Correspondence
Discrepant Results for Smoking and Cessation Among Electronic Cigarette Users We read with interest the article by Borderud et al, which claimed as the main finding of the study, “. . .E-cigarette users were twice as likely to be smoking at the time of follow-up as nonusers (odds ratio, 2.0; 95% confidence interval, 1.2-3.3).”1 That claim is opposite what appeared in Table 2 in their article and the associated text, which stated that electronic cigarette (Ecigarette) users were twice as likely to have achieved “self-reported 7-day point prevalence abstinence at 6 months,” the outcome specified in the table’s title and defined in the footnotes.1 Such discrepancies appeared throughout the article. Based on nonsystematically reported numbers presented in the text, it appears that the error is most likely in the table (the authors are obligated to produce evidence that resolves this). Proceeding on that assumption, we point out other problems. The authors failed to acknowledge the considerable selection bias from studying only smokers, and thus only E-cigarette users who still smoke. Instead, they claimed to find “no evidence that E-cigarette use was associated at follow-up with superior cessation outcomes,” based on studying a population that systematically eliminated people for whom E-cigarettes may have worked, which renders the conclusion unjustifiable. Despite the above-noted bias due to subject selection, the complete case analysis model revealed that E-cigarette users had a slightly lower smoking rate (odds ratio, 0.95; 95% confidence interval, 0.53-1.72). However, the main finding is based on what the authors misleadingly call intent-to-treat analysis. Intent-to-treat analysis is meant to address artifacts in randomized controlled trials resulting from investigator-generated and subject-generated bias, not from loss of follow-up, but the authors used the term to describe the assumption that subjects lost to follow-up were continuing smokers. They claim that this tactic represents “the full range of conservative and liberal interpretations of smoking abstinence outcomes,”1 but we believe it was an unfounded and biased assumption regarding missing data. Smokers who were Ecigarette users were twice as likely to be lost to follow-up as the other smokers (66% vs 32%), and therefore the 2286
assumption has huge consequences. The conclusion that E-cigarette users were twice as likely to be smoking is purely an artifact of the assumption. An equally plausible counterassumption is that dropouts left the program because they had quit smoking, and the sharply contrasting result from this assumption should have been reported for balance. To assess the “full range of interpretations,” regarding subjects lost to follow-up, the investigators should have assumed that all such E-cigarettes users had quit smoking whereas none of the others did, and vice versa. Table 1 in the article included the characteristics of 1074 patients enrolled in the Tobacco Cessation Program out of 4504 referred patients. Of those enrolled, 293 patients were ineligible for follow-up, 82 patients died, and 285 patients were otherwise lost to follow-up.1 To fully explore selection bias, Table 1 should be revised to include and compare descriptive statistics for each of these groups: those who were ineligible for follow-up (293 patients), those who died or were lost to follow-up (367 patients), and the patients who actually completed the study (414 patients; 9% of all referred patients), as well as the target population of 4504 patients to the extent possible. The authors “speculate that patients may be drawn to E-cigarette use for harm reduction,” and they noted that the increase in current (within the past 30 days) Ecigarette use shown in Figure 2 reflects secular trends.1 However, the magnitude of current prevalence among patients with cancer at Memorial Sloan Kettering Cancer Center in 2012 and 2013 (10.6% and 38.5%, respectively) dwarfed the prevalence of current use in the general population for those years (1.3% and 2.6%, respectively),2 which strongly suggests that harm reduction is a motivator for E-cigarette use among patients with cancer. We look forward to the resolution of these issues, because this article has the potential to provide valuable information regarding the contribution of E-cigarette use to continued smoking and cessation among patients with cancer. FUNDING SUPPORT No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURES Drs. Rodu and Plurphanswat are supported by unrestricted grants from tobacco manufacturers to the University of Louisville, and by
Cancer
July 1, 2015
Correspondence
the Kentucky Research Challenge Trust Fund. Dr. Phillips is partially supported by an unrestricted grant from British American Tobacco. The terms of the grants ensure that the sponsors are unaware of this work, and thus had no scientific input or other influence. The authors have no financial or other personal relationship with regard to the sponsors.
REFERENCES 1. Borderud SP, Li Y, Burkhalter JE, Sheffer CE, Ostroff JS. Electronic cigarette use among patients with cancer: characteristics of electronic cigarette users and their smoking cessation outcomes. Cancer. 2014; 120:3527-3535. 2. King BA, Patel R, Nguyen KH, Dube SR. Trends in awareness and use of electronic cigarettes among US adults, 2010-2013. Nicotine Tob Res. 2015;17:219-227.
Brad Rodu, DDS Department of School of University of Louisville,
Medicine Medicine Louisville Kentucky
Nantaporn Plurphanswat, PhD James Graham Brown Cancer Center University of Louisville Louisville, Kentucky
Carl V. Phillips, PhD Consumer Advocates for Smoke-free Alternatives Association Springfield, Virginia DOI: 10.1002/cncr.29307, Published online March 4, 2015 in Wiley Online Library (wileyonlinelibrary.com)
Reply to Discrepant Results for Smoking and Cessation Among Electronic Cigarette Users Dr. Rodu and colleagues have made several comments regarding our article1 that merit a response. First, Rodu et al identified an inconsistency in the reporting of our analyses examining whether electronic cigarette (E-cigarette) users and nonusers differed in their smoking cessation outcomes. Shortly after the online publication of our article, we discovered a formatting error (reference group mislabelings) in Table 2, resulting in a discrepancy between the text and table reporting of the findings.1 An erratum was submitted and published on November 25, 2014. The text and table are now consistent in reporting the equivalence of follow-up smoking cessation outcomes for e-cigarette users and nonusers using the complete case analysis (CCA) and the higher abstinence rate for nonusers with the intent-to-treat (ITT) analysis. We regret this inadvertent error. Second, Rodu et al raise concerns regarding selection bias. We are aware that selection bias may be a concern in Cancer
July 1, 2015
observational studies in which an important predictor is not randomized. In this specific case, Rodu and colleagues critiqued that the patients’ current use of E-cigarettes was not randomized. In an effort to address these concerns, a straightforward propensity score adjustment was performed and it made no substantive difference in our main conclusions. Specifically, the propensity score model fitted log odds of current E-cigarette use as a function of age, sex, cancer diagnosis, and other factors found to be associated with the use of E-cigarettes. We then added this propensity score covariate to the logistic regression models reported in Table 2.1 This adjustment made slight changes in the odds ratios, but no difference in the statistical significance of the covariates at the .05 level. In future studies, more sophisticated matching algorithms to address potential selection bias may be applied. Next, Rodu et al questioned our presentation of CCA and ITT analyses. Consistent with recommended standards for reporting smoking abstinence in treatment programs and clinical trials,2-4 we present both CCA and ITT analyses. For the ITT analysis, we included all patients who enrolled in the tobacco treatment program in the outcome analyses and assumed all patients whose follow-up smoking status could not be determined to be smoking cigarettes. We acknowledge the limitations of these analytic models,5,6 appreciate the challenge of dealing with missing binary data when reporting cessation outcomes (particularly within the context of differential rates of follow-up), and chose to present both the CCA and ITT models. This method is consistent with current abstinence reporting practices2,3 and readily allows for comparisons between our outcomes and those of other prior studies. In a recent analysis of missing data models, Witkiewitz et al concluded that missing data did not appear to significantly influence smoking outcomes after controlling for nicotine dependence level and therefore we included nicotine dependence in our analyses.7 The clinic coordinator responsible for the follow-up assessment was blinded with regard to patients’ E-cigarette use status. The follow-up data collection efforts were identical for E-cigarette users and nonusers. We know of no peerreviewed recommendation for reporting smoking abstinence that assumes that patients lost to follow-up be treated as nonsmokers. This assumption is not consistent with the high rates of persistent smoking reported among patients with cancer. Rodu et al further requested that we more fully explore potential selection bias in current E-cigarette use by revising Table 1 in our study1 to include, in addition to the 414 completers, dropouts (367 patients who were lost 2287
Discrepant Results for Smoking and Cessation Among Electronic Cigarette Users We read with interest the article by Borderud et al, which claimed as the main finding of the study, “. . .E-cigarette users were twice as likely to be smoking at the time of follow-up as nonusers (odds ratio, 2.0; 95% confidence interval, 1.2-3.3).”1 That claim is opposite what appeared in Table 2 in their article and the associated text, which stated that electronic cigarette (Ecigarette) users were twice as likely to have achieved “self-reported 7-day point prevalence abstinence at 6 months,” the outcome specified in the table’s title and defined in the footnotes.1 Such discrepancies appeared throughout the article. Based on nonsystematically reported numbers presented in the text, it appears that the error is most likely in the table (the authors are obligated to produce evidence that resolves this). Proceeding on that assumption, we point out other problems. The authors failed to acknowledge the considerable selection bias from studying only smokers, and thus only E-cigarette users who still smoke. Instead, they claimed to find “no evidence that E-cigarette use was associated at follow-up with superior cessation outcomes,” based on studying a population that systematically eliminated people for whom E-cigarettes may have worked, which renders the conclusion unjustifiable. Despite the above-noted bias due to subject selection, the complete case analysis model revealed that E-cigarette users had a slightly lower smoking rate (odds ratio, 0.95; 95% confidence interval, 0.53-1.72). However, the main finding is based on what the authors misleadingly call intent-to-treat analysis. Intent-to-treat analysis is meant to address artifacts in randomized controlled trials resulting from investigator-generated and subject-generated bias, not from loss of follow-up, but the authors used the term to describe the assumption that subjects lost to follow-up were continuing smokers. They claim that this tactic represents “the full range of conservative and liberal interpretations of smoking abstinence outcomes,”1 but we believe it was an unfounded and biased assumption regarding missing data. Smokers who were Ecigarette users were twice as likely to be lost to follow-up as the other smokers (66% vs 32%), and therefore the 2286
assumption has huge consequences. The conclusion that E-cigarette users were twice as likely to be smoking is purely an artifact of the assumption. An equally plausible counterassumption is that dropouts left the program because they had quit smoking, and the sharply contrasting result from this assumption should have been reported for balance. To assess the “full range of interpretations,” regarding subjects lost to follow-up, the investigators should have assumed that all such E-cigarettes users had quit smoking whereas none of the others did, and vice versa. Table 1 in the article included the characteristics of 1074 patients enrolled in the Tobacco Cessation Program out of 4504 referred patients. Of those enrolled, 293 patients were ineligible for follow-up, 82 patients died, and 285 patients were otherwise lost to follow-up.1 To fully explore selection bias, Table 1 should be revised to include and compare descriptive statistics for each of these groups: those who were ineligible for follow-up (293 patients), those who died or were lost to follow-up (367 patients), and the patients who actually completed the study (414 patients; 9% of all referred patients), as well as the target population of 4504 patients to the extent possible. The authors “speculate that patients may be drawn to E-cigarette use for harm reduction,” and they noted that the increase in current (within the past 30 days) Ecigarette use shown in Figure 2 reflects secular trends.1 However, the magnitude of current prevalence among patients with cancer at Memorial Sloan Kettering Cancer Center in 2012 and 2013 (10.6% and 38.5%, respectively) dwarfed the prevalence of current use in the general population for those years (1.3% and 2.6%, respectively),2 which strongly suggests that harm reduction is a motivator for E-cigarette use among patients with cancer. We look forward to the resolution of these issues, because this article has the potential to provide valuable information regarding the contribution of E-cigarette use to continued smoking and cessation among patients with cancer. FUNDING SUPPORT No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURES Drs. Rodu and Plurphanswat are supported by unrestricted grants from tobacco manufacturers to the University of Louisville, and by
Cancer
July 1, 2015
Correspondence
the Kentucky Research Challenge Trust Fund. Dr. Phillips is partially supported by an unrestricted grant from British American Tobacco. The terms of the grants ensure that the sponsors are unaware of this work, and thus had no scientific input or other influence. The authors have no financial or other personal relationship with regard to the sponsors.
REFERENCES 1. Borderud SP, Li Y, Burkhalter JE, Sheffer CE, Ostroff JS. Electronic cigarette use among patients with cancer: characteristics of electronic cigarette users and their smoking cessation outcomes. Cancer. 2014; 120:3527-3535. 2. King BA, Patel R, Nguyen KH, Dube SR. Trends in awareness and use of electronic cigarettes among US adults, 2010-2013. Nicotine Tob Res. 2015;17:219-227.
Brad Rodu, DDS Department of School of University of Louisville,
Medicine Medicine Louisville Kentucky
Nantaporn Plurphanswat, PhD James Graham Brown Cancer Center University of Louisville Louisville, Kentucky
Carl V. Phillips, PhD Consumer Advocates for Smoke-free Alternatives Association Springfield, Virginia DOI: 10.1002/cncr.29307, Published online March 4, 2015 in Wiley Online Library (wileyonlinelibrary.com)
Reply to Discrepant Results for Smoking and Cessation Among Electronic Cigarette Users Dr. Rodu and colleagues have made several comments regarding our article1 that merit a response. First, Rodu et al identified an inconsistency in the reporting of our analyses examining whether electronic cigarette (E-cigarette) users and nonusers differed in their smoking cessation outcomes. Shortly after the online publication of our article, we discovered a formatting error (reference group mislabelings) in Table 2, resulting in a discrepancy between the text and table reporting of the findings.1 An erratum was submitted and published on November 25, 2014. The text and table are now consistent in reporting the equivalence of follow-up smoking cessation outcomes for e-cigarette users and nonusers using the complete case analysis (CCA) and the higher abstinence rate for nonusers with the intent-to-treat (ITT) analysis. We regret this inadvertent error. Second, Rodu et al raise concerns regarding selection bias. We are aware that selection bias may be a concern in Cancer
July 1, 2015
observational studies in which an important predictor is not randomized. In this specific case, Rodu and colleagues critiqued that the patients’ current use of E-cigarettes was not randomized. In an effort to address these concerns, a straightforward propensity score adjustment was performed and it made no substantive difference in our main conclusions. Specifically, the propensity score model fitted log odds of current E-cigarette use as a function of age, sex, cancer diagnosis, and other factors found to be associated with the use of E-cigarettes. We then added this propensity score covariate to the logistic regression models reported in Table 2.1 This adjustment made slight changes in the odds ratios, but no difference in the statistical significance of the covariates at the .05 level. In future studies, more sophisticated matching algorithms to address potential selection bias may be applied. Next, Rodu et al questioned our presentation of CCA and ITT analyses. Consistent with recommended standards for reporting smoking abstinence in treatment programs and clinical trials,2-4 we present both CCA and ITT analyses. For the ITT analysis, we included all patients who enrolled in the tobacco treatment program in the outcome analyses and assumed all patients whose follow-up smoking status could not be determined to be smoking cigarettes. We acknowledge the limitations of these analytic models,5,6 appreciate the challenge of dealing with missing binary data when reporting cessation outcomes (particularly within the context of differential rates of follow-up), and chose to present both the CCA and ITT models. This method is consistent with current abstinence reporting practices2,3 and readily allows for comparisons between our outcomes and those of other prior studies. In a recent analysis of missing data models, Witkiewitz et al concluded that missing data did not appear to significantly influence smoking outcomes after controlling for nicotine dependence level and therefore we included nicotine dependence in our analyses.7 The clinic coordinator responsible for the follow-up assessment was blinded with regard to patients’ E-cigarette use status. The follow-up data collection efforts were identical for E-cigarette users and nonusers. We know of no peerreviewed recommendation for reporting smoking abstinence that assumes that patients lost to follow-up be treated as nonsmokers. This assumption is not consistent with the high rates of persistent smoking reported among patients with cancer. Rodu et al further requested that we more fully explore potential selection bias in current E-cigarette use by revising Table 1 in our study1 to include, in addition to the 414 completers, dropouts (367 patients who were lost 2287
