The study took place in the Thames regions of south east England and four areas to the north (Trent, West Midlands, West Yorkshire, and southern Scotland). The total catchment population (28.4 million) comprised 48.3% of the UK population. All areas followed a common protocol, with identical methods of case ascertainment and data collection with controls randomly sampled from general practitioner lists. The south east slightly differed in its method of control selection and the age range covered.
Cases were ascertained from multiple sources, including hospital departments (neurosurgery, neuro-oncology, neuropathology, neuroradiology, neurology) and cancer registries. Patients aged 18-69 years (northern centres) or 18-59 years (south east) lived in the study areas and had a first diagnosis between 1 December 2000 and 30 June 2003 (northern) or 29 February 2004 (south east) with a glioma (ICD-O-3 (international classification of diseases for oncology)12
topography: C71, morphology: 9380-9411, 9420-9460, 9480, 9505). Data on site, laterality (left, right, central) and grade of tumour (WHO grade high III-IV; low I-II13
) were abstracted from scan and pathology reports.
In the most recent year we have published data for (1992) an estimated 98% of the UK population was registered with a general practitioner.14
Controls were randomly selected from general practitioners' lists by a preset algorithm. In the south east the controls were frequency matched to reflect the age, sex, and geographical distribution of cases. In the northern centres one control per case was individually matched on age, sex, and general practice after the patient with glioma was interviewed. Nonparticipating controls were replaced. Parallel case-control studies of meningioma, acoustic neuroma, and other brain tumours were carried out with identical methods and questionnaires; the controls for these cases were included in the present analyses.
Consultants or general practitioners approached eligible participants personally or by an invitation letter. The study was introduced as an investigation of risk factors for brain tumours, without emphasising mobile phones. With the participant's informed consent, trained interviewers conducted a computer assisted personal interview. For 69 patients with glioma (7%), interviewers conducted proxy interviews, mainly with spouses.
During the interview, if participants reported that they had ever made one or more calls each week on average for a period of six months or longer, they were asked a detailed set of questions on mobile phone use. For such participants, all makes and models of phone were recorded, with a comprehensive repertoire of photographs to prompt recall. For each phone, the interviewer recorded the network operator, start and stop year, and the number and duration of calls made and received. If participants were uncertain about the calendar years or amount of use, a range was reported with the mean value taken for analysis. Additional details were gathered on which side of the head the phone was mostly used (50% or more of the time) and factors influencing emitted power levels to the head, including use of hands-free kits (start and stop dates of use and proportion of time used) and whether the phone was used mainly in an urban or rural area or equally in both.15
In the analysis, we defined regular phone use as use for at least six months in the period more than a year before diagnosis. We defined diagnostic date as the date of diagnostic pathology (n = 935, 97%) or date of the first diagnostic scan (n = 31, 3%) if diagnostic pathology was unavailable. We assessed exposure to mobile phones using the number of years from first regular use of a phone until diagnosis or equivalent reference date for controls, lifetime years of regular use, lifetime cumulative use (hours), and lifetime cumulative number of calls. Long term users (> 10 years) were dichotomised into heavy (3
113 hours) and light (< 113 hours) cumulative hours of use in the period 10 or more years before diagnosis, with the cut off point based on the median hours of use among control participants. We used data collated by Interphone11
to classify phones as either analogue or digital, based on make and model of phone, year of use, and network operator capabilities during that period.
The exposure period for people with glioma was calculated up to a year before the date of diagnosis. An equivalent reference date was required for control participants that allowed for the increase in mobile phone use over the study period and as controls tended to be interviewed after the patients with glioma. For each area (south east, northern), we constructed case strata by single calendar year of interview and single year interval between diagnosis and interview (“interview lag time”). Control participants interviewed in each calendar year were randomly allocated to strata of interview lag time, proportionally to the distribution of the cases in the same calendar year, to obtain a similar distribution of lag time as in the cases. We then calculated the reference dates for controls by subtracting the mean interview lag time in cases in that stratum from the interview dates of the controls. Exposure indices were calculated up to a year before this reference date for controls (that is, a one year latency time was used). Additional analyses were carried out with a five year latency time.
For statistical analysis we used unconditional logistic regression (StataCorp, College Station, TX) adjusted for nine regions (five regions within the south east and the four northern regions), age at reference date (five year categories), sex, deprivation (Townsend score16
), and combinations of interview year and lag time to account for the fact that controls were, on average, interviewed later in the study period than patients with glioma. We derived odds ratios for cumulative use with and without modification for reported use of headsets or hands-free sets in a vehicle, or both,7,17
with and without proxy case interviews, and separately for high and low grade tumours and urban versus rural use. We also performed a conditional logistic regression analysis on the matched northern case-control dataset.
We used two methods to assess the risk of a tumour ipsilateral or contralateral to side of phone use. Firstly, we took two groups of patients with right and left sided tumours17
and randomly assigned controls to each group and considered them to have a tumour on that side (50% left, 50% right) for the analysis. The odds ratio for risk of an ipsilateral tumour was based on the results of a logistic regression analysis where ipsilateral phone use was use on the same side of the head as the tumour for cases or the assigned side for controls. We adjusted the analysis for the side of the tumour as well as the variables adjusted for in the main analysis. If the phone was used on the opposite side to the side of tumour/allocated side the participant was classified as unexposed. Those who reported using the phone on both sides of the head were considered exposed on both the left and right sides.6,17
As the allocation of controls was based on a random assignment, the logistic regression analysis was repeated 500 times; the results for the analysis that gave the median odds ratio for ipsilateral regular phone use are presented. A similar analysis of contralateral use (use on the opposite side of the head to the tumour or allocated side) was performed. The second method calculates a relative risk for laterality of reported side of use in relation to tumour laterality but only in cases.4