Variation in bupropion findings is not due to differences in measurement: comment on Teneggi et al.

Abstract

I read with interest the paper by Teneggi et al. (2005) on the effects of bupropion (BP) on tobacco craving and withdrawal. As the authors note, the design of their study was nearly identical to a study conducted by my group, assessing two doses of BP (Shiffman et al. 2000), but yielding different findings. Specifically, Teneggi et al. found BP effects on craving, but we did not. Teneggi et al. suggest that the absence of BP effects on craving in our study might be attributable to unreliability in our measurement of craving. However, this explanation is contradicted by data. The six-item craving scale used in our study demonstrated high reliability of 0.95 (Cronbach’s α), which compares favorably with the reliability of the brief questionnaire of smoking urges (Cox et al. 2001) used by Teneggi et al. Moreover, our study assessed craving five times per day (in contrast to the three times a day by Teneggi et al.), which would further enhance the reliability of the measurement. Finally, the craving assessment in our study was able to reliably detect significant increases in craving as deprivation time increased on the first day of abstinence—much as seen by Teneggi et al. (Figs. 2 and 3 in Teneggi et al. 2005)—validating and verifying the sensitivity of the assessment. Teneggi et al. also wondered whether our use of repeatedmeasures ANOVA to contrast smoking and nonsmoking craving ratings could have degraded reliability through its reliance on difference scores. (See Sayette et al. 2000 and Shiffman et al. 2004 for discussion of analytic approaches.) However, a reanalysis of craving data without difference scores, either using covariance analysis or contrasting 300 mg BP and placebo without adjusting for ratings during smoking, yields the same null result. Thus, it appears that both studies used reliable assessments and analyses of craving, making the explanation by Teneggi et al. of the differences as due to measurement or statistical properties highly unlikely. This is important because it suggests the need to look for other potential sources of differences. As Teneggi et al. suggest, differences in populations, assessments, and other methodological factors may influence craving responses. Particularly, as Teneggi et al. suggest these laboratory methods as models for assessing drug effects, it is important to consider a range of differences—including some in often-ignored factors—that might influence craving responses. Recent work by Sayette (Wertz and Sayette 2001; Sayette et al. 2003) and by Tiffany (Carter and Tiffany 2001) suggests that smokers’ expectations of being able to smoke may influence craving intensity. Although both experimental designs called for fixed periods of abstinence, it is possible that subtle variations in instructions or expectancies could yield differences in craving. Craving is also highly responsive to environmental cues (Carter and Tiffany 1999). It’s been suggested that craving is minimized when smokers are studied in “sterile” environments devoid of smoking cues (Hatsukami et al. 1984), which appears to characterize the study environments in both studies. However, again, subtle, unintended, and undocumented differences in the research environments could account for differences in findings. Attribution of differences in the findings of the different studies to different psychometric qualities of measures is not supported by the data and may obscure a search for other sources of variability, which might be productive in helping to develop robust paradigms for laboratory assessment of craving. The author provides consulting services regarding smoking and cessation exclusively to GlaxoSmithKline Consumer Healthcare, and has an interest in a new smoking cessation product.