60 / 208
Chapter 3
58
Measurements
VCO
2
values were simultaneously collected over 1 min intervals from IC (Deltatrac II® Datex-
Ohmeda, Finland) and ventilator (Servo-I® with the Capnostat-III sensor, Maquet, Rastatt,
Germany) over at least a 10-min period during steady state (less than 10% fluctuation in VCO
2
and VO
2
by IC). Before each study, the calorimeter was calibrated with a reference gas mixture
(95% O
2
, 5% CO
2
, Datex Division Instrumentarium Corp.) The properties of the Deltatrac®
metabolic monitor have been described before
11
. Per-minute measurements from IC with an
RQ < 0.67 or >1.3 or ventilator-derived VCO
2
values of 0 were discarded, since these values are
physiologically impossible.
REE by IC was calculated using the modified Weir formula (REE (kcal/day)=[3.941*VO
2
+
1.106*VCO
2
]*1440
8
). For the Servo-I®-derived VCO
2
values REE was calculated using the
following formula: REE=5.5* VCO
2
(L/min)*1440
10
.
REE was calculated using the following predictive equations: Schofield-weight, Schofield-
weight/height
3
and the WHO (based on weight)
4
. The following clinical data were recorded
from the Patient Data Management System (PDMS) for all patients: sex, age, weight and height,
diagnosis category, ventilation mode and settings, ICU stay and survival, FiO
2
, temperature,
PRISM score on admission, use of catecholamines/sedatives/muscle relaxants and beta
blockers and length of stay at moment of measurement.
Statistical analysis
Descriptive statistics were expressed as means ± standard deviations (SD) in case of normally
distributed data; otherwise data are expressed as medians with interquartile ranges (IQR).
Relative differences between IC and other methods were calculated as follows: ((value
IC -
value other method)/ value
IC)*100%. Paired samples t-tests were performed to check if there
was a difference in mean values between methods. Spearman’s correlation coefficient (ρ) was
used to describe the association between methods of measurement in case of non-normality.
This correlation coefficient was also used to describe the association between patient weight
and the absolute value of the relative difference between methods. Linear regression analysis
was performed to detect proportional and fixed bias between Servo-I®-derived VCO
2
values
and IC-derived VCO
2
values. This method was chosen because the predictor (IC-derived VCO
2
values) is expected to be free of error (due to steady state measurements).
Bland-Altman analysis was used to assess the agreement 1) between Servo-I® and IC-derived
VCO
2
values and 2) between IC-derived REE values and calculated REE values (Servo-I® and
predictive equations)
12
. Accuracy was also quantified by the proportion of comparable
measurements, defined as a relative difference ≤10%between values derived from the Servo-I®
or predictive equations, and those of IC, to be clinically useful. Inaccuracy was quantified by the
proportion of measurements with a relative difference >30%, to determine the prevalence of
large errors
13
. Differences between the children with and without comparable measurements
were analysed using independent samples t-tests, Mann-Whitney tests or chi-square tests,
depending on the outcome used.