paediatric weight estimation by age in the digital era ... · web viewpaediatric weight estimation...
TRANSCRIPT
Paediatric weight estimation by age in the digital era:
optimising a necessary evil
Nicholas Appelbaum (1), Jonathan Clarke (1), Ian Maconochie (2) and
Ara Darzi (1)
Author Affiliations :
1 – Department of Surgery and Cancer, Division of Surgery, Imperial College
London
2 – Department of Emergency Medicine, Division of Medicine, Imperial
College London
Word counts:Abstract: 236 words
Manuscript: 2986 words (excluding acknowledgements)
Full contact information for corresponding author:
Nicholas Appelbaum
Department of Surgery and Cancer
Division of Surgery
Imperial College London
10th Floor, QEQM Building
St Mary’s Hospital
Praed Street, London, W2 1NY
Tel: +44 (0) 20 3312 1310
Fax: +44 (0)20 3312 6950
Mob: +44 7470 2588885
Web: www.imperial.ac.uk/ighi
www.helixcentre.com
Abstract
Background: Age-based weight estimation methods are regularly used in
paediatric emergency medicine despite their well-established inaccuracy.
Aim: Determine the potential improvement in accuracy achievable by the use
of a new mobile application, based on CDC/WHO weight-for-age centile data,
which incorporates a gender assignment, a body habitus assessment, and
which is capable of an age-in-months based calculation.
Methods: A theoretical, simulated validation study, comparing the
performance of the widely used APLS / EPALS formulae against two
contemporary habitus-adjusted methods, and the Helix Weight Estimation
Tool. 1,070,743 children from the 2015/2016 UK National Child Measurement
Program dataset, aged between 4 and 5 and 11 and 12 years, had age-based
weight estimates made by all five methods.
Results: Primary outcomes were the percentage of weight estimations within
10%, 20%, and those greater than 20% discrepant from actual weight for
each method. Our theoretical, gender-dependent, habitus-adjusted method
performed better than all other methods across all error thresholds. The
overall number of estimations within 10% was 70.4%, and within 20% was
95.45%. The mean percentage error was -1% compared to actual weight.
Conclusion: The use of a digital tool incorporating a subjective assessment
of body habitus, gender assignment, and the ability to estimate weight based
on age-in-months might be able optimise the process of paediatric weight
estimation by age, making this practice as safe and accurate as possible for
the occasions when weight estimation by age is chosen over length-based
methods.
Manuscript Text
Introduction
Background
Why do we estimate bodyweight?
Drugs for children are generally prescribed on the basis of age and on body
weight. However, frequently a critically ill child arrives in the Emergency
Department resuscitation room with his/her weight unknown to the care team,
under conditions where it is not possible to weigh the patient before
commencing emergency care. Examples of conditions which make weighing
the child unfeasible include spinal immobilisation, ongoing cardiopulmonary
resuscitation, emergency airway management, or severe pain that inhibits
movement of the child (1-3).
The accurate estimation of weight is important in paediatrics, e.g. for the
calculation of drug doses, the determination of equipment size for each child,
and the energy levels required for defibrillation. Inaccurate and imprecise
weight estimation techniques contribute to the high incidence of drug errors in
paediatric emergency medicine (4, 5).
Current methods of weight estimation
Estimation of weight by age
In spite of the fact that age-based formulae have consistently demonstrated
poor predictive accuracy, particularly in older children (6-8), there has been a
constant appetite to improve these methods, and from the 1950’s onwards, at
least 22 age-based formulae have been derived to estimate the bodyweight of
children. Some of these calculations require complex workings which, in a
stressful clinical environment, provide additional opportunities for erroneous
arithmetic to contribute to drug error (9).
What methods offer acceptable accuracy?
There is some consensus in the literature that an ‘accurate’ method would
demonstrate accuracy as 60-70% of weight estimations within 10% of actual
weight (PW10) and a moderate critical error rate as having 90-95% of
estimations within 20% of actual weight (PW20) (10, 11).
It is well established that the most accurate methods of paediatric weight
estimation are length-based methods rather than age-based methods (6, 8,
12-14), and only length-based methods have ever achieved this level of
accuracy.
Five age-based formulae have been validated in the UK in at least sixteen
studies we are aware of. The best performing formulae have been the
Luscombe-Owens Formula (2) (derived in Sheffield), with PW10=43%, and
the Tinnings Formula (15), PW10=43.7% - both in a 2011 validation by
Marlow et al. (6). The Luscombe-Owens formula has been incorporated in
part into the new Advanced Paediatric Life Support (APLS) formula.
Paediatric weight estimation and the childhood obesity pandemic
The increasing incidence of childhood obesity (16) has recently led to the
reinvigorated interest in methods that can adjust for the variability in body
habitus in children which was first introduced with the Derived Weight
Estimating Method (DWEM) (17) in 1986. This has until recently been
predominantly applied to length-based methods, and examples of this include
the Paediatric Advanced Weight Prediction in the Emergency Room
(PAWPER) tape (13), Mercy method (18, 19) (using mid upper arm
circumference and humeral length), Yamamoto Obesity Icon system (20), and
the waist-circumference modified Broselow system (21).
Two studies have evaluated how the addition of an assessment of body
habitus could improve the accuracy of age-based weight estimation methods.
These are (where Z = age in years):
The Erker formula (22)
Wt=(2×Z)+6 For ‘tall ‘n thin’ children
Wt=(3×Z)+6 For ‘normal’ children
Wt=(4×Z)+6 For ‘tiny ‘n thick’ children
The Wells ‘derived formula’ (7)
HS1 : Wt = (1.9×Z)+5.8
HS2 : Wt = (2.3×Z)+5.8
HS3 : Wt = (2.4×Z)+7.5
HS4 : Wt = (2.9×Z)+8.3
HS5 : Wt = (3.7×Z)+8.1
In the PAWPER system, Wells et al (13) have added to the area of subjective
assessment of habitus (improving upon the Yamamoto method) by developing
a 5-point (later 7-point) (10) visual scale of body habitus scores (HS), and
these scores have been used in the multipart formula above.
Both of these methods have only been the subject of investigation in one
validation study (in a South African population, n=963 for Erker, n=635 for
Wells) where neither demonstrated ‘acceptable accuracy’, or 60% of
estimations within 10% of actual bodyweight. Additionally, without the use of a
digital tool, these increasingly complex equations pose the risk of
miscalculation error.
To our knowledge, no habitus-adjusted age-based methods have ever been
validated in the United Kingdom.
Why do age-based methods continue to be used?
Despite the availability of more accurate methods, weight based formulae are
used throughout the world, and taught on all paediatric life support courses.
Principally, however, it is because age-based methods do not require any
specific equipment that their use endures.
There is therefore a need to examine the performance of the newest age-
based weight estimation methods in English children, and to see if the
development of a simple digital tool might be able to add accuracy and safety
to weight estimation by age where this method is chosen over length-based
methods.
Objectives
1. Determine the accuracy of the APLS / European Paediatric Advanced
Life Support (EPALS) formulae and two age-based, habitus-adjusted
body weight estimation methods in a large, 2 age-banded cohort of
English children.
2. Determine the potential improvement in accuracy achievable by the
use of a new mobile application, based on CDC/WHO weight-for-age
centile data, which incorporates a gender assignment, a body habitus
assessment, and which is capable of an age-in-months based
estimation.
It was our hypothesis that this experimental method would demonstrate
improved accuracy of body weight estimation over currently taught and used
methods.
Methods
Study Design
This was a theoretical, simulated validation study, comparing the performance
of the widely used APLS / EPALS formulae with the Wells and Erker formulae,
and a new mobile application based method using WHO/CDC centile data, a
correction for body habitus, and a gender assignment. All estimates and
comparisons were made based upon and compared with English data from
the 2015/2016 UK National Child Measurement Program (NCMP) dataset
(23).
Data sources
NHS Digital publishes publicly available data from the annual NCMP survey
on every child across the UK, as each has their height and weight measured
in Reception Year (age 4-5), and again in Year 6 (age 10-11). In addition to
multiple anthropometric parameters derived from comparison to the UK 1990
Growth Standard (24), each entry contains the school local authority code.
WHO and CDC centile datasets are freely available from the US National
Center for Health Statistics (25). For this study, only CDC centile data were
required as the CDC recommendation is to use WHO centiles for children
under the age of two, and all children in the dataset used for this two age-
banded validation were older than two.
Data management
The NCMP dataset is made available after suppression in line with the NHS
Anonymisation Standard. Extreme outliers of age-for-weight above the
99.995th percentile (3192 records) and below the 0.005th percentile (631
records) were removed from the dataset. In addition to this, 89260 records
were suppressed where the local authority code and a locally small population
might have allowed for identification of an individual.
Overall, 1,076,908 records were available for analysis.
The data were processed using the Python programming language (Python
Software Foundation. Python Language Reference, version 3.6), and the
pandas library (version 0.20.2, http://pandas.pydata.org), in the Jupyter
Notebook computational environment (https://jupyter.org).
To map the NCMP data onto the CDC data, child age-in-months was rounded
to the nearest half-month. Limited secondary data cleaning, removing extreme
outliers for age was performed: in reception year, age in months 48.5, 49.5,
and 70.5, and for Year 6 children, 120.5, 141.5, and 142.5 months. Table 1
shows the overall study population characteristics.
The total sample size analysed was 1,070,743 children.
Analysis of the APLS, EPALS and habitus-adjusted formulae
For each child in the dataset, their predicted weight was calculated using the
current APLS formula, the current EPALS formula, the Erker formula, an
estimate based on the CDC 50th centile weight for age, and the Wells formula.
For the Erker formula, we considered ‘tall ‘n thin’ children as those with a
height-for-age above the 90th percentile, and a weight-for-height below the 10th
percentile. We considered “short ‘n thick” children as those below the 10 th
percentile of height-for-age, and above the 90 th percentile for weight-for-
height. All other children were considered “normal”.
For the Wells formula, habitus scores between HS1 and HS5 were allocated
to the median BMI Z-scores from the original paper (7). All subjects in the
dataset were then allocated to the nearest HS according to their actual BMI.
The Helix weight estimation method
We chose to use 7 habitus scores, each mapping closely onto the habitus
scores used by Wells et al in the PAWPER-XL system (10). We assigned
CDC weight-for-age and BMI-for-age centiles to each habitus score as
follows: HS1=10th, HS2=25th, HS3=50th, HS4=75th, HS5=90th, HS6=95th, and
HS=97th centiles.
Each child in the dataset was allocated to the CDC BMI-for-age centile (10 th,
25th, 50th, 75th, 90th, 95th, 97th) closest to their actual BMI from the NCMP
dataset, and the corresponding CDC weight-for-age centile was used to
determine an estimated weight.
Determination of accuracy and precision
Each of these estimates was compared to the child’s actual weight from the
NCMP dataset. For each weight estimate, the percentage error (PE) and
absolute percentage error (APE) were calculated. W e is estimated weight, and
W t is actual patient weight.
PE=(W e−W t
W t)×100
APE=|(W e−W t
W t )|×100
The percentage of weight estimations within a threshold range (0% - 100% of
estimations) was calculated. Estimates within 10% (PW10), 20% (PW20) and
those being greater than 20% from actual weight (PW>20) were calculated
from these results.
The overall accuracy of the models was tested by calculating the mean
percentage error (as a measure of estimation bias) and the root mean square
percentage error (a measure of the standard deviation around the true value)
(11).
MPE=1n∑e=1
n
(W e−W t
W t)
RMSPE=√ 1n∑e=1n (W e−W t
W t)2
95% limits of agreement were calculated as a measure of precision.
Assessing the geographic burden of error for the APLS formula
We studied the geographical variation of the error burden across England for
the APLS method alone, as it is the most commonly used method. We pooled
children by their local authority code, and calculated the mean percentage
error, as well as PW10, PW20 and PW>20 with for each local authority. This
data was visualised using Tableau Desktop (Tableau Software, Seattle
Washington, USA).
Statistical analysis
Frequencies and proportions were reported for categorical variables, medians
and IQRs for continuous variables. The median absolute percentage error for
each estimation method was compared using the Mann-Whitney U-test. A
p<0.0001 was considered significant. Previous studies have examined the
mean percentage error as a primary outcome measure, however, this
provides little insight, as two methods with very different error distributions but
the same mean error would appear identical. By taking the absolute
percentage error and comparing errors non-parametrically it is possible to
assess the overall precision of the methods used. Statistical analysis was
performed using Stata (Stata-Corp. 2015. Stata Statistical Software: Release
14. College Station, TX: StataCorp LP)
The primary outcome was the performance of the various age-based weight
estimation methods compared to actual weight. A PW10>60% and a
PW20>90% was chosen as representing safe performance.
Results
Figure 1 shows the pooled performance (as PW10, PW20 and PW>20) for all
methods across the entire study population. Figure 2 shows the performance
of each method stratified by age and gender.
The performance of the APLS and EPALS formulae (these are the same
formula in the reception-year band) was poor in both age ranges. The pooled
PW10 for the APLS formula was 36.5% for the EPALS formula, 28.1%.
Not only were these formulae highly inaccurate, but their error distribution was
such that the percentage of estimations in excess of 20% discrepant from
actual weight was particularly high. This was more pronounced in the older
children, rising from 25.5% in Reception Year boys to 36.6% in Year 6 boys
for the APLS formula, and 59.9% for the EPALS formula in the older boys.
The performance and error distribution of the APLS formula varied
considerably by local authority. Figure 3 shows the mean percentage error for
the APLS formula as a ‘heatmap’ across England. The MPE, PW10, PW20
and PW>20 results from two different local authorities have been shown to
demonstrate the degree of the heterogeneity. There is, however, a small
contribution to this error burden made by the average time during the school
year that children are weighed. Underestimations will be exaggerated in local
authorities where schools, on average, weigh late in the school year.
The two habitus-adjusted formulae performed better than the non habitus-
adjusted formulae and performed better in the younger children. The pooled
PW10 for the Erker and Wells formulae were 53.7% and 59.4% respectively.
The performance of the Erker and Wells formulae was similar to that reported
in a recent validation (7).
The CDC median method was used for comparison only and is not used
clinically anywhere to our knowledge.
Our theoretical, gender-dependent, habitus-adjusted method performed better
than all other methods across all error thresholds. The overall PW10 was
70.4%. As with all methods, performance was better in the younger children. It
was, however, in the older children, in particular the older girls (who had a
higher median BMI), where this new method showed the most significant
improvement over the other formulae. The PW10 for Year 6 girls was 66.2%,
compared to 48% for the Erker formula, and 46.5% for the Wells formula. The
drop in PW10 between the Year 6 boys and girls was 3.1% for the
experimental method, and 12.7% for the best performing formula.
Table 2 shows the accuracy and precision measures of the model stratified by
habitus score and age. The model performed ‘accurately’ (PW10>60%) in all
categories except for Habitus Score 7 in the Year 6’s, where PW10=55.9%
MPE +5.6%.
The overall estimation bias (MPE) was -1%. The limits of agreement around
the MPE were wide, as is the case with all age-based methods.
Discussion
This is a first, theoretical validation of a weight estimation method which
employs a body habitus correction, an estimation based on age-in-months,
and which is gender dependent.
Our group is affiliated with the Helix Centre, a multidisciplinary team of
clinicians, designers and technologists at St Mary’s Hospital in London.
Although we argument above that it is precisely because of the lack of a need
for specific equipment that weight estimation by age persists, almost all
emergency clinicians have on their person a device capable of safely making
almost any calculation: their mobile telephone. This model has thus been
packaged into a simple mobile application, which will be made freely available
to interested researchers and parents as part of an invitation to participate in
an open, worldwide validation of this methodology.
Limitations
The principal limitation to this study is that the NCMP dataset only includes
children of ages 4-5 and 11-12 years, albeit with a considerable sample size.
These two age-bands sit on the weight-for-age growth curve at the point
which the growth curves begin to widen, and where habitus assessment is
more critical to model accuracy. Assumptions cannot be made about the
accuracy of this model outside of these age ranges.
The models assume perfect allocation of subjects to habitus scores and error
has not been modeled for any of the methods. There is considerable
subjectivity in visual assessment of habitus. Correlation between evaluator
assessment of patient habitus and actual BMI-for-age centile has been found
by Garcia et al, to be suboptimal in one study, although for a considerably
simpler tool (11).
Because it may be difficult to distinguish between adjacent body habitus
scores when examining patients, misclassification will almost certainly occur
for both the experimental model and the Wells formula, with a consequent
decrease in performance. Additionally, considering the fact that a short child
weighs less than a tall child of similar age and habitus, a visual best-guess
would take this into account, whereas a habitus adjusted model would
overestimate the weight of a shorter than average, moderately overweight
child. This is a fundamental shortcoming of weight-by-age methods.
It is possible that this analysis has imposed restrictions that might have
negatively affected the performance of the Erker formula compared to actual
life. By segmenting the allocations by children who were both ‘tall and thin’, or
‘short and thick’, the number of reallocations was relatively low. In clinical
practice, it is likely that an obese child would be allocated to the ‘tall and thin’
formula as it is obvious this formula will yield the greater estimated weight.
Pharmacological considerations
The addition of a body habitus correction to an age-based methodology more
accurately predicts total body weight than ideal body weight.
It has been reported that there is a bias towards hydrophilic drugs (e.g.
adrenaline, amiodarone) in paediatric CPR (26), which should ideally be
dosed against an ideal bodyweight, as this gives a better estimate of the
volume of distribution. A method returning a total bodyweight estimate in
overweight children could theoretically result in overdosing of medications that
do not distribute widely into adipose tissue.
There are many drugs, however, that are used regularly in paediatric
emergencies (e.g. benzodiazepines) that are hydrophobic, and where a total
body weight estimate is appropriate.
Computerised decision support systems will likely be developed in the future
where individualised dosing against pharmacokinetic considerations is
possible, even in the resuscitation environment. It is important to consider the
bearing weight estimation methods will have on this.
Conclusions
Weight estimation by length, and even better, habitus-adjusted length-based
models will almost always outperform age-based methods, and so these
methods should be considered the first line method of choice. However,
weight estimation by age is a persistent phenomenon despite the well
documented inaccuracy of the various methods.
By incorporating a subjective assessment of body habitus and a gender
assignment into a simple digital tool capable of an age-in-months based
calculation in the case a date of birth is known, it may be possible to optimise
weight estimation by age, making it at least as safe and accurate as possible
for the occasions when weight estimation by age is chosen over length-based
methods.
This mobile application needs to be validated in the clinical environment, and
our research team looks forward to working with interested collaborators in
order to do so.
Acknowledgements
We are grateful for the time and assistance we have received from the
following people:
Dr Philip Pratt, Dr Eduardo D’Aguilar and John Morrell for their advice and
technical assistance. Sara Vrbinc for her work on the user interface and
constant advocacy for usability in all of our work.
References
1. Wells M. Weight prediction in children in the emergency department. (Masters thesis) Johannesburg: University of the Witwatersrand; 2009.
2. Luscombe MD, Owens BD, Burke D. Weight estimation in paediatrics: a comparison of the APLS formula and the formula 'Weight=3(age)+7'. Emerg Med J. 2011;28(7):590-3.
3. Luscombe M, Owens B. Weight estimation in resuscitation: is the current formula still valid? Arch Dis Child. 2007;92(5):412-5.
4. Selbst SM, Fein JA, Osterhoudt K, Ho W. Medication errors in a pediatric emergency department. Pediatr Emerg Care. 1999;15(1):1-4.
5. Kaushal R, Bates DW, Landrigan C, McKenna KJ, Clapp MD, Federico F, et al. Medication errors and adverse drug events in pediatric inpatients. JAMA. 2001;285(16):2114-20.
6. Marlow RD LD, Walton LJ. . Accurate paediatric weight estimation by age: mission impossible? Arch Dis Child. 2011;96(A1-A2).
7. Wells M, Goldstein LN, Bentley A. It is time to abandon age-based emergency weight estimation in children! A failed validation of 20 different age-based formulas. Resuscitation. 2017;116:73-83.
8. Young KD, Korotzer NC. Weight Estimation Methods in Children: A Systematic Review. Ann Emerg Med. 2016;68(4):441-51 e10.
9. Gavriel Salvendy (Ed.): Handbook of human factors and ergonomics (3rd ed.). Universal Access in the Information Society. 2007;5(4):421-.
10. Wells M, Goldstein, L., Bentley, A. A validation study of the PAWPER XL tape: accurate estimation of both total and ideal body weight in children up to 16 years of age. Trauma Emerg Care. 2017;2(4):1-8.
11. Garcia CM, Meltzer JA, Chan KN, Cunningham SJ. A Validation Study of the PAWPER (Pediatric Advanced Weight Prediction in the Emergency Room) Tape--A New Weight Estimation Tool. J Pediatr. 2015;167(1):173-7 e1.
12. Wells M, Kramer E. Optimizing emergency drug dosing in children. Acad Emerg Med. 2008;15(12):1325; author reply 6.
13. Wells M, Coovadia A, Kramer E, Goldstein L. The PAWPER tape: A new concept tape-based device that increases the accuracy of weight estimation in children through the inclusion of a modifier based on body habitus. Resuscitation. 2013;84(2):227-32.
14. Black K, Barnett P, Wolfe R, Young S. Are methods used to estimate weight in children accurate? Emerg Med (Fremantle). 2002;14(2):160-5.
15. Tinning K, Acworth J. Make your Best Guess: an updated method for paediatric weight estimation in emergencies. Emerg Med Australas. 2007;19(6):528-34.
16. Ogden CL, Carroll MD, Kit BK, Flegal KM. Prevalence of childhood and adult obesity in the United States, 2011-2012. Jama. 2014;311(8):806-14.
17. Garland JS, Kishaba RG, Nelson DB, Losek JD, Sobocinski KA. A rapid and accurate method of estimating body weight. Am J Emerg Med. 1986;4(5):390-3.
18. Abdel-Rahman SM, Paul IM, James LP, Lewandowski A, Best Pharmaceuticals for Children Act-Pediatric Trials N. Evaluation of the Mercy TAPE: performance against the standard for pediatric weight estimation. Ann Emerg Med. 2013;62(4):332-9 e6.
19. Abdel-Rahman SM, Ahlers N, Holmes A, Wright K, Harris A, Weigel J, et al. Validation of an improved pediatric weight estimation strategy. J Pediatr Pharmacol Ther. 2013;18(2):112-21.
20. Yamamoto LG, Inaba AS, Young LL, Anderson KM. Improving length-based weight estimates by adding a body habitus (obesity) icon. Am J Emerg Med. 2009;27(7):810-5.
21. Tanner D, Negaard A, Huang R, Evans N, Hennes H. A Prospective Evaluation of the Accuracy of Weight Estimation Using the Broselow Tape in Overweight and Obese Pediatric Patients in the Emergency Department. Pediatr Emerg Care. 2016.
22. Erker CG, Santamaria M, Moellmann M. Size does matter--age-related weight estimation in "tall n' thin" and "tiny n' thick" children and a new habitus-adapted alternative to the EPLS-formula. Resuscitation. 2014;85(9):1174-8.
23. Report of the National Childhood Measurement Program for the 2016/2016 School Year. NHS Digital 2016. [Available from: https://digital.nhs.uk/catalogue/PUB22269, accessed 1/5/2017]
24. Cole TJ. Growth monitoring with the British 1990 growth reference. Arch Dis Child. 1997;76(1):47-9.
25. Clinical Growth Charts: National Center for Health Statistics; [Available from: https://www.cdc.gov/growthcharts/clinical_charts.htm, accessed 1/4/2017]
26. Carasco CF, Fletcher P, Maconochie I. Review of commonly used age based weight estimates for paediatric drug dosing in relation to the pharmacokinetic properties of resuscitation drugs. Br J Clin Pharmacol. 2015.