Making Sense of CEFALO

CEFALO, a multi-center European project, is the first study to look specifically at the impact of self-reported mobile phone use on risk of childhood brain tumors.

It involved 352 children with brain tumors, compared with 646 healthy kids matched kid_cell_phonefor age, gender, and geographic location. Based on logistic regression analysis, CEFALO’s authors conclude that children who regularly used cell phones were no more likely to have brain tumors than non-users.

CEFALO essentially dismisses the notion that cell phone EMF might be a risk factor for brain cancer in children. The study, funded in part by telecom companies, was published in the Journal of the National Cancer Institute.

The Environmental Health Trust (EHT), an education and advocacy group focused on controllable environmental health risks, contends that CEFALO has many data discrepancies, omissions, and methodological flaws, and that the study’s stated conclusions are contradicted by some of the actual data (see CEFALO: Mixed Signals on the Cell Phone–Brain Tumor Issue)

What follows are some of EHT’s key concerns (in plain text), and point-by-point responses (in italics) from Dr. Martin Roosli, CEFALO’s official spokesman at the Department of Epidemiology & Public Health, University of Basel.

• In the subset of kids for whom actual utilization data were available from service carriers, there was, in fact, a correlation between brain tumor risk and time elapsed since mobile phone subscriptions were started. There was a more than doubling of brain cancer 2.8 years after the first subscription for a cell phone began (OR=2.15, 95% CI=1.07 to 4.29) along with a 99.9% confidence of a trend that the longer the subscription the higher the risk.

However, there is no exposure-response association with amount of use. If this were a true risk, the brain tumor incidence would have been increased by 50% during the last 10 years. This was not the case. There are differences how operator data were made available for cases and controls. One explanation might be that patients were more often successful than controls in making operator data available for more distant time periods.

Also, cases may have changed subscriptions or phone numbers less often than controls. Thus, mobile phone operator data would cover more distant time periods in cases than controls. This would lead to an erroneous notion that cases started to use mobile phones earlier than controls. Another possibility is the presence of prodromal symptoms before diagnosis in some case patients. To provide frail children with better protection, their parents may have given them a mobile phone subscription for use in case of emergency.

• When the duration of cell phone subscription was more than 4 years (the highest exposure) there was a greater than a 3-fold risk of ipsilateral brain cancer (OR=3.74, 95% CI=1.19 to 6.71), and close to a 3-fold risk among the kids making the largest number of calls (OR=2.91, 95% CI=1.09 to 7.76). Despite these findings, clearly presented in the study’s tables, the authors state there was no association between risk and amount of use.

At the same time, risk was reduced for persons who had centrally located tumors or did not have a preferred head side for mobile phone use. This is biologically implausible and indicates that cases are more likely to state a preferred head side of use.

• There were more brain tumor cases in “central or unknown” brain sites (n=215) than in confirmed ipsilateral (n=208) or contralateral (n=190) sites. Lack of tumor location data is a major flaw in this report. Such key data should have been available from hospital records.

We had available all hospital records. There were only a few cases with unknown locations. Most had a central location or were found on the non- preferred side of the head (for reported mobile phone use).

• In statistical analysis, the terms “Ipsilateral,” “Contralateral” and “Central or Unknown” should be mutually exclusive. In the CEFALO report, when the cases in each category are summed, the total is substantially larger than the total number of brain tumor cases (613 compared to 352). CEFALO’s authors give no explanation for this significant discrepancy.

This summation is wrong because in the reference group (non-regular cell phone users), cases did not have an attribution of the preferred head side for cell phone use (since they did not regularly use cell phones). Those cases are thus included in all analyses because they are always unexposed.

• CEFALO reports on 423 eligible brain tumor cases and 909 eligible controls, with exclusions of 71 cases and 263 controls, giving participation rates of 83.2% and 71% respectively. Yet when reasons for exclusion are summed, the number of excluded cases was 121 (50 more than stated); the number of excluded controls was 280 (17 more than stated). This would result in case participation of 60% and control participation of 69%. The difference would likely increase statistical biases.

I do not know where these figures are from. In the manuscript we write: “The main reasons for nonparticipation were refusal to participate (18 case patients, 172 control subjects), inability to contact the subject (five cases, 70 controls), and physicians’ denial of permission to contact some patients due to the severity of their disease (19 cases). Thus, something must be wrong with this (EHT) summation.

• The text reports confirmatory mobile phone utilization data (from service carriers) for 35% of brain tumor cases and 34% of controls. Calculating the number of cases and controls, this would mean researchers had corroborative information for 123 cases and 200 controls. Yet, the authors report that they had corroborative data for 196 cases and 360 controls. The reason for this significant discrepancy is unclear.

Again this was not carefully read. We write: “Operator recorded data regarding the amount of time that had elapsed since the phone users’ first subscriptions were activated were available for 35% of case patients and 34% of control subjects who reported to have a mobile phone subscription.” Not everybody (in the tumor and the control groups) has a mobile phone subscription.

• The CEFALO report includes a table on risk from (non-cellular) cordless phone use. But, for 33% of the tumor cases and 35% of controls these data are “missing.” This is a major data gap, as cordless phones also generate significant EM radiation.

It seems very hard for people to remember the amount of cordless phone use. That is why we had so much missing data. It is preferable to omit inaccurate data instead of trying to force people to make unreliable statements. If there would be an effect from cordless phone, this would appear in the available data. Omitting data does not create a bias.

• In explaining the choice of cutoffs for exposure ranges, CEFALO’s authors state that, “… the 50th and 75th percentiles were chosen as cutoffs to allow for the skewed data distribution.” But in the table covering cordless phone exposure, they state that, “The 75th and 90th percentiles served as cutoffs because of broad categories.” They explain nothing about what these “broad categories” are, or how using different cutoffs in different sections of the study would affect the reported results.

Because we had many participants in the same category it was not possible to take the 50th and 75th percentile, which we aimed for. Basically the categorization we eventually chose represents just the few categories we obtained from the questionnaire data. There was no other possibility for categorization and thus there is no other result that could have been obtained in this analysis.

Subscribe to Holistic Primary Care