Cellular telephones and risk for brain tumors

Glioma (all grades).

In Denmark, diagnoses of glioma (International Classification of Diseases topography codes 191.0–191.9 and morphology codes 93803–94603) are confirmed by MRI, CT, or histologic review. Persons suspected of having glioma were referred to one of the five neurosurgical departments in Denmark, and one surgeon at each department reviewed all incoming cases and notified our study nurse when an eligible patient had been admitted. The study nurse or a specially trained medical student would then approach the patient for a face-to-face interview.

We identified 464 potential incident cases of glioma in persons aged 20 to 69 years who had been referred to these departments between September 1, 2000, and August 31, 2002. All cases were checked against the Danish Cancer Register²² to determine whether they were cancer-free before the date of diagnosis of the glioma, to ensure that no bias was introduced by a previously diagnosed cancer. As a result, 54 cases with prior cancer, other than non-melanoma skin cancer, were excluded. A further 44 persons were excluded because the diagnosis was found to be incorrect on histologic review, 10 were excluded because the cases were prevalent or because they were over 69 years of age at diagnosis, and 2 persons were excluded because they did not understand Danish sufficiently. Of the 354 eligible persons with glioma, 38 (11%) refused to participate, a further 32 (9%) died before interview (with no proxy available), and 32 (9%) were aphatic, partially paralyzed, or severely depressed, leaving 252 (71%) persons for interview. The diagnoses of all 252 glioma cases were confirmed by histologic examination. They comprised 171 high-grade gliomas (WHO grade III-IV) and 81 low-grade gliomas (WHO grade II). For 19 patients, the interviews were conducted with a proxy (usually the spouse or child) 3 to 6 months after the death of the patient.

The potential exposure of the patients to radiofrequency energy from cellular telephones was estimated as high, medium, or low from the anatomic location of the glioma. The most energy is absorbed by the tissues closest to the position of the telephone against the head, and absorption drops off considerably at a depth of 4 to 5 cm.²³ If exposure to radiofrequency fields were associated with the occurrence of brain tumors, a shift in tumor location from areas of low exposure to those of high exposure would be expected among regular cellular telephone users.

Meningioma.

Diagnoses of meningioma (International Classification of Diseases topography code 192.1 and morphology codes 95300–95393) were confirmed in the same way as for glioma. Surgery for meningioma is, however, performed later after diagnosis than for glioma, so histologic confirmation of the diagnoses took longer. We identified 291 reported cases of meningioma, which were checked at ascertainment with the Cancer Register²² to determine whether they were cancer-free before diagnosis, and 30 cases with a prior diagnosis of cancer, excluding non-melanoma skin cancer, were excluded. Of the total of 261 reported incident cases of meningioma, all except 8 (3%) were confirmed by histologic examination, except those that were diagnosed by either MRI or CT. We excluded 10 cases with an incorrect diagnosis, 8 prevalent cases or in persons aged over 69 years at diagnosis, and 5 persons who did not understand Danish sufficiently, leaving 238 eligible meningioma cases. Of the eligible patients, 35 (15%) declined to participate (mostly because of poor physical condition), 11 (5%) died before we could approach them, and 10 (4%) were too debilitated to complete an interview; 7 (3%) were not approached because they had not been informed sufficiently about their diagnosis, leaving 175 (74%) cases of meningioma. Three interviews were conducted with proxies.

Controls.

We selected one control person for each reported brain tumor case, frequency matched on age (within 5-year ranges) and sex. The control persons were population based and sampled randomly from the Danish Central Population Register. Like the cases, the controls were checked against the Danish Cancer Register²² to exclude any with a prior diagnosis of cancer. At the end of the study period, we found that we had interviewed more controls than cases; however, all the controls were used in the analysis. By post hoc matching on sex and age (5-year intervals), we gave each control person a reference date, which was the date of diagnosis of the corresponding case. The response rate was 64% (n = 822).

Interviews.

As part of the Interphone study, a computerized personal questionnaire was developed.¹⁹ Face-to-face interviews were conducted either by a study nurse or by a specially trained medical student, either at home after surgery or at the bedside in hospital before surgery. Respondents were asked if they had ever used a cellular telephone. If so, they were asked if they were regular users (use at least once a week for 6 months or more) and how many different cellular telephones they had used regularly. For each cellular telephone used regularly, start and stop dates of use were recorded. If the respondent was still using the cellular telephone on the day of the interview, the stop date was set at the date of diagnosis (for cases) or the reference date (for controls).

The questionnaire also contained questions on the numbers of calls made and received and on the duration of calls on each cellular telephone used by the respondent, and changes in pattern of use over more than 6 months. On the basis of this information, we estimated the lifetime number of calls and the lifetime hours of telephone use. For each cellular telephone, we obtained information on use of handsets with a microphone and use of hand-free sets in vehicles. This information was used to modify the exposure estimate (see below).

Tumor characteristics.

For all cases, clinical data were obtained from medical records to verify the diagnosis, calculate the tumor size, and determine its location. We used imaging reports and to some extent actual images to determine the locations of the tumors. Gliomas were assigned to the lobe of the brain in which most of the tumor tissue was located. We considered that the lobes with high exposure are the temporal and parietal lobes, those with medium exposure are the frontal and occipital lobes, and that the remaining part of the central area of the brain has low exposure. The location of meningiomas that indented the brain in the midline or above the hemispheres was assigned according to the lobe they indented. Cases in which this distinction could not be established were categorized as laterally or centrally located. The area of the brain defined as lateral was all that located laterally of an arbitrary vertical cranio-sacral line drawn through the orbit and posterially (figure).

• Download figure
• Open in new tab
• Download powerpoint

Figure. Cerebral CT scan of a brain showing the areas described as lateral and central.

Socioeconomic status.

For all patients and controls, we obtained information on socioeconomic status, including level of education, marital status, employment, income in the year 2001, and wealth defined as taxable assets in 2001 from the Intergraded Database for Labor Market Research at Statistics Denmark.²⁴ As this information was provided in an anonymous form, it could not be adjusted for in the logistic regression models (see below). Furthermore, the linkage between our data and the socioeconomic data from Statistics Denmark did not enable us to stratify by diagnosis (low-grade glioma, high-grade glioma, meningioma). It did allow us to compare the socioeconomic characteristics of cases and controls and study participants and non-participants.

Recall.

Recall bias is a serious threat to the results obtained in case-control studies and especially in studies of persons with serious brain diseases. For all patients and controls, we obtained detailed information about the frequency and duration of use of the most recent reported cellular telephone from the two largest cellular telephone network operators in Denmark. This information was compared with the data on the questionnaire in order to validate the information obtained during interview. We provided the operators with all the cellular telephone numbers that had been reported during interviews, and a search of the electronic files of the two providers gave information on incoming and outgoing calls for 27 patients and 47 controls. Periods and patterns of cellular telephone use reported on the questionnaire were compared with equivalent data provided by the operators to identify periods for which we had information from both sources. Subsequently, we compared the number and duration of calls from the two sources.

We also evaluated the recall of both cases and control persons by asking them to complete the Folstein Mini-Mental State Examination, which is a short, well-validated test of memory.²⁵

The best-established risk factor for tumors of the brain is high-dose ionizing radiation. We estimated the risk for meningioma, low-grade glioma, or high-grade glioma after exposure to five levels of ionizing radiation: 0 = baseline; 1 = up to five diagnostic radiologic examinations of the skull more than 2 years before diagnosis; 2 = one or more CT scans of the skull more than 2 years before diagnosis or more than five radiologic examinations of the skull more than 2 years before diagnosis; 3 = occupational exposure, as defined by UNSCEAR,²⁶ for more than 5 years and more than 2 years before diagnosis; 4 = radiation therapy more than 2 years before diagnosis.

Ethical approval.

Approval was obtained from the Danish Ethical Committee (jr. no. KF 01-105/00) and the Danish Data Protection Agency (jr. no. 2000-1200-268 and 2001-41-1290). Written material was produced in accordance with the Helsinki II Declaration. Signed informed consent forms were collected from each case and control person and saved for documentation.

Statistical analysis.

Unconditional logistic regression models for frequency-matched data sets were used to estimate the OR and its respective 95% CI (Proc Phreg in SAS 8, Reference). All analyses were stratified for sex and 5-year age groups and additionally adjusted for educational level (low, intermediate, high; see table 1), region of residence (eastern Denmark, Funen, and western Denmark; see table 1), marital status (married vs single, divorced, or widowed). We defined the following exposure metrics: being a regular cellular telephone user (yes, no), years since first regular cellular telephone use, lifetime number of calls, lifetime hours of use, hours of use in the interval between first regular use and 5 years before the reference date, and intensity of use, defined as lifetime hours of use divided by duration of use. Cumulative use of cellular telephones was modified according to use of hand-free sets by a factor that varied with the answers given. We reduced the exposure by 100% according to whether the respondents reported use of hand-free devices all the time, most of the time (75% reduction), half the time (50% reduction), or less than half the time (25% reduction). The cumulative use was multiplied by the modification factor for the periods when hand-free devices were used. The cut-off points used in table 2 were defined in advance of the conduct of the analyses. For the three cumulative measures, the cut-off points are based on the distribution among the combined controls. The median of this distribution was chosen as the first cut-off point (50% percentile) and the third quartile (75% percentile) as the second cut-off point, leading to a baseline category of non-regular users and three exposure categories. Proxy interviews were excluded from all analysis including information on number or duration of calls.

The sizes of tumors in regular cellular phone users were compared with those of non-regular users with the Wilcoxon test. To evaluate whether the distribution of tumor locations differed between regular and non-regular cellular phone users, we used the χ² test for univariate analysis and categorical data, modeling for ordinally scaled data (exposure areas in the brain, as described above), using weighted least squares estimates for cellular phone use and for age and sex as additional independent variables.

The agreement between the cellular telephone use derived from the questionnaire and that derived from the network operators was analyzed with the kappa coefficient on categorized data (both distributions divided into quartiles). The kappa statistic quantifies the extent of agreement beyond what level of agreement is expected by chance alone. Kappa takes values between 1 (perfect agreement), 0 (no agreement), and −1 (perfect disagreement), with values around 0.5 representing fair agreement. Furthermore we had intended to apply limits for agreement,²⁷ but this could not be done because of the small numbers and because of the asymmetry of the data, with a large disagreement among heavy users that did not follow a normal distribution.