Improvement of the association between self-reported pill count and varenicline levels following exclusion of participants with misreported pill count: A commentary on Peng et al. (2017).

INTRODUCTION

We previously reported poor associations between salivary varenicline and pill counts, and a substantial overestimation of adherence by pill counts in "Measures and predictors of varenicline adherence in the treatment of nicotine dependence" (Peng et al., 2017). We have since conducted supplementary analyses characterizing, and then excluding, individuals with established inaccurate pill count recall.

METHODS

Based on published varenicline pharmacokinetics (including drug levels, and the long half-life) and our detection limits, conservatively we should be able to detect varenicline in anyone who took at least one pill during the 48h prior to saliva collection; thus, those reporting 1 or more pills in this time frame but who had undetectable salivary varenicline were deemed to have inaccurate pill count recall. Correlations between pill counts and salivary varenicline, and Receiver Operating Characteristics curve analyses were conducted following exclusion of participants with inaccurate pill count recall.

RESULTS

Nearly 20% of our participants (N=67/376) had inaccurate self-reported pill counts. These participants were younger, non-white, lower income, and unmarried (evaluated using chi-square or Mann-Whitney U test). Following exclusion of these individuals, the correlations between salivary varenicline and pill count improved and the area under the curve (AUC) of pill counts for discriminating adherence improved modestly.

CONCLUSION

When the 20% of individuals with inaccurate pill count recall were excluded, an improved association between self-reported pill count and salivary varenicline was observed, albeit still weak. A substantial overestimation of adherence by pill counts relative to salivary varenicline is still observed even after exclusion of almost 20% of the group having established inaccurate reporting suggesting that these individuals, with identifiable inaccuracies, were only part of the overestimation of adherence.