October 20, 2011

mobile phones and cancer

This analysis accompanied a roundup which can be viewed here.

Title, Date of Publication and Journal

Use of Mobile Phones and Risk of Brain Tumours: update of a Danish Cohort Study, 20th October 2001, BMJ

Summary

• Very large study with a good use of linked records which ensures that people’s data is accurately matched across registers.
• Study compares subscribers of mobile phones to non-subscribers however the definition of these means that there may be a large number of people with up to 12 years of mobile phone use in the ‘non-subscriber’ group.
• Press releases focuses on the fact that this study looks at longer term follow-up but the longer this cohort is followed post 1996, the more mobile phone users will be included in the ‘non-subscriber’ group. It mentions that there no difference in the risks between those with >10 years use and the non-subscribers but as mentioned above those in the non-subscriber arm could have more than 12 years exposure themselves.
• A protective effect of mobile phone use on ‘all cancers’ was found but this has not been explained or discussed.
• The conclusion that there is no increased risk in the temporal glioma is perhaps overstated because the confidence interval around the risk is from a 4% decreased risk to a 22% increased risk. With further events in the category the result may become statistically significant.
• The press release talks about the ‘largest’ study to date which it may be but with the potential for so much mobile use in the non-subscriber arm the size of the study doesn’t mean anything. 

Objective 

To assess the risk of tumours of the central nervous system in mobile phone subscribers versus non-subscribers.

Study design

A cohort study using linked records of Danish mobile phone subscribers to records in the Danish nationwide CANULI Cancer cohort.

Methods

Since 1968 all people born in Denmark have been given a personal identification number which is used on all national registers and enables links to be made between entries on those registers. In this study the identification number has been used to appropriately link mobile phone subscription records to cancer records and socioeconomic status details.
The statistical methods used present no concerns.
There are some major limitations in the classifications of subscribers vs. non-subscribers. People who had only a corporate mobile phone and not a personal subscription were allocated to the non-subscription group as was any person who took out a subscription post 1996, as data were not available on subscriptions after this date. Given that this cohort was then followed up until 2007 and that the numbers of new subscriptions since 1996 is likely to be very high, there is probably a fair amount of mobile phone use in the non-subscriber group. Subjects could be in the ‘non-subscriber’ group even if they have had a mobile phone for 12 years.

Results

The results are clearly stated and appropriate adjustments have been made for potential confounders in the model. The data are displayed appropriately in the tables and in the text with 95% confidence intervals clearly shown.
However, there are a number of points in the text which describe an increase or decrease in risk when the confidence interval clearly crossing 1 meaning that this result is not statistically significant. Many of the confidence intervals are large due to the small number of events in the data.
Although, this is a very large study in terms of subjects, it is the number of events that is important when determining the power of this type of study to detect a difference. For most of the comparisons the power is probably adequate but in Tables 2 & 3 in particular there are some very small event rates and very large confidence intervals can be seen.
The fact that one of the results is statistically significant causes concern. In the ‘All cancers’ endpoint, there appears to be a protective effect of mobile phone use but there is very little mention of this. One would expect there to be some discussion around this fact because regardless of the numbers that are seen in the data you want your result to have a plausible biological mechanism or be explained by confounding or selection bias.

Strengths

It’s a large study using personal identication numbers to accurately link individual data across a number of databases.

Limitations

The biggest limitation is the classification of subscriber and non-subscriber. The inclusion of people who may have a corporate phone and of people who could have subscribed to a mobile phone any time from 1996 to 2007 in the non-subscriber group means that there is probably a large amount of mobile phone use in the non-subscriber arm.
The authors do acknowledge this and other limitations clearly in the paper but the conclusion and the press release do not fully take this into account.
The press release and the paper discuss how this is giving more data and longer term follow-up since the previous analysis but the longer this cohort is followed up since 1996, the more mobile phone users are likely to be included in the ‘non-subscriber’ group.

Conclusions

Although the authors acknowledge in the conclusion that more studies are needed where the misclassification of exposure is reduced, the conclusion is slightly overstated. It is true that there is no indication of increased risk in this study but the control group presents concerns. Again, it is true that there was no increased risk of temporal glioma shown in this study but the confidence interval for this comparison ranged from a 4% protective effect to a 22% increase so it may be due to a lack of events rather than truly no increased risk.

Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry (PSI) and experienced statisticians in academia and research. A list of contributors, including affiliations, is available here.