Predicting difficult intubation: a multivariable analysis

Purpose: To develop a clinically useful and valid model for predicting diff icult laryngoscopic tracheal intubation in patients with seemingly normal a irways by adhering to the principles of multivariable model development. Methods: This was an observational study performed at a tertiary-care teach ing hospital. Preoperatively, 444 randomly selected patients requiring trac heal intubation for elective surgery were assessed. In addition, 27 patient s in whom tracheal intubation was difficult, but were not assessed preopera tively, were assessed postoperatively. One assessor, blinded to the intubat ion information, collected the predictor variables. A reliable definition f or difficult intubation was used and all attempts were made to eliminate so urces of bias. Multivariable modeling was performed using logistic regressi on and the model was validated using the bootstrapping technique. Results: Of the 461 patients included in the analysis, 38 were classified a s difficult to intubate. Multivariable analysis identified three airway tes ts that were highly significant for predicting difficult tracheal intubatio n. These were: 1) "mouth opening", 2) "chin protrusion", and 3) "atlanto-oc cipital extension". Using these tests, a validated, highly reliable and pre dictive model is produced to determine the probability of difficult intubat ion for patients. At a selected probability cut-off value, the model is 86. 8% sensitive and 96.0% specific. Conclusion: A simple acid accurate multivariable model, consisting of three airway tests, is produced for predicting difficult laryngoscopic tracheal intubation. Additional studies will be required to determine the accuracy a nd feasibility of this model when applied to a large sample of new patients by multiple anesthesiologists.