Discussion
This prospective study in real-life paediatric secondary care setting
investigated the potential of exhaled volatile biomarkers in the
differentiation of noisy breathing infants presenting with wheezing or
rattling. Due to the low number of wheezing infants included in the
study, the statistical analysis focused on the differentiation of
rattling infants with different severity from recovered infants. The
severity was a classification based on the paediatrician’s diagnosis.
Collecting exhaled breath samples form infants using an in-house
adjusted commercial breath sampler was feasible and acceptable. In order
to avoid difficulties during breath sampling, placing the child on the
parent’s lap, distracting it with toys or movies and assistance from the
parent were very important.
Offline GC-MS and SIFT-MS analysis showed similar performance in
discriminating the different patient groups (recovered, mild rattling,
moderate rattling and severe rattling). The differentiation of
recovered/mild rattlers and moderate/severe rattlers showed slightly
better performance than the four group discrimination. Both the four
group and the two group discrimination improved by combining exhaled
breath data with PROs. For comparison, we also ran a
discriminant model using only PROs. This resulted in
much lower performance, particularly for the discrimination of mild,
moderate and severe rattling.
We used both GC-MS and SIFT-MS analytical techniques to explore their
diagnostic capacity. GC-MS allows the detection and identification of
compounds, and is the current golden standard for breath VOC analysis.
However, it cannot be used for real-time exhaled breath
measurements12, which limits its potential as a
point-of-care application. The instrumental compacity and simplicity of
use make SIFT-MS a candidate as point-of-care breath screening
system.13 However, SIFT-MS could not be used for
identification of VOCs. Multiple studies have reported the potential of
breath VOCs to diagnose respiratory illnesses in adults and
children.14-19 Exhaled VOCs may reflect reactions in
the target organ and therefore, reflect a person’s health status. The
fact that VOC metabolites are directly measured in the exhaled air,
makes them interesting as a potential easy to use point-of-care tool.
Some of the differentiating VOCs identified by GC-MS in the current
study, have been previously reported to be associated with asthma,
wheezing or viral infections. 2-butanone has been associated with
asthma.20,21 Pentanoic acid was reported as one of the
discriminating VOCs for the differentiation of preschool children
(1.9-4.5 years) with recurrent wheeze from children without
wheeze.22 Exhaled 2,8-dimethylundecane in adults was
shown to be associated with oxidative stress induced by viral
infections.23 A very similar compound,
3,9-dimethylundecane has been described to be altered in wheezing or
asthmatic children compared to controls.19
The comparison of the different discriminating models in the current
study demonstrated that exhaled breath markers can be of value for the
discrimination of rattling infants from recovered infants and the
differentiation based on severity. PROs also had its
value in combination with the exhaled breath markers, but on this own it
was less valuable for discrimination of the different groups. However,
PROs is subjective and depends on the knowledge of the
parents. In this pilot study all parents were specifically informed by a
paediatrician specialized in pulmonology who used sound recordings of
wheeze and rattle in infants to demonstrate the specific traits of the
sounds, and the difference. Furthermore, the paediatrician explained
carefully what the indications of burden to the child were (not eating,
not sleeping, less playful). That information may have influenced the
added value of the parent reported parameter. The quality of parent’s
reporting might be lower in a non-informed population. This would be
especially the case in primary care where general physicians (GP) are
not always aware of the difference between wheezing, rattling, and its
severity.
The main limitation of this study was its small sample size which may
result in overfitting of the statistical models used for classification.
To avoid this, validation of the discriminant model is mandatory i.e.
ideally, the sample size should be large enough to split the dataset
into a training set (used to train the classification method) and
validation set (used to test the trained classification
model).24 Instead of splitting the data, we applied
leave-one-out cross validation which is usually used when the number of
available samples per class is low (around 20 per
class).24 Another limitation, was the identification
of VOCs measured by GC-MS based on the National Institute of Standards
and Technology (NIST) library. Although the NIST standard reference
database is one of the most popular mass spectral databases for
metabolite identification, results show a high rate of false
identifications of metabolites.6 Therefore,
identification of VOCs is not 100% certain and standards should be used
in follow up studies to check retention time and mass spectrum in order
to confirm identification.
Confounders known to effect breath VOC composition are food
intake25, age and gender.26Especially the timing of food intake is difficult to control in the
infant population. These confounders may also have played a role in our
population. The children were between 5 and 16 months old at the time of
inclusion. This age difference may cause differences in exhaled
metabolites due to differences in metabolization and food pattern.
Medication is also a known confounder27,28, however
parents were instructed not to give their child medication on the day of
the paediatrician visit. Blanchet et al. demonstrated that although the
VOC breath profile of males and females are different, the separation
between the two groups is not very marked and the difference is not
sufficient to discriminate males and females.26Additionally, not accounting for confounders resulted in the recruitment
of a study population that represents the infant population presenting
to the real clinical practice as close as possible.