Background Many clinicians use a global visual interpretation of patient appearance to decide if a patient looks sick or not. For patients with suspected acute pulmonary embolism (PE), we tested the relationship between visual appearance of a happy patient facial affect and probability of PE+ on CT pulmonary angiography (CTPA).

Methods Eligible patients were selected by usual care to undergo CTPA, the criterion standard for PE+ or PE−. Prior to CTPA result, trained study personnel obtained physician pretest probability using the gestalt method (visual analogue scale, 0%–100%), the Wells score (0–12) and physicians’ impression of whether the patient smiled during the initial examination (smile+). Patients' faces were also video recorded and analysed with an automated neural network-based algorithm (Noldus FaceReader) for happy affect.

Results Of the 208 patients enrolled, 27 were PE+ and smile+ was more frequent in patients with PE+ than PE−, a finding confirmed by the Noldus. The diagnostic sensitivity and specificity of smile was low, and physicians overestimated presence of an alternative diagnosis more likely to PE with smile+ than smile− patients in patients with true PE. As a result, the area under the receiver operating characteristic curve (AUROC) was lower for the Wells score in smile+ patients. However, the physicians’ mean gestalt estimate of PE did not differ with smile status, nor did smile status affect the AUROC for gestalt.

Conclusions In patients with suspected PE, physician recollection of patients’ smile+ was more common in PE+ patients, and was associated with a less accurate Wells score, primarily because physicians overestimated probability of alternative diagnosis. However, the overall diagnostic accuracy of physicians’ gestalt did not differ with perceived smile status. These data suggest that the patients’ smile had less effect on the numeric gestalt pretest probability assessment than on the binary decision about an alternative diagnosis.