Machine learning approaches for predicting difficult airway and first-pass success in the emergency department: A multiple prospective observational study (Preprint)
BACKGROUND There is still room for improvement in the modified LEMON criteria for difficult airway prediction and no prediction tool for first-pass success in the ED. OBJECTIVE We applied modern machine learning approaches to predict difficult airway and first-pass success. METHODS In a multicenter prospective study that enrolled consecutive patients who underwent tracheal intubation in the 13 EDs, we developed seven machine learning models (e.g., random forest model) using routinely collected data (e.g., demographics, initial airway assessment). The outcomes were difficult airway and first-pass success. Model performance was evaluated by c-statistics, calibration slope, and association measures (e.g., sensitivity) in the test set (randomly-selected 20% of data). Their performance was compared with the modified LEMON criteria for the difficult airway and with a logistic regression model for the first-pass success. RESULTS Of 10,741 patients who underwent intubation, 543 patients (5%) had a difficult airway, and 7,690 patients (71%) had first-pass success. In predicting the difficult airway, machine learning models—except for k-point nearest neighbor and multilayer perceptron—had a higher discrimination ability compared with the modified LEMON criteria (P<0.01). For example, the ensemble method had the highest c-statistic (0.74 vs 0.62 in the modified LEMON criteria; P <0.01). For the first-pass success, machine learning models—except for k-point nearest neighbor and random forest models—had a higher discrimination ability. In particular, the ensemble model had the highest c-statistic (0.81 vs 0.76 in the reference regression; P <0.01). CONCLUSIONS Machine learning models demonstrated a greater ability in predicting difficult airway and first-pass success in the ED.