Brain tumour risk in relation to mobile
telephone use: results of the INTERPHONE
international case–control study
The INTERPHONE Study Group*
5 Corresponding author. Elisabeth Cardis; CREAL, Doctor Aiguader 88, 08003 Barcelona, Spain. E-mail: ecardis@creal.cat
*List of members of this study group is available in the Appendix.
Accepted 8 March 2010
Background The rapid increase in mobile telephone use has generated concern
about possible health risks related to radiofrequency electromagnet-
10 ic fields from this technology.
Methods An interview-based case–control study with 2708 glioma and 2409
meningioma cases and matched controls was conducted in 13 coun-
tries using a common protocol.
Results A reduced odds ratio (OR) related to ever having been a regular
15 mobile phone user was seen for glioma [OR 0.81; 95% confidence
interval (CI) 0.70–0.94] and meningioma (OR 0.79; 95%
CI 0.68–0.91), possibly reflecting participation bias or other meth-
odological limitations. No elevated OR was observed 510 years
after first phone use (glioma: OR 0.98; 95% CI 0.76–1.26; meningi-
20 oma: OR 0.83; 95% CI 0.61–1.14). ORs were <1.0 for all deciles of
lifetime number of phone calls and nine deciles of cumulative call
time. In the 10th decile of recalled cumulative call time, 51640 h,
the OR was 1.40 (95% CI 1.03–1.89) for glioma, and 1.15 (95% CI
0.81–1.62) for meningioma; but there are implausible values of re-
25 ported use in this group. ORs for glioma tended to be greater in the
temporal lobe than in other lobes of the brain, but the CIs around
the lobe-specific estimates were wide. ORs for glioma tended to be
greater in subjects who reported usual phone use on the same side
of the head as their tumour than on the opposite side.
30 Conclusions Overall, no increase in risk of glioma or meningioma was observed
with use of mobile phones. There were suggestions of an increased
risk of glioma at the highest exposure levels, but biases and error
prevent a causal interpretation. The possible effects of long-term
heavy use of mobile phones require further investigation.
35 Keywords Brain tumours, mobile phones, radiofrequency fields
Introduction
Mobile phone use has increased dramatically in many
countries since its introduction in the early-to-mid
40 1980s. The expanding use of this technology has
been accompanied by concerns about health and
safety. In the late 1990s, several expert groups critic-
ally reviewed the evidence on health effects of
low-level exposure to radiofrequency (RF)
Published by Oxford University Press on behalf of the International Epidemiological Association
� The Author 2010; all rights reserved.
International Journal of Epidemiology 2010;1–20
doi:10.1093/ije/dyq079
1
electromagnetic fields, and recommended research
into the possible adverse health effects of mobile tel-
ephony.1–4 As a result, the International Agency for
Research on Cancer (IARC) coordinated a feasibility
5 study in 1998 and 1999, which concluded that an
international study of the relationship between
mobile phone use and brain tumour risk would be
feasible and informative.5,6
INTERPHONE was therefore initiated as an interna-
10 tional set of case–control studies focussing on four
types of tumours in tissues that most absorb RF
energy emitted by mobile phones: tumours of the
brain (glioma and meningioma), acoustic nerve
(schwannoma) and parotid gland. The objective was
15 to determine whether mobile phone use increases the
risk of these tumours and, specifically, whether RF
energy emitted by mobile phones is tumourigenic.
This article presents the results of analyses of brain
tumour risk in relation to mobile phone use in all
20 INTERPHONE study centres combined. Analyses of
brain tumours in relation to mobile phone use have
been reported from a number of cohort7–9 and case–
control studies, including several of the national com-
ponents of INTERPHONE.10–25 No studies, however,
25 have included as many exposed cases, particularly
long-term and heavy users of mobile phones, as this
study.
Methods
Study design
30 The INTERPHONE study is an international, largely
population-based case–control study. The common
core study protocol is described in detail elsewhere.5,26
Sixteen study centres from 13 countries (Australia,
Canada, Denmark, Finland, France, Germany, Israel,
35 Italy, Japan, New Zealand, Norway, Sweden and the
UK) were included. To maximize statistical power, the
INTERPHONE study focussed on tumours in younger
people, 30–59 years of age, as they were expected to
have had the highest prevalence of mobile phone use
40 in the previous 5–10 years, and on regions likely to
have the longest and highest use of mobile phones
(mainly large urban areas).
Eligible cases were all patients with a glioma or
meningioma of the brain diagnosed in the study re-
45 gions during study periods of 2–4 years between 2000
and 2004. Cases were ascertained from all neuro-
logical and neurosurgical facilities in the study re-
gions (except in Paris and Tokyo where some did
not agree to participate), and in some centres also
50 from cancer registries. All diagnoses were histologi-
cally confirmed or based on unequivocal diagnostic
imaging. To facilitate interviews soon after diagnosis,
cases were ascertained actively within treatment facil-
ities wherever possible. Completeness of ascertain-
55 ment was checked through secondary sources, such
as population- or hospital-based cancer registries,
medical archives and hospital discharge or billing
files.26
One control was selected for each case from a locally
60appropriate population-based sampling frame, except
in Germany where two controls were chosen. The
sampling procedure involved individual matching in
seven centres (Canada – Ottawa, Canada – Vancouver,
France, Israel, Japan, New Zealand and UK North)
65and frequency matching elsewhere. The matching
variables were age (within 5 years), sex and region
of residence within each study centre. In Israel, the
subjects were also matched on ethnic origin. Where
stratified matching had been used, individual match-
70ing was conducted post hoc, with cases being assigned
one control (two in Germany), interviewed as close as
possible in time to the case, from those who fitted the
matching criteria.
Detailed information on past mobile phone use was
75collected during face-to-face interviews with the study
subject, or a proxy, if the subject had ever been a
regular user of a mobile phone (had an average
of at least one call per week for a period of
56 months).26 A proxy was sought when the study
80subject had died or was too ill to be interviewed. The
interviews were conducted by a trained interviewer
using a computer-assisted questionnaire, except in
Finland where a paper version was used. The ques-
tionnaire also included sections on socio-demographic
85factors, occupational exposure to electromagnetic
fields and ionizing radiation, medical history (sub-
ject’s and family), medical ionizing and non-ionizing
radiation exposure and smoking. For cases, informa-
tion was also collected on the anatomic location
90and histological type of the tumours. Where possible,
location data were obtained from magnetic resonance
imaging (MRI) reports or images; they were otherwise
obtained from surgical records or clinical notes.
Details of the specific source for each case were not
95recorded in the INTERPHONE database. Those collect-
ing the data did not know the reported mobile phone
use of individual cases.
Statistical methods
Data from countries with multiple centres were com-
100bined for the analyses, except in the UK where the
UK South and UK North, each with large numbers of
subjects, were kept separate. The word ‘centre’ in the
remainder of this article is used to refer to the 14
analytic entities (12 countries, UK North and UK
105South). All analyses were carried out for all centres
combined and for each centre separately. Formal tests
for heterogeneity of risk across centres were con-
ducted by allowing for an interaction between
centre and the exposure variables.
110The analyses presented here focus on past mobile
phone use as reported by or for the study subjects.
The main analyses were based on conditional logistic
regression for matched sets.27 The date of diagnosis of
the case was used as the reference date for cases and
2 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
controls in each matched set. For the main analyses,
the reference category for odds ratios (ORs) was the
set of subjects who reported that they had never been
regular users. Exposure variables included ever having
5 been a regular user (as defined above), time (years)
since first regular use, cumulative number of calls and
cumulative duration of calls. To allow for a latency
period of 1 year, the year before the reference date
was included in the reference category for time since
10 first regular use and all other exposure variables were
censored at 1 year before the reference date.
Cumulative number and duration of calls were ana-
lysed as categorical variables, based on deciles of the
distribution of these variables among all controls who
15 were regular users, including those matched to pa-
tients with an acoustic neuroma or a parotid gland
tumour, so that the same cut-off points are used in
all analyses.26 Cumulative use excluded use of mobile
phones with hands-free devices: for all time periods
20 for which the subject reported the use of hands-free
devices the amount of use was reduced by 100, 75, 50
or 25% depending on whether hands-free devices
were used always or almost always, more than half,
about half or less than half of the time, respectively.
25 For ease of presentation, some results are shown for
the following grouping of deciles: 1, 2–5, 6–7, 8–9 and
10, chosen post hoc to reflect the spread of the highly
skewed distribution of these variables. For conveni-
ence, we will systematically use the term ‘regular
30 user’ in text and tables to refer to ever having been
a regular user.
The reference group for these analyses, never regular
users, included people who had some mobile phone
use but never as much as one call a week on average
35 for 56 months (�32% of meningioma and 26% of
glioma cases, and 30% of meningioma and 26%
of glioma controls) and people who had never used
a mobile phone (�11% of meningioma and 9% of
glioma cases, and 8% of meningioma and 6% of
40 glioma controls). These percentages are approximate
because never use and never regular use were defined
at different dates; the reference date and the date of
interview, respectively. We are not able to determine
whether inclusion of subjects with some occasional
45 mobile phone use in the reference group had a ma-
terial effect on our results because this difference in
definition dates prevented us from distinguishing par-
ticipants with only occasional use from those with no
use at all at their reference dates. Moreover, because
50 numbers of never users at the date of interview were
small, particularly in certain age- and gender-specific
sub-groups (such as young men), never users were
not a suitable reference group for this analysis.
All analyses were adjusted for educational level; an
55 a priori decision had been made to adjust for it as a
surrogate for socio-economic status (SES). Creation of
consistent educational levels across the 13 countries is
described elsewhere.26 In practice, this adjustment
had little impact on OR estimates, changing their
60values by 42% in most instances and in all cases
by <5%. Using a 10% change-in-estimate criterion
for confounding,28 no other covariate among those
collected (see list above) was included in the main
analyses. The interval between the start date of inter-
65views in the study centre and the date of each sub-
ject’s interview was modelled by fitting the
interaction of this interval with study centre.
A common protocol was applied to impute missing
data for cases and controls.26 The study questionnaire
70allowed ranges to be given instead of exact answers to
a number of questions, including number and dur-
ation of calls and dates of start and end of mobile
phone use; in such instances, the main analyses in
this article were based on the mid-point of the re-
75ported range.
Because absorption of RF energy from mobile
phones is highly localized,29 three different types of
analyses were conducted to account for tumour loca-
tion. First, analyses were conducted by the anatomical
80lobe of the brain in which the tumour occurred.
Secondly, separate analyses were conducted for the
subjects who reported using the mobile phone
mainly on one or the other side of the head, and
the preferred side was compared with the side on
85which the tumour occurred. For this, each control
was assigned the location of the tumour of his or
her matched case. Exposure was considered to be ip-
silateral if the phone was used predominantly on the
same side as the tumour or on both sides of the head,
90and contralateral if used mainly on the side of the
head opposite to the tumour. Laterality was not as-
signed if the tumour was reported to be centrally
located (i.e. it crossed the midline of the brain);
these cases were excluded from laterality analyses.
95Thirdly, case–case analyses were carried out on the
concordance between tumour side and laterality of
phone use using the method proposed by Inskip and
collaborators.18
Sensitivity analyses
100To complement these primary analyses, we undertook
sensitivity analyses to try to determine whether any of
the following might have biased the results: (i) any
study centre; (ii) required mention of mobile phones
in the introductory letter to subjects in some centres;
105(iii) centres with a hospital-based design or particu-
larly low participation rates; (iv) respondents whose
interviews were considered by the interviewer to be of
poor quality; (v) subjects for whom proxies provided
the responses or a telephone interview was given; (vi)
110interviewers who had little experience or who had
unbalanced case to control workloads; (vii) difference
between the interview dates of cases and their
matched controls (on average, each control was inter-
viewed 3 months later than its matched case26 and
115mobile phone use was increasing rapidly during the
study period); (viii) subject’s choice between two
ways of responding to call time questions (time per
INTERPHONE INTERNATIONAL CASE–CONTROL STUDY 3
day, week or month, or time per call); (ix) subjects
who reported implausibly high amounts of mobile
phone use (by excluding them or by retaining them
and truncating their use at a specific lower value
5 when they reported a higher one); (x) method of
calculating accumulated call time; (xi) use of match-
ing and conditional analysis; (xii) the choice of a par-
ticular imputation algorithm; and (xiii) adjustment
for possible confounders.
10 Results
During the study period, 3115 meningioma and 4301
glioma cases, and 14 354 potential controls were iden-
tified. Interviews were completed with 2425 meningi-
oma cases (78%; range 56–92%), 2765 glioma cases
15 (64% participation; range by centre 36–92%) and
7658 controls (53%; range 42–74%; Appendix 1,
Table 1, Supplementary data are available at IJE
online). The most common reasons for non-
participation were subject refusal (11% of men-
20 ingiomas, 11% of glioma cases and 30% of controls);
illness, death or physician refusal (4% of meningi-
omas, 20% of gliomas and 1% of controls); and inabil-
ity to contact the subject (7% of meningiomas, 5% of
gliomas and 15% of controls).
25 The main analyses, based on matched sets only,
included 2409 meningioma cases with 2662 matched
controls and 2708 glioma cases with 2972 matched
controls. Among meningioma cases, 24% were men
and 76% women; among glioma cases, 60% were
30 men and 40% women (Table 1). Although the
median age of meningioma cases was only slightly
older than that of glioma cases (51 and 49 years, re-
spectively), 23% of glioma cases were diagnosed
before the age of 40, compared with 13% of meningi-
35 oma cases.
The proportion of proxy interviews was higher in
glioma cases (13%) than in controls (1%) or meningi-
oma cases (2%). Whereas 17% of glioma cases who
were regular users had imputations because of miss-
40 ing information in at least one of their mobile
phone-related variables, the corresponding fractions
were 9% among regular user meningioma cases and
8% among regular user controls. The proportion of
subjects who answered questions about mobile
45 phone use by giving a range of values rather than a
particular amount of use (for any of the use dimen-
sions) was very similar (�42%) for meningioma cases,
glioma cases and controls.
The prevalence of regular mobile phone use 1 year
50 before the reference date was 52% for meningioma
cases (ranging from 34 to 73% across study centres)
and 56% in matched controls (35–78%). It was higher
for glioma cases (62% overall, range: 42–80%) and
controls (64% overall, range: 45–84%), reflecting the
55 different sex distributions of these tumours.
The majority of subjects in the study were not heavy
mobile phone users; the median lifetime cumulative
call time among meningioma controls using mobile
phones was �75 h, with a median call time of
60�2 h/month and a median lifetime number of calls
about 1500. Corresponding values for glioma controls
were �100 h lifetime, 2.5 h/month and about 2000
calls. The distributions of time since start of mobile
phone use and cumulative call time were highly
65skewed, with few long-term and heavy users, and
varied across study centres and by age and sex (not
shown).
Relation between mobile phone use and
risk of brain tumours
70Meningioma
A reduced OR of meningioma was found for regular
mobile phone use in the past 51 year, OR 0.79 [95%
confidence interval (CI) 0.68–0.91; Table 2]. There
was some suggestion of heterogeneity of risk across
75centres (P¼ 0.08) with ORs <1.0 except in Canada,
Denmark, Germany and Italy (data not shown). ORs
were <1.0 for regular users in all categories of time
since start of use and cumulative number of calls.
Analyses by cumulative call time showed ORs <1.0
80in the first nine deciles and an OR of 1.15 (95% CI
0.81–1.62) in the highest decile. Analyses of cumula-
tive call time among recent-, medium- and long-term
users (Table 3) showed no indication of excess risk
except in the highest call time category in those who
85started phone use 1–4 years before the reference date:
OR 4.80 (95% CI 1.49–15.4).
There was no appreciable effect modification by age
or sex on any of these results (data not shown).
In analyses by anatomical location of the meningi-
90oma, the OR for temporal lobe tumours with regular
use was 0.55 (95% CI 0.36–0.82) and the ORs were
<1.0 in all categories of time since start of use, cu-
mulative call time and cumulative number of calls.
ORs for other lobes were also mostly <1.0 (Table 4).
95The OR for mainly ipsilateral use among regular
users was 0.86 (95% CI 0.69–1.08), and that for
contralateral use was 0.59 (95% CI 0.46–0.76;
Table 5). The ORs by time since start of use were
<1.0 in all categories of ipsilateral and contralateral
100use. When analysing by any of the exposure metrics
in Table 5, the ratios of the ORs for ipsilateral use
to contralateral use were always one or above one
regardless of level of exposure and they were highest
(�2 or 3) for the two highest categories of cumulative
105call time and the second highest category of cumula-
tive number of calls. A case–case analysis, based on
Inskip’s method, showed an OR of 1.07 (95% CI 1.00–
1.16; Appendix 1, Table 2, Supplementary data are
available at IJE online) for ipsilateral use.
110The OR for those who reported regular use of only
an analogue phone was 0.81 (95% CI 0.65–1.03) and
for only a digital phone it was 0.79 (95% CI 0.68–
0.92). Focussing on the highest decile of cumulativ