Aydin, D.; Feychting, M.; Schuz, J.; Andersen, T. V.; Poulsen, A. H.; Prochazka, M.; Klaeboe, L.; Kuehni, C. E.; Tynes, T., and Roosli, M. Impact of random and systematic recall errors and selection bias in case-control studies on mobile phone use and brain tumors in adolescents (CEFALO study). Bioelectromagnetics. 2011 Feb 3.
In case-control studies of mobile phones and brain tumors, the assessment of exposure is primarily based on self-reported past mobile phone use. Self-reported data are prone to error termed “recall error”. Recall error can occur at random, in which case the average of the errors is zero. It can occur in a systematic way leading to under- or overestimation of the exposure. If the direction and the extent of the recall error are different for cases and controls, it is termed “differential”. Another potential source of error is selection bias that occurs if the probability to participate in the study is related to the health status and to the exposure.
The objective of this study was to investigate “the effects of random and systematic recall error and of selection bias in case–control studies of mobile phone use and brain tumor risk in children and adolescents.”
Sensitivity analyses using Monte Carlo computer simulations were performed based on data from CEFALO, an international case-control study of the association between mobile phone use and brain tumor risk in 7-19-year old children and adolescents. Self-reports on mobile phone use were compared with data from mobile phone network operators.
Results, Interpretation and Conclusion
Participation rates were 83.2% for cases and 71.1% for controls. Both cases and controls overestimated the cumulative number of calls and the cumulative duration of calls. An unexpected finding was that controls overestimated their mobile phone exposure to a greater extent than cases. The opposite pattern is more common in epidemiological studies. Because cases seek an explanation for their disease, they may think more intensely about past exposures and thus tend to overestimate their exposures. In the CEFALO study, cases overestimated the number of calls on average by 9% and controls – by 34%; the difference was not statistically significant. Overestimation of the duration of calls was 52% and 163% for cases and controls, respectively; the difference was of a borderline statistical significance (p=0.07). Computer simulations have shown that, in a variety of possible scenarios that may occur in a case-control study, the combined impact of recall error and selection bias on odd ratios is complex. The results suggest that error and bias are unlikely to produce a false association in the CEFALO study. Also, if a detectable association exists, it is likely to show up in the analyses.