——> Follow us on TWITTER

The Interphone study and limits of the case-control design

Studies that relate common exposures with life-threatening diseases tend to be received with great interest from public media and shape the public’s conception of epidemiology. Interphone is one of these studies and its first part of the combined Interphone study on mobile phone use and brain cancer was recently published in The Journal 1. The results are difficult to interpret and apparently even the authors disagree among themselves. Most of the findings show reassuring results, in fact protective ORs below one but these are disregarded by the authors. For intensive use over longer time periods there are indications of an association between the exposure and the studied brain cancers (glioma and meningioma), supported by the location of the brain tumors in relation to where the phone was most frequently held. Some of the authors believe this is due to bias, especially information bias. The vast majority of readers probably just look at the actual data; most associations close to the null or even below and think the issue is closed and funding agencies may well have this idea too. Further, the study will probably not impact on consumers’ enthusiasm for mobile phones. Now we have more than 4 billion users worldwide. It is one of the most successful new technologies spreading to all segments and the population and to all countries. If the non-ionizing radiation the phones produces cause health problems with long delays we will be in for a serious surprise in the years to come.

It has been well documented that talking in the phone while driving increases the risk of accidents but here the exposure of interest is not the radio frequency (RF) exposure but the distraction caused by the conversation. Some readers, and authors, of the Interphone study will claim we cannot conclude anything from the IJE publication. Others think that this study, together with other evidence, exclude a strong excess risk of the studied brain cancers of say a magnitude of 2, which is probably within the limit of what the users will accept for a rare disease like brain cancer, but why this fundamental uncertainty? Have we crossed our limit for what epidemiologists can do with the tools they have?

First of all we have to agree that we cannot stretch our conclusions beyond what has been studied; use of mobile phones in up to 15 years with the frequency of use and types of phones as were used at the time of the study. It is important to pay attention to Saracci’s and Sament’s commentary in the same issue of IJE 2; they remind us that had we conducted studies on smoking and radiation within the same time period since start of use as for the Interphone studies, these studies would have been negative, not showing the causal links we now know exist.

Secondly we have to recognize that choosing the right design is of crucial importance. A case control study was chosen, usually a quick and inexpensive way of harvesting the population experience, and the design of choice if certain requirements are fulfilled. The most important of these requirements are:

  1. There is a strong reason to focus on one particular disease (or a few diseases) related to this exposure.
  2. It is possible to reconstruct past exposure in a valid way for both cases and controls.
  3. Cases can be identified and their source population provides a workable sampling frame for selecting controls in the population that gave rise to the cases.
  4. It is possible to persuade cases and controls to take part in the study, or at least to do so without taking their exposure history into consideration.
  5. And finally when these studies are based on recall the hypothesis should not be known to the participants, at least not when they are recruited and being interviewed.

For Interphone most of these requirements are not fulfilled. Especially the ability to reconstruct a very common exposure back in time is very difficult to do with sufficient accuracy, as we know from studies on dietary factors, intake of medicine etc. Furthermore it is also getting to be difficult to get more than 50-65% of those approached to take part in epidemiologic research and case-control studies are especially vulnerable to selection bias, especially when a well published hypothesis is being examined. When studies are based on recall we at least want to have symmetry in recall between cases and control which will be difficult to obtain when cases have brain cancer. One could have feared that cases would exaggerate their exposure and exposed case would be more motivated to take part in the study than exposed control and both sources of bias would inflate our risk estimates. That is not what we see. Well conducted detailed method studies within Interphone do show that recall of mobile phone use is poor and differential, especially for duration of calls and participation in the study is strongly correlated with use of mobile phones for both cases and controls. Had these methodological papers been done as part of the piloting, as they should have been, a case-control design would not have been the design of choice. Furthermore a case-control study has a closed exposure time period under study. A cohort study could add additional exposure time for as long as the cohort could be maintained. A cohort design would further not limit the outcome to specific endpoints for which we have no strong reason to prefer rather than other diseases. If you are a student of a particular exposure you should be interested in all possible endpoint related to this exposure. You have an ‘exposure looking for a disease’ situation; not a disease ‘looking for a cause’. We had not seen any increase in brain cancers related to the timing of mobile phone use. The increases that have been reported can probably better be explained by improvements in diagnostic tools. There are no strong indications in the descriptive epidemiologic data yet to indicate that all research should be focused on brain cancer.

The Interphone study was not cheap nor was it ‘quick’ and it did cross the limits for what we can do by using a case control method in this case. Money and time could probably have been better spent on setting up large scale cohort studies that could have provided better scientifically supported results with time. Such a cohort study could have been supported by a series of ecological studies in countries with good public available data on phone use and diseases.

What could happen now is that the Interphone study may have dried out funding resources and enthusiasm for examining this exposure. If use of mobile phones is a health hazard causing other disease or ill health, or causes brain cancers with longer induction and latency times than Interphone covered, we may not have the studies to show it. It will then be only a small comfort that Interphone provided a set of very interesting method papers that will be a most valuable source for teaching case-control methodology in the future.

Jorn Olsen, Neil Pearce, Shah Ebrahim, Cesar Victora

References

  1.  Interphone Study Group. Brain tumour risk in relation to mobile telephone use: Results of the Interphone international case-control study. Int J Epidemiol 2010; 39: 675-94
  2.  Saracci R, Samet J. Commentary: Call me on my mobile phone…or better not? A look at the Interphone study results. Int J Epidemiol 2010; 39: 695-8
Share Button

Comments are closed.