Brain Tumour Risk in Relation to Mobile Telephone Use: Results of the INTERPHONE International Case-control Study

The INTERPHONE Study Group


Int J Epidemiol. 2010;39(3):675-694. 

In This Article


The INTERPHONE study is the largest case–control study of mobile phones and brain tumours conducted to date, including the largest numbers of users with at least 10 years of exposure and the greatest cumulative hours of use of any study. An exhaustive analysis of this large data set involved estimation of hundreds of ORs; rather than focus on the most extreme values, the interpretation should rest on the overall balance of evidence. The null hypothesis of no association would be expected to produce an approximately symmetric pattern of negative and positive log ORs. A skewed distribution could be due to a bias or to a true effect. Our results include not only a disproportionately high number of ORs <1, but also a small number of elevated ORs. This could be taken to indicate an underlying lack of association with mobile phone use, systematic bias from one or more sources, a few random but essentially meaningless increased ORs, or a small effect detectable only in a subset of the data.

For meningioma, there is little evidence to counter a global null hypothesis, and we conclude that INTERPHONE finds no signs of an increased risk of meningioma among users of mobile telephones.

For glioma, an increased OR was seen in analyses in the highest decile of cumulative call time, including tumours in the temporal lobe and subjects who reported having used the mobile phone mainly on the same side as where the tumour occurred. Still, the evidence for an increased risk of glioma among the highest users was inconclusive, as the increase could be due to one or more of the possible sources of error discussed below.

In the following sections, we explore possible explanations for the apparently decreased risk of meningioma and glioma for regular users compared with never regular users, and the apparently increased risk of glioma in a subset of users.

Decreased Risk with Ever Regular Use of a Mobile Phone

An apparently decreased risk of brain tumours with ever regular use of a mobile phone (relative to never regular use) has been seen in other studies.[18,23] Putting aside a genuine protective effect as implausible, we have considered other reasons for these observations.

Sampling Bias In all but two centres, a population-based design was used. This requires that the cases in the study were representative of all cases in the respective population and that the controls represented all non-cases, within matching strata. In practice, it is difficult to demonstrate that these conditions have been fulfilled in any case–control study. Cases may be missed due to lack of detection, misdiagnosis or incomplete registration (such problems may be more likely for meningioma than for glioma). It is uncertain whether the sampling frames used to select controls represented the study base in some countries. To the extent possible, we conducted sensitivity analyses that examined the effects of different recruitment strategies between centres; they did not show substantial changes in the results (Table 6).

Levels of Participation Constrained by the requirements of ethical review committees and facing the population's increasing reluctance to participate in interview studies, we attained participation rates of 78% among meningioma cases, 64% among glioma cases and 53% among controls.[26] Although such proportions are not unusually low, they raise the possibility of selection bias with respect to mobile phone use.

Controls in 11 centres and cases in 9 centres who refused the full interview were asked to respond to a brief non-respondent questionnaire on mobile phone use. The cases and controls who complied with this short inquiry reported a lower lifetime prevalence of ever regular use of a mobile phone than did respondents to the full interview, implying that information from those who participated in the full interview may overestimate prevalence among all eligible subjects. Because participation and refusal differed between cases and controls, such non-representativeness may have distorted the OR estimates.[30] Although caution is required in extrapolating from the findings of the sub-study, we estimated, in the more plausible scenarios, that non-participation bias may have led to a reduction in the ORs for regular use of 5–15%,[30] which is less than the observed reductions below the null in the ORs in ever regular mobile phone users for meningioma (21%, 95% CI 32–9) and glioma (19%, 95% CI 30–6; Table 2).

Prodromal Symptoms Prodromal symptoms of a brain tumour could dissuade subjects from becoming phone users or reduce their use before diagnosis (reverse causation). Glioma is typically diagnosed quite soon after first symptoms. Although prodromal symptoms might result in lowered ORs among very recent users (e.g. <2 years since starting use), these are unlikely to explain the reduction in ORs observed among the vast majority of the users in our study population who started using mobile phones 2–10 years before disease onset.

Timing of Interviews As the use of mobile phones has become more common over time, the later interview dates of controls could have spuriously increased the prevalence of exposure in the control group. However, restricting analyses to matched sets in which the cases and controls were interviewed within 1 month of each other resulted in very little change in the OR for regular use ≥1 year in the past (Table 6) and hence seems unlikely to explain the low ORs overall. Further, the use of a common reference date for each case and its matched control should have minimized any bias induced by differential timing of interviews.

Confounding Higher socio-economic status has been associated with a higher risk of brain cancer in some but not all relevant studies,[31,32] and with mobile phone use, particularly when the technology was new.[9] We adjusted for education level in all analyses, but acknowledge this is an imperfect indicator of SES. Otherwise, there are few well-established risk factors for brain tumours; analyses adjusting for measured potential confounders had little impact on the ORs (Appendix 1, Table 4, Supplementary data are available at IJE online).

Low Overall Risks among Mobile Phone Users The reduced OR for regular users compared with never regular users seems unlikely to reflect a genuine protective effect and makes our results difficult to interpret.[33] It could result from the sources of error discussed above, although based on the evidence we have regarding their magnitude and effects[30,34] they may not account fully for the observed reduction in risk.

It might be possible to correct, at least crudely, for assumed downwards bias in the ORs for mobile phone use by undertaking a series of analyses using the lowest category of users as the reference category for OR estimates in higher categories. Results of such an analysis of the mobile phone use variables in Table 2 are shown in the Table of Appendix 2 (see Supplementary data available at IJE online), accompanied by a discussion of the strengths and weaknesses of this approach. We have also done some work to characterize possible sources of bias[30,34] and are currently exploring the possibility of correcting the OR estimates mathematically for their effects.

Elevated Risks of Glioma among Heavy Users

There was some evidence of an elevated risk of glioma in the highest decile of cumulative call time, with the highest point estimates seen for tumours in the temporal lobe and for subjects who reported having used their mobile phone mainly on the same side as that on which the tumour occurred. We explore here possible interpretations of these findings.

Biases Related to Possible Differential Quality of Exposure Data When compared with controls, glioma cases had a higher proportion of proxy respondents, a higher number of imputations for missing values, and a higher proportion of subjects judged by their interviewer to be non-responsive or having poor memory (data not shown). However, sensitivity analyses showed that these differences, on their own, did not explain the results seen in the highest decile of cumulative call time (Table 6).

Differential error between cases and controls in reporting of mobile phone use could substantially affect our results; such information bias could arise from several sources. First, a brain tumour, particularly in the frontal or temporal lobes, may adversely affect cognition and memory.[35] Secondly, cases may be more motivated to recall and report a publicized potential risk factor for their disease.

To investigate the accuracy of self-reported phone use, two validation sub-studies were conducted in some of the INTERPHONE centres. Amongst healthy volunteers using software-modified phones (recording number and times of calls), phone use in the past year was reported with substantial random error; with over- and under-estimation both frequent.[36] Errors were larger for duration of calls than for number of calls, and phone use was under-estimated by light users and over-estimated by heavy users. In another sub-study, records of mobile phone use up to 6 years previously were obtained for some participants in three INTERPHONE centres, allowing us to compare the interview responses with the records.[37] Overall, there was little evidence that recall quality differed between cases and controls, but there was some indication of greater over-reporting by cases than by controls for the period 3–5 years before interview. These sub-studies provide no information regarding differential reporting error for periods more distant than 5 years before interview.

Some subjects reported very high daily average call times and this was more common among cases than controls. Thirty-eight cases and 22 controls reported >5 h use/day and 10 cases and no controls reported ≥12 h/day. There is reasonable doubt about the credibility of such reports. Excluding all subjects who reported >5 h use/day reduced the ORs in the highest decile of cumulative time from 1.40 to 1.27 (95% CI 0.92–1.74). In contrast, truncating the average call time to 5 h/day had little effect on the OR. It is not clear which of these two approaches (if either) is more appropriate. However, the key question is whether these cases with unreasonably high values reflect a general tendency for cases to overestimate more than controls, which could contribute to the apparent excess risk in the highest decile. As noted earlier, there is evidence that cases tended to overestimate their past exposure more than controls did.[37]

Non-differential error (random variability or uncertainty in the exposure estimates) may also affect the findings. With dichotomous exposure indicators such bias is towards the null, but for polytomous variables the effect is difficult to predict.[38–40]

Location of Tumours and Laterality of Use of Phones Absorption of RF energy from mobile phones is highly localized.[29] Thus, an association of phone use with tumours occurring near the location of the phone would constitute stronger evidence for aetiology than an association with more distant tumours.

Ipsilateral ORs were almost always greater than contralateral ORs. There was no consistent pattern with regard to level of exposure, although a trend towards a stronger effect of ipsilateral use relative to contralateral use with increasing exposure was observed for cumulative number of calls. Results of case–case analyses (using Inskip's method[18]) also suggested higher risks of gliomas with ipsilateral phone use, but again no consistent trend with increasing exposure. The observation of an unlikely ipsilateral effect in low exposure categories suggests that cases might have over-reported use on the side of the tumour.

There is, though, evidence of lack of such reporting bias from a sub-study. In three centres (Australia, Canada and Japan), participants (172 glioma and 160 meningioma cases and 340 controls who were regular users) were asked at the end of their interview to put a mobile phone to their ear as if answering a call. The concordance between the reported side of use of the phone and the side where it was held was lower for cases (72% glioma cases, 66% meningioma) than controls (95%). The greater degree of concordance among controls suggests differential reporting quality. Among cases, however, there was as much discrepancy in the contralateral direction (52 instances) as in the ipsilateral direction (48 instances). Thus, it is possible that the ipsilateral effect is a true effect, is due to reporting bias or is a mixture of both.

Few studies have related field strength to anatomic structures, but a recent investigation of 110 phone models found that exposure is generally highest in the temporal lobe.[29] While laterality analyses may be biased by the respondent's knowledge of the side of the tumour, results for tumours in different lobes are probably less susceptible to reporting bias. ORs for glioma in the highest exposure categories were higher for tumours in the temporal lobe than in other lobes, but the CIs around the lobe-specific estimates for each measure were wide.

Coherence and Consistency The strongest evidence of an increased risk of glioma was found for cumulative call time, which is a function of the number and duration of calls. Conceptually, cumulative call time might be the most relevant measure of exposure. However, in validation studies, the number of calls was recalled more accurately than the duration of calls.[36,37] For the cumulative number of calls, the ORs, while highest in the highest deciles, were consistently below one. In the absence of a known biological mechanism, it is hard to know whether more weight should be put on results from the more accurate or the conceptually preferred exposure measure.

The apparently increased risk of glioma for cumulative call time was restricted to the top decile, ≥1640 h. There was no upward trend across the first nine deciles of cumulative call time. In contrast with the excess risk seen on the scale of cumulative call time, risk did not appear to be increased by length of time since first exposure or by duration of exposure. The pattern of point estimates of ORs in the high call time categories in three strata of time since exposure started—3.8 in the most recent and 1.3 in the more distant ones (Table 3)—is not what one would expect if there were a causal association; although the CI in the newest users was wide and encompassed the point estimates for heavy use in the two longer use groups. By analogy with known carcinogens, the lack of a consistently increasing risk with dose, duration of exposure and time since first exposure weigh against cause and effect. Nevertheless, given the uncertainty surrounding possible effects of RF on the brain, no strong case can be made for the plausibility or implausibility of any observed exposure response pattern.

Comparison of Meningioma and Glioma Results

While the ORs for meningioma were lower than that for glioma in high exposure subgroups, there were some similar patterns. First, the OR for all regular users compared with never regular users was very similar. Secondly, there was no trend in relation to cumulative call time except for an elevated OR in the highest decile. Thirdly, the increase in the last decile was more pronounced for cumulative call time than number of calls. Fourthly, the highest OR for cumulative call time was seen among subjects who had recently started regular use. Fifthly, the ORs were greater for ipsilateral than contralateral use and the ratios of ipsilateral ORs divided by their corresponding contralateral ORs were of a similar magnitude. However, while there was evidence of a higher risk of gliomas in the temporal lobe than elsewhere with several different exposure metrics, there was no such evidence for meningioma. Although ORs for meningioma were generally lower than that for glioma, the otherwise similar patterns of associations of mobile phone use with meningioma and glioma could indicate shared aetiology or shared bias.

Interpretation of these Findings

We have no certain explanation for the overall reduced risk of brain cancer among mobile phone users in this study, although selection bias is almost certainly a contributor. There is some evidence that very high users experienced excess risk of glioma, but that evidence is inconclusive because of possible bias. Further light may be shed on dose–response relations by work now being undertaken with the INTERPHONE data using precise coordinate localization of tumours within the brain in relation to estimates of absorbed RF energy.

The possibility of raised risk in heavy users of mobile phones is an important issue because of their ever-increasing use. Moreover, few subjects in our study had used mobile phones for >12 years; therefore, our results are uninformative with respect to lag periods longer than this.

Consistency with Previous Research

Our results are consistent with most of the research published to date. A large Danish cohort study of mobile telephone subscribers,[8,9] with an average follow-up time of 8.5 years, found no increased risk of brain tumours in subscribers of ≥10 years. The first case–control studies conducted included cases diagnosed in the mid-to-late 1990s and therefore could only address possible risks among short-term mobile phone users.[10,12,18,23] In addition, the highest cumulative call times in these studies were much less than in ours. Generally, these studies reported 'negative' results. In contrast, increased risks of malignant brain tumours at higher levels of accumulated use of analogue and digital mobile phones and cordless desktop phones were reported from a sequence of three case–control studies from the same authors with cases in the last diagnosed as late as 2003.[13–15] However, the methods of these studies have been questioned.[41]

Some of the INTERPHONE centres have published their results for brain tumours[11,16,17,19,22,24,25] and two pooled analyses from Northern European centres have also been published.[20,21] Most cases in these reports are included in the present analyses and constitute 69% of gliomas and 57% of meningiomas. The centre-specific analyses are consistent with our all-centre results.

Much biological research has been done in recent years on possible biological effects of RF fields. This work covers in vitro and in vivo exposure, alone and in combination with other physical or chemical agents, and has found no evidence that RF fields are carcinogenic in laboratory rodents or cause DNA damage in cells in culture.[42] Possible effects of RF fields on other biological endpoints are still being explored.

The possible effects of long-term heavy use of mobile phones on risk of brain tumours require further investigation, given increasing mobile phone use, its extension to children and its penetration worldwide. The problems presented by selection and information bias in this and probably other studies suggest that new studies should, in general, only be done if they can substantially reduce or eliminate selection bias, obtain detailed and high-quality exposure information over the full period of use and offer sufficient statistical power to detect comparatively small effects in people with heavy or long continued exposure. Monitoring of age- and gender-specific incidence rates may also be valuable, particularly if informed by good longitudinal data on mobile phone use by age and sex, and having regard to features such as brain tumour location that may allow more specific inferences about possible mobile phone use effects.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.