Cell Phone Radiation and Cancer

Christopher Labos, Kenneth R. Foster

The issue of cell phones and cancer is in the news again since the National Toxicology Program (NTP) study has released its results. Keeping track of the NTP results can be difficult. In 2016, they released the partial findings of their study (Report of Partial Findings 2016), which showed an association between cell phones and two types of cancer (cardiac schwannomas and brain gliomas). The full data was released in February 2018 (Wyde et al. 2018), and while the cardiac schwannoma association remained statistically significant, the brain glioma association was seen as more equivocal. Then in March, the NTP study results went through peer review where an eleven-member panel reviewed and voted on whether to accept or modify the study’s recommendations. The peer review panel (Actions from Peer Review 2018) voted to label the cell phone cardiac schwannoma association as demonstrating “clear evidence” of carcinogenicity and the glioma association as showing “some evidence.” (These reports are all online at the NTP website at ntp.niehs.nih.gov.)

Keeping track of this evolving evidence base can be confusing, and the NTP will issue a final report sometime this fall. But it’s worth examining why different people can come to such different conclusions about the study’s findings.

The NTP study was designed to expose rats and mice to different levels of radio frequency radiation (RFR). One group was a control group and three other groups were exposed to 1.5W/kg, 3W/kg, and 6W/kg of RFR. Researchers also tested two forms of signal modulation, reflecting two major access technologies employed by cellular telephones: Code Division Multiple Access (CDMA) and Global System for Mobiles (GSM). Both technologies transmit data in the form of modulated signals, but GSM is much less uniform in its power output than CDMA. Even though the average exposure level over time may be the same, hypothetically there could be a difference in biological effects, though there is no credible reason to expect any such differences.

The rats and mice being studied had their entire bodies exposed to RFR for nine hours every day for two years. The exposure also started in utero, not at birth. The whole-body exposure levels were far above the whole-body exposure limits for humans but were comparable to exposure limits set for very small regions of the body near a cell phone antenna. Consequently, the animals were being exposed to RFR in a way that is very inconsistent with the actual exposure to a human user of a cell phone, both in the particulars of exposure and duration of exposure. Notwithstanding these limitations, it is worth looking at what the data actually demonstrated.

The association between malignant gliomas and cell phones has been of primary interest, and it’s the one conclusion that has been subject to the most revisions. The idea that cell phones may cause brain cancer is not a new concern. The INTERPHONE series of studies (interphone.iarc.fr) is often cited as supportive evidence for this association even though the actual conclusions of the study were that “no increase in risk of glioma or meningioma was observed with use of mobile phones.” There was one statistically significant association though. Those who used their cellphones most (defined here as the top 10 percent of users) seemed to have an increased risk of glioma. But the authors noted that there were “implausible values of reported use in this group” and that bias and error in the measurement prevented a causal interpretation. Accurately measuring RFR exposure over years is extremely difficult. The combination of weak (and generally negative) results, coupled with the difficulty of accurately measuring exposure, has led health agencies to consider this evidence unpersuasive one way or the other.

Therefore, the results of the NTP were particularly eagerly anticipated. The NTP study did show an association between RFR and gliomas. However, the association was seen only in rats and not in mice. Also it was seen only in male rats and not female rats. Finally, it was seen with CDMA signal modulation but not the GSM signal modulation.

There are some important limitations to this analysis. Notwithstanding the obvious issue that animal studies do not necessarily translate to humans, it is hard to understand why the association would only be true in male rats and why it would only be true with one type of signal modulation. It is also worth noting there were very few cases of malignant gliomas in these animals. For the animals exposed to CDMA RFR, only the male rats showed an increase in gliomas—not the female rats or mice of either sex. The male rats exposed to CDMA RFR at 6W/kg had three malignant gliomas, compared to none for those exposed to 3W/kg, 1.5W/kg, or unexposed controls. For GSM RFR the 1.5W/kg, 3W/kg, and 6W/kg groups developed three, three, and two gliomas, respectively. Given the very small numbers, it becomes important to consider the possibility of random chance. A scientist would consider these results to be very fragile—if one animal in the control group had developed glioma (which is consistent with historical data for that species), the association would disappear statistically.

When it comes to cardiac schwannomas, the results are more consistent in that the association was seen for both forms of signal modulation, CDMA and GSM. But again, the results were seen only with male rats and not female rats, male mice, or female mice. Schwannomas are tumors arising from Schwann cells that produce the myelin sheath around peripheral nerves. Schwannomas are interesting because they are histologically similar to acoustic neuromas. Some studies have suggested a link between acoustic neuromas and cell phones (Hardell et al. 2013); other studies do not (Pettersson et al. 2014). Again, the results are fragile, and the evidence base is somewhat inconsistent.

Thus, any evidence linking RFR to cardiac schwannomas would seem to be possibly supportive of this link given the similarity of the tumor types. However, it is worth remembering that the rats had their whole bodies irradiated with RFR, and it is not immediately obvious why schwannomas would preferentially appear in the heart. In fact, they could have (and did) appear in any organ. Consequently, when you look at all schwannomas, not just the cardiac schwannomas, there does not appear to be a significant relationship to RFR. Therefore, for the schwannoma analysis to be positive you have to ignore the whole-body results and focus only on the cardiac findings.

Reconciling the disparate data has been made harder by the just-released study from the Ramazzini Institute in Bologna (Falcioni et al. 2018), which was rushed to publication after the NTP results were made public. This paper presented the results of a long-term rat study that suggests an increase in heart schwannomas in rats exposed to RFR. These data are hard to reconcile with the NTP. First, they used exposures about 1,000 times lower than in the NTP study, which would argue against a dose-response effect where more RFR is worse. While dose-response effects are not mandatory in science, it is difficult to understand how low and higher doses of RFR could be equally dangerous. The Ramazzini also diverges from the NTP in another way: the cardiac schwannoma association was only seen in male rats and not female ones, which makes these results far less consistent than has been reported in the media. Finally, the Ramazzini found no evidence that RFR was linked to neoplastic lesions of the brain. They claim that there was a nonsignificant trend, but this occurred in female rats as opposed to the male rats that were seen in NTP. All we can say for sure is that the NTP and Ramazzini studies are not entirely supportive of each other nor have they “settled” matters.

Given that the results are not consistent across or even within species, one must ask whether the results of the NTP could be due to chance alone. Given the small number of tumors that occurred in each group, random chance could have a significant role in these findings. We often fail to appreciate just how important random chance can be in statistical analyses. The ISIS-2 study offers up a perfect example (ISIS 1988).

The ISIS-2 study demonstrated that giving aspirin to patients after a heart attack improved outcomes. However, even though the study was overall positive, one subgroup of patients showed no benefit. That subgroup was patients born under the zodiac signs of Gemini and Libra. In fact, the authors of the ISIS-2 study purposely highlighted this rather ludicrous and totally spurious statistical finding to demonstrate that “all these subgroup analyses should be taken less as evidence about who benefits than as evidence that such analyses are potentially misleading.”

In the NTP study we have a similar problem. Remember that there were four groups of animals, which were tested against two types of signal modulation and evaluated for many different types of cancer including heart, brain, pituitary, adrenal, liver, prostate, kidney, pancreas, mammary gland, and thymus cancer among others. Thus, you have dozens of statistical analyses being run across all these many subgroups. The NTP study was an exhaustive analysis, but that thoroughness and the multiplicity of tests that were run means that you must expect some false positive results due simply to chance.

Most statistical tests are based on the assumption that you have a 5 percent false positive rate, represented by 1 – 0.95 = 0.05, or 5 percent. However, if you do two analyses the chance of at least one false positive is 1 – 0.95 = 0.0975, or 9.75 percent. Do five analyses and the chance of at least one false positive is 1 – 0.95 = 0.23, or 23 percent. Do thirty analyses and the chance of at least one false positive is 1 – 0.95 = 0.79, or 79 percent.

Therefore, the more tests you run in your study, the more likely that you will generate a false positive. And the NTP study ran a lot of tests. Consequently, they are very likely to have had false positives. Studies such as this are essentially fishing expeditions or data mining with no single hypothesis that is being tested, and it would not be surprising if the increase in heart schwannomas were just a random event.

There are statistical ways to deal with this type of multiple hypothesis testing. The Bonferroni correction is one technique that is sometimes used, and it basically amounts to using smaller p-value cut-offs the more tests you run. You basically divide 0.05 by the number of tests you intend to run. So if you perform two tests, then you should use a threshold of 0.05/2 or 0.025. If you run ten tests, then your threshold should be 0.005, and so on. The NTP study did not adjust for multiple testing.

The inherent weakness of the NTP results is their lack of consistency. We see a signal for harm in rats but not mice. We see a signal for harm in male rats but not female rats. We see a signal for schwannomas in the heart but not the rest of the body. Finally, the rats exposed to RFR actually lived longer on average than the controls. So do cell phones cause cancer while simultaneously extending survival? It is not impossible that there is some yet to be fully understood mechanism at play, but at this point random chance seems far more likely.


The NTP study is a good case history of the problems of data dredging. For more on this, see Kenneth R. Foster and Joseph Skufca, “The Problem of False Discovery,” IEEE Pulse, March/April 2016, available online at https://www.dropbox.com/s/4echhc6ez6pyn60/Foster_Skufca_2016.pdf?dl=0. and Stuart Vyse, “Moving Science’s Statistical Goalposts,” Skeptical Inquirer, November/December 2017, available online at https://www.csicop.org/si/show/moving_sciences_statistical_goal_posts.


Author Bios

Christopher Labos is a cardiologist with a degree in epidemiology. He writes regularly for the Montreal Gazette and cohosts a podcast called the Body of Evidence.

Kenneth R. Foster received his PhD in 1971. Since then he has been engaged in studies on the interaction of nonionizing radiation and biological systems, with more than 160 papers in peer-reviewed journals on topics including biophysical mechanisms of interaction, exposure assessment of RF fields in the environment, and medical applications of RF fields. In addition, he has written widely about the public controversy surrounding these issues and about broader issues related to technology and society. He is coauthor or coeditor of two books on risk assessment and the law.