Malnutrition risk in hospitalized children: use of 3 screening tools in

a large European population1–3

Michael Chourdakis,4,13 Christina Hecht,4,13 Konstantinos Gerasimidis,5,13 Koen FM Joosten,6
Thomais Karagiozoglou-Lampoudi,7 Harma A Koetse,8 Janusz Ksiazyk,9 Cecilia Lazea,10

Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018

Raanan Shamir,11 Hania Szajewska,12 Berthold Koletzko,4* and Jessie M Hulst6
4
Ludwig-Maximilians-University of Munich, Division of Metabolic and Nutritional Medicine, Dr. von Hauner Children’s Hospital, University of Munich
Medical Centre, Munich, Germany; 5Human Nutrition, School of Medicine, College of Medicine, Veterinary and Life Sciences, Royal Hospital for Sick
Children, University of Glasgow, Glasgow, United Kingdom; 6Erasmus Medical Center, Sophia Children’s Hospital, Department of Pediatrics, Rotterdam,
Netherlands; 7Clinical Nutrition Laboratory, Nutrition/Dietetics Department, Technological Education Institute, Thessaloniki, Greece; 8Beatrix Children’s
Hospital, University Hospital Groningen, Groningen, Netherlands; 9Department of Pediatrics, Nutrition and Metabolic Diseases, The Children’s Memorial
Health Institute, Warsaw, Poland; 10Clinic Pediatrics I, Clinical Emergency Hospital for Children, Cluj-Napoca (Klausenburg), Romania; 11Institute of
Gastroenterology, Nutrition and Liver Diseases, Schneider Children’s Medical Center of Israel, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv,
Israel; and 12Department of Pediatrics, Medical University of Warsaw, Warsaw, Poland

ABSTRACT
Background: Several malnutrition screening tools have been advo-
cated for use in pediatric inpatients.
Objective: We evaluated how 3 popular pediatric nutrition screening
tools [i.e., the Pediatric Yorkhill Malnutrition Score (PYMS), the
Screening Tool for the Assessment of Malnutrition in Pediatrics
(STAMP), and the Screening Tool for Risk of Impaired Nutritional
Status and Growth (STRONGKIDS)] compared with and were related
to anthropometric measures, body composition, and clinical variables
in patients who were admitted to tertiary hospitals across Europe.
Design: The 3 screening tools were applied in 2567 inpatients at 14
hospitals across 12 European countries. The classification of pa-
tients into different nutritional risk groups was compared between
tools and related to anthropometric measures and clinical variables
[e.g., length of hospital stay (LOS) and infection rates].
Results: A similar rate of completion of the screening tools for each
tool was achieved (PYMS: 86%; STAMP: 84%; and STRONGKIDS:
81%). Risk classification differed markedly by tool, with an overall
agreement of 41% between tools. Children categorized as high risk
(PYMS: 25%; STAMP: 23%; and STRONGKIDS: 10%) had a longer
LOS than that of children at low risk (1.4, 1.4, and 1.8 d longer,
respectively; P , 0.001). In high-risk patients identified with the
PYMS, 22% of them had low (,22) body mass index (BMI) SD-
scores (SDSs), and 8% of them had low height-for-age SDSs. For the
STAMP, the percentages were 19% and 14%, respectively, and for the
STRONGKIDS, the percentages were 23% and 19%, respectively.
Conclusions: The identification and classification of malnutrition
risk varied across the pediatric tools used. A considerable portion of
children with subnormal anthropometric measures were not identified
with all of the tools. The data obtained do not allow recommending
the use of any of these screening tools for clinical practice. This
study was registered at clinicaltrials.gov as NCT01132742.
Am J Clin Nutr 2016;103:1301–10.
Keywords: hospitalized children, malnutrition, nutritional screen-
ing, PYMS, STAMP, STRONGKIDS
INTRODUCTION
Malnutrition screening has been advocated as part of patients’
standard care (1–3). This recommendation is because malnu-
trition on admission or deterioration of the nutritional status
during hospitalization has been associated with a prolonged
hospital stay and adverse outcomes (e.g., increased rates of
complications such as infections), although the causality in these
associations remains to be explored (4–7). The early identifi-
cation of nutritional risk followed by an appropriate nutritional
management were proposed as part of routine clinical practice
(8). The Guidelines for Nutrition Screening by the European
Society for Clinical Nutrition and Metabolism provides recom-
mendations for adult patients but does not address pediatric
patients (9). Screening tools for the assessment of malnutrition
risk of adults have been available for many years (9–11). Similar
pediatric tools have been developed and were only tested in
small cohorts of hospitalized children (5, 7, 12–14). These tools
consist of questions related to the patient’s history and mea-
surements or clinical estimation of body size to assess risk of poor
1
Parts of the study data were previously presented at the 45th Annual
Congress of the European Society for Paediatric Gastroenterology, Hepatology
and Nutrition, Stockholm, Sweden, 27–28 April 2012, and at the 34th Annual
Congress of the European Society for Clinical Nutrition and Metabolism,
Barcelona, Spain, 8–11 September 2012.
2
Supported in part by a European Society for Clinical Nutrition and Me-
tabolism Network Grant (to BK).
3
Supplemental Figure 1 and Supplemental Tables 1 and 2 are available
from the “Online Supporting Material” link in the online posting of the
article and from the same link in the online table of contents at http://ajcn.
nutrition.org.
13
These authors shared first authorship.
*To whom correspondence should be addressed. E-mail: office.koletzko@
med.uni-muenchen.de.
Received March 10, 2015. Accepted for publication February 22, 2016.
First published online April 20, 2016; doi: 10.3945/ajcn.115.110700.

Am J Clin Nutr 2016;103:1301–10. Printed in USA. Ó 2016 American Society for Nutrition 1301
1302 CHOURDAKIS ET AL.
nutritional status (15). The tools aim to screen all inpatients and
identify those who were missed during routine admission and
whose disease outcome would improve or would not deteriorate
from a tailored nutritional intervention. However, there has been
a lack of sufficient data on the predictive value of such pediatric
screening tools on outcome and objective indexes of malnutri-
tion in large multicenter studies and of a comparative evaluation
of the various tools. Addressing these aspects may direct health
professionals on their decision to select the most suitable nu-
tritional screening tool.
We compared the risk scoring of 3 previously proposed
pediatric nutrition screening tools, [i.e., the Pediatric Yorkhill
Malnutrition Score (PYMS)14 (16, 17), the Screening Tool for
the Assessment of Malnutrition in Pediatrics (STAMP) (13), and
the Screening Tool for Risk of Impaired Nutritional Status and
Growth (STRONGKIDS) (5)] in a large multicenter study in
children admitted to hospitals across Europe. In addition, we
explored the agreement between the tools (concurrent validity)
and the relation of risk scores to anthropometric measures and
body-composition measurements as well as clinical variables
such as the length of hospital stay (LOS).
The criteria for the best outcome measure to assess the effect of
the use of a screening tool are not yet agreed upon, because it is
somewhat controversial as to whether such screening tools should
predict anthropometric measures or clinical outcomes. Therefore,
in this study, we aimed to explore the association of the scores
provided by the tools with both subnormal BMI and with the LOS.
METHODS
Study design and subjects
This prospective, European, multicenter cohort study enrolled
patients from February 2010 to July 2011 in 14 centers in 12
countries (Zagreb, Croatia; Copenhagen, Denmark; Lille, France;
Munich, Germany; Thessaloniki, Greece; Petah Tikvah, Israel;
Milan, Italy; Rotterdam and Groningen, Netherlands; Warsaw,
Poland; Cluj-Napoca, Romania; and Oxford and Glasgow, United
Kingdom). Patients (1 mo to 18 y old) who were admitted to
pediatric and pediatric surgery wards with an anticipated length
of stay .24 h were eligible to participate. Patients were con-
secutively invited to participate whenever a data collection was
possible within the first 24 h after admission. Patients who were
attending the accident and emergency department of the day
care unit were excluded.
We excluded children admitted to intensive care because of the
limited feasibility to perform detailed anthropometric measures
on the day of admission in critically ill children. To identify
children at risk of malnutrition in this group of patients is re-
dundant because all of these children are, by the nature of their
critical illnesses (e.g., unconscious and thus unable to eat), at high
risk of malnutrition and therefore should receive the respective
attention of the medical and dietetic staff. The principle of screening
14
Abbreviations used: HFA, height-for-age; LOS, length of hospital stay;
MUAC, midupper arm circumference; PYMS, Pediatric Yorkhill Malnutri-
tion Score; SDS, SD-score; STAMP, Screening Tool for the Assessment of
Malnutrition in Pediatrics; STRONGKIDS, Screening Tool for Risk of Im-
paired Nutritional Status and Growth; TSFT, triceps skinfold thickness.
is to identify individuals at risk who might normally be missed so
that they can receive adequate referral to the clinical nutritional
care team. We also excluded children admitted to day hospital
care because their expected LOSs were ,24 h. Patients with
cerebral palsy or genetic syndromes were not excluded per pro-
tocol. Details about the recruitment and protocol have been
previously published by Hecht et al. (18).
Methods
Patients were assessed with the use of a set of questions that
considered nutritional risk, and measurements of anthropometric
Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018
measures and body composition were all performed within the
first 24 h after admission. The assessors were a multidisciplinary
team that included research nurses, dietitians, medical students,
and nutritionists. A training workshop to harmonize the recruit-
ment and standardize anthropometric measures and data collection
in the different centers was held in March 2010 in Munich,
Germany.
Demographic and medical data, together with a questionnaire
for nutritional status, were collected during a structured interview
with patients and (when required) with their caregivers. The
questionnaire integrated the 4 items of the PYMS tool (16, 17),
the 3 items of the STAMP tool (13), and the 4 items of the
STRONGKIDS screening tool (5) and sorted them by item con-
tents. For each patient, the steps of each tool were completed by
the same investigator in the same order. The total score for each
screening tool was computed during the analysis of the data. The
28 assessors were encouraged not to add the scores for each tool
during data collection to avoid bias by the knowledge on cate-
gorization in a screening tool. Any decisions on providing nu-
tritional treatment were made only by attending physicians and
dieticians, not by assessors. Such decisions were made according
to normal routine procedures and were not influenced by the
study data.
Important characteristics of the PYMS (1, 16, 17), STAMP
(13, 19), and STRONGKIDS (5, 19) are reported in Supple-
mental Table 1. The PYMS and STAMP include anthropo-
metric measures (BMI compared with weight and height,
respectively); the STRONGKIDS includes a subjective clinical
assessment of nutritional status. Total scores for each tool were
computed for those age groups for whom the tools were val-
idated. PYMS was completed for patients aged 1–16 y, STAMP
was completed for patients aged 2–16 y, and STRONGKIDS was
completed for patients aged 1 mo to 18 y. For comparison of the
3 tools, only children aged 2–16 y were considered because
patients within this age range were eligible for screening by all 3
tools.
Data on height, weight, midupper arm circumference (MUAC),
and triceps skinfold thickness (TSFT) were collected. The
methods of collection have been described previously by Hecht
et al. (18). Clinical variables, including the LOS as a primary
outcome and the frequency of infectious complications (num-
ber of days with a temperature .38.58C and number of days
with antibiotic use) were derived from hospital records after
discharge.
The total score and classification of malnutrition risk (low,
medium, or high) was determined for each study participant and
screening tool. The scores obtained by the 3 screening tools were
then related to anthropometric measures, body compositions, and
 RISK OF MALNUTRITION IN HOSPITALIZED CHILDREN 1303
TABLE 1
Scoring of screening tool items for the group of children aged 2–16 y (n = 1724) who completed all tools (n = 1258)1
Scores of children completing ALL tools (n = 1258) Children aged 2–16 y (n = 1724)

Not assessed according to

Item2 0 1 2 3 Total assessed original tool questions

Item 1: current nutritional condition,3 n (%)

PYMS (0, 2) 1152 (92) — 106 (8) — 1538 (89) 186 (11)
STAMP (0, 1, 3) 967 (77) 169 (13) — 122 (10) 1474 (85) 250 (15)
STRONGKIDS (0, 1) 1031 (82) 227 (18) — — 1607 (93) 117 (7)
Item 2: weight loss,4 n (%)
PYMS (0, 1) 1036 (82) 222 (18) — — 1568 (91) 156 (9)

Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018

STAMP (NA) — — — — — —
STRONGKIDS (0, 1) 1027 (82) 231 (18) — — 1633 (95) 91 (5)
Item 3: reduced intake,5 n (%)
PYMS (0, 1, 2) 1004 (80) 228 (18) 26 (2) — 1633 (95) 91 (5)
STAMP (0, 2, 3) 913 (73) — 317 (25) 28 (2) 1633 (95) 91 (5)
STRONGKIDS (0, 1) 861 (68) 397 (32) — — 1633 (95) 91 (5)
Item 4: underlying disease,6 n (%)
PYMS (0, 1, 2) 994 (79) 255 (20) 9 (1) — 1509 (88) 215 (12)
STAMP (0, 2, 3) 670 (53) — 324 (26) 264 (21) 1529 (89) 195 (11)
STRONGKIDS (0, 2) 893 (71) — 365 (29) — 1515 (88) 209 (12)
1
NA, not applicable; PYMS, Pediatric Yorkhill Malnutrition Score; STAMP, Screening Tool for the Assessment of Malnutrition in Pediatrics; STRONGKIDS,
Screening Tool for Risk of Impaired Nutritional Status and Growth.
2
Possible scores are shown in parentheses; each tool had different possible scores for each item. The classification of risk level, determined by total
scores, differed between tools. Scores for low risk: PYMS, 0 points; STAMP, 0–1 point; and STRONGKIDS, 0 points. Scores for medium risk: PYMS, 1 point;
STAMP, 2–3 points; and STRONGKIDS, 1–3 points. Scores for high risk: PYMS, 2–7 points; STAMP, 4–9 points; and STRONGKIDS, 4–5 points.
3
PYMS: Is the BMI below the cutoff value shown in the BMI scoring guide? STAMP: Use a growth chart or the percentile quick-reference tables to
determine the child’s weight and height measurements. STRONGKIDS: Is the patient in a poor nutritional status judged by a subjective clinical assessment?
4
PYMS: Has the child lost weight recently? STRONGKIDS: Is there weight loss or poor weight gain (infants ,1 y old) during the last few weeks or
months?
5
PYMS: Has the child had reduced intake (including feeds) for at least the past week? STAMP: What is the child’s nutritional intake? STRONGKIDS: Is
one of the following items present: excessive diarrhea ($5 times/d) or vomiting (.3 times/d), reduced food intake during the last few days, pre-existing
nutritional intervention, or inadequate nutritional intake because of pain?
6
PYMS: Will the child’s nutrition be affected by the recent admission or condition for at least the next week? STAMP: Does the child have a diagnosis
that has any nutritional implication? STRONGKIDS: Is there an underlying illness with risk of malnutrition or expected major surgery?

outcome data. For the cross-tabulation of risk classification Therefore, in all data analyses except for the random-coefficient
between the tools, we decided to group the classification of model, low- and medium-risk patients for each screening tool
malnutrition risk into 2 rather than 3 categories (i.e., high were combined and presented as one group that was compared
compared with medium and low) because children allocated in with high-risk patients.
the high category were the ones that needed to be further referred
for an assessment to the dietetic and clinical team.
The study protocol was accepted by the local research or
medical ethic committees of each participating center. Before
participation, informed written consent was obtained from par-
ents and their caregivers (whenever required).

Statistical analysis
Risk scores were cross-tabulated within the 3 screening tools,
and agreement rates were computed (concurrent validity). Cohen’s
k statistic test was applied to describe the level of agreement
between the 2 tools (20) with the agreement that occurred by
chance taken into account. Baseline characteristics between groups
were compared with the use of Fisher’s exact test or Pearson’s chi-
square test for categorical data. A linear regression analysis was
applied separately for sex to adjust the association of risk of
malnutrition with TSFT and MUAC for age, chronic disease, and FIGURE 1 Malnutrition risk classification on the basis of the 3 screen-
ing tools expressed as percentages of the total number of assessed children
center. Residuals were checked for a normal distribution. In for each tool. PYMS, Pediatric Yorkhill Malnutrition Score; STAMP, Screen-
clinical practice, a substantial intervention (e.g., referral to ing Tool for the Assessment of Malnutrition in Pediatrics; STRONGKIDS,
a dietitian) will only occur in children with a high-risk score. Screening Tool for Risk of Impaired Nutritional Status and Growth.
1304 CHOURDAKIS ET AL.

Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018

FIGURE 2 Prevalence of malnutrition risk varied markedly between and within different countries with the use of the different screening tools. PYMS,
Pediatric Yorkhill Malnutrition Score; STAMP, Screening Tool for the Assessment of Malnutrition in Pediatrics; STRONGKIDS, Screening Tool for Risk of
Impaired Nutritional Status and Growth.

Age- and sex-specific BMI and weight-for-height SD-scores and further age-adequate WHO reference data were used for
(SDSs) were calculated with the use of WHO reference data; patients aged .5–18 y (http://www.who.int/growthref/en/). MUAC
WHO growth reference study data were used for children aged and TSFT SDSs that were based on WHO reference data were
1 mo to 5 y (http://www.who.int/childgrowth/software/en/) limited to patients aged 3 mo to 5 y.

TABLE 2
Cross-tabulation of risk classification between the PYMS, STAMP and STRONGKIDS1
Risk of malnutrition

Low and medium High k (95% CI) Agreement, %

PYMS compared with STAMP (n = 1308) 0.47 (0.42, 0.53) 82

Low and medium 897 121
High 118 172
STAMP compared with STRONGKIDS (n = 1318) 0.39 (0.33, 0.45) 83
Low and medium 990 32
High 187 109
STRONGKIDS compared with PYMS (n = 1490) 0.35 (0.28, 0.42) 81
Low and medium 1088 249
High 39 114
1
The horizontal rows labeled “Low and medium” or “High” represent the total n in the first tool of each pair listed; the
vertical columns represent the total n in the second tool of each pair. Intersecting Low and medium/High rows and columns
indicate where both tools agreed. PYMS, Pediatric Yorkhill Malnutrition Score; STAMP, Screening Tool for the Assessment
of Malnutrition in Pediatrics; STRONGKIDS, Screening Tool for Risk of Impaired Nutritional Status and Growth.
TABLE 3
Characteristics of children within risk groups of each screening tool1
PYMS (1–16 y old; n = 1664) STAMP (2–16 y old; n = 1374) STRONGKIDS (1 mo to 18 y old; n = 2089)

Low (n = 943) Medium (n = 305) High (n = 416) Low (n = 512) Medium (n = 547) High (n = 315) Low (n = 915) Medium (n = 968) High (n = 206)
2
Age, y 7.4 (3.6–11.3) 5.8 (3.0–11.3) 4.4 (2.0–9.9) 8.3 (4.7–12.0) 7.8 (4.1–12.0) 7.6 (3.8–12.3) 5.1 (1.3–11.2) 4.4 (1.4–10.6) 6.3 (1.9–12.6)
Age group, %
31 d to 0.9 y 0 0 0 0 0 0 21 18 15
1–1.9 y 12 13 24 0 0 0 10 14 10
2–5.9 y 30 37 34 34 39 41 23 26 24
6–12.9 y 40 32 29 49 41 38 29 26 27
13–17.9 y 18 18 13 17 20 21 17 17 24
Female, % 44 50 43 46 45 43 44 45 44
Caucasian, % 92 93 90 94 91 92 92 91 88
Acute admission, % 45 54 65 52 48 53 48 62 58
Chronic disease, % 48 49 48 36 53 75 30 48 89
Surgical, % 20 21 17 16 21 19 25 15 20
BMI SDS 0.52 6 1.233 0.28 6 1.14 20.77 6 1.58 0.46 6 1.17 0.15 6 1.23 20.30 6 1.85 0.42 6 1.25 20.04 6 1.37 21.19 6 1.61
HFA SDS 0.15 6 1.37 0.19 6 1.43 20.19 6 1.54 0.38 6 1.25 0.02 6 1.29 20.34 6 1.62 0.37 6 1.31 0.04 6 1.38 20.86 6 1.97
Nutritional support, %
Before admission 6 11 24 1 9 26 1 11 54
During hospitalization 5 12 25 1 9 27 2 11 56
LOS, d 4 (3–6) 5 (3–8) 5 (3–9) 4 (3–7) 4 (3–7) 5 (3–8) 4 (3–7) 4 (3–7) 6 (3–10)
Secondary outcomes, %
Fever4 10 21 29 10 17 19 13 23 23
Use of antibiotics5 28 44 44 28 33 41 28 43 44
1
RISK OF MALNUTRITION IN HOSPITALIZED CHILDREN

HFA, height-for-age; LOS, length of stay; PYMS, Pediatric Yorkhill Malnutrition Score; SDS, SD-score; STAMP, Screening Tool for the Assessment of Malnutrition in Pediatrics; STRONGKIDS,
Screening Tool for Risk of Impaired Nutritional Status and Growth.
2
Median; IQR in parentheses (all such values).
3
Mean 6 SD (all such values).
4
At least one event day of fever.
5
At least one event day of antibiotic use.
1305

Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018

1306 CHOURDAKIS ET AL.

A multilevel mixed-effects Poisson regression was used to
accommodate the general dependence of the LOS on the center of
the patient and the existing differences in the severity and type of
chronic diseases between centers. Thus, the center was included
as a random effect while further allowing varying effects by
chronic disease status. The association of each nutritional risk
classification by the PYMS, STAMP, and STRONGKIDS with the
LOS was tested with age, sex ,and chronic disease status in-
cluded as confounders. An interaction between chronic disease
status and nutritional risk classification was also tested.
Furthermore, percentages of children with suboptimal skinfold
thicknesses or MUACs and suboptimal BMI who were correctly
Completion of the screening tools
Because each of the 3 screening tools was developed for
different age ranges, the number of eligible children that these
tools could be applied to varied with the screening tool used.
Approximately 933 patients were either ,2 or .16 y of age and,
therefore, the STAMP could not be completed. Similarly, for
621 participants aged either ,1 or .16 y, the PYMS could not
be applied. In total, the PYMS was completed for 1664 subjects
(86% of the children in the targeted group from 1 to 16 y of age),
the STAMP was completed for 1374 study participants (84% of
children in the targeted group from 2 to 16 y of age), and the

Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018

identified or misclassified at high risk of malnutrition by each tool
were calculated and compared with each other. Also, percentages
of children classified at high risk despite a normal MUAC, TSFT,
or BMI were compared between the 3 tools. To have the same
children included for each tool, only children aged 2–5 y were
included for the analysis of SDSs for MUAC and TSFT. Data
management and statistical analyses were carried out with R
2.13.2m software (The R Foundation for Statistical Computing)
and Stata 12.1 software (StataCorp LP).
RESULTS
Patient characteristics
A total of 2567 patients (median age: 4.7 y; IQR: 1.4–11.1 y)
were enrolled in the study (80% pediatric and 20% pediatric sur-
gery patients). Nearly one-half of the study population were female
(44.9%), and 44.8% of subjects had an underlying chronic disease
and were electively admitted (18). Most study participants were of
Caucasian origin (91%) and were at home before admission (91%).
Nutritional support before admission was administered to 11.8% of
the study population. During the hospital stay, nutritional support
was given to 12.3% of participants (6.2% of subjects received oral
supplements, 6.1% of subjects received tube feeding, and 0.8%
of subjects received parenteral nutrition with a few overlaps), of
whom 76% of participants were already receiving the support
before admission. Approximately 20% of children who received
nutritional support before admission were not allocated to a nutri-
tional support regimen after admission, according to hospital data.
The median LOS was 4 d (IQR: 3–7 d). A BMI SDS of less
than 22 was present in 7.0% of the study population at hospital
admission, whereas 7.9% of the participants had a height-for-age
(HFA) SDS of less than 22 at hospital admission.
STRONGKIDS was completed for 2089 subjects (81% of the
children in the targeted group from 1 mo to 18 y of age). For
almost one-half of the study group (1258 children; 49%), all 3
tools were completed. Completion rates of each individual
component of the 3 tools are listed in Table 1. Because the
researchers occasionally found it challenging to respond to some
of the steps of the individual tools, a numbers of screens were
left incomplete.
Malnutrition risk classification
The classification of malnutrition risk of the assessed children
by the 3 screening tools showed a substantial variation in the
different tools (Figure 1). In addition, the risk-classification dis-
tribution varied markedly within and between countries (Figure 2).
Overall, the proportion of high-risk patients ranged be-
tween 5% and 51% (PYMS: 15–51%; STAMP: 9–51%; and
STRONGKIDS: 5–30%). The greatest difference between the
proportions of high-risk patients on the basis of the 3 screening
tools within one center was 32% (Greece).
Among the 1258 patients for whom all 3 tools were completed,
the distribution of risk classification according to the 3 screening
tools is shown in Supplemental Figure 1. In more detail, in this
subgroup of 1258 patients, the different tools categorized from
10% (STRONGKIDS) to 22% (STAMP and PYMS) of children in
the high-risk group. In total, only 87 participants (7% of all
patients with 3 completed tools) were jointly rated as being at
high risk of malnutrition with the use of all 3 tools. Less than
one-half of the patients (41%) were classified at the same risk
level for malnutrition with the use of the 3 different tools. This
percentage increased to 74% when children with low and me-
dium risk were grouped together and compared with the high-risk
group. The agreement between the tools, with statistical chance
accounted for, was fair to moderate (20).

TABLE 4
Relation between the LOS and nutritional risk classification with the use of a random-coefficient model1
PYMS (n = 1669) STAMP (n = 1379) STRONGKIDS (n = 2089)

Risk Coefficient (95% CI) P Coefficient (95% CI) P Coefficient (95% CI) P

Low — — — — — —
Medium 1.112 (1.05, 1.18) ,0.001 1.08 (1.02, 1.14) 0.005 1.19 (1.14, 1.24) ,0.001
High 1.38 (1.32, 1.45) ,0.001 1.37 (1.29, 1.46) ,0.001 1.82 (1.72, 1.93) ,0.001
1
Adjusted for age, sex, and chronic disease status and with the dependence within centers taken into account. LOS,
length of stay; PYMS, Pediatric Yorkhill Malnutrition Score; STAMP, Screening Tool for the Assessment of Malnutrition in
Pediatrics; STRONGKIDS, Screening Tool for Risk of Impaired Nutritional Status and Growth.
2
Comparison with low-risk category (i.e., medium-risk patients stayed 1.11 d longer in the hospital than did low-risk
patients scored with the use of the PYMS).
 RISK OF MALNUTRITION IN HOSPITALIZED CHILDREN 1307

High (n = 128)

The total includes children who completed all tools and had calculated BMI. For 5 children, no BMI could be calculated because the length value was missing. PYMS, Pediatric Yorkhill Malnutrition
A pairwise comparison resulted in a 55% agreement for the

20.88 6 1.50
PYMS with the STAMP (k = 0.31; 95% CI: 0.28, 0.35) and 58%

72
26
30
agreement for the PYMS with the STRONGKIDS (k = 0.33; 95%
CI: 0.29, 0.37). The greatest degree of agreement was shown
between the STAMP and STRONGKIDS (60%; k = 0.37; 95%
CI: 0.33, 0.40). This agreement increased to 74% when a com-

Medium (n = 550)

54.6 (36 of 66)

STRONGKIDS
bined classification of low and medium risk was compared with

0.05 6 1.39

Score; SDS, SD-score; STAMP, Screening Tool for the Assessment of Malnutrition in Pediatrics; STRONGKIDS, Screening Tool for Risk of Impaired Nutritional Status and Growth.
the high-risk group. A pairwise comparison between tool pairs

434
88
28
resulted in w80% agreement and is shown in Table 2 (PYMS
compared with STAMP: moderate agreement; PYMS compared
with STRONGKIDS: fair agreement; and STAMP compared with
STRONGKIDS: fair agreement) (21).

Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018

Low (n = 575)

0.53 6 1.26

8
518
49
Clinical characteristics of patients in the 3 risk groups for
each tool
Characteristics of children within the risk groups of each

High (n = 274)
screening tool are described in Table 3. The proportion of pa-

20.27 6 1.88
tients with an underlying chronic disease was higher for patients

46
51
177
identified as high risk than for patients identified as medium or
low risk with the use of the STAMP (75% compared with 53%
or 36%, respectively) and the STRONGKIDS (89% compared
with 48% or 30%, respectively). With the use of the PYMS,

Medium (n = 494)

22.7 (15 of 66)

patients with a chronic disease were equally classified into the 3

0.14 6 1.23
STAMP
risk categories (48% compared with 49% or 48%, respectively).

75
9
410
The administration of nutritional support, both before admission
or during the hospital stay, was higher for patients identified
with high risk than for patients identified with medium or low
risk with the use of all 3 tools. In addition, high-risk patients

Low (n = 485)

0.45 6 1.18
identified with the use of all 3 tools experienced fever more
BMI SDSs within risk groups of 2–16-y-olds for 3 malnutrition-risk screening tools (n = 1253 of 1258)1

437
42
6
frequently and were prescribed more antibiotics than were
medium-risk-patients and low-risk-patients.
The LOS increased from low- to high-risk patients as identified
with the use of all 3 tools (Table 3). This result was also supported
High (n = 274)

20.78 6 1.55

by the effect estimates of the multivariate regression analysis with

age, sex, chronic disease, and center taken into account (Table 4).
147
67
60

Risk categorization and anthropometric measures

Medium (n = 222)

Mean SDSs for either BMI or HFA differed between the 3 risk
9.1 (6 of 66)
0.23 6 1.16

groups within each tool (Table 3; for additional details, see

PYMS

Supplemental Table 2). In addition, a considerable number of

190
30
2

children with low BMI SDS (,22) were not picked up as being
high risk (and were categorized either in the low- or medium
risk-category) with the use of the 3 tools. Table 5 displays rele-
vant differences in the 3 tools for the group of children (n = 1253)
Low (n = 757)

0.50 6 1.25

who completed all 3 tools and had BMI data available.

687
66
4

The MUAC and TSFTwere measured in 2263 study participants

(88%) and 2094 study participants (82%), respectively. Linear
regression results for all 3 screening tools showed a significant
relation between malnutrition risk and MUAC for both sexes after
All values are means 6 SDs.
,22 SDSs and not categorized
in the high-risk group, % (n)

adjustment for age, chronic disease, and center. SDSs for the MUAC
,21 to at least 22 SDSs, n

and TSFT for patients 2–5 y of age in relation to the risk groups of
each screening tool are shown in Table 6.
At least 21 SDS, n

,22 SDSs, n

DISCUSSION
The aim of all 3 screening tools is to identify children at risk of
2
TABLE 5

BMI SDS

malnutrition on admission to select patients for additional eval-

1

uation and a potential intervention. However, there are differences

concerning the use of these tools because they were designed
1308 CHOURDAKIS ET AL.
TABLE 6
MUAC and TSFT SDSs for children 2–5 y old within the risk groups for 3 malnutrition risk screening tools1
PYMS (n = 407) STAMP (n = 389) STRONGKIDS (n = 401)

Low Medium High Low Medium High Low Medium High

MUAC SDSs 0.52 6 1.17 2

0.24 6 1.18 20.27 6 1.13 0.44 6 1.15 0.31 6 1.11 20.21 6 1.33 0.67 6 1.13 0.17 6 1.29 20.81 6 1.16
At least 21 SDS, n 197 75 82 119 149 69 156 173 19
,21 to at least 22 13 12 16 10 18 13 5 27 9
SDSs, n
,22 SDSs, n 4 0 8 1 1 9 0 7 5
TSFT SDSs 1.13 6 1.22 0.85 6 1.12 0.42 6 1.33 0.96 6 1.23 0.88 6 1.15 0.75 6 1.50 1.09 6 1.23 0.87 6 1.30 0.34 6 1.32
At least 21 SDS, n 192 81 84 117 150 70 140 178 23

Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018

,21 to at least 8 2 10 7 4 8 7 9 3
22 SDSs, n
,22 SDSs, n 0 1 4 1 2 2 0 4 1
1
For MUAC SDSs, calculations included all children (PYMS: n = 407; STAMP: n = 389; STRONGKIDS: n = 401) who completed the tool and had an MUAC
measurement. For TSFT SDSs, calculations included all children (PYMS: n = 382; STAMP: n = 361; STRONGKIDS: n = 365) who completed the tool and had
a TSFT measurement. MUAC, midupper arm circumference; PYMS, Pediatric Yorkhill Malnutrition Score; SDS, SD-score; STAMP, Screening Tool for the
Assessment of Malnutrition in Pediatrics; STRONGKIDS, Screening Tool for Risk of Impaired Nutritional Status and Growth; TSFT, triceps skinfold thickness.
2
Mean 6 SD (all such values).

for application by different users (e.g., pediatricians and nurses)
and in different age groups (5, 13, 17). In addition, the PYMS
and STAMP include anthropometric measures, whereas the
STRONGKIDS focuses on identifying children at nutritional risk
on admission by visual inspection of body habitus alone.
This study showed marked differences in the number of pa-
tients who could be screened with the use of the 3 tools. Also, the
scores and classification of malnutrition risk in children varied
substantially according to the tool used. Few smaller studies that
were conducted previously looked into the agreement in nutri-
tional risk classification with the use of the PYMS, STAMP, and
STRONGKIDS and also showed this agreement to be modest
(19, 22–24). A lack of agreement may be explained by the fact
that the tools are different although they contain similar steps.
Although several components within the tools are similar, there
are discrepancies in the scoring, duration of recall history, and
approaches used to assess body size.
By definition (item 1), the PYMS was expected to categorize
all children with BMI SDSs ,22 into the high-risk category.
However, this was not the case for a low number of children
(7of 96 children) with subnormal BMI that were not identified
correctly with the use of the PYMS.
In this study, we assessed the discriminant validity of the
screening outcomes of each tool against body composition and
explored their ability to predict adverse clinical outcomes. For
each tool, we showed a reverse association between malnutrition
risk with body composition and a positive association with the
LOS. In particular, for each tool, children who scored a high risk
of malnutrition stayed longer in the hospital and had lower mean
MUAC and TSFT values than did patients with low or medium
risk. It should be emphasized that sensitivity and cutoffs of the
MUAC are still debatable, and the MUAC might be a more valuable
tool in the assessment of markedly malnourished children. How-
ever, the MUAC is often considered useful in the clinical assess-
ment and follow-up of patients.
The association between the risk-score classification and LOS
was strongest with the STRONGKIDS. However, it was unclear
how much of this association was explained by disease severity
and how much was attributed to the effect of malnutrition.
It is arguable what benchmark would be best for assessing the
value of a screening tool. Amaral et al. (3) and Kyle et al. (25)
showed a significant association between the screening score of
nutrition risk screening tools and the LOS in adults, but they
stated that the LOS is also influenced by many nonnutritional
factors. However, adverse effects of malnutrition and the in-
fluence of the underlying disease interact and both affect the
LOS, which should be considered when assessing associations of
risk scores and secondary outcomes such as fever or the use of
antibiotics.
We think that it is important for the tools to agree in the
detection of high-risk patients, including those with subnormal
BMI, HFA, and skinfold-thickness measurements, which was not
the case in the current study. We considered high-risk patients to
be those who needed to be referred for a more detailed as-
sessment and were more likely to need a nutritional intervention.
Moreover, screening tools are also aiming to identify children
at risk of deterioration of malnutrition risk because of an acute
medical insult despite normal anthropometric measures at hos-
pital admission. This identification encompasses a large pro-
portion of children who are admitted in acute settings in
developed countries, and the intervention and prevention of
weight loss are probably as important as the correction of
weight loss and growth catch-up in children who are already
malnourished (26).
Some strengths of our study are its multicenter setting and
the large number of participants from different countries. To our
best knowledge, this is the first study that compares 3 different
screening tools in a large pediatric population. We used one
growth reference (the WHO growth standard) for all children and
thereby excluded the variation between different country-specific
growth charts. However, we did not use disease-specific growth
charts that were available (e.g., for cerebral palsy patients) be-
cause these charts are available for only a few selected diagnoses
and have generally not been based on pan-European patient
populations. We also acknowledge that our study may have
suffered from a sample-selection bias, because some children who
were severely sick may not have joined the study. In addition,
a substantial number of children were receiving nutritional
 RISK OF MALNUTRITION IN HOSPITALIZED CHILDREN
support at study entry, which most likely reflected the profile of
patients who regularly attended the highly specialized hospitals
that participated in this study. An additional potential limitation
of this study is the fact that we did not perform a full nutritional
assessment as a reference for the comparison of screening scores
(1, 17). Moreover, with our data, we could not account for the
effects of disease groups or severity on the association between
malnutrition risk and clinical outcomes. The power to detect
nutrition-associated infections was limited by the generally short
LOS of the patients included in the study, which reflects current
clinical practice. Large differences were shown between coun-
tries, which may have reflected differences in population char-
acteristics or clinical practice. Furthermore, our study evaluated
the screening tools in the specific study population enrolled, and
the extrapolation of results to other populations must be done
cautiously.
Although significant associations were observed for all 3
tools between high risk of malnutrition with an increased LOS
and suboptimal anthropometric measures, the agreement between
tools in the classification of the same patients at same risk of
malnutrition was modest. Although screening tools have the
potential to enhance the awareness of clinicians regarding the
importance of the nutritional status of pediatric patients (1, 23),
raising awareness in health care professionals alone is not
a sufficient justification for establishing an additional in-
vestigation in patients. Rather, a reasonable prediction of risk
of malnutrition or of an outcome with a good sensitivity and
specificity is expected as a prerequisite for the clinical routine use
of a screening tool.
Although the STRONGKIDS is not based on anthropometric
measurements, the authors who have described the STRONGKIDS
have also advocated for the measurement of weight and height as
part of the assessment of nutritional status on admission after the
initial risk screening. The PYMS and STAMP are based on an-
thropometric measures and, thus, detect the large majority of
children with abnormal anthropometric measures (26, 27). How-
ever, the use of these tools may come at the expense of too many
children being categorized as high risk. Other aspects also need to
be considered, such as the clinical performance and impact of any
selected tool on current health care resources (e.g., staff workload
and practicality).
In conclusion, the identification and classification of risk of
malnutrition varied between tools and countries. The agreement
between tools was modest, which is a finding that might be partially
attributed to the absence of broadly agreed-upon definitions and
measures of pediatric malnutrition. On the basis of these findings,
we cannot conclude that any one tool would be superior over
another. Beyond the diagnostic validity, we recommend that the
selection of the most appropriate tool for routine use on hospital
admission further depend on documented and validated clinical
performance of the tools and on the availability of and impact on
health care resources. At this time, the use of pediatric screening
tools for assessing nutritional risk in routine clinical practice cannot
be recommended.
We thank Joachim Schweizer for his input to the design of the study and Veit
Grote and Martina Weber for help with the statistical analysis. We also thank
Meetanand Baichoo, Carmen Mihaela Culcitchi, Efstratia Daskalou, Laura
Dell’Era, Diana Flynn, Frederic Gottrand, Corina Hartman, Sanja Kolacek,
Tena Niseteo, Katarzyna Olszewska, Anna Piwowarczyk, Julien Rousseaux,
Paola Pavesi, Peter B Sullivan, and Angharad Vernon-Roberts and all other
1309
members of the European Society for Clinical Nutrition and Metabolism
Network project as listed in Hecht et al (18) for their help on site.
The authors’ responsibilities were as follows—MC: coordinated intra-
group reviews and communication, helped with the statistical analyses, and
drafted the manuscript; MC, CH, KG, KFMJ, BK, and JMH: commented on
the first and subsequent drafts of the manuscript; CH: contributed to the
writing of the study protocol and the first draft of the manuscript, coordi-
nated the study, participated in conducting the study, and performed the data
entry, management, and analyses; KG: participated in the initial part of the
study design, contributed to the sample collection, and coordinated intra-
group reviews and communication; KFMJ, TK-L, HAK, JK, CL, RS, HS,
and JMH: participated in the initial part of the study design, contributed to
the sample collection, and were responsible for the data acquisition, data
Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018
interpretation, and analysis; BK: conceived the study, participated in its de-
sign, contributed to the writing of the study protocol, and helped to draft the
manuscript; and all authors: read and approved the final manuscript. The
presented data were part of a PhD thesis accomplished by CH at the Medical
Faculty of the Ludwig-Maximilians-University of Munich. KG, KFMJ, and
JMH have been involved in the development of the STRONGKIDS and
PYMS, respectively. None of the authors reported a conflict of interest
related to the study.
REFERENCES
1. Gerasimidis K, Macleod I, Maclean A, Buchanan E, McGrogan P,
Swinbank I, McAuley M, Wright CM, Flynn DM. Performance of the
novel Pediatric Yorkhill Malnutrition Score (PYMS) in hospital
practice. Clin Nutr 2011;30:430–5.
2. McCarthy H, Dixon M, Crabtree I, Eaton-Evans MJ, McNulty H. The
development and evaluation of the Screening Tool for the Assessment
of Malnutrition in Pediatrics (STAMP) for use by healthcare staff. J
Hum Nutr Diet 2012;25:311–8.
3. Amaral TF, Antunes A, Cabral S, Alves P, Kent-Smith L. An evaluation
of three nutritional screening tools in a Portuguese oncology centre. J
Hum Nutr Diet 2008;21:575–83.
4. Campanozzi A, Russo M, Catucci A, Rutigliano I, Canestrino G,
Giardino I, Romondia A, Pettoello-Mantovani M. Hospital-acquired
malnutrition in children with mild clinical conditions. Nutrition 2009;
25:540–7.
5. Hulst JM, Zwart H, Hop WC, Joosten KF. Dutch national survey to
test the STRONGkids nutritional risk screening tool in hospitalized
children. Clin Nutr 2010;29:106–11.
6. Rocha GA, Rocha EJ, Martins CV. The effects of hospitalization on the
nutritional status of children. J Pediatr (Rio J) 2006;82:70–4.
7. Sermet-Gaudelus I, Poisson-Salomon AS, Colomb V, Brusset MC,
Mosser F, Berrier F, Ricour C. Simple pediatric nutritional risk score to
identify children at risk of malnutrition. Am J Clin Nutr 2000;72:64–70.
8. Agostoni C, Axelson I, Colomb V, Goulet O, Koletzko B, Michaelsen KF,
Puntis JW, Rigo J, Shamir R, Szajewska H, et al. The need for nutrition
support teams in pediatric units: a commentary by the ESPGHAN com-
mittee on nutrition. J Pediatr Gastroenterol Nutr 2005;41:8–11.
9. Kondrup J, Allison SP, Elia M, Vellas B, Plauth M, Educational and
Clinical Practice Committee, European Society of Parenteral and
Enteral Nutrition (ESPEN). ESPEN guidelines for nutrition screening
2002. Clin Nutr 2003;22:415–21.
10. Stratton RJ, King CL, Stroud MA, Jackson AA, Elia M. ’Malnutrition
Universal Screening Tool’ predicts mortality and length of hospital stay
in acutely ill elderly. Br J Nutr 2006;95:325–30.
11. Kruizenga HM, Seidell JC, de Vet HC, Wierdsma NJ, van Bokhorst-de
van der Schueren MA. Development and validation of a hospital
screening tool for malnutrition: the short nutritional assessment ques-
tionnaire (SNAQ). Clin Nutr 2005;24:75–82.
12. Secker DJ, Jeejeebhoy KN. Subjective Global Nutritional Assessment
for children. Am J Clin Nut 2007;85:1083–9.
13. McCarthy H, McNulty H, Dixon M, Eaton-Evans M. Screening for
nutrition risk in children: the validation of a new tool. J Hum Nutr Diet
2008;21:395–96.
14. McDonald CM. Validation of a nutrition risk screening tool for chil-
dren and adolescents with cystic fibrosis ages 2-20 years. J Pediatr
Gastroenterol Nutr 2008;46:438–46.
15. Joosten KF, Hulst JM. Nutritional screening tools for hospitalized
children: methodological considerations. Clin Nutr 2014;33:1–5.
1310 CHOURDAKIS ET AL.
16. Gerasimidis K, Macleod I, Finlayson L, McGuckin C, Wright C, Flynn
D, McGrogan P, Maclean A, Love E, Swinbank I, et al. Introduction of
Pediatric Yorkhill Malnutrition Score–challenges and impact on
nursing practice. J Clin Nurs 2012;21:3583–6.
17. Gerasimidis K, Keane O, Macleod I, Flynn DM, Wright CM. A
four-stage evaluation of the Pediatric Yorkhill Malnutrition Score in
a tertiary pediatric hospital and a district general hospital. Br J Nutr
2010;104:751–6.
18. Hecht C, Weber M, Grote V, Daskalou E, Dell’Era L, Flynn D,
Gerasimidis K, Gottrand F, Hartman C, Hulst J, et al. Disease associ-
ated malnutrition correlates with length of hospital stay in children.
Clin Nutr 2015;34:53–9.
19. Ling RE, Hedges V, Sullivan PB. Nutritional risk in hospitalised
children: an assessment of two instruments. European e-journal of
clinical nutrition and metabolism 2011;6:e153–e7.
20. Landis JR, Koch GG. The measurement of observer agreement for
categorical data. Biometrics 1977;33:159–74.
21. Viera AJ, Garrett JM. Understanding interobserver agreement: the
kappa statistic. Fam Med 2005;37:360–3.
22. Wiskin AE, Owens DR, Cornelius VR, Wootton SA, Beattie RM.
Pediatric nutrition risk scores in clinical practice: children with
inflammatory bowel disease. J Hum Nutr Diet 2012;25:319–22.
23. Moeeni V, Walls T, Day AS. Assessment of nutritional status and
nutritional risk in hospitalized Iranian children. Acta Paediatr 2012;
101:e446–51.
24. Moeeni V, Walls T, Day AS. Nutritional status and nutrition risk
screening in hospitalized children in New Zealand. Acta Paediatr 2013;
102:e419–23.
25. Kyle UG, Genton L, Pichard C. Hospital length of stay and nutritional
status. Curr Opin Clin Nutr Metab Care 2005;8:397–402.
26. Milani S, Wright C, Purcell O, Macleod I, Gerasimidis K. Acquisition
and utilisation of anthropometric measurements on admission in
a pediatric hospital before and after the introduction of a malnutrition
Downloaded from https://academic.oup.com/ajcn/article-abstract/103/5/1301/4637451 by guest on 28 September 2018
screening tool. J Hum Nutr Diet 2013;26:294–7.
27. Grek J, Puntis J. Nutritional assessment of acute medical admissions
is still done badly despite ‘nutrition screening’. Arch Dis Child 2013;
98:922–3.