63 / 208
Ventilator-derived VCO
2
measurements to determine REE
61
3
Comparable measurements, defined as a difference ≤10% between Servo-I®-derived VCO
2
values and those of IC to be clinically useful, were seen in 18 children (44%). When comparing
these 18 children to the 23 children with a difference >10%, it was shown that children with
comparable measurements were significantly older (median 7.4 vs. 0.6 years, p<0.001), and
taller (median 121 vs. 65 cm, p < 0.001) with higher weight (median 23 vs. 6.3 kg, p=0.001),
suggesting that the size of the differences between the methods decreases with age and
weight (Table 1a). There was no significant difference in diagnosis, PRISM score, heart rate,
temperature, use of medication, ventilation mode or supplied oxygen fraction between these
2 groups of children. Median VCO
2
values of the Servo-I® (112 vs 33 ml/min, p<0.001) and of IC
(109 vs 46 ml/min, p<0.001) were significantly higher, as were the median inspiratory (142 vs
36 ml, p=0.002) and expiratory (143 vs 35 ml, p=0.001) tidal volumes and the respiratory rate
(Table 1b). Measurements with a difference >30% between Servo-I®-derived VCO
2
values and
those of IC were seen in 11 children (27%).
Since ventilator settings are weight-based, we plotted the relative difference between VCO
2
values derived from IC and Servo-I® by weight in kg (Fig. 1c). As shown in this figure, in children
weighing less than 15 kg (n=21), there was a substantial bias with only 19% of measurements
being comparable (difference ≤10%). In children weighing 15 kg or more (n=20), 14 children
(70%) had comparable measurements (difference ≤10%). Among this group there was no
significant correlation (Spearman’s ρ=-0.222, p=0.347) between weight and the absolute value
of the relative difference between the two methods, suggesting that the accuracy does not
depend on weight for children weighing more than 15 kg (see also Fig. 1c). Five percent of
measurements in children weighing 15 kg showed a difference >30%. Therefore a weight of 15
kg could be a clinically acceptable threshold for reliability of the Servo-I®-derived VCO
2
values.
We used this threshold in the remaining analyses.
REE values derived from IC were compared to REE values calculated from Servo-I®-derived
VCO
2
values. Correlation was high [ρ=0.954 (95%CI: 0.915 to 0.975)]. There was a mean relative
difference of 19%with wide 95% limits of agreement; -36.3% and 74.4% as shown in Figure 1b.
The linear regression analysis showed both fixed (p=0.001) and proportional bias (slope 1.09,
p=0.038).