Recent American Heart Association guidelines suggest that the diagnosis of "innocent murmur" can be made without echocurdiography if certain clinical criteria are present. Because little is known about the auscultatory skills of primary care physicians (PCPs), we studied their ability to classify several murmur features used in these guidelines.
We used a cross-sectional design in which physicians evaluated a panel of 12 "virtual patients," whose heart sounds were produced by a high fidelity simulator. A convenience sample of 64 general internists and family physicians answered multiple choice questions about the auscultatory findings in each patient. We report sensitivity and specificity for each feature, along with 95% confidence intervals adjusted to account for clustering.
Sensitivity to detect a systolic murmur (present in all patients) was 98% (96%, 99%). To detect non-ejection shaped murmur, sensitivity was 65% (59%, 72%), and specificity 79% (75%, 83%). In patients with an ejection-shaped murmur, sensitivity to detect a peak in the second half of systole was 66% (57%, 72%), and specificity 53% (46%, 62%). To detect murmurs that got louder with a "maneuver", sensitivity was 80% (75%, 85%), and specificity 93% (91%, 95%). To detect any extra heart sound, click or diastolic murmur, sensitivity was 65% (58%, 71%), and specificity 86% (82%, 89%). Using a global measure of ability to detect any marker of a pathologic murmur, sensitivity was 95% (94%, 97%), but specificity was only 39% (32%, 46%).
Looking at physician-specific, rather than observation-level results, nearly all physicians (94%) had a global sensitivity greater than 85%. However, most physicians (63%) had a global specificity below 35%. Global specificity also varied considerably across physicians, with values ranging from 0 to 100%.
Certain physician characteristics were associated with better performance. Using multi-variable logistic regression, global specificity was higher for doctors more than I0 years out of medical school compared to their less experienced peers (OR 2.3 [95% CI 1.13, 4.72]). Judgments rated "very confident" by physicians for all directly observed murmur features also had higher specificity in the logistic models.
Based on their evaluation of 12 simulated patients, the ability of PCPs to classify individual auscultatory components of systolic murmurs is inconsistent, although global sensitivity is excellent. If a single abnormality in any murmur feature would trigger echocardiography, PCPs would miss fewer than 5% of potentially diseased cases. Using this strategy, however, many normal patients would be referred for further testing.
