Defining difficult laryngoscopy findings by using multiple parameters: A machine learning approach

Background Preoperative identification of patients whose trachea will be difficult to intubate would decrease the rate of anesthesia related adverse respiratory events. Each test for airway examination may predict a separate aspect of airway. A computer-based approach is tested in this study to precisely evaluate difficult laryngoscopy. Aim of the work Aim of the work was to evaluate the efficacy and accuracy of a multiparameter computer-based system for prediction of difficult laryngoscopy. Study design 50 Adult patients presenting for non emergency surgery at Alexandria main university hospital from February 2015 to Feruary 2016 with unanticipated difficult laryngoscopy were assessed postoperatively according to selected nine airway parameters. The same was done for their matched 50 controls after full recovery from general anesthesia. All data were entered into an information–based computer system where they were converted into numerical variables. All data have been processed and analyzed using the Microsoft visual studio 2008 (C#.net) and WEKA (Waikato Environment for Knowledge Analysis) machine learning algorithms. Classification was done using J48 algorithm based on a decision tree and a “Weighter” filter was used to allow one to specify a numeric attribute to be used as an instance weight. Results Processed data have been designed as a software termed “Alex Difficult Laryngoscopy Software” (ADLS). Positive predictive value was 76%, Negative predictive value was 76%, Matthews correlation coefficient was 0.52 and area under the ROC curve was 0.79. Conclusion “Alex Difficult Laryngoscopy Software” (ADLS) is a machine learning program for prediction of difficult laryngoscopy. New cases can be entered to the training set thus improving the accuracy of the software.