large studies are needed to have enough statistical power to detect effects when exposures and outcomes are rare. Thus, while the observational nature of the epidemiologic studies makes them difficult to conduct and interpret, they remain indispensable by providing direct information on the health of people exposed to an agent.


Recent concerns about possible health effects of exposure to RF radiation from cellular phones have centered on brain cancer (Krewski 2007), and a series of epidemiologic studies looking at the association between cellular phone use and malignant and benign brain tumors, as well as salivary gland tumors, have been published since 1999 (Johansen et al. 2001; Lonn 2004; Shoemaker et al. 2005; Schuz et al. 2006; Hardell et al. 2006; Lahkola 2007). The majority of these have been case-control studies.4 Limitations of these studies have included inclusion of only a small number of heavy users, inability to account for a sufficiently long latency5 period, differential participation by cases and controls according to history of cellular phone use, inclusion of both prevalent and incident cases,6 and relatively crude exposure assessment. All of these case-control studies relied on self-reported histories of cellular phone use, which can be prone to both random and systematic errors.7 The latter could occur if cases or their proxies tended to over-report cellular phone use relative to controls or if tumor laterality8 influenced reported laterality of phone use. It also is possible that cognitive impairment would cause brain cancer cases to incorrectly report phone use.

Analyses conducted as part of the INTERPHONE study, a multi-center international case-control study of brain and parotid gland tumors, point to selection bias9 due to likely under-enrollment of nonexposed controls,


A case-control study is a type of observational study in which subjects are selected on the basis of whether they have or do not have a disease under investigation. Characteristics such as previous exposure are then compared between cases and controls.


Latency is the period of subclinical disease following exposure that ends with the onset of disease.


Prevalence is the number or proportion of cases or events or conditions present in a given population. Incident cases are those who are newly diagnosed with the disease of interest.


Random errors vary in a nonreproducible way around a limiting mean. These errors can be treated statistically by use of the laws of probability. Systematic errors are reproducible and tend to bias a result in one direction. Their causes can be assigned, at least in principle, and they can have constant and variable components.


Tumor laterality is the preference in location of tumor in one portion of the body over other locations in the body.


Selection bias occurs when the association that is observed among those who participate in a study is different from what would have been observed because individuals who were eligible to participate refused or were not selected.

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement