This one headline perfectly encapsulates why Vox.com’s coverage of tobacco policy is terrible:
That’s a lie, actually. Vox’s tobacco coverage is bad for more reasons than can be encapsulated in one headline, and it’s not really much better or worse than any other publication’s, but I’m trying to keep with the current form of writing on the internet. “Not that the old way was perfect,” writes my old school blog pal Jason Kuznicki, “but nobody fisks anymore, and for that we ought to be ashamed.” I agree. So how about some fisking? Perhaps my rank on Technorati will go up if you link to this.
Here’s Vox’s German Lopez writing on the Department of Housing Urban Development’s announcement that it will be banning smoking in all 1.2 million of the nation’s public housing units. “One major problem with this policy is it seems to single out low-income people,” writes Lopez, providing a glimmer of hope that he might think twice about intruding into 1.2 million homes of the less well off. “But there’s an easy solution to that: Indoor smoking should be banned everywhere — inside bars, restaurants, your home. Full stop.”
Never mind the casual authoritarianism behind the sentiment, or the constitutional issues it raises. (Has the Commerce Clause been interpreted so broadly as to allow this sort of thing? Eh, probably.) And never mind whether enforcing this particular ban by evicting some of the poorest members of society from their homes is a humane idea — Megan McArdle covered that here. Let’s just look at some numbers, the stock in trade of explanatory journalism. Lopez supports his advocacy for enforcing a smoking ban even in privately owned homes by noting the scope of the problem:
According to the Centers for Disease Control and Prevention’s most recent data, smoking kills 480,000 people each year. Secondhand smoke alone kills nearly 42,000 people. To put that in perspective, that’s around 8,000 more people than die to either car crashes or gun violence.
The 480,000 deaths figure got picked up in just about every story about the smoking ban in federal housing, including additional Vox posts this weekend from Julia Belluz and Dylan Matthews, from whom I’d expect better. (Matthews also seems to wish that we could ban smoking in private homes, but regrets that people would “freak out” if we tried. Imagine!) Both the total number of deaths and the number attributable to secondhand smoke are dubious, however, and none of the writers appear to have put any effort into understanding where they come from.
Let’s take the total number first. As Belluz notes, the smoking rate in the United States has dropped from 21 percent of Americans in 2005 to 17 percent in 2014. The decline gets more significant the farther back you look. Yet the CDC’s estimates of mortality caused by smoking don’t seem to reflect this. The CDC currently tells the public that smoking kills 480,000 people per year. That’s nearly 40,000 more annual deaths than it attributed to smoking from 2000-2004, when it claimed that smoking killed 443,000 people. And that’s more than the 438,000 deaths per year the CDC assigned to smoking from 1997-2001.
It’s worth asking why smoking mortality seems be increasing even as smoking rates are doing down. This is partially due to population growth and demographic changes; mortality rates would be a far more useful metric for comparison over time. And since smoking is associated with chronic diseases, some lag in the figures is to be expected. But still, the 480,000 figure is controversial.
Unlike deaths due to car accidents and gun shots, which are discrete events that can be tallied, there’s no direct way to count deaths caused by smoking (or especially deaths caused by secondhand smoke). Death certificates don’t say things like, “Bob died from heart disease made marginally worse by his cigarette habit.” Instead, researchers compare the prevalence of causes of death that are associated with smoking to the rate of smoking among different age groups, attempt to account for other confounding factors, and give their best shot at an estimate of how many people who died would still be alive in a tobacco-free world. Exactly how the current guess of 480,000 deaths is arrived at isn’t really clear.
The CDC’s numbers have been criticized in academic journals. A 2007 paper by Brad Rodu and Philip Cole in Nicotine and Tobacco Research offers a different model, which Rodu summarized in an article for Cato’s Regulation that criticizes the lack of transparency in the CDC estimates. With the Rodu-Cole model, you see the good news you’d expect from fifty years of declining smoking rates: “The U.S. mortality rate attributable to smoking declined about 35% between 1987 and 2002.”
OK, that’s just one paper, and from a source you may consider less reliable than the CDC. But in 2012 the same journal published another paper by Brian Rostron, whose affiliation is with the Food and Drug Administration’s Center for Tobacco Products. Rostron is also critical of the CDC’s methods, noting that they “have not been substantially revised since their introduction in the 1980s.” Rostron’s revised estimate for annual smoking-related deaths in 2004 is 380,000. The paper concludes, “we have estimated smoking-attributable mortality over time and found that the number of estimated deaths has peaked and finally begun to decline for both men and women in the United States.” If that’s correct, then we should have experienced a decade of declining deaths since 2004, and the CDC’s 480,000 figure is wildly off target.
[Update: German Lopez brings up a new paper in the New England Journal of Medicine, which suggests some excess mortality missed by earlier figures due to diseases that have not yet been definitively linked to smoking. Some of the relative risks are low, and there may be confounding variables, but the sample size is large. If those links bear out, then the sets of estimates above would be revised upward.]
That’s the total number of deaths attributable to smoking. How about the fraction attributed to secondhand smoke? If it’s difficult to figure out how many smokers are dying because of tobacco use every year, guessing how many people die from secondhand smoke is even more problematic. The CDC currently pegs the number at 41,000 deaths per year, of which a little over 7,000 are due to lung cancer and about 34,000 are due to heart disease.
Let’s tackle lung cancer first. If you want a rigorous, scientific indictment of secondhand smoke as a cause of lung cancer, you probably can’t do better than the 2006 report from The Surgeon General. The report concludes unequivocally that environmental tobacco smoke increases the risk of lung cancer in non-smokers. But by how much? This is expressed as a relative risk. A relative risk of 1 would indicate that people exposed to secondhand smoke are no more likely to develop lung cancer than those who are not exposed. (People who smoke habitually have a relative risk in the neighborhood of 16, which is huge.) What would you guess is the relative risk for non-smokers who are exposed to secondhand smoke over the long-term at their home or workplace?
If you guessed barely greater than 1, you’re correct. The report’s table of meta-analyses puts the relative risks of exposure at home for non-smoking spouses or at work for non-smoking employees in the range of 1.12 at the low end to 1.43 on the high end. That’s really low! Low enough to be sensitive to bias in how various studies are weighted, and low enough that it can’t be reliably detected in studies of secondhand smoke exposure. (One of the meta-analyses of childhood secondhand smoke exposure even suggests that children who are exposed to tobacco smoke are less likely to develop lung cancer. As they say on Twitter, “Whoa, if true!”)
How you interpret this ambiguity likely depends on your political priors. With that in mind it’s interesting to see the Journal of the National Cancer Institute publishing a headline like “No Clear Link Between Passive Smoking and Lung Cancer.” That’s a 2013 story about a cohort study of 76,000 women that “confirmed a strong association between cigarette smoking and lung cancer but found no link between the disease and secondhand smoke.” To go on:
The incidence of lung cancer was 13 times higher in current smokers and four times higher in former smokers than in never-smokers, and the relationship for both current and former smokers depended on level of exposure. However, among women who had never smoked, exposure to passive smoking overall, and to most categories of passive smoking, did not statistically significantly increase lung cancer risk. The only category of exposure that showed a trend toward increased risk was living in the same house with a smoker for 30 years or more. In that group, the hazard ratio for developing lung cancer was 1.61, but the confidence interval included 1.00, making the finding of only borderline statistical significance. […]
But many studies that showed the strongest links between secondhand smoke and lung cancer were case–control studies, which can suffer from recall bias: People who develop a disease that might be related to passive smoking are more likely to recall being exposed to passive smoking. […]
However, Silvestri finds some reassurance in the passive-smoking findings. “We can never predict who is going to develop lung cancer,” he said. “There are other modifiers. But you can say, with regard to passive smoke, it’s only the heaviest exposure that produces the risk. We kind of knew that before, but it’s a little stronger here.”
“We’ve gotten smoking out of bars and restaurants on the basis of the fact that you and I and other nonsmokers don’t want to die,” said Silvestri. “The reality is, we probably won’t.”
The study has not, to my knowledge, appeared in a journal, but if you’ve followed debates about secondhand smoke at all, then the candor in the commentary above is refreshing. The association between secondhand smoke and lung cancer has never been as scary as anti-smoking advocates make it out to be. Read Christopher Snowdon for a sense of the numbers; the ten studies with the largest sample sizes find risk ratios of 1.29, 1.11, .70, 1.03, 1.53, 1.10, .90, and .96. This is hardly the stuff of nightmares, and it’s difficult to imagine basing an accurate model of secondhand smoke mortality on such tiny risks.
How about deaths due to heart disease? Economist Kip Viscusi provides what I think is a fair summary in his 2002 book Smoke-Filled Rooms: “Despite the central role of lung cancer in the ETS [environmental tobacco smoke] debates, the heart disease linkage may be greater, and the risk levels much larger.” This accords with the CDC’s higher attribution of deaths due to heart disease than to lung cancer.
Studies on the short-term effect of smoking bans, however, are not encouraging. In the early 2000s, a wave of research conducted in small cities that had implemented smoking bans concluded that they could drastically reduce heart attacks. Skeptics criticized these studies for relying on data from small populations. Newer research confirms that they were right to be doubtful. The most notable example is a 2014 study of the state of Colorado in The American Journal of Medicine which was co-authored by researchers who had previously published papers showing that small towns in Colorado experienced declines in heart attacks after banning smoking. In their new research, they write:
Although local smoking ordinances in Colorado previously suggested a reduction in acute myocardial infarction hospitalizations, no significant impact of smoke-free legislation was demonstrated at the state level, even after accounting for preexisting ordinances.
These analyses support the hypothesis that small study populations may be more likely to find dramatic changes in acute myocardial infarction incidence, whereas increasing the study sample size attenuates the magnitude of the reduction. Also, review of the studies in aggregate reveals data asymmetry that suggests the potential for publication bias or heterogeneity not entirely explained by a random-effects meta-analysis. […] Overall, a review of published research shows that acute myocardial infarction RR reduction appears inversely related to sample size. […]
Available evidence suggests that acute myocardial infarction incidence has been decreasing dramatically, unrelated to smoke-free ordinances. […] This emerging evidence highlights the importance of accounting for secular trends in acute myocardial infarction incidence before definitive attribution to smoke-free ordinances can be made. […]
Overall, available evidence suggests that the decrease in acute myocardial infarction incidence associated with reductions in secondhand smoke exposure may be substantially lower than originally estimated.
The decline in hospitalizations due to heart attacks is an important factor to consider. From the same paper:
Data from the Centers for Disease Control National Environmental Public Health Tracking Network recently evaluated secular trends in 20 Network states from 2000 to 2008 using a longitudinal linear mixed effects model. The authors documented a statistically significant overall decrease in age-adjusted acute myocardial infarction hospitalization rates, with most states showing more than a 20% decline during the period. This temporal reduction in acute myocardial infarction incidence is of a magnitude that exceeds the reduction observed in many smoking ordinance studies. Despite this, some analyses have not accounted for secular trends.
This reduction in heart attacks doesn’t get much attention, but it comes up in Rostrom’s paper as well: “Deaths from ischemic heart disease for persons aged 65 and over decreased from 201,000 in 2000 to 158,000 in 2007 for U.S. men, and from 233,000 in 2000 to 170,000 in 2007 for U.S. women.”
And you see similar downward trends in the UK. A study of Scotland from 2002-2010 found that deaths from heart disease declined by 43%. Nearly half of this decline was credited to improved treatment. Changes in the rate of smoking were credited for only 4% of the decline. One has to wonder: Given all the advances of the past decade or so, why is the CDC’s estimate of secondhand hand smoke deaths caused by heart disease still so high?
The decline in heart attacks and deaths due to heart disease has occurred over a similar time period as the spread of smoking bans. The best source for information on the adoption of smoking bans in the United States is Americans for Nonsmokers Rights, who maintains a database of smoking restrictions. According to their tracking, the number of jurisdictions in the United States with 100% smokefree laws in all workplaces, restaurants, and bars has increased from two in 1993 to 790 in October 2015. Estimates of the number of deaths caused by secondhand smoke, however, have been strangely consistent. The current estimate from the CDC is 41,000. That’s a little less than the 2006 Surgeon General’s report estimate of nearly 50,000. And if you go back to 1990, you can learn from The New York Times that:
The newer understanding of the health hazards of passive smoking were underscored in a report at a world conference on lung health in Boston last week. Dr. Stanton A. Glantz of the University of California at San Francisco estimated that passive smoke killed 50,000 Americans a year, two-thirds of whom died of heart disease.
Sound familiar? It’s almost as if the claim that secondhand smoke kills around 50,000 Americans per year is based on its political utility rather than any firm grounding in epidemiology.
To review: In the past two decades, the rate of smoking has substantially declined. Treatment and prevention of heart disease, which allegedly causes the vast majority of deaths associated with secondhand smoke, has improved dramatically. Doubts about the magnitude of the association between secondhand smoke and both lung cancer and heart attack incidence have spread to mainstream academic journals. And exposure to secondhand smoke has been greatly reduced thanks to declining smoking rates, the proliferation of smoking bans, and changing social norms. Yet the number of Americans dying from secondhand smoke exposure has, supposedly, barely nudged downward over twenty-five years. These facts don’t hang together.
So how many deaths does secondhand smoke cause per year? I’ve been writing about tobacco policy intermittently for nearly a decade and in-depth for the past year, and I’m reluctant to commit to a number. I can tell you that I’m extremely skeptical of the CDC’s figure of 41,000, and I think that any responsible journalist ought to be skeptical too.
This brings up problems with contemporary reporting on tobacco policy and with explanatory journalism in general. In the old days of tobacco reporting, a policy announcement such as the ban on smoking in public housing would have gotten just as much coverage as it did this week. But reporters covering it might have also sought comment from pro-smoking sources. A lot of what those sources said would have been total spin and bullshit of the sort satirized in Christopher Buckley’s Thank You for Smoking, but some of what they said might have been valid criticism that pointed writers to legitimate weaknesses in their stories. That dynamic has been much reduced since the Master Settlement Agreement of 1998 that dismantled pro-tobacco organizations.
A couple years ago I interviewed Michael Siegel, a professor at the Boston University School of Health who formerly worked for the CDC’s Office on Smoking and Health, and he was blunt about the deterioration of scientific integrity in the anti-smoking movement. Siegel is generally in favor of indoor smoking bans, but has become a vocal critic of how anti-tobacco groups’ exaggerate the dangers of secondhand smoke. He blames this in part on the loss of an adversarial process:
The current state of tobacco control I would describe, quite sadly, as misguided. It is now guided more by ideology and politics than by science. Ironically, I think one of the reasons it has lost its way is that some time around 2000 or so, the tobacco industry relinquished its watchdog role. Organizations in tobacco control used to be very careful because they knew the tobacco industry was watching and would call them on it if they exaggerated or distorted the truth. But after around 2000, the tobacco companies stopped playing this role and basically allow the tobacco control groups to say anything they want.
Good journalism is more than regurgitating a scary-sounding number from an authoritative source. It’s understanding the motivation and reliability of your sources, and seeking out potential opposition to see if there’s a counterargument that they failed to mention. Reporters know to be skeptical of tobacco companies. They haven’t learned to be skeptical of anti-smoking sources, and given the changes in the regulatory landscape of tobacco it’s long past time that they do. It’s the only way to avoid uncritically reporting that a smoking ban can decrease heart attacks by 60% in just six months, or that smokers are “contaminated” and “actually emit toxins,” or that people who use e-cigarettes are “inhaling Chinese-made antifreeze,” or that 41,000 Americans are dying from secondhand smoke exposure in 2015.
The past few days of coverage at Vox, with three different writers repeating the same dubious statistic and none of them investigating it, doesn’t raise my confidence in their model of explanatory journalism. This is explanation divorced from skepticism. The current media environment makes it easy to find numbers to support one’s political view with just a few minutes of Googling, appearing to write from a perspective of data-driven empiricism, and harvesting those all-important clicks. (Is Vox’s “The case for banning smoking indoors — even in your home” any less dumb than Slate’s recent piece arguing that spooning is sexist?) But when those numbers are taken at face value and without context, the writers are just one bad statistic away from calling in the cops to search for ash trays on your kitchen counter.
I like Vox. I really do, even if I find myself turning more often to their entertainment writing than their policy pieces. I view the site as a generally useful source of information about complex topics that it’s difficult for any one person to know in detail. Yet occasionally the site covers a topic that I do know in detail. And when it does, I have to wonder how much of their other coverage is equally superficial and credulous.