As some of you know, during a recent ongoing thread concerning risk of cancer to home dwellers that are exposed to radon, Mr. Newman referenced three studies which he believed supported his contention that the risk was real and significant and conclusively and directly demonstrated that radon was a causative factor in lung cancer.
I promised Mr. Newman that if he provided me with the studies, I would review them, and post a critical review here on this board. Since such reviews take a big portion of my time, I cannot review them all at once – so here is the second review, and I will get to the last one in the next couple of weeks or so.
Here it is: (…ahem…)
I have reviewed the paper titled Radon in homes and risk of lung cancer: collaborative analysis of individual data from 13 European case-control studies (Darby S; Hill D; Auvinen A, et all.) as it was provided to me and appeared in the downloaded form from www.bmj.com
In the hierarchy of epidemiological studies, the type of “study” that traditionally carries the least weight, and least credibility is known as a “meta-analysis.” Meta-analyses are just as the name implies – “meta-“ a change in position; behind, or after. They are reviews of a collection of other people’s work, wherein the author has no first hand knowledge of the work, and the author has no quality control over what was originally published. Very often (as was the case here), the author does not explore confounder or bias. A good meta-analysis can be hammered out in an afternoon.
The paper presented for my review, however, is even lower on the “weight of evidence” scale in that not only was it a meta-analysis but, included in the collection of reviewed studies was another (at least one) meta-analysis. In fact, the paper presented for my review by Mr. Newman wasn’t even a meta-analysis at all, but rather it was an abstract of a meta-meta-analysis. (2 “metas”)
The paper was like paying an hundred bucks for a ticket to a piano recital and upon arrival, learning that the pianist is the kid next door. After sitting through his third attempt at “Twinkle, twinkle, little star” you grow suspicious that the little guy has more aptitude in choppin’ than Chopin. It was truly painful to read.
The nature of the paper can be thought of less than a study, and more of a “book report.” The structure of the paper, complete with 26 (twenty six) authors for a three page review suggested that the paper was probably a school project (albeit university level) wherein the first author was probably the university professor, the middle 24 authors, were his students, and the last author, listed merely as “R Doll” was (I am guessing) the preeminent epidemiologist, Sir William Richard Shaboe Doll.
I would speculate that the students alternate the order of their names and repetitively submit the same paper, with minor changes to various journals, so they each get a chance to be one of the top authors. I will also speculate that the last author, R Doll, never even read the paper.
In any event-
As I mentioned, the paper is essentially a “book-report” that looked at 13 European studies; NONE of which, according to the authors supported a link between radon exposure in homes and lung cancer. So it is interesting that Mr. Newman selected this particular paper thinking it supported his position when the overwhelming information referenced in the paper actually opposes his position.
The students conclude that although the 13 studies they selected do NOT show an appreciable hazard from residential radon, they contend (but don’t explain why they so contend) that the reason why the studies do not show an appreciable risk due to radon is because although…
Studies to estimate directly the risk of lung cancer associated with residential radon exposure have been conducted in many European countries. Individually, none has been large enough to assess moderate risks reliably.
In this opening statements the students acknowledge two things 1) their foundational studies don’t support an appreciable risk of cancer from residential exposures to cancer, and 2) if the risks are there, they are thought to be “moderate.”
The students then go on to say:
Greater statistical power can be achieved by combining information from several studies, but this cannot be done satisfactorily from published information.
And yet, this is precisely what the students then go on to attempt. In truth, the comment lacks precision since [FONT=Arial]sometimes, under some circumstances, the data from some studies may be combined to gain greater statistical power. However, in this paper, the students do not go on to explain why that is the case here and since the alternative statement is equally true; that is: “since sometimes, under some circumstances, the data from some studies may be combined to multiply confounders, and therefore obtain less statistical power…” … in the absence of knowing why the students believe their foundational studies are appropriate candidates for combination, the whole thing remains a big mystery.[/FONT]
A foundational statement for the students is:
Air pollution by radon is ubiquitous. Concentrations are low outdoors but can build up indoors, especially in homes, where most exposure of the general population occurs. The highest concentrations to which workers have been routinely exposed occur underground, particularly in uranium mines. Studies of exposed miners have consistently found associations between radon and lung cancer. 2,3.
And the tacit association is that these exposures and risks to miners can be extrapolated to residential exposures and risks (which is not what Mr. Newman thought the paper did, the presented papers were supposed to directly demonstrate a causal association at a significant rate). But if we go and look at the student’s references, we see that the references do not really support what the students necessarily think they are saying. In fact, the students, then immediately begin to back-peddle and say:
Extrapolation from these studies is uncertain but suggests that residential radon, which involves lower exposure to many more people, could cause a substantial minority of all lung cancers.
A substantial minority of all lung cancers? What on earth does “a substantial minority” translate to? That sounds like me trying to get out of a tough spot while selling vacuum cleaners.
But what the students don’t say is that the uncertainty is so large, those same studies suggest there is NO appreciable risk from residential radon. (And what about the 13 studies they used for their meta-analysis which, according to the students, show NO appreciable risk?). This paper is painfully trying.
Another point the students overlooked was the question of cohabitating matched pairs wherein one (a case) comes down with lung cancer, but the other matched for residency (a control?) doesn’t. How come?
The students, however, do go on to explain that they didn’t really have a lot of actual measurements for some of their foundational studies, so, they decided to make some up! (How do you think the original authors would respond to that? Why do you suppose the original authors didn’t include those data or estimates?) To do this they “estimated” exposures. However, they didn’t really explain the validity of their estimates or how they came about them, and the explanation they did give sounds more like they created data (or “guessed” if one wants to be more charitable), and so they guessed what a lot of the exposures were in their paper. And yet, in their paper they consistently refer to the made-up exposures as “measurements.” It’s painful reading.
The students don’t explore or explain the possible selectional bias of each of the 13 studies and how multiplication of those biases may occur. The students don’t go on to explain the potential association or interrelation of any of the studies to any other study, something that may profoundly effect a meta-analysis. Here’s why:
Dr. Greene sets out to study some anecdotal evidence that storks cause babies. He looks a birth rates, and compares them against stork populations, and discovers a clear and consistent correlation between the two; he reports that he has established an “association” between storks and babies.
Dr. Blue, includes by reference, Dr. Greene’s findings, and reproduces the raw data and the conclusions.
Dr. Black gets bored one day and decides to study a long-gone fable about storks bringing babies. He does a literature search and finds a paper on the subject matter by Dr. Green, and another paper, by some completely different, unassociated, guy named Dr. Blue. Dr. Black decides to perform a comparative “study” to see how closely Dr. Blue’s data fit Dr. Green’s data. Dr. Black reports that Dr. Blue’s data is “statistically consistent” with Dr. Greene’s study.
Dr. Grey, finds the whole thing amusing and reports Black’s findings in the Journal of Creative Skepticism.
Jason O’Silverstein, a student of journalism, decides to do a class project on medical science vs. folklore and chooses the old Eastern European chestnut about storks and babies. Jason trawls the internet and finds four studies by four different MDs and decides to perform a meta-analysis on their studies. Since he never was too good at epidemiology, he really didn’t want to get into the problem of compounding confounders. In his final report, he entirely fails to realize that he doesn’t have four studies, he has one report that was repeated three different times. He tells his readers that
Studies to estimate directly the contribution of storks to birth rate have been conducted in many European countries. Individually, none has been large enough to assess the association reliably. Studies by medical doctors have consistently found associations between storks and babies. 2,3. Extrapolation from these studies is uncertain but suggests that storks could cause a substantial portion of all minority babies. Greater statistical power can be achieved by combining information from several studies, but this cannot be done satisfactorily from published information. So that is exactly what I’ve done.
Painful. I’m telling you, this paper was painful. [FONT=Times New Roman][FONT=Arial]This paper, if anything at all, is an exercise in tautology. [/FONT][/FONT]I can tell you this, I was never permitted to put out this sort of stuff (Thank you Dr. R).
There are some gems:
We assessed the association between radon and lung cancer in two ways. Firstly, a model was fitted in which the additional risk of lung cancer was proportional to measured radon.
OK – This is interesting, since most of the time, we use models to predict an outcome based on known inputs (i.e. we attempt to predict the number of cancers based on the radon exposure). Here, the students seem to be fitting the model to the outcome (Making the inputs fit the outcome). How can you do this? Well, hell it’s easy, since you made up the exposures in the first place, just keep making up different exposures until ya’ get the thing to fit!
Painful. Painful. Painful. ](*,) (I think, I may know what may have killed Dr. Doll last year… it may have been this paper.)
OK, then they say …
No I can’t do it. Sorry, it’s too painful. They say their model showed that smoking didn’t have an effect on lung cancer. But if smoking is a major cause of lung cancer, and radon (according to the students) only [FONT=Arial]couldcause a substantial minority of all lung cancers but, but, but… smoking exhibits less of a risk than radon… I can’t … it’s too painful. #-o [/FONT]
This paper is an abstract of a book report of a meta-analysis of at least one other meta-analysis that never should have seen the light of day, and is not likely to be used by anyone who may have to take the stand and defend it.
Oh; by the way… their arguments on linearity? Read ‘em, they’re painful.
The paper did not represent a study, and did not contain any new research. The paper was an abstract of a report on a meta-analysis that was based on 13 studies (at least one of which was another meta-analysis) that showed that there was no appreciable risk of cancer from radon exposure.
The authors (all 26 of them :shock: ) reported that although none of the studies in their meta-study showed an appreciable risk of lung cancer due to radon exposure, their work (based on those studies) indicate that radon is responsible for 2% of ALL cancers in Europe (This would delight the asbestos industry! But they will never reference this paper for fear of being ridiculed by it’s inclusion.)
Painful. Painful. Painful.
The paper failed to deliver on Mr. Newman’s contention that the paper conclusively and directly tests and demonstrates that radon exposure, as seen in houses, significantly increases the risk of cancer.
I will have to conclude that either Mr. Newman did not read this paper, or if he did, did not understand it (which would not be too surprising since it’s not a particularly understandable paper anyway.)
Although my dinner at the Brown Palace is safe for the moment, I nearly lost mine while reviewing this paper.
I will get to the third study as time allows.
Cheers, All!! Have a great weekend. (My computer just made a strange “pa-ching” noise… either my wife programmed in some kind of “beer alert,” or this old system found the whole thing too painful).
Caoimhín P. Connell
Forensic Industrial Hygienist
(The opinions expressed here are exclusively my personal opinions and do not necessarily reflect my professional opinion, opinion of my employer, agency, peers, or professional affiliates. The above post is for information only and does not reflect professional advice and is not intended to supercede the professional advice of others.)
I edited this post at 07:04 MST 01/27/07 fixing misspellings and missing words.