This is the next in a series of guest blogs on BRHP. The opinions expressed in it are of Martin Röösli himself. Publication of these opinions in BRHP does not mean that BRHP automatically agrees or endorses these opinions. Publication of this, and other guest blogs, is an attempt to start an open debate and free exchange of opinions on RF and health.
Over the years I had interesting and exciting discussions with Martin about epidemiology and especially the Danish Cohort. Clearly, our ideas about the scientific value of Danish Cohort differ. Those interested in my opinions can find them in several posts here, on BRHP, and in my column in The Washington Times Communities and, finally, in my opinion piece in The Scientist Magazine.
Here, below, is what is Martin’s opinion on the Danish Cohort.
Martin Röösli: Epidemiology, I like!
I love epidemiology because it is about numbers. It is rarely black and white; it is about shades, it is about magnitude and about likelihoods. I love epidemiology because it is complex. It deals with the whole heterogeneity that life offers us. For real life questions such as long term health risks, conclusions can be rarely drawn in a simple way. One has to find and put together all relevant puzzle pieces from different sources in a clever way to obtain the full picture.
That is, why I love the “infamous” Danish Mobile Phone Subscriber Cohort Study. I would not claim that the study is superior to all other mobile phone brain tumour studies. Admittedly, it has weaknesses, adequately discussed in the paper by the authors. Most importantly, however, the study has strengths and offers a different perspective on the relevant question whether long term mobile phone use causes brain tumours.
The strengths of the Danish cohort studies are minimal selection bias and objective exposure data, which cannot be offered by case-control studies to date. The used exposure surrogate – duration of a mobile phone subscription in years – is crude. But is it really useless as often claimed?
Let us first look at an eye-striking finding in the Schüz et al, 2006 paper (JNCI). Male mobile phone subscribers had a decreased risk for smoking related cancers whereas female mobile phone subscribers had an increased risk for these sites and in addition an increased risk for Cervix uteri cancer. Is this due to mobile phone radiation? This is very unlikely but plausibly explained by different social behaviour and gender different smoking patterns of early mobile phone subscribers compared to the general population. This is an illustrative example of confounding for a textbook in epidemiology, which also convincingly demonstrates that the Danish subscriber cohort study is sensitive enough to detect cancer relevant exposure differences between the two study groups. In the most recent paper from Frei et al. (2011) the analysis could be adjusted for educational level and income and these confounding effects were somewhat reduced.
The most important question is thus, whether to be an early mobile phone subscriber is a representative exposure proxy for long term mobile phone use. It has been argued that this is not the case, because more than 300,000 business users could not be identified and were included in the comparison group, and this has downward biased the risk estimates. This is true but as an epidemiologist I have to think in numbers! Let us assume for glioma a true relative risk of 1.5 for an average long term mobile phone user and that 300,000 subscribers are erroneously included in the unexposed Danish population (4,100,000). Let us further be conservative and assume that these business users were heavy users and had actually a higher relative risk of 2.0. Thus the relative reference risk of 1.0 of the non-subscribers would increase to 1.07 [=(4,100,000*1+300,000*2)/4,400,000]. The resulting observed risk in the study would then be 1.40 (=1.5/1.07). As you can see, there is some underestimation but is completely unlikely to explain the actually observed relative risk for glioma of 0.98 for female subscribers and 1.08 for male subscribers. As for alcohol in the blood, the extent of dilution matters. This is an illustrative example on the impact of sensitivity and specificity of epidemiological exposure assessment on the study result; a quite complex issue for students in epidemiology.
Obviously the exposure surrogate used in the Danish Cohort Study is a very valid and discriminant long-term exposure surrogate if glioma risk starts to increase exactly after an induction period of 12 years. Under this assumption mobile phone subscription after 1995 would be irrelevant for the follow-up period until end of 2007. However, if the induction period is much longer, has a large between subject variability, or if the glioma risk is rather related to the amount of use than the duration since start of regular use, the used exposure proxy may not be informative. Nevertheless, indications were found that early subscribers still use the phones more often than other subscribers (Frei, et al, 2011), and that early subscribers were four times more likely to be regular mobile phone users than non-subscribers in an Interphone validation study (Schüz et al, 2007). In addition, early mobile phone users were more likely to having used analogue phones with higher output power than later subscribers. Altogether, it seems very likely that early subscribers received a higher lifetime microwave dose compared to the rest of the population.
In summary, the Danish Subscriber Cohort Study allows to reject some of the hypotheses about mobile phone use and cancers. It cannot solve everything and needs to be considered in the context of all available research, which includes case-control studies and ecological time trends analyses. All these epidemiological study designs have their limitations and strengths and only a balanced and comprehensive evaluation of all aspects will produce adequate conclusions. Most importantly, epidemiology provides data from real populations consisting of numerous genetic variations and susceptibilities and not only for one single genetic variation. That is what eventually matters for the population! It has been demonstrated all over again, if an exposure is relevant for the population’s health, epidemiology will reveal it. More difficult is to detect small risks or risk related to rare events (e.g. histological subtypes or genetic variants). Needless to say that epidemiology cannot predict the future. If induction period of mobile phone effects lasts several decades, the impact on population level is not (yet) noticeable. This uncertainty remains and can be addressed with other research disciplines. But as I said, epidemiology is about likelihoods, it is about quantities, it is complex, it is about real life and it is about the population’s health. That is why I became an epidemiologist.