A quantitative risk assessment for chronic traumatic encephalopathy (CTE) in football: How public health science evaluates evidence

ABSTRACT

How should science and policy interpret the recent finding that 110 of 111 former National Football League (NFL) players had brain pathology known as chronic traumatic encephalopathy (CTE) at autopsy? Some physicians view this (and related epidemiologic and mechanistic evidence) skeptically, emphasizing that the association between repeated head trauma (RHT) and CTE may be artifactual, that this “incidence” is biased by self-selection of players with cognitive or emotional symptoms, and that even if RHT causes CTE, the lesions themselves may be inconsequential. Public health scientists look at this emerging evidence quite differently; in particular, they tend not to fall prey to certain illogical arguments justifying inaction. We present a quantitative risk assessment showing that even accounting for the non-representativeness of the 110 cases, the risk of CTE in the NFL workforce amply meets both parts of the test for “a significant risk of material impairment of health” that would permit the U.S. Occupational Safety and Health Administration to intervene to reduce RHT exposure. We further conclude that according to available evidence, CTE is a public health problem, and that lawyers and physicians need to understand that this conclusion is based on standards of evidence at least as long-standing and robust as their own.

KEYWORDS:

• epidemiology
• OSHA
• quantitative risk assessment
• head trauma
• science-policy
• chronic traumatic encephalopathy (CTE)

Acknowledgments

The first author thanks Chris Deubert and I. Glenn Cohen (both formerly with the Harvard Football Players Study), John Evans, and Sidney Shapiro for helpful comments during the drafting of this article; we also appreciate the improvements suggested by two anonymous reviewers.

Funding

Dr. Bieniek is supported by the Florida Department of Health, Ed and Ethel Moore Alzheimer's Disease Biomedical Research Program (7AZ08) and the Mayo Clinic Younkin Scholars Program on Synaptic Biology and Memory. This publication was in part made possible by CTSA Grant Number UL1 TR002377 from the National Center for Advancing Translational Sciences (NCATS), a component of the NIH. Its contents are solely the responsibility of the authors and do not necessarily represent the official view of NIH.

Notes

¹ Note that we array these questions according to the time-honored paradigm first codified in the early 1980s (National Research Council 1983), in which quantitative risk assessment begins with the “hazard identification” stage (what health effect(s) can the substance or agent induce?), proceeds in parallel through exposure assessment (how much exposure to the hazard do humans undergo?) and dose–response assessment (what is the function relating exposure to probability-of-harm?), and concludes with the “risk characterization” stage (how large and how severe are the consequences of the population's exposures?).

² As a reviewer pointed out, this observation is both unremarkable (NFL players surely benefit from the “healthy worker effect” and are expected to live longer than average merely because they were once fit enough to make an NFL team) and incomplete (it says nothing about any quality-of-life differences between NFL players and the general population).

³ For a recent review article discussing the evidence linking RHT to CTE as well as some of the possible weaknesses in the current knowledge base, see Asken et al. (2017).

⁴ In this case, it may be that the most parsimonious division lies between physicians whose specialty is not neuropathology (e.g., concussion specialists, psychiatrists, orthopedists) and those who specialize in neuropathology and whose approach to evaluating evidence may tend to be similar to the public health orientation.

⁵ For a prominent example of this, consider the case of smoking and lung cancer: 54 years after the landmark Surgeon General's Report (US Dept of HEW 1964), we still do not know with certainty which constituent(s) of tobacco smoke cause lung cancer, which target(s) in the genome are affected, the molecular mechanism by which any constituent exerts its toxic effect, whether genetic correlates place specific individuals at much higher or lower risk per unit of exposure, or whether any individual smoker, no matter how intense her exposure, got her disease there or from something else. All science can say is that it's reasonable to believe that widespread reductions in smoking have been and would be associated with widespread reductions in lung cancer, even though one could also rightly say that “the science of tobacco and lung cancer is still in its infancy.”

⁶ The article ultimately recommended, however, that various “soft law” mechanisms would likely be more useful and constructive here, particularly the model of an “enforceable partnership” wherein OSHA, the NFL, and the players' union would jointly craft a code of conduct to reduce RHT. In such a model, OSHA would disclaim any ambition to write and enforce a regulation, but the private-sector parties would in turn agree that adhering to the code of conduct would constitute the employers' “general duty” under the OSHA statute and that OSHA could issue citations for failure to perform that duty.

⁷ There are two fundamentally complementary types of epidemiologic studies. “Cohort studies” compare persons exposed versus unexposed to a putative hazard, to see if the incidence of a disease is greater in the former group. “Case-control studies” look at persons with a disease and a matched control group of healthy persons, to see if the former group is made up of persons more likely to have been exposed to the hazard. Since CTE can only be diagnosed definitively at autopsy, not enough time has elapsed for any true cohort studies to have been mounted – the best epidemiologic evidence at present comes from studies like Bieniek et al. (2015) (CTE incidence in subgroups with and without prior RHT exposure).

⁸ Several final OSHA health standards, including those for methylene chloride (62 Federal Register, 1494–1619), methylenedi-aniline (57 Federal Register, 35,631), and ethylene oxide (49 Federal Register, 25,734), are based solely or primarily on animal bioassay data, with little or no human evidence.

⁹ Confounding is probably the most important flaw affecting epidemiologic studies. However, true confounding is not as common as laypeople may assume, based on how frequently this criticism is invoked. For a factor to be a true confounder, it must be associated both with exposure and with effect. For example, smoking can only create a false-positive relationship between a chemical exposure and lung cancer if smoking is associated with lung cancer (which it certainly is), and if for some reason smoking is significantly more common in those exposed to the chemical compared to those unexposed (or those more highly versus those less highly exposed).

¹⁰ Note that the seven workers were exposed to DBCP at approximately 0.3 parts per million; the OSHA standard required employers to lower concentrations to below one part per billion.

¹¹ In one of the most renowned examples of vindicated precautionary action outside the workplace, the FDA delayed approval of the sedative thalidomide circa 1960, largely on the basis of a very limited case report from an Australian obstetrician who delivered three babies with phocomelia (a rare constellation of birth defects) and recognized maternal ingestion of thalidomide as the common factor. Very few U.S. babies were affected, in contrast to tens of thousands in Germany and the U.K. where the case reports were not heeded.

¹² Section 6(b)(5) of the Occupational Safety and Health Act of 1970 (P.L. 91–596) gives OSHA the authority to set standards only when employees are in danger of suffering “material impairment of health or functional capacity.”

¹³ See, e.g., the newly promulgated final rule for exposure to respirable crystalline silica, where OSHA determined that the mildest possible abnormality on a chest X-ray (a divided grade of “1/0” for small opacities) would be counted as silicosis, even though at this point in the disease process the employee would have no symptoms whatsoever (81 Fed. Reg. 16826, 16837 (March 25, 2016). See also 82 Fed. Reg. 2470, 2547 (Jan. 9, 2017) promulgating a final rule for exposure to beryllium, in which OSHA concludes that early-stage chronic beryllium disease (CBD), an asymptomatic period during which small lesions and inflammation appear in the lungs, is material impairment because of “evidence and expert testimony that early-stage CBD is a measurable change in an individual's state of health that, with and sometimes without continued exposure, can progress to symptomatic disease” (emphasis added).

¹⁴ AD also involves aggregation of hyperphosphorylated tau, but because its primary hallmark is the accumulation of fibrils of amyloid protein within plaques, it is considered a “secondary tauopathy.” CTE per se (when not co-existing with AD) does not display amyloid accumulation.

¹⁵ But see the discussion below of the subsequent paper by Gao et al. (2017), which may provide the first case report of CTE in one individual with no history of head trauma. Also see the discussion below of the findings by Tagge et al. (2018) suggesting that a single severe impact can lead to pathologic changes consistent with early stages along the pathway to CTE.

¹⁶ However, it would not be unprecedented even if none of the rodent studies eliciting head trauma could reproduce the changes seen in human brains; although rodents are considered excellent models for chemical carcinogenesis, for example, one of the most notorious human carcinogens (tobacco smoke) was considered a “negative rodent carcinogen” for many decades. See this quote from Coggins 2007 (at p. 331): “Until recently, the published literature on inhalation studies with laboratory animals and cigarette smoke consisted entirely of negative findings, as far as neoplastic disease is concerned.”

¹⁷ In re Nat'l Football League 2015: “Beyond identifying the existence of abnormal tau protein in a person's brain, researchers know very little about CTE.”

¹⁸ Indeed, Montenigro et al. found a strictly monotonic dose-response relationship for six of six different adverse outcomes, as a function of six different levels of estimated lifetime G-forces sustained in sport impacts. A public health regulatory agency would consider the consistency of these demonstrations to be further evidence of a true association between impacts and symptoms.

¹⁹ On the other hand, at least one recent study (Hay et al. 2016) has suggested that although neurocognitive deficits following a single moderate or severe traumatic brain injury have generally been assumed to represent Alzheimer's disease, the pathology in some of these patients may in fact more closely resemble CTE. Therefore, the “linear above threshold” function posited by Montenigro et al. may not apply when a single impact with substantial force is involved (there may be a threshold in terms of cumulative G-force, but perhaps not if the exposure is measured in number of impacts).

²⁰ Gao et al. (2017) wrote that “To date, repetitive traumatic brain injury has been shown to be associated with no neuropathological changes, with CTE alone, with CTE and another neurodegenerative disease, or with non-CTE neurodegeneration.” But this is no different from saying “To date, smoking has been shown to be associated with perfect health, with NSCLC alone, with NSCLC plus bladder cancer, and with bladder cancer alone.” In neither case does the existence of individuals who fit any of these three patterns cast any doubt on the association at the population level between exposure (to RHT or tobacco) and a specific lesion (CTE or NSCLC).

²¹ More importantly, even if there is a genetic component, public health agencies never abandon their interest in exposure while awaiting word about the genetic contribution. Public health regulatory agencies like OSHA cannot influence people's genetics, but sometimes can influence their exposures. Therefore, they bear in mind the tenet that “genetics may load the gun, but environment pulls the trigger” (this tenet was apparently originally coined by Prof. Judith Stern of the University of California at Davis, circa 2000) – their mission is to explore cost-effective ways to reduce the population prevalence of disease by controlling one or more factors that will do so, regardless of whether other factors might also do so. Even where science strongly suspects that persons with a genetic variant are at much higher risk per unit of exposure than others are (e.g., for chronic beryllium disease), regulatory agencies concentrate on controlling the exposure, which of course is the most direct means of reducing for risk for both those who may be susceptible and those who may be “resistant.” And as a policy matter, OSHA and other agencies even regulate the exposures of both women and men to toxicants suspected of causing damage primarily to the developing fetus (cf. lead); male workers are presumably wholly “resistant” to these effects, but the employer is not permitted to establish male-only areas of the establishment where exposures can be higher.

²² This point was made particularly well by eminent epidemiologist Sir Richard Peto in 1977: “It is a common misconception that because many smokers do not develop lung cancer there must be constitutional or environmental reasons why they do not… Even in A.D. 3000 when all the details of cellular susceptibility and environmental and metabolic peculiarity have been elucidated, a complete and full description of the process of cancer induction will still require that good and bad luck be invoked to explain why my brother got cancer and I did not… There will still be unexplained stochastic variation which it will be scientifically worthless to investigate, in exactly the sense that it would be scientifically worthless to ask, once the molecular basis of Mendel's laws is properly understood, what was the reason why a particular child born to two parents each with one recessive gene for red hair did not, in fact, have red hair.” (emphasis added)

²³ See the 1991 decision in UAW, et al., v. Johnson Controls Inc. (499 U.S.187), which upheld OSHA's right to require lead exposure reductions that protected pregnant women as well as all other workers, on the grounds that excluding women from high-lead areas was not an acceptable means of control.

²⁴ In cases where genetic susceptibility may play a major role, as exemplified by chronic beryllium disease, observational studies sometimes find higher incidences of disease among subgroups exposed to lower exposures, presumably because these subgroups were systematically (or randomly) “enriched” with more susceptible individuals. So applying the margin-of-safety concept to a subgroup with a measurable increase in disease may fail to protect those exposed to lower “safe” levels, because these levels may in fact still be capable of yielding unacceptably high incidences among susceptible individuals.

²⁵ Although a definitive diagnosis of CTE, based on confirming the characteristic pathology, can only occur post-mortem, this does not mean that the disease cannot be “diagnosed” via the standard judgments and heuristics that define the art of differential diagnosis. Indeed, since along with signs, symptoms, and medical history, there exist in vivo tests for several of the diseases that could explain a patient's condition, negative results on these tests along with clinical judgment can allow for a reasonable diagnosis of CTE during life.

²⁶ Using fully 45 years of player experience is a conservative (biases low) choice, but it is reasonable in light of the fact that the player within the CTE numerator who was probably the first to appear in the league (John Mackey) began his career in 1963, and perhaps the youngest and most recently diagnosed player with CTE (Paul Oliver) entered the League in 2008. In other words, the 110 cases in the numerator all occurred among players who were in the league between 1963 and 2008, although virtually all of the 110 presumably involve players who were active during a shorter window within this 45-year period.

²⁷ We also estimated risk more simply, by observing that at any given time, there are approximately 1000 “regular players” active in the NFL: 22 players on the offensive plus defensive squads, plus perhaps 10 others from the 24 remaining players (46 players are dressed for each game) who garner significant playing time, for each of the 32 teams (32 × 32 = 1024). So if the average career is 7.1 years, there are (1024/7.1) players starting or ending their careers each year, or about 150 players, for a total of 6,490 player-careers of experience over the 45-year period; if there is more “churn” in the NFL (average career of 3.5 years), the total number of player-careers would instead be 13,166. Even using the larger denominator, the incidence of CTE could be no lower than 110/13,166, or 0.0084, which is more than eight times higher than the 1/1000 benchmark (and using the league's estimate of churn, the incidence would be 16.9 per thousand). These figures are very similar to those we derived using player-snap data.

²⁸ A reviewer suggested a very worthwhile alternative calculation: that we try to calculate the proportion of deaths in which CTE was found at autopsy versus the total number of deaths from all causes among the relevant subgroup of NFL players (namely, all players who entered the League between 1963 and 2008). The calculations in the main text estimate CTE deaths among all players, the vast majority of whom are still living and who may therefore add to the numerator, so the alternative presented in this note assumes that when all of these players eventually die, the proportion of them who will have CTE at autopsy would be the same as the proportion found to date. This likely introduces some unknown amount of upward bias, in that CTE very likely caused many of the 110 documented cases to die before their normal life expectancy, “enriching” this subgroup with CTE cases. According to Nixon (2014), 14,658 players entered the League between 1963 and 2008; of these, 711 had died as of June 2014. Therefore, 15.5% (110/711) of the deaths in this cohort involved CTE cases. This corresponds to a lifetime risk (in the sense of “what will your cause of death eventually be?”) of 155 per 1000, which is many-fold higher even than the substantial working-lifetime estimates presented in the main text. Note that the larger figure of 1300 deceased NFL players cited in Ward, Williams, and Manchester (2017) is not the proper reference group, since many of these players were active long before 1963 and were never suspected of having CTE or autopsied – they are part of the denominator but not the numerator.

²⁹ In addition, Ling et al. note that even some of the cases without documented head trauma may have experienced trauma, since one of the major diseases selected for in this brain bank (progressive supranuclear palsy) leads to “early and frequent falls” that themselves could have caused CTE pathology.

³⁰ In addition, the pathology Ling et al. depict in their manuscript is not characterized according to consensus standards for describing CTE, in that the authors only illustrate perivascular glial pathology that in and of itself is not specific to CTE and can be found in the elderly (in the elderly it is known as “aging-related tau astrogliopathy,” or “ARTAG”; see Kovacs et al. 2016). The mean age of death for the “CTE” cases in Ling et al. is 81, which is much older than other CTE studies and raises the issue of whether they are not measuring CTE but rather ARTAG.

³¹ It is also not completely clear whether OSHA can, or in the future could, consider collegiate student-athletes as “employees” for the purposes of OSHA jurisdiction (see Willborn 2014–2015, who makes a case that they are employees for the purposes of the National Labor Relations Act). In that case, the proper excess risk estimate might construe risk as above the background rate among those unexposed to RHT before college.

³² OSHA might well be especially concerned that behavioral symptoms (especially violence towards oneself or others) possibly associated with CTE may put others at risk. OSHA has in the past paid special attention to risk factors that can leave the workplace and be “taken home,” particularly certain toxic substances like lead, beryllium, and asbestos that can increase risks among cohabitants (OSHA 2014).

³³ In 2012, OSHA revised its 20-year-old Hazard Communication Standard to align with the Globally Harmonized System (GHS) developed by the United Nations. The GHS requires that containers of known human carcinogens have warning labels on them stating “Danger: May Cause Cancer.” This is a true but inane statement, as it replaces information about something that definitely causes cancer in people with the reassuring information that it only “may” cause cancer in you, the individual reader of the label. There are virtually no substances, even at extremely high levels of exposure, that “will” cause cancer in every single person so exposed, so the “may” information abruptly changes a playing field that U.S. companies had understood for decades from a public health warning to a clinical warning.

³⁴ For example, a person might have to drink the proverbial “case of diet soda per day” to match the exposure of rodents in toxicology studies, but if the cancer incidence among the rodents is 50%, an exposure 1/1000 of that is still a very large risk (assuming the dose-response can be modeled as linear in this range) (Finkel 1994).