Lahkola A, Auvinen A, Raitanen J, Schoemaker MJ, et al. (2007):
This paper gives the results obtained in 5 countries that participated in the Interphone case-control study. The countries were Denmark, Finland, the Southern and Middle parts of Norway, 4 regions of Sweden, and the Thames region of Southeast England. Incident cases of glioma were identified from relevant medical departments. The authors have previously reported their findings on acoustic neuroma (Schoemaker et al., 2005). The age range was 20-69 in the Nordic countries and 18-59 in England. Nordic controls were selected from national population registers with matching for age, sex, and region. In England they were randomly selected from general practitioner lists. Trained interviewers conducted the personal interviews, mainly at hospital or in the home. Some were conducted by phone. The interview was computer-assisted, except for Finland, where a paper questionnaire was used. All exposure within one year of the reference date was ignored, except when calculating the years since first use of mobile phones. For cases, the reference date was the date of diagnosis, and for controls was the date of interview.
Out of 2,530 eligible cases, 1,521 (60%) participated. For controls the numbers were 6,581 eligible, 3,301 participating (50%). The main reasons for nonparticipation of cases were illness or death (18%), refusal (8%) and inability to contact (7%), and for controls were refusal (33%) and inability to contact (15%).
The Odds Ratio (OR) for glioma in relation to regular use of a cell phone was 0.78 (95% CI 0.68, 0.91). There was no evidence of increase in OR with analogue as opposed to digital phones, or with long-term use. Similar results were obtained with men and women. The OR for regular ipsilateral use of cell phones was 1.13 (CI 0.97, 1.31), but for ≥ 10 years, the OR was 1.39 (CI 1.01, 1.92). The OR for contralateral use for ≥ 10 years was 0.98 (CI 0.71, 1.37). The p trend for duration of ipsilateral use was 0.04.
The authors discuss the types of bias that could influence the results - mainly selection bias and recall bias, and the possibility of random error in reporting past use.