months. 123 (86%) of 143 children had scores. Children who lived in homes with soil floors
at least one G lamblia episode during the first 2 years of scored lower than children who lived in homes with
life. G lamblia incidence ranged from 0 to 48 episodes cement floors. We saw no association between primary
per year at risk during infancy. caregiver and WISC-R scores.
267 (3%) of 9959 stool samples tested positive for Stunting prevalence rates were negatively correlated
C parvum. 77 (54%) of 143 children had at least one with cognitive test scores. An increase in prevalence of
episode of C parvum. The estimated median age at onset stunting corresponded to a decrease in WISC-R score
of first C parvum infection was 161 months. (table 3). Children with 2 months or more of stunting per
year had lower scores than those with little or no
Univariate analysis stunting. Children stunted for the first time between ages
Children in grades 2 and 3 scored lower on the WISC-R 6 and 11 months, and 12 and 17 months, achieved lower
test than children in grade 5 (table 2). Children in grade WISC-R scores than children never stunted during
4 had similiar scores to those in grade 5. Private school infancy (table 3). WISC-R scores of children stunted for
attendees and smaller class size were asssociated with the first time before age 6 months and after 17 months
higher scores. Children who were held back in school at did not differ from those of children who were never
least once and who completed homework assignments stunted in the first 2 years.
infrequently had lower scores. Persistent stunting into late childhood was also
Table 2 shows socioeconomic covariates that were associated with cognitive test scoreschildren who
significantly associated with WISC-R scores. Children of suffered persistent stunting scored lower on the
mothers and fathers who had had a college education WISC-R than children who were never stunted. Catch-

THE LANCET Vol 359 February 16, 2002 www.thelancet.com 567

For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES

Children Unadjusted Multivariate adjusted*

(n=143)
 (95% CI) p  (95% CI) p
Stunting prevalence (months per year) 143 12 (20 to 04) 0005 06 (13 to 01) 0162
<2 38 Reference .. .. ..

2 105 74 (120 to 29) 0002 42 (80 to 04) 0032

Age at first stunting event (months)
Never stunted 97 Reference 0011 .. 0173
05 6 14 (-87 to 115) 0779 01 (95 to 93) 0983
611 18 60 (122 to 01) 0055 20 (-73 to 33) 0458
1217 18 91 (153 to 30) 0004 65 (116 to 14) 0014

18 4 74 (48 to 197) 0232 08 (94 to 109) 0882

Never or stunted other age 107 Reference .. .. ..
617 36 79 (125 to 33) 0001 43 (83 to 05) 0031
Stunting persistence, infancy until late childhood
Never stunted 91 Reference 0010 .. 0129
Persistent 19 84 (145 to 24) 0009 46 (97 to 05) 0077
Catch-up 27 18 (71 to 35) 0503 21 (65 to 23) 0347
Late incident 6 90 (11 to 191) 0081 54 (29 to 137) 0200
Stunting severity
None or not severe 133 Reference .. .. ..
Severe 10 91 (171 to 11) 0027 54 (124 to 16) 0129
011 months (none or not severe) 138 Reference
Severe 5 101 (213 to 11) 0077 01 (107 to 109) 0989

12 months (none or not severe) 107 Reference

Severe 10 132 (215 to 48) 0002 100 (175 to 24) 0011
Diarrhoea prevalence by age-groups (days/month)
05 months 143 00 (06 to 06) 0953 02 (08 to 04) 0530
611 months 142 04 (-03 to 10) 0272 03 (05 to 11) 0508
1217 months 116 15 (27 to 02) 0026 01 (12 to 11) 0888

18 months 72 15 (16 to 05) 0293 01 (10 to 11) 0918

011 months 143 02 (06 to 10) 0606 00 (09 to 08) 0951

12 months 116 17 (29 to -04) 0008 02 (14 to 09) 0670

Giardia incidence
1 episode 66 Reference .. .. ..
>1 episode per year at risk 77 53 (94 to 13) 0011 41 (80 to 02) 0039
*Multivariate models for stunting variables have been adjusted for intertester variability, paternal education, school type, ever held back, and grade level (grades 3, 4,
and 5, combined as reference category). Models for diarrhoea and Giardia included these variables as well as severe stunting in the second year of life. p-values from
ANOVA F test. Wald test. Severe stunting is defined as height-for-age Z score below 3 SDs of the NCHS median. Analysis of severity was done by year of life
because too few children had severe stunting in each of the first two semesters of life. ||Multivariate adjusted model has 117 children or less as indicated due to
inclusion of stunting severity in the second year of life. For continuous variables, regression coefficient () is the change in WISC-R for a one unit increase in the
variable.
Table 3: Relations of stunting, diarrhoeal disease and G lamblia infection to WISC-R test scores

up and late incident children did not significantly differ parity, main source of income, career type, floor type,
from those who were never stunted. Children who were and language spoken at home while the child was
severely stunted had lower WISC-R scores compared learning to speak, were all accounted for by paternal
with moderately stunted and non-stunted children (table education. We believe that paternal education had a
3). Children who were severely stunted in the second stronger association with cognitive test scores because it
year of life scored much lower on the WISC-R test than is the most comprehensive indicator of household
children moderately stunted and non-stunted during that socioeconomic status in this community. Fathers were
period. more likely to achieve higher educational levels than their
Diarrhoea prevalence in the second year of life was wives or partners and were generally the main source of
significantly associated with cognitive test scores at age 9 income in the household. These patriarchal patterns are
(table 3). Between 12 and 24 months, a 1-day increase in common in Peru and many less-developed countries.
diarrhoea per month was associated with a decrease in Although conceptually each variable may have a specific
WISC-R scores. effect, statistically they are all highly correlated. Paternal
G lamblia incidence was significantly associated with education was the only predictor that was independently
cognitive test scores (table 3). Children with more than associated with cognitive test scores after adjustment for
1 episode per year at risk had lower WISC-R test scores maternal education and birthplace, parity, principal
than children with one or fewer G lamblia episodes per source of income and career type, floor type, and
year at risk. Neither symptomatic nor symptomless language spoken at home while the child was learning to
episodes and stool samples were associated with WISC-R speak.
scores. No associations were recorded between C parvum Multivariate models for stunting were adjusted for four
infection and cognitive test scores according to the significant covariates: paternal education, school type,
number of episodes, incidence and prevalence, and grade retention, and grade level. The models for
symptomatic infections. diarrhoea prevalence and G lamblia incidence were also
adjusted for severity of stunting in the second year of life.
Multivariate-adjusted models After multivariate adjustment, stunting prevalence and
Class size and homework habits were no longer age at first stunting event were still significant predictors
significant predictors of cognitive test scores after of cognitive test score, whereas persistent stunting was no
adjustment for school type, grade repetition, and grade longer significant in multivariate analysis. Adjustment
level. In addition, maternal birthplace, level of education, only modified slightly the effect of severe stunting in the

568 THE LANCET Vol 359 February 16, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet Publishing Group.
 ARTICLES

=  coefficient = 95% CI between Peruvian and Filipino children in the proportion

Parental of children who were severely stunted (22% vs 43%).
88
education This finding shows that the negative effect of linear
School type 69 growth retardation on cognitive function is not restricted
to settings with a high proportion of severely stunted
School children.
retention 144
Our results also show that prevalence and timing of the
School grade 32 first stunting event are related to a decrease in
Severe intelligence quotient after adjustment for other significant
40 covariates. Sample size prevented us from examining the
stunting
Giardia independence of these effects from those of severe
25 stunting. However, if independent effects of stunting
incidence
severity, timing, and persistence on cognitive ability are
20 15 10 5 0 5 10 15 20 shown, then the burden of growth retardation will reach
 (95% CI) even greater levels.
We did not assess iron status in this study. Surveillance
Factors associated with WISC-R test scores studies26 in this community and other shanty towns in
CIs are the change in mean WISC-R test scores associated with the Lima have reported that about 25% of children under
presence of each variable. Values at the right of the graph are the 4 years of age were anaemic. Since iron deficiency is a
percentage of the variance explained by each factor.
risk factor for cognitive deficits in school-aged children,27
second year of life on WISC-R scores. This analysis was we cannot rule out the possible effects of iron deficiency
repeated with data from 117 children who had at least on WISC-R scores.
12 months of follow-up from birth. The effects of the In univariate analysis, diarrhoea prevalence in the
four adjustment covariates were unchanged compared second year of life was associated with lower cognitive
with the models that included all 143 children. test scores. Although children with higher diarrhoea
Diarrhoea prevalence in the second year of life was not prevalence rates were more likely to be severely stunted
significantly associated with WISC-R scores after adjusting in the second year of life and have parents who were not
for paternal education, school type, grade retention, grade college educated, when children with high and low
level, and severe stunting in the second year of life. diarrhoeal burden were assessed separately, children who
However, children with higher diarrhoea prevalence rates were severely stunted scored worse than those who had
were more likely to be severely stunted in the second year never been stunted. These results suggest that the effect
of life and have parents who were not college educated of severe stunting is present even in children with low
(p<0001 and p=0008, likelihood-ratio test). diarrhoeal burdens, and supports the hypothesis that the
G lamblia incidence remained significantly associated effect of stunting is independent of the action of
with WISC-R scores in the multivariate adjusted model diarrhoeal disease. In addition, the effect of diarrhoea
after controlling for socioeconomic status, schooling, and disappeared after adjustment for paternal education,
severe stunting. whereas the effect of stunting remained significant.
We were able to explain 40% of the variability in Because stunting and paternal education are both
WISC-R scores in a model including both severe stunting significantly associated with WISC-R scores, and
in the second year of life and giardia incidence, as well as stunting is not a unique consequence of diarrhoea, they
socioeconomic and schooling variables (figure). No act as confounders by artificially creating the association
significant interactions were seen between any variables between diarrhoeal disease and cognitive test scores. The
in our multivariate-adjusted models. effect of diarrhoeal disease on WISC-R was not found
after modelling the relation between stunting and
Discussion socioeconomic factors, and WISC-R.
We have shown that stunting during infancy has a strong Guerrant and colleagues15 have reported an association
adverse effect on cognitive function in late childhood. between diarrhoeal disease and poorer cognitive test
Severe stunting in the second year of life was associated scores, after adjustment for socioeconomic status and
with a 10-point reduction in intelligence quotient scores maternal education. However, it is unlikely that with a
after adjustment for socioeconomic, schooling, and other sample of only 25 children, appropriate adjustment for
significant factors. Previous studies have been limited by multiple confounding factors could have been done. In
use of shorter, less comprehensive cognitive tests that are our study, a larger sample allowed us to model the
restricted to specific domains of intellectual ability and association of multiple factors with cognitive function,
have relatively weak internal validity. Mendez and Adair4 even for variables with prevalence rates that were lower
reported a similar association between severe stunting than 8%. Although parents in general are better educated
and cognitive function in Filipino children using a local in the Peruvian Pampas than in the Brazilian Favela, our
test of non-verbal intelligence. However, because the larger sample size allowed us to adjust for paternal
construct validity of this test was not described, the education while taking into account other significant
correspondence of the reported differences in test scores covariates, namely schooling and stunting. In addition,
(1433 points) with actual intelligence deficits cannot the effect of stunting on cognitive function described in
be established. Our use of the full-scale WISC-R, a our results was not assessed in the Brazil study25 and
widely used, comprehensive, and reliable test of cognitive might also explain their finding.
ability, greatly increases the internal validity of our results C parvum infections were not related to decreased
and extends our conclusions to a wider range of cognitive function, despite the reported effect of
intellectual skills than those examined in other reports. cryptosporidiosis on linear growth faltering.11,12 The
In our analysis, severe stunting in the second year of absence of a detectable effect might stem in part from the
life explained 4% of the variance in cognitive test scores. low incidence and the late age of onset of C parvum in
This figure is similar to the results reported by Mendez these children. Similarly, Guerrant and colleagues15 in
and Adair (3%6%),4 despite the striking difference Brazil recorded no correlation between C parvum and

THE LANCET Vol 359 February 16, 2002 www.thelancet.com 569

For personal use. Only reproduce with permission from The Lancet Publishing Group.
ARTICLES

cognitive dysfunction. Other studies2830 have recorded an children who experienced catch-up growth. Children
association between cognitive function and helminth who either recovered early or became stunted late in
infections such as hookworm and Trichuris trichiura. infancy were relatively unaffected. Our results suggest
However, the dry desert conditions present in the that strategies to promote the cognitive function of school
Pampas surrounding Lima are not conducive to survival age children in less-developed countries should focus on
or transmission of geohelminths. Current data suggest securing the nutrition and wellbeing of children early in
a low prevalence from 0 to 08% of geohelminths such life.
as Ascaris lumbricoides, T trichiura, hookworm, and
Strongyloides stercoralis in the Pampas, and that when Contributors
infections do arise, severity is minimum. D Berkman designed the research methods, gathered data, managed data
entry, devised the analytical framework for the study, did the statistical
We have reported a small effect of incident giardia analysis, and wrote and edited the paper. A Lescano assisted in devising
episodes on WISC-R test scores. The effect of giardiasis the analytical framework for the study and doing the statistical analysis,
on WISC-R scores remained significant after adjustment wrote most of the statistical analysis, and edited the paper. R Gilman
for paternal education, schooling factors, and severe participated in study design, advised in data collection, guided the
scientific process, contributed to writing and editing of the paper,
stunting, reducing the chance that residual confounding provided advice on the epidemiology of diarroheal diseases and
could explain these findings. Giardia infections reflect the G lamblia infection, and assisted with funding. S Lopez organised,
occurrence of only a few specific events of faecal- managed, and did most of the data collection. M Black participated in
oral contamination, making diarrhoeal disease a study design, guided the scientific process, contributed to writing and
editing of the paper, and provided advice on anthropometrics, and the
more comprehensive indicator of environmental assessment and interpretation of cognitive performance in children.
contamination. However, in this study, WISC-R scores
were associated with giardiasis and not with diarrhoeal Conflict of interest statement
disease, suggesting that G lamblia infection could be a None declared.
better indicator of those aspects of environmental
contamination that are associated with deficits in Acknowledgments
cognitive function. There is no evidence or rationale for a This study was funded in part by ITREID and the RG-ER fund for the
advancement of research in tropical medicine, and by the SSB fund.
specific aspect of the pathogenesis of G lamblia infection We thank Richard L Guerrant for his stimulating lecture and
to have deleterious effects on cognitive development. comments which provoked this study, Elissa Malkin for her support, and
Nevertheless, giardiasis can lead to zinc and other J B Phu and D Sara for technical assistance.
micronutrient deficiencies that have been associated with
deficits in cognitive development.31 Nash and Mowatt32
reported that G lamblia variant specific surface proteins References
bind zinc and other heavy metals in the intestine. Giardia 1 Bronfenbrenner U. The ecology of human development:
can also cause malabsorption of fats.24 experiments by nature and by design. Cambridge: Harvard University
To keep the risks of confounding to a minimum, we Press, 1979.
2 Connolly KJ, Kvalsvig JD. Infection, nutrition and cognitive
modelled the effects of each variable significantly performance in children. Parasitology 1993; 107 (suppl): 187200.
associated with WISC-R scores. Analysis of multiple 3 Brown JL, Pollitt E. Malnutrition, poverty and intellectual
expressions of each variable, excluding those that were development 1996. Sci Am 1996; 274: 3843.
colinear or redundant in their relation with cognitive test 4 Mendez MA, Adair LS. Severity and timing of stunting in the first
scores in univariate analyses, were made. This analytical two years of life affect performance on cognitive tests in late
strategy enabled us to obtain a highly parsimonious childhood. J Nutr 1999; 129: 155562.
5 Powell CA, Walker SP, Himes JH, Fletcher PD,
ANOVA model that explained 40% of the variance in Grantham-McGregor SM. Relationships between physical growth,
WISC-R scores. Mendez and Adair4 reported very similar mental development and nutritional supplementation in stunted
results using at least ten more variables and a sample size children: the Jamaican study. Acta Paediatr 1995; 84: 2229.
15 times greater than ours. Our results show that a 6 Black RE, Brown KH, Becker S. Malnutrition is a determining factor
reduced set of carefully selected variables can in diarrheal duration, but not incidence, among young children in a
longitudinal study in rural Bangladesh. Am J Clin Nutr 1984; 37:
appropriately describe the sources of variability in 8794.
cognitive test scores and represent the most relevant 7 Sepulveda J, Willett W, Muoz A. Malnutrition and diarrhea: a
domains affecting intelligence. longitudinal study among urban mexican children. Am J Epidemiol
The combined effects of stunting and giardia infections 1988; 127: 46576.
account for an intelligence quotient deficit of almost 15 8 Guerrant RL, Kirchhoff LV, Shields DS, et al. Prospective study of
diarrheal illnesses in northeastern Brazil: patterns of disease,
points. These results are important for re-evaluating the nutritional impact, etiologies, and risk factors. J Infect Dis 1983; 148:
current estimates of the burden imposed by symptomatic 98697.
and symptomless infections and malnutrition during 9 Farthing MJ, Mata L, Urrutia JJ, Kronmal RA. Natural history of
infancy. Our finding of an association between giardia Giardia infection of infants and children in rural Guatemala and its
infections and decreased WISC-R scores is interesting impact on physical growth. Am J Clin Nutr 1986; 43: 395405.
and deserves further attention in a more rigorous study 10 Guerrant RL, Schorling JB, McAuliffe JF, de Souza MA. Diarrhea as
a cause and effect of malnutrition: diarrhea prevents catch-up growth
designed to account for multiple parasitic, bacterial, and and malnutrition increases diarrhea frequency and duration.
viral enteric infections. Because of the high prevalence of Am J Trop Med Hyg 1992; 47: 2835.
stunting in children living in less-developed countries, 11 Checkley W, Gilman RH, Epstein LD, et al. Asymptomatic and
estimated to be as high as 40% in children younger than symptomatic cryptosporidiosis: their acute effect on weight gain in
Peruvian children. Am J Epidemiol 1997; 145: 15663.
5 years of age,33 our results can be used in design of
12 Checkley W, Epstein LD, Gilman, RH, Black RE, Cabrera L,
intervention programmes seeking to have the greatest Sterling CR. Effects of cryptosporidium parvum infection in Peruvian
effect on the adverse effects of stunting. children: growth faltering and subsequent catch-up growth. Am J
In our study, children who were severely stunted in the Epidemiol 1998; 148: 497506.
second year of life had the most substantial impairments 13 Lunn PG, Erinoso HO, Northrop-Clewes CA, Boyce SA. Giardia
in cognitive function. We also showed that children who intestinalis is unlikely to be a major cause of the poor growth of rural
Gambian infants. J Nutr 1999; 129: 87277.
had persistent stunting into late childhood were 14 Fraser D, Bilenko N, Deckelbaum RJ, Dagan R, El-On J, Naggan L.
considerably more likely to be stunted for the first time Giardia lamblia in Israeli Bedouin infants: risk factors and
after age 6 months and before 17 months compared with consequences. Clin Infect Dis 2000; 30: 419-424.

570 THE LANCET Vol 359 February 16, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet Publishing Group.
 ARTICLES

15 Guerrant DI, Moore SR, Lima AA, Patrick PD, Schorling JB, 24 Ortega YR, Adam RD. Giardia: overview and update. Clin Infect Dis
Guerrant RL. Association of early childhood diarrhea and 1997; 25: 54549.
cryptosporidiosis with impaired physical fitness and cognitive 25 Brookmeyer R, Crowley J. A confidence interval for the median
function four-seven years later in a poor urban community in survival time. Biometrics 1992; 38: 2941.
northeast Brazil. Am J Trop Med Hyg 1999, 61: 70713. 26 Asociacion Benefica PRISMA. 1994. Vigilancia Nutricional Regional
16 Stephensen CB. Burden of infection on growth failure. J Nutr 1999; Lima 19911992: Informe Final.
129 (suppl): 53438. 27 Halterman JS, Kaczorowski JM, Aligne CA, Auinger P, Szilagyi PG.
17 Gilman RH, Marquis GS, Ventura G, Campos M, Spira W, Diaz F. Iron deficiency and cognitive achievement among school-aged
Water cost and availability: key determinants of family hygiene in a children and adolescents in the United States. Pediatrics 2001; 107:
Peruvian shantytown. Am J Public Health 1993; 83: 155458. 138186.
18 Fernandez-Concha D, Gilman RH, Gilman JB. A home nutritional 28 Nokes C, Grantham-McGregor SM, Sawyer AW, Cooper ES,
rehabilitation programme in a Peruvian peri-urban shantytown Robinson BA, Bundy DA. Moderate to heavy infections of Trichuris
(pueblo joven). Trans R Soc Trop Med Hyg 1991; 85: 80913. trichiura affect cognitive function in Jamaican school children.
19 Sterling CR, Gilman RH, Sinclair NA, Cama V, Castillo R, Diaz F. Parasitology 1992; 104: 53947.
The role of breast milk in protecting urban Peruvian children against 29 Watkins WE, Cruz JR, Pollitt E. The effects of deworming on
cryptosporidiosis. J Protozool 1991; 38: 2325. indicators of school performance in Guatemala. Trans R Soc Trop Med
20 Gilman RH, Marquis GS, Miranda E, Vestegui M, Martinez H. Hyg 1996; 90: 15661.
Rapid reinfection by Giardia lamblia after treatment in a 30 Sakti H, Nokes C, Hertanto WS, et al. Evidence for an association
hyperendemic Third World community. Lancet 1988; 1: 34345. between hookworm infection and cognitive function in Indonesian
21 Arrowood J, Sterling CR. Comparison of conventional staining school children. Trop Med Int Health 1999; 4: 32234.
methods and monoclonal antibody-based methods for 31 Black MM. Zinc deficiency and child development. Am J Clin Nutr
Cryptosporidium oocyst detection. J Clin Microbiol 1989; 27: 1998; 68 (suppl): 46469.
149095. 32 Nash TE, Mowatt MR. Variant-specific surface proteins of
22 Vidal MF, Gilman RH, Ungar BL, et al. Detection of Giardia lamblia Giardia lamblia are zinc-binding proteins. Microbiology 1993; 90:
antigen in children living in a Peruvian periurban shantytown (pueblo 548993.
joven). J Clin Microbiol 1991; 29: 63637. 33 de Onis M, Monteiro C, Akre J, Glugston G. The worldwide
23 Wechsler D. Wechsler Intelligence Scales for Children-Revised. San magnitude of protein-energy malnutrition: an overview from the
Antonio (TX): The Psychological Corporation, Harcourt Brace WHO Global Database on Child Growth. Bull World Health Organ
Jovanovich Inc; 1974. 1993; 71: 703-12.

Uses of error
Assumptions
Martin Barratt

A Commentary on our trial comparing medical or for the treatment of the first episode of nephrotic
surgical treatment for children with vesicoureteric reflux syndrome in childhood. Corticosteroids were introduced
(VUR) and nephropathy picked up on a statement which in 1950, and by the time I entered paediatric nephrology
was not strictly accurate. We have shown that there are in 1967, it had become sufficiently obvious that they
no differences in the outcome of children with severe induced remission in most children with minimal change
VUR using medical or surgical management strategies. nephropathy, such that a prospective controlled trial was
It would have been more correct to say that we failed to thought to be unethical. There was uncertainty about
demonstrate a significant difference in outcome between how long treatment with high dose steroids should
the two groups, but that if there were a benefit from continue after remission. I wanted to minimise steroid
surgery, we could be 95% confident that it was not more toxicity and advocated withdrawal as soon as the urine
than 20% at 4 years, and 30% at 10 years of follow-up. was free of protein. This policy was the basis of our local
We could well have missed a benefit from surgery, management protocol, and was repeated in lectures,
apparent with more cases, longer follow-up, better reviews, textbooks and consensus statements. However,
selection of patients, or less than ideal medical others believed that longer courses of steroids were
management (the trial effect). We could have needed to prevent relapse, and in the end, several
prematurely closed the book on surgery, discouraging controlled trials proved them to be correct. Our desire to
others from embarking on long-term clinical studies, or protect children from the side effects of prolonged
research foundations from funding them, and dissuading treatment resulted in a higher relapse rate and more
purchasers from approving the procedure. steroids in the long run. The mistake was not to have
If this was a mistake, it was not my biggest one. recognised earlier that a controlled trial would have
I supported the use of short-course high-dose prednisolone resolved this dilemma.

Institute of Child Health, University College, London WC1N 1EH, UK (Prof M Barratt FRCP)

THE LANCET Vol 359 February 16, 2002 www.thelancet.com 571

For personal use. Only reproduce with permission from The Lancet Publishing Group.