BACKGROUND

Previously we demonstrated that a computer algorithm based on bedside airway examinations and facial photographs accurately classified easy and difficult airways. The extent of the ability of anesthesiologists to perform the same task is unknown. We hypothesized that providing photographs would add to the predictive ability of anesthesiologists over that achieved when provided only with the Mallampati (MP) score and the thyromental distance (TMD). We further hypothesized that human observers would implicitly bias their predictions toward more sensitive determination of difficult airways, rather than more specific determination of easy airways.

METHODS

Residents, fellows, and attending anesthesiologists with varying levels of experience (N = 160) were presented with MP and TMD information from 80 Caucasian men subjects. The same subjects' data, accompanied by 3 facial photographs in head-on and right and left profiles, were also presented. Anesthesiologists classified the airways as easy or difficult according to specified criteria ("easy" defined as a single attempt with a Macintosh 3 blade resulting in a grade 1 laryngoscopic view; "difficult" defined as >1 attempt by an operator with at least 12 months anesthesia experience, grade 3 or 4 laryngoscopic view, need for a second operator, or nonelective use of an alternative airway device). Accuracy, sensitivity, and specificity were calculated for each anesthesiologist. We further developed a cost function to quantify a relative bias toward avoiding an unexpectedly difficult intubation versus overpreparing for an easy intubation.

RESULTS

One hundred sixty respondents completed the study. Presenting photographs improved respondents' sensitivity and accuracy in classifying airways, though specificity decreased slightly. Overall accuracy when given photographs was 61.6% (95% confidence interval, 60.8%-62.4%), which was significantly lower than the computer's performance of 87.5% (t test, P < 0.0001). Presentation of photographs, compared with MP and TMD alone, caused anesthesiologists to change their prediction from easy to difficult more frequently if the patients were obese (weight or body mass index), despite not having data on weight or height available. The cost function demonstrated that anesthesiologists strongly preferred to enhance sensitivity (detecting difficult airways) as compared with specificity (detecting easy airways), with a ratio of 6.5:1 (95% confidence interval, 4.9:1-8.4:1).

CONCLUSIONS

Anesthesiologists can derive useful information from facial appearance that enhances the prediction of a difficult airway over that achieved when presented with MP and TMD data alone. Anesthesiologists implicitly bias their predictions toward detection of difficult airways, compared with the true incidence of difficult airways, at the expense of accuracy and specificity. This behavior may be rational for cognitive tasks in which the costs of failure are strongly asymmetric.