Doubt: How Industry's Assault on Science Threatens Your Health - Chapter 6 Notes
Tricks of the Trade: How Mercenary Scientists Mislead You
The regulatory systems should use the best evidence available at the present time, not wait for absolute certainty.
Industry exploits the nature of science, where knowledge is accumulated over time and understanding evolves.
Scientists seek the truth but do not deal in absolute certainties, instead relying on the "weight of the evidence".
Human disease has complex causes. Scientists cannot feed toxic chemicals to humans, so they use "natural experiments" and animals, extrapolating from the evidence to recommend protective measures.
In public health, absolute certainty is rare because regulatory questions can only be answered imperfectly.
Asbestos causes lung cancer even at low levels, but the precise risk associated with every exposure cannot be stated with certainty.
The best available evidence must be sufficient for regulatory programs to be effective.
There is a growing trend of demanding proof over precaution in public health.
Environmental activists can also use scientific uncertainty to advance policy aims through an overzealous application of "the precautionary principle."
Demanding assurance that a policy will result in no harm can delay scientific advances and public health interventions.
Food irradiation is an example of a technology that may genuinely improve the human condition but has been disparaged and delayed.
Epidemiology and Uncertainty
Epidemiology is susceptible to uncertainty campaigns because large epidemiological studies require complex statistical analysis.
Judgment is required throughout the process, so disciplined integrity is mandatory in epidemiology.
Epidemiology is more likely to produce false negative results than false positive ones.
Epidemiologists cannot state that a specific chemical exposure has definitely caused cancer in a specific patient, except in rare instances like mesothelioma caused by asbestos.
Epidemiology establishes probabilities that reliably pertain to a given population.
Environmental Exposures and Worker Studies
Much of what we know about the toxic effects of environmental exposures comes from studies of workers.
Workers often make up the discrete population with the greatest exposure to a given chemical.
Studying health effects among workers is often preferable to isolating those effects in the general population.
Studies of embalmers, mortuary workers, pathologists, and anatomists find a higher than expected risk of leukemia as a result of formaldehyde exposure.
Studies of workers in the rubber and shoe industries, who have been exposed to higher levels of benzene, find higher than expected rates of leukemia, which can be extrapolated to other populations with lower exposures.
Studies of distinct worker populations are called "historical prospective" or retrospective cohort studies.
Cohort studies identify a population in the past, follow them forward in time, and examine the illnesses they developed or died from.
Cohort studies can have inaccurate knowledge of the study subjects' exposure histories, so crude estimations must be made.
Epidemiologists analyze the results of "natural experiments" that have occurred in the real world, such as the long-term effects of radiation exposure to humans from studies of the survivors of Hiroshima and Nagasaki.
Study Duration and Nefarious Intent
Well-designed workplace studies require populations that are followed for at least twenty years, preferably thirty or more, because the cancers that most chemicals cause usually require such long periods of time to show up.
For most chemicals, cohort mortality studies that examine workers whose exposures began less than twenty years earlier will not show an effect, so it is reasonable to suspect a nefarious reason for conducting such a study.
Data Collection and Analysis
Thirty-year mortality studies do not actually require thirty years to conduct. Epidemiologists use extant historical records to reconstruct those years.
Epidemiologists use tracking systems, including Social Security data and the National Death Index, to determine who is alive and who is dead at the end of the period under study, when they died, and the cause of death.
Industrial hygienists construct job-exposure matrices, which assign estimated exposure levels for the chemical in question to different job titles and different locations within the workplace.
The work history of each participant is plugged into this matrix to derive a rough estimate of individual historical exposure.
Disease rates among study participants with different exposure histories are compared with those that would be expected of members of the general public.
Death is an extreme outcome, but often the only one for which data are available.
Mortality data are less helpful when studying diseases that are not likely to cause death, such as bladder cancer.
Cancer registries and reports to state health departments can provide data on illnesses in the United States.
A study in a state with a cancer registry can examine the incidence of a particular cancer; otherwise, that study will be restricted to mortality.
Study Size and Comparison
The larger the study, the better.
Ideal is the comparison of workers who have high exposures to a given chemical with workers in the same or a similar facility with lower or no exposure.
More often, epidemiologists compare the mortality experience of a worker cohort with a standard population.
Standardized mortality ratio (SMR) is the ratio of actual deaths with expected deaths. If there are twice as many actual as expected deaths, the SMR equals 2, suggesting that these workers have twice the risk of dying from that cause.
Small studies are inherently suspect because they do not have the statistical power to detect a real increase in disease risk in populations.
Competent epidemiologists use the number of people and the age distribution to determine the power of a study in advance.
Industry-commissioned mortality studies may look at a relatively small group of workers for a relatively short period of time and fail to find any negative impact.
Industry scientists sometimes add large numbers of unexposed workers to a study population, but their inclusion is largely uninformative and may help mask real effects.
Study Quality and Bias
A poorly conducted study is more likely to result in a false negative than in a false positive.
For the results from a negative study to be taken seriously, the study must be large and sensitive and gather accurate exposure data.
Even when a study is large enough, covers a sufficient period of time, and has access to a cornucopia of exposure data, other factors can still undermine epidemiologists.
Selection bias occurs when the worker cohort under study is not representative of the general population from which it comes and with which it will be compared.
The most common selection bias is the "healthy worker effect," which reflects the fact that the worker population was "selected" because it was healthier to begin with.
Almost every study that compares the death rate of workers to that of a geographically appropriate comparison population finds that the workers have the lower overall risk.
Information bias, which can take many forms, although the most common is the misclassification of exposure estimates.
Exposure misclassification results in a lower degree of risk than in fact exists.
Systematic Error and Dilution
Systematic error related to exposure misclassification is the more simple effect of dilution.
This results when we do not have good exposure information and groups of workers with different exposures are lumped together.
When a small group of heavily exposed workers is diluted in a large group of other, less heavily exposed workers, a large excess for the heavily exposed workers can seem smaller than it actually is or even disappear entirely.
Confounding
Confounding is the existence of a factor that is related to both the disease and the factor under investigation.
Industry uses confounding to blame confirmed health risks on an unaccounted-for confounder, such as smoking.
Smoking is nearly impossible to account for precisely.
In order to be a confounder, the most highly occupationally exposed workers in a population would have to be the heaviest smokers (an unlikely scenario).
In some cases cigarettes are not confounders at all; asbestos workers who also smoke have lung cancer rates far higher than either nonsmoking asbestos workers or smokers with no asbestos exposure at all.
If two chemicals are present in the same plant, the estimated effects of one may be confounded by that of the other.
Judgement Calls and Actual Results
Scientists use information they import from other sorts of studies, particularly animal studies, to make judgments.
Statistical tests help decide whether the findings more likely reflect a true causal relationship or just a chance finding.
With some signature diseases, the occurrence of just a few cases in one place is enough to establish a problem.
Identifying instances of "too many cases" is not nearly so easy with the more common diseases or causes of death.
The identification of a genuinely increased risk depends on the size of the increase and that of the population under study.
If exposure to a given chemical triples the risk of leukemia, three leukemia cases in a cohort of 100 workers in which only one case would be expected would not likely be statistically significant.
There is a chance distribution as the cause of the two excess cases.
If the population is 1,000 workers, not 100, and we find thirty cases instead of the expected ten, it is very unlikely that the excess would be attributable to chance.
Animal Studies
Scientists have been exposing animals to toxic products to predict what will happen when humans are exposed to the same substances.
All mammals have similar tissues, organs, and biochemical systems.
Bad news for a lab rat is bad news for all other mammals, including us.
Animal studies can help explain the results of the "natural experiments" that epidemiologists study.
They can also predict whether substances that we cannot study epidemiologically might cause cancer in humans.
Animal studies can help answer questions about the effects of different exposure levels, different exposure times, the interaction of multiple exposures, or the effects on the very young, the very old, the fetus, and new chemicals.
They cannot ethically subject humans, even informed volunteers, to doses of known carcinogens and powerful toxins.
The ability of a carcinogen to produce cancer at low levels of exposure was confirmed in the fabulously expensive mega-mouse and mega-rat experiments.
There is no threshold, no minimum dose, required to induce cancer.
There is a dose-response relationship, which shows that the risk of disease increases as the exposure increases.
Cancer may require many years and even decades to develop in humans, which is much longer than the natural lifespan of most of the small mammals used in toxicology studies.
Instead, we use a smaller number of lab animals and give them large doses, knowing that a substance that does not cause cancer does not cause cancer, period, not even at the highest doses.
Defenders of a substance that has been found to be carcinogenic in animal studies may offer as an excuse the fact that the exposure was far more than a human would ever confront.
Interpretation and Synthesis of Findings
Public health and environmental protections are based not on the results of individual epidemiological or animal studies but rather on an interpretation or synthesis of the findings of multiple studies and multiple types of studies.
Experts look for the weight of the evidence and attempt to synthesize the entire picture, then make a pronouncement about causation or risk based primarily on the studies to which they have accorded more weight.
"Weight of the evidence is a subjective approach."
Other approaches to data synthesis involve combining the results of several studies to provide numerical risk estimates.
Scientists and regulators are drawn to these methods since they provide the illusion of precision, but the reality is that the results of these studies are also shaped by the assumptions and beliefs of the investigator.
Meta-Analysis
A meta-analysis is a study in which the results of several similar studies are combined to provide a result that should have more statistical power because it includes far more study subjects than any of its component studies.
Meta-analyses can be useful when based on well-designed smaller studies, none of which would be large enough to detect a small effect by itself.
Meta-analyses are susceptible to the "garbage in/garbage out" principle.
Model Building
Model building is another quantitative approach to data synthesis.
One particular mathematical model is the "risk assessment," which is based on a combination of data and assumptions, and it is has become the coin of the regulatory realm.
Some risk measurements are relatively straightforward exercises that use information on the known health effects associated with higher exposure levels to predict the effects at lower exposure levels.
In the absence of powerful epidemiologic studies, risk assessments that attempt to measure the effects of chemical exposures are by necessity more complex, more opaque, and, as a result, more controversial.
The devil here is definitely in the details.
Risk Assessment Exercise
In a 1991 exercise on risk assessment conducted by the Commission of European Communities, eleven teams of scientists and engineers estimated the accident risk at a hypothetical small ammonia storage plant. The risk estimates for an accident ranged from 1 in 400 to 1 in 10 million.
Manipulation of Risk Assessments
Change a few parameters that are buried deep in a mathematical model, and a hazardous chemical can be miraculously transformed into one that is not very dangerous at all.
William Ruckelshaus: "Risk assessment data can be like a captured spy: if you torture it long enough, it will tell you anything you want to know."
Case Study: Benzene
Industry can take advantage of the inherent uncertainty in epidemiological studies in order to forestall regulatory action.
Benzene is a very important chemical even though it has in many uses been replaced as a solvent by less toxic substances.
It is one of the contaminants at the majority of the nation's toxic waste sites.
Benzene is also a constituent of gasoline and a product of combustion.
Benzene exposure can cause life-threatening aplastic anemia, and even low levels cause leukemia.
In 1948 the American Petroleum Institute's "API Toxicological Review of Benzene" discussed "reasonably well documented instances of the development of leukemia as a result of chronic benzene exposure."
The report concluded that "it is generally considered that the only absolutely safe concentration for benzene is zero."
In 1973 Dr. Robert Eckardt wrote, "[The] accumulation in the literature of cases of leukemia following benzene exposure leads to the inevitable conclusion that benzene is a leukemogenic agent."
NIOSH Study of Benzene
NIOSH was able to employ the emerging state-of-the-art tools of epidemiology in industrial settings.
NIOSH scientists had their pick of America's industrial facilities for factories with one predominant chemical exposure, minimal confounding factors, a stable workforce, and good records.
Two Goodyear Tire and Rubber plants in Ohio fit the bill.
The main product of these plants was synthetic rubber, with benzene as the dominant chemical in the production process.
This landmark NIOSH study of twelve hundred workers quantified the leukemia risk and found a doubling of risk among workers exposed for up to four years, a fourteenfold excess risk among those exposed from five to nine years, and a thirty-threefold increase in those exposed for at least ten years.
Published in 1977, the NIOSH study was a major factor in OSHA's decision to lower the eight-hour average exposure standard for benzene workers from 10 ppm to 1 ppm.
Immediately challenged by the industry, the standard was set aside by the Supreme Court, which ruled that OSHA had not shown that its standard would achieve substantial reduction in risk.
This ruling established the new standard for all OSHA regulations.
The results of the revised NIOSH study confirmed those of the original research and found increased risk of leukemia ranging from 1 ppm to as high as sixtyfold for the highest exposure levels.
In 1987 OSHA reissued the new exposure standard of 1 ppm.
Industry Attacks on NIOSH Study
The oil industry spent tens of millions of dollars to cast doubt on the NIOSH study with a series of analyses and reanalyses.
To this day, the oil industry is still spending major money to attack the NIOSH epidemiology.
This entire incident constitutes a textbook example of some of the tricks of the trade, as well as the uses and misuses of epidemiology in the regulatory arena and in litigation.
The "divide and conquer" strategy looks at smaller units in order to find differences among them, then uses these differences to cast general doubt on the overall results.
Oil companies have produced epidemiological studies on other, less heavily exposed worker populations their own.
These studies were essentially guaranteed not to be as informative as the NIOSH study of the Goodyear workers because they were diluted.
Diluted studies commonly show little risk effect, especially when dealing with a disease such as leukemia, which is not a very common cause of death to begin with.
The oil studies were both diluted and underpowered.
putting together numerous diluted cohorts yielded only a much larger but still very diluted cohort-lots of people, but few with significant exposure.
Guaranteed result: no excess leukemia.
Low Level Exposure Claims
One last-ditch recourse is to claim that the disease effect is real only at the highest levels, while lower levels yield no increased risk.
The oil companies commissioned a slew of analyses that claimed to detect a threshold, or safe level, for benzene exposure.
It's a game, and everyone knows it, but OSHA must, by law, analyze the proffered studies, file answers, analyze the answers to the answers, and so on ad infinitum.
You selectively remove or censor cases, thereby turning a positive result excess risk-into a negative result-no excess risk.
The risk estimate is a fraction in which the numerator is the number of disease cases and the denominator is the study population.
Lowering the numerator by a fairly small number can make all the difference.
Scientists who were working for Shell Oil and British Petroleum reanalyzed the NIOSH study and looked for disease cases that could be eliminated from the numerator.
Paustenbach Reanalysis
The American Petroleum Institute brought in ChemRisk's Dennis Paustenbach to reanalyze the landmark work yet again.
Dr. Paustenbach and his colleagues changed the exposure estimates, replacing the original ones with worst-case scenarios and eventually assuming the occurrence of levels so high that one would assume that such extreme exposures would produce an epidemic of serious benzene poisonings, which had never occurred.
The new exposure estimates successfully shifted the dose-response curve.
mercenary reanalyses and critiques would never be enough to roll back the OSHA standard.
The oil industry needs to be ready should OSHA decide to lower the standard to 0.5 ppm or even 0.1 ppm.
For the moment this is a highly unlikely prospect, however; the more immediate industry concern is the EPA.
Production and use of gasoline inevitably leads to the release of benzene-and if benzene were shown to cause disease at ultralow exposure levels, the EPA could force the oil companies to spend huge sums of money to reduce emissions.
Studies of no value whatsoever in the regulatory arena can be quite valuable for corporate defendants in the courtroom.
A jury might be impressed by a one-hundred-page peer-reviewed article that claims that all of the government science must be wrong, must be junk science, whereas the industry's own sound science proves that benzene did not cause this individual's leukemia.
ChemRisk Assessment
ChemRisk produced estimates that the exposures generally did not exceed the old OSHA standard.
One is their excessive length, and it’s target audience is not scientists but jurors and judges, who may be favorably impressed by the excessive length.
Chinese Studies
China offers a natural experiment that epidemiologists dream of: a growing industrial sector with a wide range of exposure levels for benzene, easy access to the factories, extensive exposure monitoring, and few workers lost to follow-up.
Early results from this study document a doubling of the leukemia risk at exposure levels averaging only 10 ppm.
Additional results from China documented blood disease altered white blood cell and platelet counts in workers exposed to benzene levels below 1 ppm.
These results should put an end to the claims of the American Petroleum Institute.
The stunning Chinese result also provides further evidence that the existing OSHA standard of 1 ppm is not sufficiently protective and raises the question of whether there is any safe level of exposure to benzene at all.
Oil industry product defense consultants have published papers criticizing the China study. The criticisms were not very convincing, so the industry has taken the bolder step of producing its own series of studies in China.
The planned research is expected to:
Provide strong scientific support for the lack of a risk of leukemia or other hematological disease at current ambient benzene concentrations to the general population.
Establish that adherence to current occupational exposure limits (in the range of 1-5 ppm) do [sic] not create a significant risk to workers exposed to benzene.
Refute the allegation that Non-Hodgkins lymphoma can be induced by benzene exposure.
How could the sponsors know what the research is expected to find? "This study will have fewer benzene-exposed workers than the original China study, so the effects at low exposure levels will be harder to see than those at higher levels."
Scientists and their sponsors will wave their results and say, "Look, we did our own study in China and didn't find the same effect."
They only need to manufacture some uncertainty by raising questions about the accuracy and validity of the studies that do find an effect.
Defending Secondhand Smoke
Industry communications must contain less of "But that study is wrong" and more of "Look what this study shows."
The industry could not produce that new and better science because the consensus science was correct.
New headache for the industry included passive smoking, involuntary smoking environmental tobacco smoke or ETS, secondhand smoke. Such pollutant was becoming a veritable migraine for the industry because it necessarily caught the attention of both EPA and OSHA.
A 1978 industry report warned that a campaign by antismoking forces targeting secondhand smoke would be the most dangerous development to the viability of the tobacco industry that has yet occurred. In 1981 the first important epidemiologic study was published showing that nonsmoking women whose spouses smoke have a higher rate of lung cancer than those married to nonsmokers.
On the primary issue of smoking per se, it could still try to hide behind the "personal choice" defense, but this would not work with secondhand smoke.
Internal tobacco industry documents attributed as much as 21 percent of the geographic variation in cigarette consumption to public smoking restrictions.
The Whitecoat Project
Europe, where smoking is more prevalent, smokers are more tolerated by nonsmokers, and workplace restrictions are loosely enforced, the industry set up the Whitecoat Project as a way to get some friendlier science in front of the public and restore smoker confidence.
Dr. Myron Weinberg, founder of the product defense firm the Weinberg Group, was brought on board.
scientists were asked whether they were interested in problems of "indoor air quality". Lawyers were involved in this vetting.
*Philip Morris [would] then expect the group of scientists to operate within the confines of decisions taken by PM scientists to determine the general direction of research, which apparently would then be 'filtered' by lawyers to eliminate areas of sensitivity. Their idea is that the groups of scientists should be able to produce research or stimulate controversy."
For every PhD there is probably an equal and opposite PhD somewhere.
The credibility Problems
The industry's white coats were not nearly so white and bright and freshly starched as those sitting on the EPA's Scientific Advisory Board.
The industry felt it had a better chance fighting the scientific evidence against secondhand smoke; scientists who opposed the industry on the basic smoking question might be more willing to join them on this one. The industry argued that its studies were the target of “publication bias,” banned from frontline journals with an antismoking stance.
Law Suits Against Big Tobacco
The litigation against the asbestos companies was also in full swing. RJR scientists recognized that epidemiologic studies of workers exposed to asbestos and other environmental toxins could be used to shift the blame back in that direction, and the company went to court and successfully demanded the raw data underlying the famous asbestos studies.
Integrated Exposure and Hazard Assessment Initiative
Shift a portionately [sic] higher amount of risk (maybe all) to the asbestos defendants, particularly if plaintiff's asbestos is high; or alternatively, if smoking dose is low…. An example would be a case where the plaintiff's lung cancer is more likely to have arisen in another tissue and metastasized to the lung. In this case, every effort should be made to eliminate, or drive as low as possible, asbestos exposure since current evidence suggests that asbestos tumors arise principally in the lung, be they classic or mesothelioma. By contrast, if plaintiff's cancer is clearly primary to the lung, it is imperative that every effort be made to maximize occupational exposure not only to asbestos but also to other agents in the workplace."
Dr. Michael Ginevan to "develop a rationale for attribution of a greater proportion of lung cancer to radiation than heretofore claimed."
Action on Smoke and Health (ASH)
OSHA denied the petitions, with backing from some within the agency who believed that, even though indoor air quality standards were warranted, it was politically impossible to put them in place and the ill-fated effort to do so would be an incredible drain on the agency's limited resources.
The court sided with the agency.
In March 1992, when the AFL-CIO petitioned OSHA to issue an overall indoor air quality standard, OSHA said it would consider the matter a good way to deal with the issue in an election year.
Attacking indoor air quality rule
Big Tobacco saw this action as potentially do or die; it was absolutely imperative to discredit the link between workplace secondhand smoke and disease.24 A July 1994 document described Philip Morris's plan to convert the promulgation process from bureaucratic fiat to political dogfight. Over the next month, if we have anything to do with it, this opposition is going to intensify and we're going to give the poobahs at OSHA a taste of what democracy is really like.
Dr. Roth may be the premier, all-purpose, pro-industry reanalyzer.
Procalimations by H. Daniel Roth
*OSHA neglected to include some published studies in its analyses.
*OSHA misrepresented the findings in other studies to suggest an ETS effect in cases where no such effect is indicated.
*OSHA failed to recognize that many of the studies it cited are of poor quality.
*OSHA failed to recognize that many of the studies it relied upon failed to adjust for confounding factors, an omission which in all likelihood led the Administration to overestimate the effects of ETS.
*OSHA failed to correct for the tendency of individuals to mischaracterize the smoking habits of their household members and coworkers.
*OSHA presented no scientifically defensible calculations to support its findings.
*OSHA failed to test whether the data from different studies are homogenous and could be aggregated to analysis.
Ancillary benefit of any uncertainty campaign is that it is guaranteed to buy some time. This point was well made in a conference call involving Philip Morris executives, their lawyers, and Myron Weinberg. Overwhelm OSHA.
*The uncertainty campaign worked.24 The tobacco industry's well-funded strategy “to put the bureaucratic machinery on overload” stymied OSHA's efforts
Enviromental protection Agency (EPA) and the tobacco Industry
The EPA was categorized secondhand smoke as a Group A carcinogen-a chemical that causes cancer in humans.
One industry document argued that the industry could not win a 'credibility fight' with the EPA so don't even try.
Acting on this advice, the industry tried to enlist as many other regulated industries as possible to front for the cause in the name of “sound science."
Center for Indoor Air Research
CIAR claimed that it could prove that the case against secondhand smoke was flawed.
How I read that contorted rhetoric: CIAR doesn't influence studies because CIAR selects only those studies that won't need the influence.
The manipulation and selective publication of data by Big Tobacco's scientists resulted in a distortion of the literature now widely known as the “funding effect, a term used to describe the close correlation between the results desired by a study's sponsors and the results reported.”The CIAR had a special program to support, publish, and promote studies that found secondhand smoke harmless.41 When researchers at the University of California examined 106 review articles, they found more than a third concluded that secondhand smoke was not harmful. Three-quarters of these dissenting reviews had authors who were affiliated with the tobacco industry.*
The early studies investigated the risk among nonsmoking spouses of smokers, specifically. The study by Takeshi Hirayama, chief epidemiologist of the National Cancer Center Research Institute in Tokyo, was the most prominent of these. If it could be discredited, some of this regulatory mess might disappear, so the industry went after Dr. Hirayama's work.
Attacking Hirayama's Study
This study was conceived and supported by the cigarette makers, but by working through Covington and Burling, a prominent Washington, D.C., law firm, they were able to conceal their intimate involvement in every aspect of the job.
project was directed by Dr. Michael Ginevan, the same fellow who had worked on radon.
arrived of this new work on secondhand smoke posed potential conflict-of-interest problems at the consulting firm as well, but its scientists badly wanted to work out a suitable arrangement They have made every effort not to deal with other clients that represent a conflict of interest, but in turn, expected to be supported by the [tobacco] industry. In short, they want a role in ETS
Louisiana State University and Fontham studies.
One result of this research was explosive the nonsmoking wives of male smokers had an increased risk of lung cancer of 30 percent but the second was apocalyptic for the industry: Tobacco smoke in workplaces and other locations outside the home increased lung cancer risk by 40-60 percent./
Fontham did not care to watch the industry hirelings twist her results and make her findings disappear; she refused the entreaties of the tobacco companies to give up her data.
Shelby Amendment, which requires all federally supported researchers to give up the their raw data
General Epidemiological Principles (GEP)
together with