R r r i n n g g! Who's Calling?

The study that we've all been waiting for for 10 years to tell us whether cell phones cause brain tumors is.....inconclusive! As reported in the International Journal of Epidemiology, the World Health Organization's International Agency for Research on Cancer (IARC) conducted a large case-control study of 13,000 people in 13 countries, with interviews of 2700 users with a type of brain tumor called glioma, another 2400 with meningioma, and 7500 controls with no cancer. They chose people most likely to have been the heaviest cell phone users in the last 5-10 years, people in urban areas and aged 30-59. Controls were matched on age (within 5 years), sex and region of residence within each country. (Phone image from freefoto.com.)

Subjects were asked to recall details about their cell phone usage over 10 years -- average call duration, use of hands-free devices and so on -- and the questionnaire also asked for information about socio-demographic variables (though they ultimately used educational level as a surrogate for socioeconomic status), occupational exposure to electromagnetic fields and ionizing radiation, medical history, medical exposure to x-rays or MRI, and smoking. Information about the location of the tumor was also asked.

The majority of subjects in this study were not heavy users of cell phones. Among meningioma cases, the prevalence of regular cell phone usage was 52% during a single year, and 56% among controls, and among meningioma controls, lifetime usage was about 75 hours, or about 2 hours a month.

The odds of having meningioma or glioma with regular cell phone use compared with controls (those without cancer) was 0.79 (95% confidence interval: 0.68-0.91) for meningioma cases and 0.81 (CI: 0.70-0.94) for glioma, in the entire study population. The odds ratio (OR) varies somewhat by hours of usage, but is always below or around 1 (an odds ratio of 1 means that the test variable, in this case cell phone use, was not found to effect risk of disease), which, if taken literally, suggests that risk of brain tumor was reduced with cell phone usage, though the researchers believe that some systematic bias rather than an actual protective effect of cell phones is more likely to explain this result. Age and sex seemed to make no difference to the results. An important point is that these OR values are specific to this particular sample, and are only as good as the sample itself was 'representative' (and of what).

The authors calculated hundreds of odds ratios, broken down by age, sex, location of tumor, SES, hours of phone usage, and so on, and some were greater than 1. They found, for example, that the OR for glioma for those with the greatest call time (over 1640 hours cumulative use) was greater than 1. That some ORs were greater than 1, even if there was no actual effect, is not surprising, since with a 95% confidence level 5% of the results are expected to be positive by chance alone, so whether or not they have found any real association is still open to question. But, the fact that this is just the association that was being tested in this study would mean that more weight should be given to this finding, though the OR didn't gradually increase with use, which argues against it being a conclusive result. 

Our results include not only a disproportionately high number of ORs less than 1, but also a small number of elevated ORs. This could be taken to indicate an underlying lack of association with mobile phone use, systematic bias from one or more sources, a few random but essentially meaningless increased ORs, or a small effect detectable only in a subset of the data.

This was a $24.4 million study -- a grant almost as big as a Verizon bill -- with 1/4 of the funding coming from cell phone industry sources, because Verizon or Vodafone really, truly want to know the truth (don't they?). They of course welcomed the lack of clear findings -- as though that meant definitely there was no risk. Go on, call someone and tell 'em the good news!

According to one story reporting the study:

The British-based GSM Association, which represents international cell phone firms, said IARC's findings echoed "the large body of existing research and many expert reviews that consistently conclude that there is no established health risk."

In fact, that wasn't the conclusion of the study (rather it was that there may be a link of heavy usage with glioma, but that more research was needed -- a shocking surprise that they would ask for more grant funding).

So, apart from the clear conflict of interest in the industry funding of and reporting on the results of the study, do we actually know more than we did before the $24 million was spent, and if not, why not?

This was a retrospective study, meaning people were asked to recall behavior from years ago, which is unavoidably unreliable, particularly the farther back one goes. The authors of the study themselves discuss a number of other reasons that the study could be faulty, particularly selection bias, or a refusal to participate by some people who were contacted, which may be systematically related somehow to risk, and information bias, or a differential error in the reporting of past phone usage by cases or their proxies (who were interviewed when a subject with a brain tumor was too ill to answer questions or had died) vs controls. Proxies would presumably not know as much about the subject's usage, and it could be that cases were more motivated to remember cell phone use if it could explain their disease.

Prospective studies, which would follow people at some starting point for a certain length of time and measure as many relevant factors as possible would be a more accurate way to look at this question, and in fact a 30 year prospective study is apparently in the works. But cell phone technology is changing all the time so whether results in 30 years will be relevant to then current technology is rather doubtful.

In any case, the results of the studies done to date pretty convincingly show that the risk, if any is very small, and this raises the major problem with prospective studies: they are very costly, their conclusions don't arrive til your grandchildren are nagging for their first cell phones, and unless they're huge they will only find a very few cases. That might mean we should dismiss the risk, and if the risk were just of hang-nails that would be a reasonable conclusion. But the risk is cancer, and as with other studies of cancer risk, notoriously including cancer risks from radiation, nobody thinks we should just zap away and not worry. So what to do?

There's an obvious reasonable and rational solution: simply to accept that there may be a small, but serious risk, just as we know there are risks of driving cars, flying in planes, or getting dental x-rays. And here, a law can easily be passed that would require that headsets be part of any new mobiles. That would actually be good for Vodafone (not necessarily because of civic responsibility but at least because there'd be more gear to sell) and would obviate the problem unambiguously. If there is a radiation risk to whatever's near your pocket where the phone is stored, that would be even more difficult to detect, since the tissue is denser and thicker there than at your ear.

These large studies use as their null hypothesis that cell phones are safe -- they entail no risk. Instead, they should use as the null the current best-estimate of the risk, and try to reject that hypothesis. This is a different kind of statistical approach (related to what is known as Bayesian analysis, in which you take what you think you know, and adjust your new findings accordingly).

Since there is evidential reason to think cell-phones might be dangerous, but the risk (if any) is clearly small, the burden of proof of new studies should be to reject the idea that the phones are dangerous, not that they're safe, and estimate the maximum risk consistent with the data (if the proper null cannot be rejected). Another statistical issue is the probability ('p-value') used in evaluating the evidence. If the nominal 0.05 value is used, that's rather liberal, since it's not likely that much that's convincing can be said, in terms of small risks, at such a level. But since the risk (if any) is quite serious -- brain cancer -- the proper p-values should be very conservative -- bending over backward statistically, to say that one won't dismiss risk unless the evidence is very strong that there isn't any.

Simply requiring head-sets would be a sound, safe, and sane public health policy. Headsets would also save many lives from accidents due to distraction, as in driving. Such a policy, like fluoridating toothpaste, would have benefits but not inconvenience anybody -- well, hardly anybody. It would greatly inconvenience the epidemiological establishment by depriving it of hundreds of millions of dollars of funds, to support their habit for decades, to prove that......cell phone risk is very small, if any.