The magic 25%

A handful of anti-smoking extremists have long hoped that smoking is linked to breast cancer. The pink ribbon breast cancer campaign is arguably the best-publicised and best-funded initiative in pubic health. Because breast cancer is the most common form of cancer amongst women, even a small association with smoking would allow tobacco control advocates to claim that millions of cases could be prevented by stamping out tobacco.

The problem is that there really isn't any reason to think the two are related. Sixty years of epidemiological research has failed to find a link and, unlike with diseases of the lung and airways, there is no obvious causal mechanism. As recounted in Velvet Glove, Iron Fist (pp. 236-38), neither the International Agency for Research on Cancer (IARC) nor the American Cancer Society believe there is a link and even the otherwise outré Surgeon General's report of 2006 didn't claim smoking to be a cause of breast cancer.

Outside of California, it is generally accepted that breast cancer is not a smoking-related disease. Inside California, things are always a little different. From his pulpit at UCSF, Stanton Glantz has been insisting on a connection for years, and the California Environmental Protection Agency (Cal-EPA) conducted a meta-analysis in 2004 which found an association between breast cancer and passive smoking. When the American Cancer Society expressed reservations about this meta-analysis (amongst other flaws, it excluded a notable cohort study which would have wiped out the association), Glantz went berserk and referred to doubters as "religious fanatics", thus displaying an extraordinary lack of self-awareness.

Glantz has been at it again this month following a review of breast cancer risks conducted by the Institute of Medicine. Getting rather excited at the prospect at linking arms with the pink-ribbon campaign, he overstated the conclusions of the IOM report and announced:

It's time for the large breast cancer advocacy groups to join the tobacco control community.

Glantz seems to think that the IoM report implicated smoking (and passive smoking) as a cause of breast cancer. That is not how I read it, nor is it how the New York Times read it. What the IoM actually found was this:

The evidence also indicates a possible, though currently less clear, link to increased risk for breast cancer from exposure to benzene, 1,3-butadiene, and ethylene oxide, which are chemicals found in some workplace settings and in gasoline fumes, vehicle exhaust, and tobacco smoke.

This was the only reference to tobacco in a 700 word press release. In the report itself, the IoM say that they cannot rule out a link, but that the evidence is equivocal. Tobacco remains a "possible" cause in the same way that mobile phones were found to be a possible cause of brain cancer in a recent IARC report. In other words, the collated evidence does not suggest a causal link, but some studies have found an association.

There are two interesting aspects of the breast cancer/smoking hypothesis. The first is that there was barely a hint of a link for the first 40 years of epidemiological research, as the IoM acknowledge:

Before 1993, more than 50 epidemiologic studies examined the relationship between breast cancer and exposure to tobacco smoke. Although the quality of studies was highly variable, the better conducted studies did not suggest a causal relationship (Palmer and Rosenberg, 1993). An IARC review published in 2004 included studies conducted before 2002, and it relied heavily on a pooled analysis of 53 case–control and cohort studies by the Collaborative Group on Hormonal Factors in Breast Cancer Study (2002) that contended that apparent associations with smoking were confounded by alcohol consumption. The IARC (2004) conclusions were that neither active nor passive smoking was associated with increased risk of breast cancer.

In any other field of research this would be enough to put the matter to bed, but tobacco control was flooded with money in the 1990s and so it continued. This coincided with the rise of ultra-low risk epidemiology and cherry-picked meta-analyses which, in turn, was accompanied by the burden of proof being relaxed in the science to the point where statistically insignificant findings were taken seriously.

Breast cancer is a very common disease and smoking is a very common behaviour. Given these facts, any association between the two should have been evident very early on (by the 1950s, if not even earlier). That no one found an association despite smoking being the most studied risk factor of the twentieth century strongly suggests that none exists. "If smoking was a major cause of breast cancer, we would have found it by now," says Dale Sandler, chief of the NIEHS Epidemiology Branch.

Those who say that smoking (active or passive) causes breast cancer are making an extraordinary claim and, despite efforts being redoubled in the last fifteen years, there is no extraordinary evidence and very little ordinary evidence.

From the IoM report:

Active smoking 

The summary risk ratio was 1.10 (95% CI, 1.07–1.14), indicating a weak association with increased risk for early initiation of smoking. For women who smoked only after a first pregnancy, the summary risk ratio was 1.07, but it was not a statistically significant increase in risk (95% CI, 0.99–1.15). A subsequent report from the NHS found a statistically significant increase in risk associated with greater smoking intensity (i.e., pack-years of smoking) from menarche to a first birth (p for trend <0.001) (Xue et al., 2011). At 1–5 pack-years of smoking before a first birth the hazard ratio (HR) is 1.11 (95% CI, 1.04–1.20); for 16 or more pack-years, the HR is 1.25 (95% CI, 1.11–1.40).

No increase in risk was evident for pack-years smoked from after a first pregnancy to menopause. For 31 or more pack-years, the HR was 1.05 (95% CI, 0.92–1.19). However, pack-years of smoking after menopause may be associated with a slight reduction in risk (p for trend = .02) (Xue et al., 2011). For 16 or more pack-years of postmenopausal smoking, the HR was 0.88 (95% CI, 0.79–0.99).

... For women who started smoking between ages 15 and 19, the HR was 1.21 (95% CI, 1.01–1.44); whereas those who initiated smoking after age 30, the HR was 1.00 (95% CI, 0.76–1.32).

Brown et al. (2010) concluded that their data did not show a consistent association between smoking and significant increases in breast cancer risk among U.S.- or foreign-born Asian women. For example, the results for current smokers showed an OR of 0.9 (95% CI, 0.6–1.3) while ex-smokers had an OR of 1.6 (95% CI, 1.1–2.2).

A study that examined risk for triple-negative breast cancer found no statistically significant increase in risk over nonsmokers based on smoking status, age at initiation, or duration of smoking (Kabat et al., 2011). By comparison, women with estrogen-receptor- positive cancers (ER+) were at significantly increased risk with earlier initiation (< age 20: HR = 1.16, 95% CI, 1.05–1.28) and longer duration of smoking (≥30 years: HR = 1.14, 95% CI, 1.01–1.28).

These relative risks are low or non-existent and even the positive findings are often not statistically significant. The most interesting thing about these associations is that they are actually lower than the associations claimed for passive smoking.

Passive Smoking

A 2005 review by the California Environmental Protection Agency of various health hazards associated with exposure to secondhand smoke included a meta-analysis of 19 epidemiologic studies of breast cancer ... The meta-analysis produced an overall estimate for exposed women of RR = 1.25 (95% CI, 1.08–1.44) (CalEPA, 2005; also reported in Miller et al., 2007). When the analysis was restricted to five studies with more comprehensive exposure assessment, the overall estimate was RR = 1.91 (95% CI, 1.53–2.39).

In 2006, the U.S. Surgeon General’s report The Health Consequences of Involuntary Exposure to Tobacco Smoke, which included consideration of many of the same studies as the California review, concluded, “The evidence is suggestive but not sufficient to infer a causal relationship between secondhand smoke and breast cancer” (HHS, 2006, p. 13). The conclusion was based on a review of the findings from seven prospective cohort studies, 14 case–control studies, and a meta-analysis of all of these studies. The meta-analysis found that women who had ever been exposed to secondhand smoke (10 studies) were at increased risk of breast cancer (RR = 1.40, 95% CI, 1.12–1.76).

The idea that passive smoking is more dangerous than active smoking is patently absurd, but that didn't stop ASH (USA) hyping Cal-EPA's meta-analysis with this headline in 2005:

Secondhand Tobacco Smoke More Dangerous Than Smoking Itself

It is fitting that an organisation that endorses so much flim-flam should wind up embracing the principles of homeopathy, but any reasonable person understands that the dose makes the poison. In its understated way, the IoM acknowledges that it is a tad unlikely that people who inhale less than 1% of the dose inhaled by smokers would be at greater risk.

For most other smoking-related diseases, the relative risks are much stronger for active smoking than passive smoking. Thus findings of equivalent or stronger relative risks for breast cancer with passive smoking than with active smoking are difficult to explain mechanistically.

And yet these perverse findings exist and they require explanation. At first glance, it seems that the epidemiological research into breast cancer and tobacco don't tell us very much at all. Certainly, they don't tell us very much about the environmental causes of breast cancer, but I think they tell us quite a bit about the state of epidemiology. They show how easy it is to find a relative risk of around 1.25 (ie. a 25% increase) in an observational study. It takes only moderate recall bias or deficiencies in a study's design to come up with such associations. In the case of secondhand smoke and breast cancer we can surmise that the associations are false because there is no link with active smoking, but it is curious that the claimed associations with other diseases also fall in the same ultra-low bracket, regardless of the magnitude of the risk from active smoking.

Smokers are around 1,000 to 2,000% more likely to develop lung cancer. The passive smoker's excess risk is said to be around 25%.

Smokers are around 70% to 100% more likely to develop coronary heart disease. The passive smoker's excess risk is, again, around 25%.

Smokers are not any more likely to develop breast cancer, but the passive smoker's excess risk is said to be—you guessed it—25%.

Despite huge variations in the effects of smoking, the effects of secondhand smoke—if we are to take the epidemiological studies at face value—are remarkably consistent. Consistent with each other, that is. Not consistent with the rest of science.

Share

Tweet