5%. Using these adherence cut points, sensitivity was comparable across measures (see Table 4); however, pill count specificity (80% specificity, 95% CI = 30.0�C99.0) sellekchem was markedly greater than that of 3-day recall and VAS (60% specificity, 95% CI = 17.0�C93.0 for both measures). The probability that adherence��plasma varenicline levels greater than or equal to 2.0 ng/ml��was correctly identified across measures (i.e., positive predictive value) was high and ranged from 0.96 to 0.98, whereas the probability that pill count or self-report indicating nonadherence��plasma varenicline levels less than 2.0 ng/ml��was accurate (i.e., negative predictive value) was lower, ranging from 0.33 to 0.55. Table 4.

Optimum Adherence Cutpoints and Related Sensitivity, Specificity, PPV, and NPV of Pill Count, 3-Day Recall, and Visual Analogue Scale for Detecting Adherence When Compared With Plasma Varenicline Concentrationa Figure 1. Area under the receiver operating characteristic curve for pill count, 3-day recall, and VAS. Discussion In this study, we found significant concordance among the varenicline adherence self-report measures, pill count, and plasma varenicline concentration. Of the pill count and two commonly used self-report adherence measures (3-day recall and VAS), pill count had the strongest association with plasma varenicline concentration, the largest area under the ROC curve, and also demonstrated the best combined sensitivity and specificity. Specifically, among participants categorized as adherent by their plasma varenicline concentration, pill count correctly classified them 88% of the time (sensitivity).

While 3-day recall and VAS also demonstrated high sensitivity, the specificity of pill count was greater. Specifically, for participants categorized as non adherent by their plasma varenicline concentration, pill count correctly classify them 80% of the time compared with only 60% of the time with 3-day recall and VAS (specificity). Across all measures, the probability that a result indicating adherence would meet the threshold of 2.0 ng/ml for varenicline plasma adherence was uniformly high (positive predictive value). However, given the high rate of adherence in our sample, the probability that participants categorized as nonadherent on pill count, 3-day recall, or VAS had varenicline plasma concentrations of less than 2.

0 ng/ml was lower (ranging from 33% for VAS to 50% for 3-day recall; negative predictive value). Few studies have examined pill count as a method for assessing adherence to smoking cessation pharmacotherapy, although it has been used widely across other health domains and could be reasonably Batimastat incorporated into tobacco dependence treatment (Bangsberg et al., 2001; Kalichman et al., 2007, 2008; van Onzenoort et al., 2009).