Monday, September 29, 2008

Interphone exposed in the Economist

Mobile madness

Sep 25th 2008
From The Economist print edition

The threat to human health from mobile phones, if any, is much disputed. A shame, then, that a massive multinational study on the question has ended in chaos

Science Photo Library

“LONG-TERM mobile-phone use increases risk of benign tumours!” “Clean bill of health for the mobile!” “Mobile phone-cancer link not proven!” Those who have followed the saga of whether or not mobile phones are damaging people’s brains are used to contradictory headlines. A decade of coverage has left readers and viewers more confused than enlightened, with news reports alternating between alarming claims and soothing reassurances. Yet even by the standards of modern news, it is unusual to see such contradictory headlines about the same piece of research. Which is why a study, called Interphone, provides a cautionary tale.

Interphone began in 2000, ended in 2006, cost $30m and involved around 50 scientists working in 13 countries on 14,000 people. It has, however, still to come to a settled conclusion. A draft of its supposed findings was circulated in June, and Elisabeth Cardis of the Centre for Research in Environmental Epidemiology in Barcelona, who led it, thought until recently that a final paper would be submitted this month. Now, though, it looks as if that will not even happen this year.

The contradictory headlines are the result of national research teams releasing single-country reports, despite the fact that these inevitably involve smaller samples. The results from nine of the 13 single-country studies have been made available in this way, and the consequence is a farrago of misinformation. Many of the national reports suggest, for example, that ever having been a regular mobile-phone user offers statistically significant protection against some brain tumours. This finding is so counter-intuitive that it has led most of the people involved to acknowledge serious flaws in the study’s design.

Please press “recall”

One problem was what statisticians call selection bias. Interphone began by gathering a group of people who had had the cancers of interest (glioma, meningioma, acoustic neurinoma and parotid gland tumour) and questioning them about their past use of mobile phones. The researchers then approached a number of healthy people in order to compare them with the cancer patients, and find out if there was a systematic difference in mobile-phone use between the two groups. Some of those approached agreed, and some declined. Of those who agreed to take part, 59% were regular mobile-phone users as defined by the study’s protocol. Later on, those who had declined were recontacted and asked about their mobile use. Among this group, only 34% were regular users. That meant those in the control group were more likely than average to be regular users, and therefore were not representative of the population at large.

Moreover, the definition of “regular mobile-phone use” was itself questionable. Anyone who had used a phone just once a week for at least six months qualified. That is a pretty low rate of usage. If phones really do cause cancer, but only at high exposure, employing such a generous definition of regular use means that the effect might be diluted into undetectability.

Another potentially serious flaw is that participants asked in 2001-02 about their mobile use a decade earlier will have been using analogue, not digital, handsets. That would lead to a different pattern of exposure and therefore of potential risk.

How the shift from analogue to digital phones might have changed things is anybody’s guess. But both the selection bias and regular-phone-use definition might be expected to mask any adverse effect of phones. The most serious flaw in the study, however, is generally reckoned to be recall bias, which would tend to work in the opposite direction, by suggesting problems with phones that do not, in fact, exist.

Recall bias happened because the study was retrospective rather than prospective. In other words it looked at what people had done in the past rather than following their behaviour into the future. In practice, that meant asking them about past behaviour, and relying on the accuracy of their memories.

Even a healthy person would probably have difficulty recalling exactly how often he used his mobile phone a decade ago, and which ear he routinely held it to. Someone subsequently diagnosed with a brain tumour might easily be biased, consciously or unconsciously, to exaggerate the former and misstate the latter. And that would be enough to account for the fact that several of the single-country studies found that ten years’ mobile use was associated with an increased risk of brain tumours on the side of the head that people said was exposed to their handsets. It would also—more significantly—account for the studies’ weird suggestion that the same amount of use stops tumours developing on the other side of the head.

The scientists involved in Interphone are, it must be said, well aware of these biases, and efforts to work out their effects and adjust for them were always part of the plan. In May Dr Cardis and her colleagues published a validation study in the Journal of Exposure Science and Environmental Epidemiology. This compared the recollections of about 500 Interphone subjects with their actual mobile use according to the records of the network operators. The researchers found a high level of recall errors. Participants underestimated the number of calls they had made by an average of about 20%, and overestimated call duration by 40%. Ominously for the statisticians, the recall errors of those with tumours increased with time.

Suitable (and legitimate) statistical massaging of all the data may be able to offset these biases in a way that could not so easily be done for the smaller data sets from individual countries. Nevertheless Dr Cardis admits the delays in releasing the report have been due to “the difficulty of interpreting the findings due to potential biases” and to the “conducting of additional analyses to try and disentangle the potential impacts of selection and recall errors on the risk estimates”.

The Interphone researchers are split into three camps. One believes any increased incidence of tumours shown in the study is purely the result of the biases. Another thinks it really has found increased risks of certain tumours and wants to call for precautionary measures. A third group is just keeping quiet. One person who knows many of the scientists, but prefers not to be named, describes the relations between members of the three groups as “strained”—harsh language in the world of scientific research.

Regardless of what eventually gets published, some people have already decided the findings warrant action. Ronald Herberman, the director of the University of Pittsburgh Cancer Institute, made headlines in July with a memo to 3,000 members of his staff urging them to limit their use of mobile phones, to refrain from keeping their handsets near their bodies at night and to avoid using them when reception is weak as a phone will then boost the strength of its signal in order that the network can hear it. He based his warnings on “early unpublished data”, and is believed to have been referring to Interphone.

Whatever the outcome, though, at least one lesson has been learnt. Follow-up studies now in the planning stage are expected to use prospective as well as retrospective data. In other words they will pick people at random and see what happens to them. That method takes longer to come to a conclusion—but it is more likely to be one you can trust.