Background The continuous monitoring of partial pressure of blood carbon dioxide (pCO2) in premature babies has proven to be challenging. Spot measurements of pCO2 can be performed by taking a blood sample. However the frequency of such measurements is limited by their invasiveness.
Aim We aim to develop a continuous non-invasive method of predicting pCO2 using features of the preterm electroencephalography (EEG) signal.
Methods A regression model was trained on eight 12 hour EEG recordings that contained 22 blood gas measurements in total. All measurements were obtained from babies born before 28 weeks’ gestation and less than 72 hours old. The duration of EEG quiescence (interburst interval) and relative power of delta EEG frequency band values surrounding the point pCO2 measurements were averaged using a specified smoothing window.
Results It is shown that by combining the measurements of both a defined period of EEG interburst interval and the relative power of delta EEG frequency band using a multivariate linear regression model, a prediction of pCO2 can be performed. The automatic removal of mechanical artefact and artefact due to other external influences is demonstrated. A regression coefficient (R2) of 0.64 is obtainable using both the interburst and delta relative power as predictors for pCO2. All variables are significant to within p<0.05. A section of continuous prediction of pCO2 using EEG showing correlation with simultaneous transcutaneous carbon dioxide measurement is demonstrated.
Conclusion The ability to provide a novel non-invasive continuous monitoring of pCO2 in newborn preterm babies is discussed.