To the Editor:
Lönn et al1 recently presented results on mobile phone use and the risk of acoustic neuroma. However, statistical calculations in this paper need to be clarified.
At a recent conference on the question of whether radiofrequency fields increase the risk of cancer (http://www.cost281.org), power in statistics was raised as an issue to be considered in health risk assessment. The alpha level controls the probability of falsely rejecting the null hypothesis, and the power is 1 minus the probability of missing a real effect.
Lönn et al showed no effect from radiofrequency mobile phone exposure on acoustic neuromas for all 148 cases as compared with 604 controls, but they reported elevated risks for long-term mobile phone use ipsilateral to the location in the head of acoustic neuroma. Limiting the analysis to ipsilateral exposure, for regular use of >10 years’ duration, the odds ratio (OR) was 3.1 (95% confidence interval = 1.2–8.4) and for time since first regular use, the OR was 3.9 (1.6–9.5). However, these results are based on small numbers of long-term users (Table 1).
Furthermore, these reported results were 2 of 37 post hoc subgroup comparisons. Lönn and colleagues do not discuss the power of their statistical calculations, and the statistical sensitivity of the results could not be evaluated. There are pros and cons concerning utilization of multiple-test corrections in epidemiologic studies.2 However, without an appropriate correction3 for post hoc multiple comparisons in subgroups, the door for chance findings is wide open and the assessment of statistical power is compromised.
Lönn et al have performed 37 comparisons at alpha level 0.05. Significant results expected by chance are 0.05* 37 = 1.85. As predicted, there were indeed 2 significant results observed. We calculate the power to detect a relative risk of 4.0 to be 39% for duration of regular ipsilateral use ≥10 years and 53% for time since first ipsilateral use ≥10 years (Table 1). The Interphone criterion4 set out in 1999 was a power of 80% or more to detect a RR of 1.5 of disease for mobile phone exposure. This criterion is not met for the multiple comparisons.
We respectfully request clarification and validation of the statistical methodology of Lönn et al. Published papers are available on other aspects of the methodology of the Interphone study.4,7,8 We suggest that a peer-reviewed publication of the statistical protocol (including the protocol for multicomparisons in subgroups) for individual and the Meta-Interphone study results could clarify and guide future presentation and interpretation of results.
Sheila A. Johnston
Independent Neuroscience Consultant, London, UK, firstname.lastname@example.org
GSF-Forschungszentrum fur Umwelt und Gesundheit, Oberschleissheim, Germany
1. Lönn S, Ahlbom A, Hall P, et al. Mobile phone use and the risk of acoustic neuroma. Epidemiology
2. Aickin M. Other method for adjustment of multiple testing exist. BMJ
4. Cardis E, Kilkenny M. International case–control study of adult brain, head and neck tumours: results of the feasibility study. Radiation Protection Dosimetry
5. Lehmann EL. Testing Statistical Hypotheses
, 4th printing. New York: John Wiley & Sons, Inc; 1966.
6. Scherb H. Determination of uniformly most powerful tests in discrete sample spaces. Metrika
7. Berg G, Schuz J, Samkange-Zeeb F, et al. Assessment of radiofrequency exposure from cellular telephone daily use in an epidemiological study: German validation study of the international case-control study of cancers of the brain—INTERPHONE-Study. J Expo Anal Environ Epidemiol
8. Lönn S, Forssen U, Vecchia P, et al. Output power levels from mobile phones in different geographical areas; implications for exposure assessment. Occup Environ Med