No, We Can’t Say Whether Cancer Is Mostly Bad Luck

Two years ago, Time wrongly reported that “Most cancer is beyond your control.” The Guardian incorrectly wrote: “Two-thirds of adult cancers largely ‘down to bad luck’ rather than genes.” And the BBC misleadingly said: “Most cancer types ‘just bad luck.’” All of these deceptive headlines arose from a widely misinterpreted study that looked at the role of random chance in initiating cancers. That paper was itself criticized for a slew of methodological flaws, and spawned more than a hundred rebuttals.

Its authors are now back with a follow-up, which reads like a weird blend of doubling-down, clarification, and mea culpa.  Although they’ve gone some way towards addressing the problems of their first paper, their critics still say they’ve made several of the same conceptual mistakes. And once again, their work has led to similarly botched headlines.¹

Ultimately, this story reveals less about why people do or don’t get cancers, and more about how hard it is to talk or think about these diseases.

*****

In 2015, Cristian Tomasetti from Johns Hopkins Bloomberg School of Public Health and Bert Vogelstein from the Johns Hopkins Kimmel Cancer Center were trying to work out why some parts of the body, like the skin or large intestine, are so much more prone to cancer than others, like the brain or small intestine. By looking at U.S. data on 31 types of cancer, they found a clue. The lifetime risk of developing cancer in a particular tissue was strongly correlated with how often the stem cells² in that tissue divide.

Which made perfect sense. When a cell divides, it has to duplicate all of its DNA. Every time this happens, it picks up a few mutations—typos, created when DNA is copied imperfectly. Most of these mutations are harmless, but some will disrupt crucial genes. And if stem cells collect enough mutations in the wrong genes, they start dividing uncontrollably, creating a tumor. So tissues whose stem cells divide more frequently should indeed be more susceptible to cancer.

That would have been completely uncontroversial, had Tomasetti and Vogelstein not framed their results in terms of “bad luck.” Some cancer-causing mutations are inherited, while others are inflicted upon our DNA by environmental risks like tobacco, sunlight, alcohol, or asbestos. Tomasetti and Vogelstein argued that the random mutations arising in dividing stem cells represent a third group—distinct from the other two, more important, and unlikely to be preventable. When they published their results in the journal Science, they wrote:

“These results suggest that only a third of the variation in cancer risk among tissues is attributable to environmental factors or inherited predispositions. The majority is due to “bad luck,” that is, random mutations arising during DNA replication in normal, noncancerous stem cells.”

The paper triggered a hailstorm of criticism. Some scientists chastised the methods. Why did they ignore common cancers like breast and prostate? Why did they only focus on the U.S.? Others accused the duo of undermining public health. Many personal choices, from quitting smoking to staying lean, can dramatically reduce one’s risk of cancers, but why would you bother if you read headlines saying that these diseases are “largely down to bad luck?”

Such headlines were disastrously wrong. For a start, Tomasetti and Vogelstein looked at the differences between body parts, not people. Their data explained why tumors are more likely to strike the bowel than the brain, but not this bowel versus that one. As oncologist Vinay Prasad tweeted: “[Their] paper does not explain to cancer patients why they got cancer. [It] explains why cancer doctors get more colon cancer consults than sarcoma consults.”

Last week, the controversial duo returned with another co-author and a second paper, which provides more data for their “bad-luck hypothesis.” This time, they looked at 17 cancers, including breast and prostate. They also went beyond the U.S., collating data from 69 countries that vary greatly in their cancer rates and their exposure to environmental risks. And despite that variation, everywhere the team looked, they found the same strong correlation between a tissue’s cancer risk and its rate of stem cell divisions.  

Next, they used data from tumor-sequencing projects and epidemiological studies to sort cancer-causing mutations into three buckets,³ depending on their origin. Overall, they calculated that 66 percent of cancer mutations are due to random, unavoidable replication errors (R), 29 percent are due to environmental factors (E), and 5 percent are inherited (H). But those proportions vary a lot between cancers: in lung cancer, just 35 percent of mutations are randomly acquired, compared to 77 percent for pancreatic cancer.

But the team have clearly emphasized that these numbers don’t tell us what proportion of cancers are preventable. It can take several mutations to trigger a case of cancer, so even if one of those was due to an avoidable environmental factor, that cancer could still have been prevented.⁴ That’s why, as the team stressed, their finding that 66 percent of mutations are randomly acquired is totally consistent with other estimates that 42 percent of cancer cases are preventable. Mutations don’t equate to cases.

Try telling that to the Daily Mail, Sun, and Forbes, which all ran headlines ascribing the majority of cancer cases to bad luck. The British charity⁵
Cancer Research UK made the same error in a (since-corrected) post describing the team’s work. Stat avoided that pitfall, but when Scientific American reprinted their story, they bowdlerized the headline into “Most Cancer Cases Arise from ‘Bad Luck.’” Most egregiously, The Daily Telegraph said “Two thirds of cancers are unavoidable even if you live a healthy life.”

As for the paper itself, “it’s certainly more balanced than their earlier one,” says Rebecca Siegel, an epidemiologist at the American Cancer Society, “but it seems to me that they’re still missing the boat.”

For Siegel, it comes down to how you interpret the so-called R-mutations—the ones that supposedly arise during normal DNA replication. Imagine a group of rapidly dividing stem cells. Sure, on their own, they would spontaneously develop a lot of random cancer mutations. But they would also amplify any mutations they picked up from an environmental trigger, producing many daughters that carried the same scars. Both routes produce the same pattern—greater cancer risk in more rapidly dividing cells—but only in the former are mutations initiated by unlucky stem cells. In the latter, they are secretly environmental in origin. For similar reasons, it seems weird to divide mutations into heritable ones, environmental ones, and replication errors. All three categories are deeply connected. An inherited mutation might hobble a cell’s ability to repair typos in its DNA, while an environmental trigger might make the cell divide more rapidly. Both events would increase the frequency of replication errors.

But Tomasetti actually agrees. He has defined the R-mutations as those that occur in a normal tissue that’s not beset by anything else; if a mutation is caused by extra replication due to an environmental trigger, he puts it in the E bucket. The R-mutations represent the absolute baseline, in a hypothetical world where no carcinogens or cancer genes exist.

Still, Song Wu, a statistical geneticist at Stony Brook University, says that Tomasetti’s team has likely overestimated the proportion of R-mutations. They first estimated the proportion of mutations caused by known cancer genes, and known environmental risk factors—and subtracted those from the total. The rest, they say, are R-mutations. But that’s only true if we think we already know everything about the genes and risk factors that lead to cancer, “and I doubt anyone working in cancer believes that’s even close to the truth,” says Wu. For example, virtually every case of cervical cancer is caused by a virus called HPV. If this study had been done before HPV was discovered, it would have concluded that almost all cervical cancer mutations were R-mutations, and that the disease wasn’t preventable. As it is, we have a vaccine for it.

“We can’t use things that are unknown,” counters Tomasetti. “If new environmental factors will be discovered,” maybe the daunting 66 percent figure will go down. Then again, it could also go up since the Western population is ageing (and age brings even more stem cell divisions), and since public health policies might reduce the contribution of environmental factors.  

Wu and others argue that Tomasetti hasn’t even taken account of known unknowns—sources of environmental risk that are clearly present, but still undefined. Consider prostate cancer. The team calculated that a whopping 95 percent of prostate cancer mutations are R-mutations, which would make the disease almost entirely unpreventable. “That just doesn’t make sense,” says Yaniv Erlich, a geneticist at Columbia University and the New York Genome Center. There are substantial differences in prostate cancer rates between different countries. If the disease was mainly caused by random replication errors, rates should be the same everywhere you look—and they clearly aren’t. Indeed, immigrants who move from countries with low rates to those with high ones tend to pick up the higher risk of their new homes. The environment clearly matters for prostate cancer; it’s that we don’t yet know which factors are important. “The ‘environment’ is inherently harder to study than [inherited] mutations or replicative errors,” says Clarice Weinberg, a statistician at the National Institute of Environmental Health Sciences.  “We know quite a lot about carcinogenic effects of smoking, obesity and certain occupational exposures, but not much about other environmental factors experienced during  life or prenatally.”

The rates of the other common Western cancers—breast, lung, prostate, and colorectal—also vary considerably between countries. Many others, including thyroid, kidney, and liver cancers, have seen their rates increase over time. Yet others are substantially influenced by known risk factors: The majority of esophageal cancers are caused by tobacco and alcohol, while most skin cancers are caused by sun exposure. All of this argues against the dominance of R-mutations in fueling cancer. Tomasetti’s papers are based on mathematical models—interesting, but we need to gauge their claims against the reality of hundreds of epidemiological studies. “Their hypothesis just doesn’t jive with epidemiological evidence that we know to be true,” says Siegel.

And even if a particular cancer is entirely caused by R-mutations, it might not be unpreventable, as sites like Smithsonian and NPR have reported. Aspirin, that most familiar of drugs, might help to reduce the risk of colorectal cancer, as well as other types like esophageal and pancreatic. Some scientists are looking to drugs like aspirin as ways of halting the evolution of cancer, by reducing the rate at which mutations occur in the first place. And as cancer scientists move beyond their traditional focus on mutated genes, it may become possible to prevent tumors by targeting surrounding cells, reducing inflammation, or stimulating the immune system.

In fairness, Tomasetti’s group say in their paper that “R mutations appear unavoidable now, but it is conceivable that they will become avoidable in the future,” and he told me on the phone that “this is definitely something we should focus our research on.” But Erlich is concerned that exactly this kind of research will be stymied by the focus on dominant and supposedly unavoidable R-mutations. “I think it’s unfortunate,” he says. “This is a high-profile paper by famous people in the field. Others will look at this and say: This is the endpoint in the war against cancer?”

The odd thing about the two papers is that, for all their controversy, they don’t seem very radical. As Otis Brawley, chief medical officer for the American Cancer Society, told CNN: “[It] doesn’t tell me anything I hadn't known for the last 20 years.” “I think it’s very well-known that chance plays a role,” adds Siegel. “If you look at lung cancer, we know that 80 percent of U.S. cases are to do with smoking, but only 20 percent of smokers develop lung cancer. Chance is a huge factor.”

But Tomasetti argues that chance is understated. “You can check the website of any major institution, and there’s no mention of this,” he says. “I’m not claiming that 66 percent is the right number, but this is a component that was rarely mentioned and never measured. And it’s here to stay.”

If that’s the case, we had better improve in how we talk about it. The “bad luck” rhetoric is unhelpful, especially when it’s equated with “replicative errors”. Ultimately, it all comes down to luck. Puff on a cigarette and the carcinogens within may or may not disrupt an important gene. If you inherit a gene that predisposes you to cancer, you may or may not develop the disease—even the highest-risk genes are not guarantees. Cancers are all about probabilities, and this is one of the hardest things about the diseases to convey. You can do everything “wrong” and slip through the net. You can do everything “right” and still get come up short.

Tomasetti and Vogelstein have said that they hope to alleviate the guilt felt by patients—and especially parents of children with cancer—who read that many cancers are preventable and feel that they’re to blame for their poor health. That is a noble goal, but there’s also a risk of demotivating people who could do something about their cancer risk. This is one of the two great challenges of cancer communication: walking the fine line between empowerment and guilt, between hope and despondence.

The other challenge lies in going from the abstract world of statistical models and population-wide studies to the concrete world of individuals and patients. In reporting the recent paper, CNN wrote that “bad luck mutations increase cancer risk more than behavior” and that “dumb luck plays a more significant role than either environmental, lifestyle or hereditary factors in causing this disease.” That’s arguably accurate when you’re talking about mutations in a statistical model. But readers will look at that and think about themselves in their daily lives. They’ll hear that their personal cancer risk is determined more by the vagaries of fate than by their own choices. And they’d be wrong to do so.

I can rattle off statistics about what proportion of lung cancers are caused by smoking versus other causes, or what proportion of mutations are environmental, hereditary, or otherwise. But I cannot tell someone whether their cancer was down to the wrong carcinogen hitting the wrong gene, or random errors in a dividing stem cell. That fundamental, heartbreaking, existential question—Why me?—has no answer.