A Gene for Power-Line Leukemia?

Some people believe that living near high-voltage power lines raises the risk of childhood cancer. Most people are skeptical. A Chinese group have just published a paper in the journal Leukemia and Lymphoma, claiming that a genetic polymorphism in the XRCC1 gene, which has been previously linked to various cancers, raises the risk of electromagnetic field (EMF)-related leukemia. People who believe in EMF-related leukemia are happy. The Daily Mail report on this study quoting no less than three such people.

What's the real story? The authors took 123 childhood leukemia patients living near Shanghai. They took blood samples for DNA analysis and asked the parents to report on a wide range of possible environmental risk factors, not just EMF:

The mothers of the patients were interviewed at the hospital by specifically trained medical doctors using a questionnaire. Visits to the present (or previous) residential areas of 66 cases were arranged, and the actual values of magnetic field intensities were measured using an EMF detector (TriField Meter, AlphaLab, USA). Questionnaires covered information about the parents’ sociodemographic characteristics, the children’s pre and postnatal characteristics and the familial history of cancer and autoimmune diseases. The questions related to environmental exposure covered pregnancy and the period from birth to diagnosis and detailed information including: Was there a television set/refrigerator/ microwave oven in the children’s rooms? Did you regularly use insecticides at home? Did you use gardening chemicals such as, fertilisers, herbicides, insecticides, fungicides, others? Were there chemical factories/telecommunication transmitters/electric transformers/power lines around your house?

Relying on self-report like this raises the risk of recall bias, but to be honest, this doesn't seem like a major problem. Certainly there is a much bigger problem with this study (see below). The authors genotyped the children for six different SNPs (genetic variants) which have been previously implicated in cancer

The MassARRAY technology platform (Sequenom, San Diego, California, USA) was used to detect the SNPs in hMLH1 Ex8–23A4G (rs1799977), APEX1 Ex5þ5T4G (rs1130409), MGMTEx7þ 13A4G (rs2308321), XRCC1 Ex9þ16G4A (rs25489), XPD Ex10–16G4A (rs1799793) and XPD Ex23þ61 T4G (rs13181)

See the problem that's developing here? Six SNPs, who knows how many different environmental factors (the paper isn't clear, but it seems to be at least seven, see below) - that's a textbook example of multiple comparisons. Any statistical comparison has a chance of giving a positive result just by chance. If you do enough comparisons, you will find something, just by chance.

The authors do not report making an attempt to correct for this (although there are plenty of ways of doing so). They never even acknowledge the problem. They simply report on their only positive result - an association between the XRCC1 risk allele and "proximity to electrical transformers and power lines" - and relegate all the negative results to a brief summary

No significant interactions between the proximity of the electric transformers and power lines and other genotypes were observed. No significant interactions were observed between genotypes and the presence of television sets, refrigerators or microwave ovens in children’s rooms, pesticides use or the presence of chemical factories or telecommunication transmitter within 500 m of the houses.

The positive result was that out of the children with leukemia, those living within 100m of electrical transformers and power lines were more likely to carry the XRCC1 risk allele than those not living within this proximity. Those living within 50m were slightly more likely than that. Under the assumption that genotype is not correlated with environment in the general population (a reasonable assumption, and they did test this in a control sample), this indicates a G x E interaction for leukemia / lymphoma risk, with p below 0.01.

One such result from what seems like at least 42 such comparisons is not especially impressive. It's certainly not proof of an interaction between XRCC1 and EMF, it's not even "suggestive evidence", it's at best a prompt for further research. Even being generous, and assuming that they would not have reported on an association with any risk factor other than proximity to power lines, this is still 6 comparisons with different polymorphisms (more if you count the fact that children living at differing distances from power lines were tested seperately).

Postscript: I hope that I'm wrong about this. It would be great if XRCC1 raised the risk of childhood cancer, because it would mean that we could prevent some childhood cancers by keeping at-risk children away from power-lines. This post is just something I hacked together in an hour and a half on a Sunday morning, and I'm not a statistician - it would be awful if I've just spotted a serious problem with an important paper which went un-noticed by journal editors and peer reviewers. So if someone wants to disagree with me please, please do - I'll provide the PDF of the paper on request if you need it. Until then, I think that this is especially bad example of the problem of multiple comparisons and a tragic case of sloppy science which could end up having serious consequences for health, in terms of acting as a red herring distracting from more valuable research.

[BPSDB]

You Yang, Xingming Jin, Chonghuai Yan, Ying Tian, Jingyan Tang, Xiaoming Shen (2008). Case-only study of interactions between DNA repair genes (hMLH1, APEX1, MGMT, XRCC1 and XPD) and low-frequency electromagnetic fields in childhood acute leukemia Leukemia and Lymphoma, 49 (12), 2344-2350 DOI: 10.1080/10428190802441347