Disclaimer Disclaimer Disclaimer Disclaimer Home           Home Home           Home About  site   About  site About  site   About  site About  me    About  me About  me    About  me Current reading   Current reading Current reading   Current reading CONTACT   CONTACT CONTACT   CONTACT Links                  Links Links                  Links Book shelves      Book shelves Book shelves      Book shelves
Blog Archive    Blog Archive Blog Archive    Blog Archive Recent posts  Recent posts Recent posts  Recent posts
I recently published a paper with the exciting title 'Analyses of temporal and spatial patterns of glioblastoma multiforme and other brain cancer subtypes in relation to mobile phones using synthetic counterfactuals' in the peer-reviewed journal Environmental Research. At the time of writing this is still available as open access here [link], but because this research did not receive additional funding from anywhere it will be placed behind a paywall soon [update 12/12/2018: it seems the journal has gone open access, so there may be a chance this remains available to everyone]. This work continued from previous work on the likelihood of mobile phone use as a risk factor for brain cancer published earlier here as open access [link], but looked specifically at trends in (1) an aggressive form of brain cancer, glioblastoma multiforme (GBM), in different brain regions and (2) different cancer subtypes, including GBM, in one particular region in the brain; the temporal lobe. The first study indicated a higher-than-expected incidence of cancers in this anatomic region, while a recent descriptive study that received a lot of media attention showed that specifically GBMs have been increasing in the temporal lobe [link]. Because the temporal lobe is one of the brain regions that receives the most radiation when using a mobile phone for calling, the current study was interesting additional work to see if this increase could indeed likely be associated with mobile phone use. Both papers were picked up by Microwave News, and in both cases I was asked to provide further background and comment on this. Also in both cases, I felt that what was eventually put on that website did not reflect the (quite extensive) explanations and comments I had made. I cannot force people to put stuff on their website, so there is not much I could do to balance that information better, but I do have my own website; you are looking at it as we speak. So, although my intention has always been to not use The FunPolice as a vehicle for my own work, I am making an exception now to provide a bit more background (and some interesting additional analyses) to that paper. First of all, it is only fair to provide link to the Microwave News article (related to this new paper only, the older paper is water-under-the-bridge). The article, entitled 'Location, Location, Location Aggressive Brain Tumors Tell a Story' can be found here [link], so please read it for yourself. Unfortunately, by referring to ‘round 1’ and ‘round 2’ Microwave News decided to pitch different views on the importance of mobile phones in increases in GBM trends in England as a battle between myself and Alisdair Philips, a mobile phone and cancer campaigner who runs Powerwatch, and who is also the lead author for the other paper on GBM trends in England. Just to be clear on this. There is no battle of such kind: Alisdair and I used the same data (the official UK Cancer Registry data) and we show exactly the same thing (obviously, one would almost say). The incidence of Glioblastoma multiforme has been increasing since about the middle of the 1990s in England, and has been increasing faster in certain anatomic regions, most notably the temporal and frontal lobes, than in others. That is what the descriptive data show... The question though, is why. Alasdair's argument is straightforward. This increase started roughly when a significant number of people started to use mobile phones, and is primarily shown in those brain regions that receive the highest radiation exposure when calling: ergo it has to be mobile phones that caused it. Or, in other words: 'If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.' The problem with this, of course, is that this conclusion is based on eyeballing trends only. Although you can speculate about causes of any changes in these trends, it is not possible to base any strong conclusions on this. Also, it is worth mentioning in this particular situation that just because a number (incidence in this case) increases over time doesn't necessarily mean something is weird and unexpected is going on. As epidemiologists we try to do better than this, and we do our best to strengthen any claims on causality as much as we can. For example, we combine information from these timeseries with information from other studies and evidence from animal and mechanistic studies. With respect to mobile phone radiation and cancer however, that evidence is far from convincing: there are no randomized controlled trials in humans investigating this (for obvious reasons I hope), evidence from observational studies is contradictious, and although studies in animals and cells provide some evidence for biological effects, they do not convincingly demonstrate risk of cancer. For example, despite the results from the large randomized and controlled NTP study in mice and rats providing some evidence of carcinogenic risk, there are reasons to be a bit careful with basing any conclusions on risk in humans in these studies (as discussed in detail by Prof Kabat here [link]). Another, quite new, way of improving the claims that be drawn from these kind of timeseries of annual incidence (of brain cancers) is to try and estimate their counterfactual trends and compare these to what really happened. Counterfactuals are basically 'what if' scenarios: 'What would have happened with the time trends of the annual incidence of brain cancers in England in the case where mobile phone radiation has absolutely no effect on brain cancers?“ If we can create these counterfactual scenarios and we can compare them to what really happened, we can get a better idea of the possible impact of radiation. After all, if for this question of GBM trends in England the counterfactual and the real time trends are similar, then it there is no need for mobile phone use to be brought up as a risk factor. In other words, and specific to this GBM example, despite an increase in the annual number of new cases from the 1990s, these may actually be exactly as expected (or conversely, where large differences between both are observed some other factor, or factors, is needed to explain the trend). I did, of course, not just outline this concept of counterfactuals for entertainment value, but also because this is what I did in my paper. To do this, I used what Microwave News calls an 'arcane method' (it is not anymore though, I just explained it) and constructed those counterfactuals based on time trends in other factors that could affect the brain cancer trends, and then compared them with the real brain cancer trends. For the interested, because of another ‘arcane method’ called Bayesian model averaging, the final results are relatively insensitive to the specifics of the set of other factors. The results of these analyses are really interesting. The results from the first paper indicated that, although GBM annual incidence increased (as all the data show) this was almost exactly as expected based on the counterfactual timeseries. However, brain cancers in the temporal lobe did increase much more than expected; by about 30ish percent. Moreover, not only did the brain region correspond with where you would expect it if radiation from mobile phones caused cancer, the time when this should occur after first use (i.e. the lag) also corresponded with mobile phone as the cause. In fact, including mobile phone usage in the models showed that this could well explain a large part of the trend. Although this seemed to suggest that mobile phones could cause brain cancers in the temporal lobe specifically, it did not point to GBM as the target cancer subtype: a hypothesis favoured by some people with an interest in this area, including those mentioned above. A logical next step therefore was to see whether these additional numbers of brain cancers in the temporal lobe could be GBM specifically, or whether it was the location regardless of cancer subtype that was the main target. *** And that brings us to the latest, second paper [link]. The initial results confirmed other people's suspicions. The target was indeed GBM specifically, but only in the temporal and frontal lobes; the frontal lobe being the other brain region with high radiation exposure during calling with mobile phones! Not just that, but when I looked at different age categories this also seemed consistent with mobile phones being the cause. "Gosh…." It’s important, especially with these kind of ecological analyses, to check a lot of things before making strong claims. I decided to look closer at the effects in different age groups to see if these results would still hold when looking at older and younger groups. And that is when it became clear that the largest effects (relatively speaking) are actually observed in children who had been too young to have used mobile phones during the relevant, early time period and people who had been too old to have been amongst the first people to use mobile phones (the over 75s and over 85s now, so ~65+ at the earlier and relevant time period). For other groups the trends were similar to those counterfactuals or opposite to what would be expected if mobile phones caused cancer. And that is the bottom line of that paper. There has indeed been an increase in the annual number of new GBM cases, especially in the frontal and temporal lobes, but in the relevant age groups where effects should have occurred these trends can be modelled (forecasted) very well from trends in other factors. There is no, or hardly any, role to play for mobile phone use. Where an effect is observed, this is in the very young and very old, from which we can conclude that we need to broaden our thinking and look at other factors than mobile phones that could explain this. Some of these are mentioned in the paper, and they may all play a role (or not), while there may other ones not included. Nonetheless, one potential explanation for the increased incidence could be that over that same time period medical detection and diagnosis of these brain cancers has improved greatly; especially in older people. Given the age group effects discussed above, this seems a reasonable contributing factor, if probably not the only one. In the Microwave News article this hypothesis is dismissed by saying that any effect would be observed in all brain regions, not just in the temporal and frontal lobe. That is a fair point, but given that just over half of the GBMs occur in  these brain regions (with the other 50% scattered through all other brain regions or classified as ‘unspecified’), one would expect the clearest signals in those regions as well. Nonetheless, the percentage of GBMs in the temporal and frontal lobes relative to all other regions seems to have been increasing slightly since the 1990s Unfortunately, I do not have the data that would be required to investigate such a hypothesis in detail, but with the data I do have we may be able to get some rough idea whether improvements in medical detection and diagnosis may be a viable possible (partial) explanation. Below are the time series of annual new cases of GBM in the temporal and frontal lobes as well as those classified as ‘unspecified’; the latter can be viewed as an indication of quality of diagnosis because, if these had been better classified at the time, the brain region would have been reported. Note that I added some random scatter to these plots (plus don’t give you the axes) because I don’t think I am allowed to share the actual numbers. A quick look at these trends indicates that indeed there seems to be a decrease in the annual incidence of ‘unspecified GBMs’ from about 2006; accompanied by what may be an additional increase in the incidence of GBM in the temporal and frontal lobes.    Now here is the idea….. If improvements in the quality of diagnosis is a (partial) explanation, then we can hypothesize that GBMs in the frontal and temporal lobes PLUS those classified as ‘unspecified’ together should follow a temporal pattern that follows the predicted pattern. We can do these analyses using the same statistical methods used in the original paper. So I did that, and the results are shown in the Table below. The original analyses from the paper, had I compared the temporal and frontal lobes show exactly what was shown in the paper and indicate an excess risk of 32%, with a certainty of -1% to 65% and a “statistical significance” (it is not really, but we can use it as such) of 0.03; confirming evidence of an excess number of GBM cases when and where one would expect them if mobile phone were the cause (remember that age group specific analyses indicated it probably was not though). Now we add the GBMs with unspecified location to the total number of new cases and re-run the analysis. I have made a couple of different assumptions with this, because it is not clear how much of these unspecified GBMs should be added: 1. The average number of GBMs in the frontal and temporal lobes relative to all GBMs over the whole time period is 59%, so I also assumed 59% of unspecified GBMs would have been in those lobes had they been diagnosed correctly. 2. As mentioned above this percentage increased by year, so I did the same thing as above, but instead of the 59% I used annually calculated percentages. 3. Maybe there was a specific issue with diagnosis in the temporal or frontal lobes, so I also assumed all GBMs with unspecified location were all really located in one of these lobes had they been diagnosed correctly. An overview of the results of these analyses are shown in the table below.  The main conclusion from these analyses is that if unspecified GBMs are allocated to the temporal or frontal  lobe, then there is no evidence of an excess number of new cases anymore. All associations are non-  significant and the relative effect sizes have reduced from 32% to  8%-18% depending on the assumption.   I do not know which one of these alternative models is the most plausible one, but relatively unimportant  given that they all indicate the same thing: improvements in diagnosis may be a plausible explanation, at  least in part, for the excess number of new cases of GBM in the frontal and temporal lobes. Secondly,  although one could argue that there is still an 8-18% excess, this method is not sensitive enough to detect if  this is different from expected trends (it is non-significant so you will), while also even if true this implies  the radiation from mobile phones is nowhere near as important as sometimes claimed; if a risk factor at all.   I can do these analyses specifically for different age groups, and I may do this in future, but given that this is  not really the best data to investigate this particular hypothesis, I will not do these now. I think the main  points from these analyses are clear, and indicate that mobile phones use is increasingly unlikely as an  important risk factor for the observed trends in glioblastoma multiforme while they further are suggestive of  improvements in medical procedures over the last 30 years being a plausible cause for the observed effects,  at least in part. As a more general conclusion though, these analyses, as well as those in the original paper, demonstrate why  strong conclusions on causal risk factors on eyeballing timetrend data is a bad idea.   
Serving the community with a smile.......... ................in a Public Healthy kind of way
Back Back Back Back
Thinking about time-series......and mobile phones