Tobacco Scam: Smokefree Restaurants: Secondhand Smoke - The Issue is Health

What's statistical significance?

British American Tobacco's misrepresentation of the WHO/IARC lung cancer study, claiming the results showed "no risk," is in line with long-standing tobacco industry attempts to exploit public unfamiliarity with "statistical significance" and "confidence intervals."

When a study reaches "statistical significance," it means the researchers are very confident (usually 95% confident, a widely used convention) that the change observed in that study was not a fluke of chance. How can they be so confident?

1.An epidemiological study will examine a random sample of people exposed and people not exposed to a toxin (such as secondhand smoke). It will compare the rates of disease (heart disease, lung cancer)of the exposed individuals to the unexposed individuals.

2.This comparison is expressed as a ratio (x/y). If the ratio of disease in exposed individuals (x)versus unexposed individuals (y)exceeds 1.0, we conclude the risk is "elevated." For example, the ratio of 1.17 found by the IARC study indicates that secondhand smoke in the workplace increases an individual's risk of lung cancer by 17%.

Problem? Just as in public opinion polls, results still depend on the specific individuals who end up in the study. If you did the study again, you'd get slightly different results because you'd study different people. Random selection and other procedures are designed to eliminate biases, but the process of random sampling itself introduces uncertainties.

Statisticians estimate this inherent uncertainty with something called a "confidence interval." Newspapers reporting opinion polls usually call this the "margin of error." Confidence intervals depend on sample size and the confidence level you're after � say, 95%. A smaller sample produces a larger margin of error; a larger sample a smaller margin of error. The confidence interval is the range in which we can be reasonably confident �95% confident, in fact � the true health risk lies.

EXAMPLE The 17% increased risk of lung cancer for workplace secondhand smoke exposure detected in the IARC study had a confidence interval of 0.94 - 1.45.

If the 95% confidence interval contains 1.0 you can't, with 95% confidence, exclude the possibility that secondhand smoke did not change the risk of lung cancer � based solely on the evidence in that one study. But Big Tobacco, distorting statistics, takes that single study and claims that it shows there's no proven risk from breathing secondhand smoke.

Big Tobacco ignores the fact that the true risk is equally likely to be anywhere in the confidence interval. It is as likely to be at the upper end as at the lower end. Thus, the tobacco industry is correct in stating that the risk in the IARC study is not "statistically significant." But they don't mention the study's results are equally consistent with the conclusion that lung cancer risk is elevated as much as 45% (the upper end of the confidence interval) based on measures of workplace exposure in Europe.

EXAMPLE An earlier American Cancer Society paper reported a 27% higher risk of lung cancer for non-smokers exposed to secondhand smoke (95%CI, 0.85 - 1.89) that the industry used as part of its attack on the Hirayama study. The Tobacco Institute again focused on the lower bound of the confidence interval and ran newspaper ads claiming the study showed that secondhand smoke had an "insignificant" effect on lung cancer in non-smokers � even though the risk might just as well be elevated as much as 89% for a given individual.

Remember, the range of the confidence interval is a function of sample size, not just the risks due to secondhand smoke. Failure to reach "statistical significance" can be due to a study not being big enough. Look at ALL the evidence... The important thing is that the IARC study's results are consistent with all of the other evidence linking Secondhand smoking and lung cancer. Why is that critical?

In assessing scientific evidence, one must look at ALL the evidence, not just one study at a time. By looking at all the evidence scientists are able, in effect, to make the study sample much larger thereby narrowing the confidence interval or "margin of error." The sheer consistency in estimates over the years increases the overall confidence we can have in the conclusion that secondhand smoke statistically significantly increases the risk of lung cancer. Major evaluations of lung cancer risk in passive smokers:

Organization	Year	Nation	Relative Risk	Confidence Interval*
WHO/IARC	1998	7 European Nations	1.16 (spousal) 1.17 (workplace)	0.93 - 1.44 0.94 - 1.45
Scientific Committee on Tobacco and Health	1998	U.K.	1.20 - 1.30	N/A
California Environmental Protection Agency	1997	U.S.	1.20	N/A
National Health and Medical Research Council	1997	Australia	1.32	1.10-1.69
U.S.EPA	1992	U.S.	1.19	1.01-1.39
National Research Council	1986	U.S.	1.34	1.18-1.53
Surgeon General	1986	U.S.	1.53	N/A
*Confidence intervals are two-tailed 95%, except U.S.EPA which is one-tail 95% (two-tail 90%). The EPA was estimating how confident it could be that secondhand smoke INCREASED the risk of lung cancer rather than whether it CHANGED the risk. In other words, EPA � unlike the London Sunday Telegraph � did not consider the possibility that breathing secondhand smoke protects people from lung cancer.