Automatic detection of CO2 rebreathing during BiPAP ventilation

Abstract

Abstract CO2 rebreathing significantly influences respiratory drive and the work of breathing during BiPAP ventilation. We analyzed CO2 movement during BiPAP ventilation to find a method of real time detection of CO2 rebreathing without the need of CO2 concentration measurement sampled from the circuit (method expensive and not routinely used). Methods: Observational study during routine care in 15 bed university hospital ICU. At 18 patients who required BiPAP ventilation, intubated or by mask ventilation, during weaning period airflow, pressure and CO2 concentration signals were registered on both sides of venting port and 17 respiratory parameters were measured or calculated for each of 4747 respiratory cycles analyzed. Based on CO2 movement (expiration-inspiration sequences) 3 types of cycle were identified, type I and II do not induce rebreathing but type III does. To test differences between the 3 types ANOVA, t-tests, and canonical discriminant analysis (CDA) were used. Then a multilayer perceptron (MLP) network, a type of artificial neural network, using the above parameters (excluding CO2 concentration) was applied to automatically identify the three types of respiratory cycles. Results: Of the 4747 respiratory cycles, 1849 were type I, 1545 type II, and 1353 type III. ANOVA and t-tests showed significant differences between the types of respiratory cycles. CDA confirmed a correct apportionment of 93.9% of the cycles; notably, of 97.9% of type III. MLP automatically classified the respiratory cycles into the three types with 98.8% accuracy. Conclusions: Three types of respiratory cycles could be distinguished based on CO2 movement during BiPAP ventilation. Artificial neural networks can be used to automatically detect respiratory cycle type III, the only inducing CO2 rebreathing.