This prospective 'methods comparison' study was conducted in a level III neonatal intensive care unit from September 2020 to June 2021. The study protocol was approved by the Institutional ethics committee. A written, informed consent was obtained from the parents. The study was performed in line with the principles of the Declaration of Helsinki. All preterm neonates < 30 weeks' gestation, < 7 days old, and required invasive mechanical ventilation in the NICU were included in the study. Those with cardiorespiratory malformations, circulatory instability, hypothermia in the preceding 12 hours, on high-frequency ventilation, and where parents refused consent were excluded.
ETCO2 was measured using the Microstream capnography ETCO2 module and capnoLine tube integrated into the Intellivue MX800 monitors (Microstream CO2 Extension, Philips India Limited, Gurgaon, Haryana, India). As per the manufacturer's recommendations, the ET module was calibrated each time before use to get a reliable respiratory cycle waveform and respiratory rates. The trend interval was set to the lowest possible averaging time. Arterial or arterialized CO2 (PacCO2) was measured in a blood-gas analyzer (Radiometer India, Kurla, Mumbai, India). The analyzer uses a Clark electrode which measures the change in the current flowing through a reaction chamber where oxygen (O2) is reduced to OH− ions by a change in voltage. All samples were processed within 30 minutes of sampling. Following pre-analytical precautions were taken: use of pre-heparinized syringes and capillaries, expelling the air bubbles by the standard recommended procedures, closing the syringe and capillary tips with caps, rotating the syringes and capillaries for uniform mixing of the heparin, entering the study subject's temperature and the fraction of inspired oxygen (FiO2) and repeated inversion of the syringe/capillary before inserting into the machine for processing.
As the study primarily assessed the correlation between one continuously monitored numerical variable (ETCO2) and another intermittently monitored numerical variable (PacCO2), the samples were time matched to ensure that they were representative of the physiological changes that took place in the internal milieu and were comparable on a time scale. For time matching, the actual time when the arterial/arterialized blood started to flow within the syringe or the capillary (depending on the method of blood sampling for arterial gas estimation) was noted down and documented to the accuracy of seconds. To facilitate reliable capturing of the actual time of the artery/capillary sampling, a second person (staff nursing officer) noted down the time. The values of ETCO2 corresponding to the arterial sampling time were noted down for analysis to the accuracy of ± 30 seconds. While the samples were drawn for ABG analysis, the ETCO2 values were documented.
Blood gas analyses were performed at the treating clinician's discretion, which depended primarily on the underlying lung disease, acute or convalescing lung disease, and settings on the ventilator. All the collected samples were analysed, and each pair of measurements of ETCO2 and PacCO2 were available for analysis (no missing data, equivalent to the intention to treat analysis in an intervention trial). PacCO2 was analysed independent of ETCO2. This meant that the PacCO2 assessor (principal investigator) was not aware of the ETCO2 value till the blood sample was processed in the blood gas analyser.
Sample size and statistical analysis
As the correlation coefficient (r) between PacCO2 and microstream ETCO2 (primary outcome variable) was not known, we assumed no correlation exists between them (null hypothesis). To reject the null hypothesis and for an expected correlation coefficient of 0.5 (assumed) with a two-sided alpha error of 0.05 and beta error of 0.1, we required 38 samples*. To account for missing data at random of ~ 20%, a total sample size of 46 was required.
*N=[(Za + Zb)/C]2 + 3; (Za is 1.96 for an alpha error of 0.05; Zb is 1.282 for a beta error of 0.10; C = 0.5Xln[(1 + r)/(1-r)])
N is the total number of subjects required, and 'r' is the expected correlation coefficient (assumed as 0.5 in the current calculation).
Data were analyzed using the R language for statistical computing and specific packages. Descriptive measures such as mean, median, range, and standard deviation (SD) were calculated for continuous data. Normality assumption was tested with an appropriate test. Due to more than one pair of measurements from each subject (repeated measures), inter- and intra-rater reliability could not be assessed by calculating the intra-class correlation coefficient (ICC) as it will be erroneous. Hence, a correlation between ETCO2 and PacCO2 was examined by repeated-measures correlation coefficient (RM - Correlation). Locally weighted smoothing regression (LOESS) method was used for scatter plot regression lines. Bland-Altman plots for agreement were created, and mean difference (agreement) and their 95% confidence intervals were calculated. As the data had multiple levels (hierarchical data) as well as repeated measures with unequal time points of measurements, a mixed linear model was used for regression. A two-tailed p-value of < 0.05 was considered as significant.