#600: Powerwatch commentary on the Danish cellphone industry study
The weblog version of this message is at:
#600: Powerwatch commentary on the Danish cellphone industry study
While most of the world’s media blindly follow the bidding of the cellphone industry by mis-reporting the Danish cellphone study. Following is a commentary put up on the Powerwatch site. Please connect with the link:
http://www.powerwatch.org.uk/news/20061206_danish_phones_cancer.asp
Summary:
A large new study from Denmark has found no increase in cancer risk amongst 451,679 users from 1982-1995. Instead, for select cancers, it has found statistically significant decreases in cancer.
The conclusions and recommendations taken from the paper abstract are as follows:
“In conclusion, we found no increased risk of brain tumors, acoustic neuromas, salivary gland tumors, eye tumors, leukemias or overall cancers in this large, nationwide cohort study of cellular telephone subscribers in Denmark. ”
We have some serious issues with a number of aspects of this paper, and believe that due to the setup of the research itself it could not possibly have found an effect even if one was present. On top of this there are certain findings that we find suspiciously likely to be false results.
Firstly, the most obvious problem is how the “regular mobile phone users” are selected. There is no problem in the logic they have used for the 450k people in their study, and we would agree that these people are likely to be regular phone users (that is, if we accept that the level of “regular” should be determined by 1 phone call per week for 6 months). However, they have compared it to the national cancer statistics (without the 450k) as a base, which to be truly accurate would imply that they are comparing it to an effectively “unexposed” group if we are to judge the relative increase in risk.
Looking at the beginning of their “Study and Methods” section, we find that they originally had 723,421 records, taken from all cellular telephone subscriptions during the period 1982-1995. Of these, 102,819 records were excluded because of duplication (change of address for example), leaving 620,602 people in their sample size. They then excluded a further 200,507 subscriptions under the basis that they were “corporate” subscriptions and the actual person using the phone could not be identified. So out of the original recordset, over 30% (which by their own admission may be the “most active users”) were excluded. They also ignored all mobile phone users that started their contracts after 1995 (most mobile phone users in the country will now fit into this category) and ignored all non-contract phone usage (pay as you go).
As a result, the reference group will include at least as many (and likely considerably more unless there really is less then 16% of the Danish population that use a mobile phone) mobile phone users as the supposedly exposed group. There will of course be non-users as well, but having this many mobile phone users in the reference group will push the expected OR of such a sample size considerably close to 1.0 (no increased risk) anyway – it is little surprise that that is what they found.
Secondly, they didn’t actually find that the phones were having no effect, they found a protective effect for certain types of cancer. Out of the 18 different cancer types examined, it was found that prolonged mobile phone usage actually protected the user from 7 of the categories, and that this protective effect was statistically significant. More interestingly still, these protective effects were found in cancers that would not normally be associated with phone usage (lung cancer, liver cancer, pancreatic cancer for example). Without something to give some indication (either scientifically or logically) that this should be happening, such a large protective effect should be triggering some alarm bells – for example, maybe the “expected” values of cancer rate were too high across the board.
To analyse what the possible outcome of this would be (if true), we adjusted the ORs to get the smoking related cancers as close to 1.0 as possible, and looked at the outcome. To do this we multiplied the OR and the range bounds by 1.27 (not very scientific, as the aim was merely to look for a trend), so that the average of the ORs for lung, pharynx, oesophagus, liver and pancreas were around 1.0 (in fact 1.00076). When we did this we found that a number of cancers became statistically significantly increased, the most prominent of which were brain cancer, leukaemia, bladder cancer, prostate cancer and testicular cancer. Of course, this is entirely non-scientific with regards to accuracy, but it is interesting to see that the increases are in the cancers typically associated with mobile phones (brain, prostate and testicular) – if the latter two are theoretically increased because of the phone sitting on your belt, it is interesting to see that bladder cancer would also make sense as an increase.
This isn’t to say that these increases are real, but it strikes us as an extraordinary coincidence that the most commonly accused cancers were the ones with most significant increases after our adjustments – it certainly adds credance to the argument that there may have been something at fault with their reference cancer risk levels.
All in all this appears to be an extremely flawed piece of research, that either by design, incompetence or unavailability of data seemed destined to find no effect from the very beginning.
Leave a reply →