Detection and prediction limits for identifying highly confusable drug names from experimental data

Confusions between drug names that look and sound alike are common, costly, harmful, and difficult to prevent. One prevention strategy is to screen proposed new drug names for confusability before approving them. Widespread acceptance of preapproval tests of confusability is compromised by the lack of experimental designs and statistical methods to support valid inferences about whether a proposed new name is unacceptably confusing. One way of identifying confusing names is to conduct memory and perception experiments on a set of drug names which would include both the new name and a set of control names (e.g., names already on the market). The experiment would yield an observed error rate for every name. Inferences about the acceptability of the new name can be made by comparing the error rate of the new name to the distribution of error rates of the control names. We describe four memory and perception experiments on drug names, carried out using clinicians as participants. Each experiment included drug names designated as test and control names. We demonstrate how to use a combination of logistic regression, Poisson prediction limits, and highly assured credible intervals to identify and apply a threshold for identifying unacceptably confusing names. Our models show an excellent fit to the data. These experimental designs and analytic methods should be useful in the preapproval testing of proposed new drug names and in similar regulatory scenarios where it is necessary to draw inferences about the comparative safety or effectiveness of new vs. old products.