I got Poe'd by Rod Adams!
Look at the title:
Systematic Bias Discovered in Atomic Bomb Survivor Cancer Comparisons
Oh, the ironies (keeping the title in mind):
1. Rod is showing bias. He doesn't like the scientific consensus so he welcomes anything to refute it. That is bias. In science, we go where the evidence takes us without an agenda other than objectivity. LNT is the consensus. If you don't like that, that is bias (unless you have evidence to overturn it, and we both know you don't.). Accept reality, no big deal.
2. This particular "study" addresses an earlier atomic bomb survivor study published in Radiation Research. An unbiased researcher who wanted to criticize the original study would publish in the same journal. Did that happen here? Of course not! It's published in the pseudo-scientific journal of the International Hormesis Association, whose bias is to find hormesis! This is like a Creationist writing a "study" on evolutionary biology and publishing it in a creationist journal after reading about a new fossil find. Guess what? The creationist's "study" will conclude there's something wrong with the fossil find.
Oh, sorry..they're not called "creationists" anymore, that became unpopular, "intelligent design advocate" sounds much more sophisticated. And, oh....I almost forgot. The International Hormesis Association changed their name to the International Dose-Response Association. Hmmm...
In real science we don't have journals AIMED at supporting an agenda. Journals publish studies on certain subjects (biology, health physics, anthropology, etc). We collect data and try to explain what's going on in the Universe. We want to intentionally challenge what we think is true. We don't have the answer first and start a journal on it! Hey, I imagine Jesus (hormesis) exists, I'm starting a journal. Please send me stuff which appears to support his (its) existence! Let's massage the data and the narrative to support our conclusion even if a more rational explanation is plausible!
In real science we don't have journals AIMED at supporting an agenda. Journals publish studies on certain subjects (biology, health physics, anthropology, etc). We collect data and try to explain what's going on in the Universe. We want to intentionally challenge what we think is true. We don't have the answer first and start a journal on it! Hey, I imagine Jesus (hormesis) exists, I'm starting a journal. Please send me stuff which appears to support his (its) existence! Let's massage the data and the narrative to support our conclusion even if a more rational explanation is plausible!
3. So let's see now, we've got a biased blogger (in search of anything to refute the science), promoting a biased "study" (the author chose a different journal than original), that was published in a biased journal (seeks evidence of hormesis)! Refer to the blog title! AAHHHHH!
4. Let's go to the "study", but first, let's review the earlier A-bomb study. You need to realize that there is no such thing as a perfect epidemiological study. The best type of study is called a double-blind randomized trial. This is usually used for medical efficacy (is a new medicine better than an existing one or a placebo?). One group is given the something (neither they nor the administrator knows if it's the new medicine or not) and the other is given the alternative (they don't know either). The DOSES ARE WELL KNOWN, people are chosen randomly and objectively on the front end. We don't use this type of study for toxicology because we're dealing with toxins! It would be unethical to intentionally expose people to them.
--Aside--If you watch the first 3 minutes of this video, you'll see that Ed Calabrese (head of the hormesis society and journal) doesn't understand the difference between medical efficacy (why we don't accept homeopathy) and toxicology (mechanisms underlying radiation toxicity). Poor boy, he just doesn't get it!---
For toxins we have to wait until an unplanned exposure occurs (researcher has no control) and then ESTIMATE DOSES. If the researcher gets on the job immediately after an unplanned exposure the study is a prospective study, he collects data into the future (a researcher can also study long after an exposure by doing a retrospective study, but he loses data accuracy). The A-bomb study is mostly prospective (good) with a little retrospective (it started something like 5 years after the blasts, so they had to look back in time). It is a closed-population study. This means we want to follow everyone in it until they die. No one enters the study since it started and no one leaves unless they die (in practice this may not be 100% possible). This means the study IMPROVES with each passing year. Rod has complained about funding this "until they all die", but that's what we want! I could go on and on about pros and cons, but the bottom line is this is the best study we have, though not perfect. And as of today, it just got better!
At the start of this type of study the epidemiologists have to determine how they're going to "stratify" the data. What are the important differentials and what sorts of factors do we expect to not matter much or average out over time?
The more a study slices and dices and the more factors we track the more the study costs. There are a number of things we aren't stratifying on, but the expectation is, those things are not very important to understanding the dose response effect.
Now to the pseudo-study. The author is saying that because background cancer rates change over time depending on prefecture, that the A-bomb study has systematic bias.
Do NOT be distracted by "differences in prefectures". The focus is might there be something that will influence the dose response effect between the exposed and unexposed groups. Things which really have a chance of doing so are called confounders. The people in the A-bomb study are living throughout Japan. Some may move every few years. They may live in one prefecture and work in another,, etc. So these background cancer variations, BASED ON PREFECTURES are expected to be washed out overall. One can imagine a worse case scenario where all the people just happen to be located in the worse possible locations throughout the entire study, such that we get a significant systematic error, but not too likely.
But of course we know what the author's agenda is. He wants to show hormesis (bias), so out of lots of different potential confounders, he magically picks one (cancer rate differences by prefecture), and gets the graph to look just the way he originally expected it should to begin with. If he couldn't do so, he would have concocted some other underlying difference, and tried again. That's agenda-based, cherry picking. Can you see the irony in this, being the subject is about bias? This is basically numerology, not science. This is how people find Biblical hidden messages...they apply different coding patterns to the text until by chance every here and there some word is formed that meets their expectations, and viola, Jesus has hidden messages!
His mathematical approach is way too simplistic. It basically modifies the ERR (a dose vs. excess cancer metric) with a metric based on background incidence rate over time. It is a "fudge factor" approach and inconsistent with the probabilistic bias analysis techniques of modern epidemiology. But it does get him the results he wants.
There are many potential confounders, but they are expected to wash out. No one is accounting for differing medical radiation exposures. But the expectation is that on average the exposed (to A-bomb) population will have about the same medical radiation exposures as the unexposed (to A-bomb) population.
Want another? What if a person (exposed or unexposed) dies of something other than cancer. But what if the person had an undiagnosed cancer at the time of death? In the wash.
Want another? How about Fukushima? Etc., etc.
This is why we like large numbers of people in an epidemiological study, and the A-bomb study has large numbers. Most other radiation studies do not.
If the pseudo-science author really wanted to show that this prefecture effect mattered, he could actually go and correctly analyze the real data for it. That would be evidence versus conjecture. That would be worthy of publication in the real scientific literature (if done objectively and correctly - he'd be expected to explain his motivation). But there is no good reason to expect that "background cancer difference by prefecture over time" will have a significant confounding effect between the exposed and unexposed groups. THIS IS WHY IT ISN'T DONE IN ANY STUDY! (It's the long standing consensus, that it's insignificant) But hey, if you think otherwise, do the analysis. That's how real science operates.
Beware - the consensus will be very skeptical. (But have no fears you will have an alternative publication avenue. Thanks, Ed!)
5. Rod titles his blog "Systematic Bias Discovered...". No, there has been no systematic bias discovered! The pseudo-study is cherry-picked conjecture. Do the analysis in order to make the discovery. Heed warning above.
--Aside--If you watch the first 3 minutes of this video, you'll see that Ed Calabrese (head of the hormesis society and journal) doesn't understand the difference between medical efficacy (why we don't accept homeopathy) and toxicology (mechanisms underlying radiation toxicity). Poor boy, he just doesn't get it!---
For toxins we have to wait until an unplanned exposure occurs (researcher has no control) and then ESTIMATE DOSES. If the researcher gets on the job immediately after an unplanned exposure the study is a prospective study, he collects data into the future (a researcher can also study long after an exposure by doing a retrospective study, but he loses data accuracy). The A-bomb study is mostly prospective (good) with a little retrospective (it started something like 5 years after the blasts, so they had to look back in time). It is a closed-population study. This means we want to follow everyone in it until they die. No one enters the study since it started and no one leaves unless they die (in practice this may not be 100% possible). This means the study IMPROVES with each passing year. Rod has complained about funding this "until they all die", but that's what we want! I could go on and on about pros and cons, but the bottom line is this is the best study we have, though not perfect. And as of today, it just got better!
At the start of this type of study the epidemiologists have to determine how they're going to "stratify" the data. What are the important differentials and what sorts of factors do we expect to not matter much or average out over time?
The more a study slices and dices and the more factors we track the more the study costs. There are a number of things we aren't stratifying on, but the expectation is, those things are not very important to understanding the dose response effect.
Now to the pseudo-study. The author is saying that because background cancer rates change over time depending on prefecture, that the A-bomb study has systematic bias.
Do NOT be distracted by "differences in prefectures". The focus is might there be something that will influence the dose response effect between the exposed and unexposed groups. Things which really have a chance of doing so are called confounders. The people in the A-bomb study are living throughout Japan. Some may move every few years. They may live in one prefecture and work in another,, etc. So these background cancer variations, BASED ON PREFECTURES are expected to be washed out overall. One can imagine a worse case scenario where all the people just happen to be located in the worse possible locations throughout the entire study, such that we get a significant systematic error, but not too likely.
But of course we know what the author's agenda is. He wants to show hormesis (bias), so out of lots of different potential confounders, he magically picks one (cancer rate differences by prefecture), and gets the graph to look just the way he originally expected it should to begin with. If he couldn't do so, he would have concocted some other underlying difference, and tried again. That's agenda-based, cherry picking. Can you see the irony in this, being the subject is about bias? This is basically numerology, not science. This is how people find Biblical hidden messages...they apply different coding patterns to the text until by chance every here and there some word is formed that meets their expectations, and viola, Jesus has hidden messages!
His mathematical approach is way too simplistic. It basically modifies the ERR (a dose vs. excess cancer metric) with a metric based on background incidence rate over time. It is a "fudge factor" approach and inconsistent with the probabilistic bias analysis techniques of modern epidemiology. But it does get him the results he wants.
There are many potential confounders, but they are expected to wash out. No one is accounting for differing medical radiation exposures. But the expectation is that on average the exposed (to A-bomb) population will have about the same medical radiation exposures as the unexposed (to A-bomb) population.
Want another? What if a person (exposed or unexposed) dies of something other than cancer. But what if the person had an undiagnosed cancer at the time of death? In the wash.
Want another? How about Fukushima? Etc., etc.
This is why we like large numbers of people in an epidemiological study, and the A-bomb study has large numbers. Most other radiation studies do not.
If the pseudo-science author really wanted to show that this prefecture effect mattered, he could actually go and correctly analyze the real data for it. That would be evidence versus conjecture. That would be worthy of publication in the real scientific literature (if done objectively and correctly - he'd be expected to explain his motivation). But there is no good reason to expect that "background cancer difference by prefecture over time" will have a significant confounding effect between the exposed and unexposed groups. THIS IS WHY IT ISN'T DONE IN ANY STUDY! (It's the long standing consensus, that it's insignificant) But hey, if you think otherwise, do the analysis. That's how real science operates.
Beware - the consensus will be very skeptical. (But have no fears you will have an alternative publication avenue. Thanks, Ed!)
5. Rod titles his blog "Systematic Bias Discovered...". No, there has been no systematic bias discovered! The pseudo-study is cherry-picked conjecture. Do the analysis in order to make the discovery. Heed warning above.
No comments:
Post a Comment