A Randomized Trial of Ways To Describe Test Accuracy: The Effect on Physicians' Post-Test Probability Estimates

Some people believe that likelihood ratios provide diagnostic information that is more useful than sensitivity and specificity estimates.

To assess how physicians' estimates about probability of illness are affected by the presentation of a diagnostic test's value as an estimate of sensitivity and specificity versus a likelihood ratio or an inexact numerical graphic.

Random assignment of vignettes with different presentation formats of diagnostic test accuracy.

Auditorium at a continuing medical education conference.

183 physicians.

After estimating probabilities of 6 common illnesses described in patient vignettes, physicians reviewed pertinent test results presented in 1 of 3 formats.

Physicians' probability estimates of illness before and after receiving test information, and post-test probability estimates based on the Bayes theorem.

Absolute percentage point differences between the physicians' estimated and the Bayes-based post-test probabilities varied from −7 to 31, from −7 to 28, and from 1 to 29 for the sensitivity and specificity, likelihood ratio, and graphical groups, respectively. Mean differences of probability estimates between the sensitivity and specificity and the likelihood ratio groups were small for all vignettes (−2 to 3 percentage points; summary mean z value across the 6 vignettes, 0.04 [95% CI, −0.14 to 0.21]).

The small pool of participants (who were potentially selected) and the limited number of vignettes prevented a more detailed analysis of relationships between the interpreted strength of diagnostic evidence and estimations of illness probability.

These findings suggest that presenting diagnostic test accuracy with likelihood ratios does not affect some physicians' estimates of illness probability compared with presenting diagnostic test results as sensitivity and specificity.