P96
THE USE OF ARTIFICIAL INTELLIGENCE FOR SIZING OF PAEDIATRIC TRACHEAL TUBES
R. Vaughan, T. Kong Kam Wa, C. Holmes
CHI at Temple Street, Dublin, Ireland
Introduction and Aims
The appropriate size and type of tracheal tubes is important in paediatric anaesthesia. Under- or over- sizing can lead to complications, and repeated airway interventions increase the risk of airway trauma and aerosolisation exposure (1). Traditional sizing methods are known to be imprecise in practice with tracheal tube exchange rates in paediatric anaesthesia being reported as high as 31% (2). Our primary aim was to establish and back-test a bespoke tracheal tube sizing model using collected data and artificial intelligence to reflect our own population. Secondary aim was to observe our own institutional practice in tracheal tube selection.
Methods
Data from 508 consecutive paediatric intubations was gathered, including tracheal tube type (cuffed vs uncuffed), sizes and the nature of any changes required for satisfactory airway intubation. The final tracheal tube size used was assumed to be a correct size for the patient. Data validation was performed by Alteryx™, an integrated analytics automation platform, with removal of 11 outliers. This dataset was then analysed using an artificial intelligence program, DataRobot™ to model an algorithm designed to accurately predict tracheal tube sizing. The accuracy of the artificial intelligence derived method was compared to other traditional formulae (Khine (0 – 8 years), Motoyama (greater than 2 years), Cole and Penlington (all ages)) for this same cohort.
Results
For our primary aim, DataRobot™ produced a Support Vector Machine (SVM) regressor model with specifically determined weightings of two main variables – age and weight – to produce our best fit model. The artificial intelligence model was then applied to our 508 intubation dataset. It retrospectively predicted the final size with an accuracy of 51% in cuffed and 55% in uncuffed tracheal tubes. This was higher compared to Motoyama 48%, Khine 23%, Penlington 46% and Cole 43%. For our secondary aim, the data set showed 3.9% of our cuffed tracheal tubes initially attempted required a further reintubation due to incorrect sizing, compared to 20.5% of uncuffed tracheal tubes.
Discussion and Conclusion
All tracheal tubes sizing models used exhibited lower than expected accuracy, likely due to retrospective application with the assumption of only one single size as satisfactory. Nevertheless, superior accuracy of AI model prediction suggests that use of artificial intelligence can aid correct first-time size selection of paediatric tracheal tubes. Additionally, using an uncuffed tracheal tube increased the likelihood of requiring reintubation with an alternative tracheal tube more than 5-fold.
References:
- Gálvez JA, Acquah S, Ahumada L, Cai L, Polanski M, Wu L, et al. Hypoxemia, Bradycardia, and Multiple Laryngoscopy Attempts during Anesthetic Induction in Infants: A Single-center, Retrospective Study. Anesthesiology. 2019 Oct;131(4):830–9.
- Weiss M, Dullenkopf A, Fischer JE, Keller C, Gerber AC, European Paediatric Endotracheal Intubation Study Group. Prospective randomized controlled multi-centre trial of cuffed or uncuffed endotracheal tubes in small children. Br J Anaesth. 2009 Dec;103(6):867–73.