Application of receiver operating characteristic analysis to refine the prediction of potential digoxin drug interactions

Abstract

In the 2012 Food and Drug Administration (FDA) draft guidance on drug-drug interactions (DDIs), a new molecular entity that inhibits P-glycoprotein (P-gp) may need a clinical DDI study with a P-gp substrate such as digoxin when the maximum concentration of inhibitor at steady state divided by IC₅₀ ([I₁]/IC₅₀) is ≥0.1 or concentration of inhibitor based on highest approved dose dissolved in 250 ml divide by IC₅₀ ([I₂]/IC₅₀) is ≥10. In this article, refined criteria are presented, determined by receiver operating characteristic analysis, using IC₅₀ values generated by 23 laboratories. P-gp probe substrates were digoxin for polarized cell-lines and N-methyl quinidine or vinblastine for P-gp overexpressed vesicles. Inhibition of probe substrate transport was evaluated using 15 known P-gp inhibitors. Importantly, the criteria derived in this article take into account variability in IC₅₀ values. Moreover, they are statistically derived based on the highest degree of accuracy in predicting true positive and true negative digoxin DDI results. The refined criteria of [I₁]/IC₅₀ ≥ 0.03 and [I₂]/IC₅₀ ≥ 45 and FDA criteria were applied to a test set of 101 in vitro-in vivo digoxin DDI pairs collated from the literature. The number of false negatives (none predicted but DDI observed) were similar, 10 and 12%, whereas the number of false positives (DDI predicted but not observed) substantially decreased from 51 to 40%, relative to the FDA criteria. On the basis of estimated overall variability in IC₅₀ values, a theoretical 95% confidence interval calculation was developed for single laboratory IC₅₀ values, translating into a range of [I₁]/IC₅₀ and [I₂]/IC₅₀ values. The extent by which this range falls above the criteria is a measure of risk associated with the decision, attributable to variability in IC₅₀ values.

Fig. 1.

Receiver operating characteristic curve incorporating all company data. The sensitivity on the x-axis represents the TPR, whereas 1-specificity on the y-axis represents the FPR, where the specificity is the TNR.

Fig. 2.

Digoxin DDI decision tree.