November 30, 2018

Stanton A. Glantz, PhD

How real is the “reality check” on the relationship between youth vaping and smoking?

Recently David Levy and colleagues published “Examining the relationship of vaping to smoking initiation among US youth and young adults: a reality check” in Tobacco Control.  In this paper they used data from all the surveys over time that measured youth and young adult e-cigarette use and smoking and concluded

There was a substantial increase in youth vaping prevalence beginning in about 2014. Time trend analyses showed that the decline in past 30-day smoking prevalence accelerated by two to four times after 2014. Indicators of more established smoking rates, including the proportion of daily smokers among past 30-day smokers, also decreased more rapidly as vaping became more prevalent.

The inverse relationship between vaping and smoking was robust across different data sets for both youth and young adults and for current and more established smoking. While trying electronic cigarettes may causally increase smoking among some youth, the aggregate effect at the population level appears to be negligible given the reduction in smoking initiation during the period of vaping's ascendance.

The good news is that Levy and colleagues are finally accepting the overwhelming evidence that kids who start with e-cigarettes are more likely to end up smoking cigarettes, the so-called “gateway effect.”

Now they have fallen back to arguing that the gateway effect is not big enough to overcome the benefits of e-cigs as substitutes for cigarettes.

The approach they used, interrupted time series analysis, estimates the declining trend in cigarette use over time, then test whether this trend changes (in this case, they generally tested for a slope change) after the advent of e-cigarettes. 

imageInterrupted time series is a well-established method for analyzing changes in trends.  Indeed, Lauren Dutra and I used interrupted time series to do a similar analysis of the effect that the advent of e-cigarettes on cigarette smoking among youth using the National Youth Tobacco Survey from 2004 to 2014.  Based on data over that time, we found that the advent of e-cigarettes (using a start date of 2009 for e-cigs) did not affect the declining trend in cigarette smoking, but led to an increase in total tobacco product use (figure).   

Lauren and I also used individual-level data (something Levy and colleagues did not consider in their analysis) and found that about one-third of the kids using e-cigarettes had risk profiles that made them unlikely to start using nicotine with conventional cigarettes.  Thus, e-cigarettes are expanding the tobacco market.

Levy and colleagues expand the time period of analysis up to 2016 or 2017 (depending on the data source).  This is an important addition because, as they note, e-cigarette use has continued to grow since 2014 (their Figure 1).

I have several concerns about the analysis and interpretation of the data that Levy and colleagues present. 

The assumption in interrupted time series analysis as they (and we) do it is that the underlying trend is linear (a straight line).  The figures in the supplementary file suggest that many of the time histories are curved.  Failing to account for this curvature can distort the results and also make the results highly sensitive to the break year (i.e., where the line bends) in the analysis.

A related concern because of the curvature in the data is the break year they use in their analysis is 2014.  They justify this by arguing that 2014 is when e-cigarette use took off.  But, if you look at their Figure 1, you could also argue for using 2009 as the break year because that is where the data they have on e-cig use extrapolated back to zero.  While e-cig use was lower before 2014, an increasing effect of e-cigs would be captured in the slope change in an interrupted time series model (assuming that the linear assumption is met).

The specific shape of the data curve is especially important in most recent years where e-cigarette use has increased so much among youth and young adults.  This fact, combined with the gateway effect, would lead one to predict that historical drops in cigarette smoking would stop or even reverse.  Indeed, looking at the detailed data in the supplementary figures shows this in several (but not all) cases:

  • The Monitoring the Future (MTF) data showed  increases in 10th grade 30 day and daily cigarette smoking (Supplementary Figures 1 and 8) and essentially flat 12th grade smoking between 2016 and 2017 (Supplementary Figure 2). 
  • The National Youth Tobacco Survey showed flat 30 day cigarette smoking from 2014 to 2017 (Supplementary Figure 3), the first time that there was not a drop in cigarette smoking since NYTS started in 2004.  (I also do not understand why Levy and colleagues did not use the NYTS back to 2004; they started in 2011.)
  • The National Health Interview Survey showed flat young adult smoking prevalence from 2015 to 2016 for both males and females (Supplementary Figures 13 and 14).

Another important omission in Levy and colleagues’ paper is that they only present data on cigarette smoking rather than total tobacco product use.  This is an exceptionally important variable because if e-cigarettes are increasing nicotine use among youth, that is a bad thing.

And that is what Lauren Dutra and I found through 2014 (figure above).  The rapid increase in e-cigarette use after 2014 in Levy’s Figure 1 reinforces this concern. 

The figure is a line chart showing the percentage of U.S. middle and high school students who currently used e-cigarettes and any tobacco product during 2011–2018.New data for 2018 released by the CDC in the November 16, 2018 MMWR reinforces how serious this problem is.  They documented continuing increases in e-cigarette use through 2018 (figure) and, more important, increases in any tobacco use

That is the real reality.



Add new comment

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.