Professional Documents
Culture Documents
Marlow 2019
Marlow 2019
Arch Dis Child: first published as 10.1136/archdischild-2018-314873 on 5 July 2018. Downloaded from http://adc.bmj.com/ on 8 July 2018 by guest. Protected by copyright.
estimation methods: a simulation study
Robin D Marlow,1,2 Dora L B Wood,3 Mark D Lyttle2,4
Arch Dis Child: first published as 10.1136/archdischild-2018-314873 on 5 July 2018. Downloaded from http://adc.bmj.com/ on 8 July 2018 by guest. Protected by copyright.
While overall the table method had the lowest frequency of
Method errors, a proportion of these were disproportionately large,
Age range Formula mainly due to month/year confusion.
Single-formula APLS 1–10 years Weight=(age+4)×2
(APLS1)
Three-formula APLS <12 months Weight=(0.5×age in months)+4
Discussion
(APLS3) 1–5 years Weight=(2×age)+8
Our novel method of assessing usability found significant differ-
ences in the speed of use of different age-based methods of weight
6–12 years Weight=(3×age)+7
estimation. Using this as a proxy for task difficulty suggests that
Modified Best Guess <12 months Weight=(age in months+9)/2
method
the use of single formula or a reference table is an easier task to
1–9 years Weight=(2×age)+10
(BGM) carry out than the other methods. However, speed did not auto-
10–14 years Weight=(4×age)
matically correlate with accuracy of application. While reference
Reference table* 0–16 years Weight=50th centile for age tables were faster and there was a non-significant trend to more
Derived from UK-WHO
growth charts16
frequent correct usage, the magnitude of errors was significantly
smaller using the three-formula APLS method.
*Reference table as provided to participants available as online supplementary file.
Many studies have examined the predicted accuracy of esti-
APLS, Advanced Paediatric Life Support.
mation methods against a population’s true weight with no
consensus of the ideal method,6 but few studies have examined
approval was not required and so was not sought. As a pilot their practical application in real-world settings. Age-based
study no power calculation was performed. calculations tend to underestimate true weight but instead
Intermethod variability of speed and accuracy were analysed for predict an ideal bodyweight which, due to the pharmacological
normally distributed data using one-way repeated measures anal- properties of most resuscitation drugs, may be more suitable.7
ysis of variance with post hoc pairwise t-tests, and for non-para- The only other study comparable to ours applied both age and
metric data we used Kruskal and Dunn tests. P values were adjusted length-based methods to a cohort of 80 children in the USA, and
for multiple testing using the Holm method. Analysis was done found a 5% error of application rate, with device-based methods
using R.5 Comparing the responses with the expected answers we (eg, tapes) having the highest rate of application errors (25%).8
categorised errors into simple arithmetical, data entry or month/ The additional choices and calculation inherent in paedi-
year confusion. Answers were considered an error if not equal atric resuscitation increase cognitive loading.9 An alternative
to the result generated by the formulae/table for the current age method to avoid weight estimation altogether by having preas-
being assessed and were analysed as rates, absolute and percentage signed equipment and drawn up medications in broad age-based
difference from correct answer. Errors more than 10 times larger categories.10 Although conceptually simpler this requires dedi-
or smaller than the correct answer were deemed input errors and cated colour-coded equipment and leads to improvement but
excluded from the accuracy analyses. not abolition of error.11 With an increasing move in medicine
to checklists for cognitive unloading, the use of a prepopulated
Results table of weights, drug doses and equipment sizes for age is highly
Fifty-seven acute paediatric staff (table 2) completed the test appealing and would seem to facilitate the optimum balance of
with a median of 1.4 attempts each (range 1–6). In practice, 71% usability and accuracy. We seem to have come full circle as these
reported routinely using the single-formula APLS method. Post- were first described in the 1980s,12 but with concerns over their
test 69% reported a preference for the reference table method, accuracy13 they fell out of favour in preference for the APLS
finding it easiest to use. calculation methods. While technological innovations including
Comparing the total time taken to complete the tests, the device applications promise the potential for more complex
reference table method was significantly quicker to use than calculations, the practicalities of touchscreens, blood and latex
other methods (p<0.001) (table 3). Comparing the accuracy gloves represent significant barriers in high-pressure clinical
of the methods, there were significantly more errors made scenarios. With critically unwell or injured children we feel the
using the single-formula APLS compared with using the refer- paradigm must be avoidance of errors through simplification.
ence table (p=0.002). However, using the single-formula APLS Poor design in aviation14 and anaesthesia15 cognitive aids has
most errors occurred when it was applied to 6 month-olds; as been linked to harmful effects. Our data suggest that reference
the formula is only recommended between the ages of 1 and 10 tables should also be designed in a way to minimise potential
years we repeated the analysis excluding the 6-month task. This picking errors.
demonstrated no significant difference in accuracy between As with all clinical medicine we have to be pragmatic; esti-
these methods (table 4). Across the 2240 answers provided there mated weight will rarely be as accurate as true weight. Many
were only 10 errors classified as input type. would argue that simple estimates provide a safe starting point
The magnitude of errors made using the triple-formula and that effect should be guided by response. More accurate
APLS method was significantly less than the other methods methods bring with them more complexity, even before consid-
(p=0.02, 0.02, 0.05), with a median percentage error of 10% erations of body composition are taken into account. But we
Arch Dis Child: first published as 10.1136/archdischild-2018-314873 on 5 July 2018. Downloaded from http://adc.bmj.com/ on 8 July 2018 by guest. Protected by copyright.
In summary, our study which stresses tested individuals applying
Method of weight estimation methods of weight estimation under time pressure revealed
APLS1 APLS3 BGM Table significant differences in task difficulty and identified unex-
Total time taken 45* (40 to 49) 55* (49 to 58) 48* (42 to 53) 30* (28 to 33) pected sources of error. With significant mistakes still made in
(s) (95% CI) 5%–10% of cases we feel usability must be an important consid-
Accuracy 90† (87 to 92) 94 (91 to 95) 94 (92 to 96) 96† (93 to 97) eration in the development of future methods of emergency
(%) (95% CI)
decision-making. Simply providing a list of weight values is not
Median % 22 (11 to 33) 10* (7 to 21) 39 (17 to 42) 66 (8 to 132)
enough to avoid errors, with design and layout as important
error (95% CI)
factors that require assessment before clinical use. We feel our
Correct 504 524 527 536
answers (n) simple tool provides a method of rapid evaluation to identify
Error type and prevent uncommon but potentially significant errors.
Arithmetic 55 (98%) 34 (94%) 27 (82%) 7 (29%)
Funding The authors have not declared a specific grant for this research from any
Input 1 (2%) 1 (3%) 6 (18%) 2 (8%)
funding agency in the public, commercial or not-for-profit sectors.
Month/year 0 1 (3%) 0 15 (63%)
Competing interests None declared.
*Denotes that a result is significantly different from the other three methods
(p<0.05) (comparing each method by pairwise analysis). Patient consent Not required.
†Marked pairs of results are significantly different from each other (p<0.05), but not Provenance and peer review Not commissioned; externally peer reviewed.
significantly different from the unmarked methods.
APLS, Advanced Paediatric Life Support; BGM, Best Guess method. Data sharing statement All data and scripts used for analysis are available on
request from the lead author.