Interpretation of Epidemiological Studies on the Relationship Between Mobile Phone Use and Cancer

In May 2011 the IARC (International Agency for Research on Cancer) classified radiofrequency electromagnetic fields as a possible human carcinogen mainly based on epidemiological studies about the association between mobile phone (MP) use and brain tumors. Considering that brain tumors have long latencies of around 30 years, it is unlikely that this association is due to an 'initiating' activity of MPs since virtually all studied brain tumor cases must have had already a covertly growing tumor when they started MP use. But there could be other adverse effects exerted by a MP when acting on later stages of malignant development. We propose that MP use acts adversely by increasing tumor growth rate and model it by an impact on the latency distribution shifting the age-incidence function to younger age.

We calculate (1) relative risks (RRs) for MP use in comparison to the meta-analytic RR estimate for glioma in adults; (2) RRs for neuroepithelial childhood brain tumors in comparison to the findings of the MOBIkids study; and (3) hazard ratios in comparison to the results of the Million Women Study (MWS).

The meta-analytical odds ratio for glioma and long-term MP use in adults of 1.22 (95% confidence-interval: 1.02-1.46) could be explained by a shift in the age-incidence function by 32% of MP usage duration. Applying a 20% shift for childhood neuroepithelial brain tumors reproduced the ORs that were predominantly less than 1 in the MOBIkids study. For glioma risk in perimenopausal women in relation to long-term MP use in the MWS we found hazard-ratios close to 1 applying a 32% shift in the age-incidence function.

The standard interpretation of relative risk estimates must be revised if exposure to the agent commenced after the malignant development has already started. All reported RR estimates of MP use can be reproduced by positing MP use increased tumor growth rate. However, since these results are obtained applying a modeling approach, further tests using epidemiological methods, which will be difficult or hardly feasible, or utilizing more promising laboratory methods are needed.