In its Morbidity and Mortality Weekly Report of October 9, 1998, the Centers for Disease Control and Prevention claimed that data from the annual National Household Survey on Drug Abuse (NHSDA) show that the incidence of first cigarette smoking ("even one or two puffs") increased by 30% among persons 12-17 years of age between 1988 and 1996, and that the incidence of daily cigarette use in this age group increased by 50% over the same period. But there are several major problems with the report.
First, the CDC's conclusion about increasing incidence, which is defined as the rate of appearance of new smokers, is inconsistent with the NHSDA's own numbers on the prevalence of smoking in this age group (prevalence is the proportion of people in this age group who are smokers). Clearly, if more and more adolescents are starting to smoke, there should be more and more adolescents who are smokers. In other words, if the incidence of smoking initiation increases over successive years, this should increase the prevalence of smokers. Yet, according to the 1994 NHSDA there was no significant change in the prevalence of smoking in the 12-17 year age group in the 15-year period from 1979 to 1994:"No significant decrease in smoking since 1979 occurred for those age 12-17 (12 percent in 1979 and 10 percent in 1994). The rate of smoking among youths has been constant since 1992, at about 10 percent. However, among those age 12-13, the rates were 1.9 percent in 1992 and 2.6 percent in 1994. Similarly, for youths aged 14-15, rates were 8.8 percent in 1993 and 10.0 percent in 1994. While these apparent increases are consistent with data from the Monitoring the Future Study [an annual classroom survey funded by the National Institute on Drug Abuse], they are not statistically significant in the NHSDA data. [Source: Substance Abuse and Mental Health Services Administration (SAMHSA), "Preliminary estimates from the 1994 National Household Survey on Drug Abuse," September 1995.]
The discrepancy between the CDC's figures (reporting growth in youth smoking initiation incidence) and the NHSDA's figures (reporting stable youth smoking prevalence over most of the same period) is glaring. How can this be?
For one thing, the figures for incidence of smoking initiation did not come from the NHSDA, which only collects prevalence data. Instead the incidence figures were synthesized by the CDC, using a method that is incomprehensibly described in its MMWR article. The only thing reasonably clear is that the CDC expresses smoking initiation incidence for a given year as the number of new smokers "per 1000 person-years (PY) of exposure." The CDC gives the example of an adolescent who starts smoking at age 15. That adolescent had 10 person-years of "exposure," according to the CDC, because 10 is the number of years between age 5 and age 15. What was the adolescent "exposed" to during those years? The CDC doesn't say, but the implication is that whatever it was, it was causal and started at age 5. Why at age 5? Why not the year before smoking began? Again the CDC doesn't say.
In any case, it's instructive to compare what the CDC says about its incidence figures with what its table actually shows. For example, the CDC says:"Among persons aged 12-17 years, the incidence of first daily cigarette use fluctuated from 1966 (42.6) to 1983 (43.8) and gradually increased from 1988 (51.2) to 1996 (77.0) (Table 1)."
Sure enough, that's what Table 1 shows. However, these "trends" are not so impressive when you take the 95 percent confidence intervals of the yearly incidence figures into account. Examination of those confidence intervals reveals year after year of overlapping error ranges, which means you could draw a line of any slope -- positive, negative, or horizontal -- through the parts of the ranges that are shared by all the years listed. To illustrate, let's look at the confidence intervals for 1988 and 1996, a period the CDC says saw a dramatically increased incidence of adolescent smoking initiation:
The 95 percent confidence interval for 1988's incidence of 51.2 was plus or minus 7.4, which means that the incidence for that year could have been, with 95 percent confidence, anywhere between 43.8 and 58.6. The 95 percent confidence interval for 1996's incidence of 77.0 was plus or minus 13.7, which means that the incidence for that year could have been, with 95 percent confidence, anywhere between 63.3 and 90.7. Thus it's quite possible that the difference between 1988 and 1996 is the difference between 58.6 (1988's upper limit) and 63.3 (1996's lower limit), which would not be impressive.
Furthermore, because the incidence values and 95 percent confidence intervals are quite similar for most of the years between 1988 and 1996, there is no clear trend at all, except for the last three years -- 1994 through 1996. The CDC says the incidence of first daily use went from 67.7 1994, to 71.8 in 1995, to 77.0 in 1996. But those figures are not based on any survey data for those years. Instead, the CDC says it "estimated" the figures for those years by using data for other years. Well now how reliable is that? The CDC doesn't say.
The CDC's number juggling on teen smoking is a model of junk science, but it got the headlines the CDC wanted, such as this one on the AP story in The Washington Post:
"Daily Smoking By Teens Has Risen Sharply"
In a pig's eye.
Comments on this posting?
Click here to post a public comment on the Trash Talk Bulletin Board.
Click here to send a private comment to the Junkman.
Copyright © 1998 Steven J. Milloy. All rights reserved on original material. Material copyrighted by others is used either with permission or under a claim of "fair use." Site developed and hosted by WestLake Solutions, Inc.
Material presented on this home page constitutes opinion of Steven J. Milloy.