To the Editor:

Myung et al¹ claimed that according to their meta-analysis there is “possible evidence linking mobile phone use to an increased risk of tumors.” We have several methodologic concerns. Myung et al applied the Newcastle-Ottawa Scale (NOS) to assess the quality of published case-control studies.² The validity of this checklist approach is at best unknown (as the inventors of the NOS state themselves²) or doubtful (as we believe). Why? First, blinding of the interviewers against case-control status is a quality item of NOS. However, blinding of interviewers against case-control status is often wishful thinking as interviewers can easily unblind themselves by simply seeing patients, by noticing mental problems as among brain cancer cases, or by the need of patients to talk about their serious disease before or during interview. Therefore, it is important to conduct highly standardized interviews by trained and certified interviewers who are regularly monitored by audiotaping the interviews.

A mere splitting of studies into blinded and unblinded ones does not adequately reflect these issues. Myung et al's subgroup findings among studies (unblinded v blinded exposure assessment) depend on the assumption that the blinding of the interviewers against case-control status is important, was reported, and was actually realized in the studies. If we ignored the highly arbitrary distinction between unblinded and blinded studies, we would not observe any association between mobile phone use and risk of cancers (Myung et al's overall odds ratio, 0.98; 95% CI, 0.89 to 1.07). Second, another quality item of the NOS is “identical nonresponse rates among cases and controls.” However, this criterion trivializes the concept of valid selection as response rates have to be identical within subgroups of cases and controls, that is, exposed and unexposed subjects, in order to minimize nonresponse bias, as can be mathematically shown.³ Third, the lack of description of any of the NOS items in the space-limited publications of case-control studies is equated with a lower quality of the studies themselves according to NOS, mixing up study quality with the quality of study reports. However, not reporting certain design issues does not mean that they were actually not performed (eg, Devereaux et al⁴). Fourth, Myung et al arbitrarily used a cutoff of 7 points to distinguish between low bias and high bias studies. If they had used a cutoff of 6 points, an additional nine studies would have been labeled as low bias. How sensitive are the results of their meta-analysis in relation to the choice of this cutoff? In summary, we challenge the validity of Myung et al's meta-analysis within subgroups of NOS quality scores.

Myung et al performed fixed and random-effect models for combing odds ratios from single studies. Besides using the arbitrary cut point of 50% of the I² index, we question the use of fixed-effects models. Using a fixed-effect model would mean that one assumes one constant true odds ratio for each and every study in the respective meta-analysis. In view of the many different cancers, very different exposure assessments, or study settings this notion of one single odds ratio is somewhat implausible. Another statistical point concerns the method of subgroup analyses (blinded v unblinded studies) that were conducted. It is widely accepted, at least in randomized trials, that such analyses should not be performed by calculating effects in subgroups separately (as Myung et al did), but checking the interaction of the subgroup factor with the exposure first.⁵

For the interest of the reader, we would like to comment on our own results related to mobile phone use and risk of uveal melanoma. After the publication of our first case-control study,⁶ we ran a large case-control study on uveal melanoma focusing on mobile phone use and used the same detailed exposure assessment as the Interphone study used. This study has not been included in the meta-analysis by Myung et al as the publication date was January 2009. We could not corroborate our previous results that showed an increased risk of uveal melanoma among regular mobile phone users.⁷ Our probabilistic multiple error sensitivity analyses to evaluate the potential of exposure misclassification bias and selection bias did not explain the null result.

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The author(s) indicated no potential conflicts of interest.