RAMP* Addition-2: "Causation," February 25, 2000
A Response to the Comment,
But Dose-Responses Do Not Prove
They Do Not Rule Out Some
* RAMP is a short name for the book,
Radiation from Medical Procedures in the
Pathogenesis of Cancer and Ischemic Heart Disease: Dose-Response
Studies with Physicians per 100,000 Population,
by John W. Gofman. Nov. 1999.
 Part 1.
|
Reasonable Biological "Mechanisms" Fortify Dose-Responses |
|
Part 2.
|
Dose-Response: Partly from Xrays, Mostly from Some Other Cause? |
|
Part 3.
|
Requirements for a Reasonable Challenge on the Causation Issue |
|
Part 4.
|
Where Is a Similar Objection to the Atomic-Bomb Study and Others? |
Some
responses to RAMP (perhaps only to its Executive
Summary) share a comment which can be paraphrased as follows: "The
analysis produces very impressive dose-responses, but dose-responses do
not prove causation by medical radiation, because dose-responses do not
rule out some other cause of the observed relationships." This issue
was addressed several times in RAMP, where it is indexed under
"Causation: Types of Evidence."
There is no need to reproduce all
those parts, which are already available in the book itself. Here,
we want to make some additions.
Part
1. Reasonable
Biological "Mechanisms" Fortify Dose-Responses
The "some other cause" comment applies to every dose-response study of free-living humans in the biomedical literature. Analysts (ourselves included) recognize the possibility of a confounding variable, meaning an unidentified and unmatched cause (or anti-cause) of the same effect. The "some other cause" comment is generic, and is equally applicable to the Atomic-Bomb Survivor Study and to every other dose-response study in the radiation literature of free-living humans. Therefore, the generic comment can be made about RAMP, no matter what unit of dose is measured along the x-axis (e.g., PhysPop, medical rads, or other).
Although a strong, positive dose-response is regarded as the "gold standard" in establishing causation, everyone recognizes that dose-response studies among free-living humans can never rule out a confounding variable with certainty. Therefore, analysts look also for supplemental support for causation. For instance, it is common to have reservations about an observed dose-response unless a biological "mechanism" for the cause-effect relationship is plausible. Indeed, the search for a plausible biological "mechanism" is central to the unsettled issue of whether exposure to microwaves initiates or promotes Cancer.
The dose-response studies in RAMP are supported, for the mechanism of carcinogenesis, by the well-accepted facts that xrays can cause mutations and that acquired mutations are a cause of Cancer when they alter cancer-related segments of a cell's chromosomes.
Can acquired mutations also act as atherogens? Research on this question has received only meagre attention, so far. RAMP (Chapter 44) discusses some evidence from the pathologists that human atherosclerotic plaques are clonal. Chapter 44 also discusses some experimental animal evidence. RAMP Addition-3 will describe some additional molecular evidence that acquired mutations are more frequent in human atherosclerotic plaques than in non-atherosclerotic tissue from the same person.
Two additional real-world observations --- facts --- support the proposition that acquired mutations have a causal role in atherogenesis.
One observation is that atherosclerotic plaques are localized, and occur adjacent to normal tissue. People have patches of atherosclerosis. Plaques do not involve the entire vessel. This observation has not been well-explained by current models of atherogenesis, but localization is very well explained if plaques develop only at the particular sites which acquire atherogenic mutations (RAMP, Chapter 45, "A Unified Model").
The second observation, from RAMP itself, is that mortality rates from Cancer and from IHD behave like each other with respect to the dose-response with medical radiation (while NonCancer NonIHD causes of death have either the opposite dose-response, or none at all). Since xray-induced mutation is the mechanism which explains the observation for Cancer, it is highly reasonable to expect that the same mechanism (xray-induced mutation) explains the same dose-response observed in the same study between radiation and mortality from Ischemic Heart Disease.
In
summary, the spectacularly strong, positive, dose-response
observations in RAMP are fortified as causal by a reasonable biological
mechanism: Acquired mutations.
|
Part
2.
|
Dose-Response: Partly from Xrays,
Mostly from Some Other Cause? |
With respect to RAMP, the generic "causation" comment may reflect a suspicion that the observed dose-responses are only partly due to medical radiation, and that perhaps they are mostly due to some unidentified cause of both Cancer and Ischemic Heart Disease (IHD) --- some shared, additional, unmatched cause which has been positively correlated with medical radiation along the x-axis. A positive correlation would be required, in order to make medical radiation appear more important than it really is. Various possibilities, for this type of confounding variable in the RAMP analysis, were explored in considerable detail in Chapter 68.
So far, no commentator has expressed the neutral concern that some unknown cause of both Cancer and IHD might be negatively correlated with medical radiation along the x-axis --- which would cause medical radiation to appear less important than it really is. This type of confounding factor, in the RAMP analysis, was explored in considerable detail in Chapter 48.
Without
mentioning or refuting our points in Chapter 68,
certain commentators have speculated on two possible "other causes" of
the observed dose-responses in RAMP. Both of those specific "other
causes" occurred to us, also. They are among the topics examined in
RAMP's Chapter 68, entitled "Is There a Reasonable Non-Radiation
Explanation for the Observations?" Our discussion can be examined
there. Here, we abbreviate. What are the two speculations?
The first speculation is this: A positive correlation may exist, by Census Divisions, between PhysPop (the x-variable in RAMP) and the density of people per 100,000 population who are already sick from Cancer and Ischemic Heart Disease. In other words, if the higher PhysPop Census Divisions also had a higher density of Cancer and IHD patients per 100,000 population than did the lower PhysPop Census Divisions, then this could explain a rising mortality rate (the y-variable) from Cancer and IHD as PhysPop values rise.
This speculation approximates a reversal of the variables: The y-variable (the rate of Cancer or IHD deaths per 100,000 population) "causes" the x-variable (the density of physicians per 100,000 population).
We dismiss this speculation because the correlation is significant and negative between PhysPop and NonCancer NonIHD causes of death by Census Divisions (RAMP Chapter 25), while the correlation is simultaneously strongly positive between PhysPop and death from Cancer and from IHD, by Census Divisions. In other words, there is no general relationship between the density of persons with fatal disorders and the density of physicians, by Census Divisions.
Moreover,
the dose-response between PhysPop and Cancer is
spectacular in 1940, when Cancer accounted for only 11% of deaths and
when IHD accounted for about 17% (an approximation, because there are
no data for IHD until 1950). In 1940, the NonCancer NonIHD causes of
death were overwhelmingly dominant. If very sick people and physicians
attract each other, then a strong positive correlation should have
occurred in 1940 between NonCancer NonIHD deaths and PhysPop, by Census
Divisions. Instead, the correlation was significantly negative (RAMP,
Chapter 25). We consider it non-credible to explain this fact by
proposing that a special attraction, pulling people across Census
Divisions, existed between physicians and patients having Cancer and
IHD, but not having other fatal diseases.
The second speculation is that a positive correlation may exist, by Census Divisions, between PhysPop (the x-variable in RAMP) and degree of "urbanization," and that "urbanization" is a cause of extra Cancer and IHD.
Suppose that a tight positive correlation, from 1920 onwards, could be documented between PhysPop and degree of urbanization by Census Divisions. And suppose that a tight positive correlation, from 1920 onwards, could be demonstrated between degree of urbanization and the age-adjusted mortality-rates from Cancer and IHD by Census Divisions. Then would "urbanization" be the main cause of the higher mortality rates from Cancer and IHD in the higher PhysPop Census Divisions?
A
neutral person, looking for the biological mechanism which
makes urban life more carcinogenic and atherogenic than rural life (if
that can be demonstrated, by Census Divisions), has certainly not
eliminated xray-induced mutations as the prime cause. If (a) PhysPop
and "urbanization" are positively correlated over decades by Census
Divisions, and if (b) "urbanization" and age-adjusted mortality from
Cancer and from IHD are also positively correlated over time by Census
Divisions, the cause of (b) would probably be the extra xrays, given
for decades to the populations of the more urbanized Census Divisions
by the extra physicians practicing in the more urbanized Census
Divisions.
Part
3. Requirements
for a Reasonable Challenge on the Causation Issue
The following review may be appropriate for persons who believe that there must be "some additional cause" (additional to medical radiation) producing the observations in RAMP. The additional cause should "jump at least three hurdles" in their minds, in order to be a credible candidate:
1) The speculative non-radiation cause must have a tight positive (not a negative) correlation over decades with PhysPop, by Census Divisions.
2) The speculative non-radiation cause must be a cause of both Cancer and IHD, but not of NonCancer NonIHD causes of death. Otherwise, it cannot explain the observations in RAMP. Smoking is a cause of both Cancer and IHD, but we have already demonstrated (Chapter 48) that the impact of smoking has a negative correlation with PhysPop in the 1940-1990 period.
3) The speculative non-radiation non-smoking cause must have its status, as a cause of both Cancer and IHD, supported by more than "just" dose-response studies. Persons, who object to RAMP because "dose-response evidence alone does not prove causation," will need to apply the same standard to any other cause of Cancer and IHD.
An
objective analyst will recognize the low probability that
"some additional cause" produces most of the results in RAMP.
Part
4. Where Is a Similar Objection
to the A-Bomb Study and Others?
The observation of a strong positive dose-response is properly regarded by analysts in biomedical research as very strong evidence of causation. The dose-response is based on real-world evidence, whereas there is only a speculative possibility that the dose-response is largely or wholly due to a confounding variable. Specifically for RAMP, Part 3 illustrates why such a speculation is unlikely to be correct.
When a positive dose-response is observed in data, faith in the causal presumption is appropriately highest when there is a plausible biological mechanism, when the database is free from potential bias, when the database is very large, and when the positive correlation is statistically powerful.
The dose-response studies in RAMP score extremely high on all those points. Indeed, the quality of the data and results in RAMP makes a far stronger case for causality than is possible ever to make from the A-Bomb Survivor Study or any other radiation study yet undertaken.
Yet RAMP is the study which elicits the objection about not proving causation. Therefore, it seems sensible to wonder: Would the objection occur if the very same data had revealed a low impact from medical radiation? Perhaps the objection is the way of some commentators to say that they doubt the validity of the findings for a different reason: The findings do not match the prevailing opinion.
The issue of prevailing opinion has been analyzed in RAMP Addition-1, where we demonstrated that prevailing opinion is not the same thing as the range of informed scientific opinion, and that the findings in RAMP do fall within the range of informed scientific expectation.