ENVIRONMENTAL CLAIMS JOURNAL 2018, VOL. 30, NO. 1, 58-70 https://doi.org/10.1080/10406026.2017.1422660
Routledge
Taylor & Francis Group
3907 Noy
BD) Check tor updates
Protecting Science and Public Health in Communities - PFOA and Possible Cancer Clusters in the Village of Hoosick Falls, Rensselaer County, New York
Robert A. Michaels
RAM TRAC Corporation
ABSTRACT
Making measurable that which is not measurable often is a critically important responsibility of public health professionals, for example to identify environmental causes of cancer clusters as a valuable step toward addressing them. In epidemiology such associations cannot be measured directly, but they can be inferred probabilistically. The appropriate approach therefore is statistical. Its implementation, however, has been flawed due to an intrinsic conflict between protecting public health vs. protecting the body of scientific knowledge. Science typically is protected via application of the 95-percent confidence criterion, whereas public health must be protected via application of the statistically less demanding precautionary principal. Public health professionals can and must serve both masters. One approach to fulfilling this mandate is exemplified here via evaluation of possible associations between PFOA (perfluorooctanoic acid) and the incidence of multiple cancer types in the Village of Hoosick Falls, Rensselaer County, New York.
Introduction
Measure that which is measurable; make measurable that which is not.
Source uncertain, attributed to Galileo (1564 — 1642)
Making measurable that which is not measurable often is a critically important responsibility of public health professionals. Often, for example, public health professionals must identify cancer clusters and determine whether environmental factors are associated causally with them as a first step toward effective risk management. Likewise, public health professionals often must determine whether or not a particular clinical condition in a population of vaccinated individuals constitutes a plausible side effect that was associated causally with their prior vaccination (Michaels 2017b). Thus, they must evaluate associations prospectively to protect
CONTACT Robert A. Michaels @ ram@ramtrac.com; www.ramtrac.com (ex) RAM TRAC Corporation, 3100 Rosendale Road, Schenectady, NY 12309-1510.
© 2018 Taylor & Francis Group, LLC
ENVIRONMENTAL CLAIMS JOURNAL @) 59
public health, and retrospectively to determine whether failure to protect has occurred, as well as to ascribe responsibility for the failure.
Although such associations cannot be measured directly, they can be inferred probabilistically. The appropriate approach therefore is statistical. Its implementation, however, has been flawed due to an intrinsic conflict between protecting public health vs. protecting the body of scientific knowledge. Science typically is protected via application of the dual criteria of replicability of experiments and application of the 95-percent statistical confidence criterion in data analysis. Public health in contrast must be protected via application of the statistically less demanding precautionary principal. From their respective positions, both viewpoints justifiably claim the mantle of ‘conservatism.
Conservatism in science protects the knowledge base from errors introduced by inadequate stringency of experimental technique and of data analysis. Public health professionals, however, simultaneously must be proactively conservative to safeguard life and health. This may require considering associations, such as cancer clusters or vaccine side effects, that probably are real, or even associations that only might be real. For example, public health professionals should not overlook a cluster of cancer cases associated with a carcinogen, or an adverse clinical effect associated with a vaccine, if the statistical confidence level for the association is only 90 percent, because that association actually might prove to be both real and causal.
Courts of law and courts of public and of government agency opinion therefore must balance opposing needs: rejecting unscientific or antiscientific claims vs. accepting claims of injury if ‘probable, even if not proved to a scientific degree of certainty. In statistics, this conflict is intrinsic, between Type 1 vs. Type 2 errors: rejecting a true null hypothesis (for example, a real cancer cluster or vaccine side effect is missed) vs. accepting a false null hypothesis (for example, a statistical fluke is interpreted as a real cancer cluster or a real vaccine side effect). In testing whether an alleged cause of injury ‘probably’ (in legalese, ‘more probably than not’) was the actual cause, courts and government agencies should adhere to scientific principles, even if applying a less stringent statistical confidence criterion than 95 percent.
As scientists, public health professionals can and must serve both masters. One approach to fulfilling this stern mandate is exemplified via evaluation of the possible causality of associations between PFOA (perfluorooctanoic acid) and the incidence of multiple cancer types in the Village of Hoosick Falls, Rensselaer County, New York. The first principle supersedes all technical principles: repudiate politics and ‘political correctness’ in the (happily confluent) interests of both science and public health.
Methods
This assessment is based upon critical examination of available documents relating to the Village of Hoosick Falls including environmental and biomonitoring data and scientific, technical, and regulatory literature. Most specifically, it evaluates the
60 ©) RAMICHAELS
New York State Department of Health cancer cluster study of Hoosick area residents (NYS DOH 2017). Methods applied are generally those of scientific peer review and synthesis of published findings to draw new conclusions.
Findings: NYS DOH cancer cluster analysis
An apparent cluster of cancer deaths was observed among co-workers at the Saint Gobain Performance Plastics Company in the Village of Hoosick Falls, precipitating discovery in the Village's drinking water of PFOA to which the co-workers might have been exposed at its source, occupationally. This contamination and anecdotal cluster of cancers in turn precipitated a NYS DOH health study (NYS DOH 2016c) of possible “unusual elevations” of incidence rates for multiple cancer types among Hoosick area residents. NYS DOH’s study design applied the stringent academic standard of 95-percent confidence for accepting cancer clusters as real, and found multiple elevations (NYS DOH 2017), but none except lung cancer reaching that stringent 95-percent significance criterion (Table 1). NYS DOH dismissed the lung cancer elevation, however, attributing it to smoking rather than to PFOA.
NYS DOH’ health study investigated possible cancer clustering, termed “unusual elevations” (NYS DOH 2016c).! In advance of the study I recommended that its scope be expanded to include non-cancer effects, and that the cancer cluster study be configured to include the following features:
- Adopt a health-protective criterion of statistical significance to trigger further investigation of cancer clusters, rather than simply a 95-percent confidence level, to assure that real clusters will not be interpreted as statistical flukes;
— Consider rare cancers, whose incidence is expected to be zero in the small population of Hoosick Falls (about 5,000), not just more common cancers reported in the New York State Cancer Registry. Reports of multiple types of rare cancers would be expected to be even more unlikely statistically, unless caused by stressors (such as PFOA), and should be considered probabilistically together, not just individually;
- Conduct a prospective health risk assessment to supplement retrospective assessment of the incidence of multiple cancer types. Retrospective observations of elevations of cancer incidence might be statistically insignificant, even in the presence of real cancer causes;
- Incorporate the time dimension into cancer incidence analysis. The 20-year time period of the NYS DOH study, 1995 through 2014, should be expanded
1 NYS DOH (2016c): “The State Health Department will conduct an investigation to see if there are unusual elevations of cancer among Village residents. The investigation will include total cancers and specific types of cancer diagnosed from 1995 through 2012. To accomplish this, the State Health Department will use data from the New York State Department of Health Cancer Registry, which receives reports on all cases of cancer occurring in New York State. The planned start date is 1995 because street address at the date of diagnosis is available in computerized form for all cases starting with that year. The planned end date is 2014 because 2014 is currently the most recent year for which data are available. The State Health Department will produce a report on the findings of the investigation.”
ENVIRONMENTAL CLAIMS JOURNAL © 61
Table 1. Smoking rates vs. lung cancer incidence.
lung cancer?* observed, Village of expected, NYS observed/ Village of Hoosick Hoosick Falls except NYC expected Falls elevation 91 65 14 +40%
observed, Village of expected, adjusted | observed/expected | Village of Hoosick
Hoosick Falls for Rensselaer Falls elevation
County smoking rate
91 74 12 +23.5% prevalence’, Rensselaer] prevalence’, New | Rensselaer County/ | Rensselaer County County York State New York State elevation 717 47.95 15 +23.8% smoking* prevalence, Rensselaer prevalence, New | Rensselaer County/ | Village of Hoosick County York State New York State Falls elevation? (percent) 20.35 17.95 11 +2.4% prevalence, Rensselaer prevalence, New | Rensselaer County/ | Rensselaer County County York State New York State elevation (percent) 20.35 17.95 11 +2.4% lung cancer eleva- Village of Hoosick Falls x16.7 tion/smoking elevation Rensselaer County x9.9
1. per 100,000 population, age-standardized, 2014;
2. Assumes smoking rate in Village of Hoosick Falls is same as smoking rate in Rensselaer County.
3. HealthData.org. County Profile: Rensselaer County, New York; http://www.healthdata.org/sites/default/files/files/cou nty_profiles/US/2015/County_Report_Rensselaer_County_New_York.pdf, 9 pages, access date 8 June 2017;
4. NYS DOH. Cancer Incidence Investigation 1995-2014, Village of Hoosick Falls, Rensselaer County, New York. Albany, New York State Department of Health, http://www.health.ny.gov/hoosick, 18 pages including Appendices, May 2017.
using ancillary data. The longer period should be subdivided into time windows to examine possible trends in the appearance of cancer cases, such as individual years as illustrated in Kulldorff et al. (1998);
— Conduct detailed investigations aimed at attributing observed cancer incidence elevations to specific causes if possible. NYS DOH’s study design failed to state the objective of elucidating the cause(s) of “unusual elevations” that might be found; and
- Consider anecdotal and other ancillary data, such as documented cancers among co-workers or among pets living in the same household. Animals may be more intimately associated than people with the water, soil, and biota in their outdoor environment. NYS DOH failed to explicate that such potentially probative anecdotal data would be sought and used.
62 ©) RA.MICHAELS
I had evaluated NYS DOH’s online study description months earlier, well before the study was initiated. I also published the above-listed recommendations for the study design (Michaels 2016; Michaels 2017a). On Monday 5 June 2017 I attended a meeting in the Village of Hoosick Falls at which NYS DOH staff “confidentially” briefed the Village Mayor and Town of Hoosick Falls Supervisor about the Agency’s cancer cluster study of the Village (NYS DOH, 2017). A summary of the study methods had been provided in advance, but not the full study (even though it was dated May 2017), nor its findings or conclusions. NYS DOH mailed the study to multiple recipients on 6 June for receipt on 7 June. Media were briefed on 6 June, and news reports embargoed until 7 June.
Discussion and conclusions
NYS DOH'’s cancer cluster study differed in significant details from the published recommendations explicated above, as follows.
Decision to exclude blood serum PFOA database from study design
For the cancer cluster study, NYS DOH gleaned 480 cancer cases from the New York State Cancer Registry for the 20-year period 1995 to 2014. NYS DOH’s report suggestively indicated that “in 2016 NYS DOH performed blood testing for Hoosick Falls area residents. Blood tests confirmed residents’ exposure to PFOA.” NYS DOH, however, did not follow up this suggestion by actually using the serum PFOA database in its study, for example by determining the cancer status of individuals in the serum database, and comparing PFOA levels of people with vs. without cancer in the database. The study instead compared observed cancers of many types with expected numbers of the same cancers in a reference population using data predating the 2016 blood serum database.
NYS DOH representatives explained that this study design is “standard” and that the Agency had used it “dozens of times,’ generally to confirm or refute residents’ perceptions of cancer clusters, but rarely to investigate a specific substance. Thus, NYS DOH applied its standard study design to the non-standard (unusual if not unique) situation of the Village of Hoosick Falls, whose residents were concerned about PFOA released to the drinking water supply and to other environmental media by the Saint Gobain Performance Plastics Company and/or its predecessors. NYS DOH did not explain why the potentially probative serum PFOA database was excluded from consideration in its health study.
A strictly statistical approach narrowly focusing on incidence data seems fraught with the peril of overlooking possible clusters that are worthy of further, detailed investigation. The challenge after recognizing a cluster is attributing a cause or probable cause to it, if possible. This is an important purpose of conducting further, detailed investigation. In short, suggestive data should be investigated further, in detail, to avoid overlooking cancer clusters and cancer cluster causes.
ENVIRONMENTAL CLAIMS JOURNAL @) 63
Failure to compare high-exposure vs. low-exposure populations within the Village
Serum PFOA levels constitute a proxy for past exposure (NYS DOH 2016a,b), and a predictor of future risk. Yet, NYS DOH in the past had declined to relate PFOA exposure to future risk, either quantitatively or even qualitatively (NYS DOH 2016a,b)”. In the present study NYS DOH compared cancer incidence in the Village with that in a larger, mainly external population, specifically New York State, excluding New York City. A more direct approach, rejected by NYS DOH, would have studied highexposure vs. low-exposure populations within the Village.
This could have been done in two ways. One was to use the blood serum database to compare PFOA levels in residents who had cancer with serum levels of the much larger number in the database without cancer. The second was to compare the serum PFOA and cancer status of people in the Village who had residential plus occupational exposure, vs. those who had only residential exposure. Either of these potentially probative approaches, or both together, could have determined whether the incidence of specific cancers, or the overall cancer incidence, in the Village was associated with serum PFOA levels, and therefore with exposure to PFOA.
Potential skewing of results due to exclusion of cancer cases
NYS DOH’s application of its standard method of comparing cancer incidence in the study area with that in an external reference population was flawed, as it included only an undetermined fraction of cancer cases. Many cancers observed in the Village were excluded from the study because their diagnosis postdated the 1995 to 2014 study period. Other cancer cases also were excluded, because they were diagnosed elsewhere, after the patients had moved out of the Village.
Selection of white, non-Hispanic reference population
NYS DOH selection of the white, non-Hispanic population of New York State excluding New York City as the reference population for comparison was appropriate. Census data (Census Reporter 2017) indicate that this group comprised 97 percent of the Village population (reported to be 3,453). Similarly, Rensselaer County was reported to be only 3.8 percent Hispanic, and 96 percent white. The comparison population, though valid, poses some analytical difficulties (relating to inclusion vs. exclusion of New York City) for interpreting the NYS DOH study.
Statistical significance barrier to cancer cluster recognition
Like others (for example Kulldorff et al. 1998), NYS DOH (2017) applied a 95-percent confidence limit as its criterion for identifying “unusual elevations” of
2 NYS DOH 2016b: Can PFOA blood levels predict the likelihood of having health problems? The measured PFOA levels only tell us about exposure to PFOA. The blood testing results provide important information about exposure to PFOA and allows for comparisons to people living elsewhere. Individual results only provide exposure information and are not cannot [sic] be used to determine of [sic] whether a person's current illness is due to PFOA or if a future illness is likely to result from PFOA.
64 © _ RAMICHAELS
cancers of each type included in the study. This high confidence limit is appropriately stringent for scientific studies, which must protect the scientific literature from introduction of errors. NYS DOH, however, simultaneously is charged with protecting public health. The stringent statistical significance criterion chosen would exclude consideration of cancer incidence elevations reaching, say, only the 90-percent confidence level, even though such elevations still are likely to be nonrandom. Thus, NYS DOH’s study design screened out potential cancer clusters, rather than identifying such potential clusters, reporting them as candidate clusters, and then following up by evaluating them further.
Dismissal of Village lung cancer cluster as irrelevant to PFOA
NYS DOH identified lung cancer as a statistically significant cluster in the Village, with 91 cases observed compared with just 65 expected (Table 1). This represents an incidence elevation of 40 percent above that expected based upon the reference population. NYS DOH dismissed this finding, however, on the rationale that lung cancer has been associated with smoking, but not with PFOA. At the 5-June briefing, NYS DOH representatives explained misleadingly that Rensselaer County has a higher smoking rate than the rest of New York State, accounting for the elevation of lung cancer incidence in the Village.
Rensselaer County’s smoking rate indeed has been reported to be elevated compared with New York State, raising the issue of whether its degree of elevation was sufficient to explain fully the elevated lung cancer incidence in the Village of Hoosick Falls compared with the State. Rensselaer County’s smoking rate was reported to be 20.3 percent for females and 20.4 percent for males, compared with 16.2 percent and 19.7 percent respectively for New York State (including New York City, in the published data; HealthData.org 2017). That is, the smoking rate in Rensselaer County is just 2.4 percent higher than the smoking rate for New York State (Table 1). As indicated in the table, this analysis shows that the elevation of lung cancer in the Village is nearly 17 times greater than the elevation of smoking in Rensselaer County (40 percent/2.4 percent) compared with the State. By extension (per NYS DOH), lung cancer elevation in the Village likewise is nearly 17 times greater than the elevation of the (assumed) smoking rate in the Village.
NYS DOH’s explanation of the elevated lung cancer incidence as reflecting an elevated smoking rate therefore conflicts with available data. The 2.4-percent difference between the smoking rate in the County (also that assumed for the Village) vs. the State would result in just nine more expected cases of lung cancer in the Village (74 instead of 65), whereas the actual number of cases above this augmented expected number is 17 more than the nine (91 instead of 74; see Table 1)°. To elevate the expected number of lung cancer cases in the Village to 91, which was the observed number, the smoking elevation in the Village over the State would have
3 Calculation: 65/0.1795 = x/0.2035, x = 65 (0.2035.0.1795) = 65 (1.134) = 74, where: 0.2035/0.1795 = 1.134 = +13.4 percent Rensselaer County vs. NYS smoking rates, and x = expected lung cancer cases in Village = 65 x 1.134 adjustment for higher smoking rate.
ENVIRONMENTAL CLAIMS JOURNAL @) 65
to have been 25.1 percent’. Based upon this analysis, NYS DOH’s dismissal of the statistically elevated lung cancer in the Village as being irrelevant to PFOA conflicts with available data.
Rensselaer County also was reported to have a lung cancer rate that is 23.8 percent higher than that for New York State (Table 1). Thus, the County’s lung cancer elevation compared with the State (23.8 percent) is 10 times higher than the smoking rate elevation (23.8 percent/2.4 percent). In contrast, the Village's lung cancer elevation compared with the State (40 percent) is 17 times higher than the smoking rate elevation (40 percent/2.4 percent), assuming (with NYS DOH) that the smoking rate in the Village is elevated to the same degree as that in Rensselaer County, in both cases relative to New York State. The particular reasons for these discrepancies are unknown, but would seem to involve factors other than smoking that are associated with lung cancer.
This elevation of lung cancer incidence in the Village beyond elevation of smoking incidence in the present context suggests a possible association of the elevated lung cancer rate in the Village specifically with PFOA, beyond the known association of lung cancer with smoking. The suggested association with PFOA is unproven, but signifies that NYS DOH’s report was unjustified in dismissing it. Further, the suggested association of lung cancer with PFOA was strengthened by other results that were reported, but that were not considered by NYS DOH.
Possible association of Village cancer incidence elevations with PFOA
NYS DOH concluded that cancer cases in the Village were unassociated with PFOA; that is, they were random rather than systematically above, equal to, or below the expected number. This means that, among cancer types exhibiting a number of cases that was not exactly equal to the number expected, randomly half should have been above the expected number, and randomly half below. A pattern of the number of cases exceeding the expected number, if observed, could indicate non-randomness, depending upon the consistency of the pattern of excess. NYS DOH failed to conduct this type of statistical analysis, which is termed ‘nonparametric. Nonparametric statistical analysis is powerful, in part because the data being analyzed need not conform toa normal (bell-shaped) distribution. Nonparametric statistics can be applied to ordinal data, that is, to the ranking of numbers, as in ‘above’ or ‘below a benchmark such as, in the present example, the expected number of cancer cases.
NYS DOH provided a table reporting expected vs. observed numbers of cases for 23 cancer types and another category of “all other types.” Among the 23 specifically named cancer types, just two exhibited a number of cases exactly equal to the expected number (19 non-Hodgkin's lymphoma cases and 14 leukemia cases). With respect to another six cancer types, data were unreported because the small number of cases (less than six of each cancer type) might compromise patient privacy. This leaves 15 cancer types for which the number of observed vs. expected cases was
4 Calculation: 65/17.95 = 91/x; x = (91 * 17.95)/65; x = 25.13 percent, where: x = Village of Hoosick Falls smoking rate (%) needed to produce 91 expected lung cancer cases; 65 = observed number of lung cancer cases in the Village, and 17.95 percent = NYS smoking rate.
66 ©) RA.MICHAELS
reported, and for which the observed vs. expected numbers were unequal. A random distribution therefore would most probably show about 7 or 8 above expectation (about half) and 7 or 8 below (also about half). In fact, 10 out of 15 cancer types showed a number of cases higher than the expected number, and only five showed a number below expectation.
The degree to which this NYS DOH result might constitute a departure from randomness may be discerned from the same statistics that apply to coin tossing. As with fair coins, each cancer type that is associated with an unequal number of observed vs. expected cases has an equal (0.5) probability of the observed number being higher than expected, and the same probability of the observed number being lower than expected, assuming randomness. The odds of 15 coin tosses producing at least 10 heads and at most five tails is about 15 percent (based upon an online multinomial calculator (http://stattrek.com/onlinecalculator/multinomial.aspx). The same is true for the odds of 15 cancer types producing at least 10 that are associated with more cases than expected, and at most five that are associated with fewer cases than expected.
The NYS DOH cancer cluster results therefore were not statistically significant at the 95-percent confidence criterion, but they do suggest that the excess of observed cases might have been non-random. Indeed, the skewed distribution was in the direction of excess, not deficit, and the probability of the observed degree of excess occurring randomly was just 15 percent. That is, the observed outcome was nearly six times more likely to be non-random than random (85 percent/15 percent). This raises the question of whether a nonrandom process of elevating multiple cancer types makes sense biologically.
Biological plausibility of cancer incidence association with PFOA
A biological explanation for PFOA raising the incidence of multiple cancer types is available. Indeed, such cancer incidence elevation by PFOA is to be expected, even if PFOA is not itself a carcinogen. This is because the critical toxic effect (the adverse health effect occurring with least exposure) exerted by PFOA is immunosuppression (see, for example, Michaels 2016 or Michaels 2017a). Immunosuppression results in reduced immune surveillance, and reduced ability to distinguish ‘self’ vs. ‘non-self’ (foreign) tissue. When cells of the body become cancerous, a normal immune system would recognize them as foreign, and work to kill them. By suppressing the immune system, immunosuppressors may permit cancer development induced by carcinogens, such as cigarette smoke. Laboratory testing of an immunosuppressor that is not also a carcinogen, however, would not reveal such cancer causation. Cancer causation would be revealed via the immunosuppression mechanism described above only in the unusual circumstance that the organisms also were exposed to carcinogens as part of the bioassay.
Rare cancers
Rare cancers are unlikely to occur in a population as small as that of the Village of Hoosick Falls, though NYS DOH indicated that rare cancers combined comprised about eight percent of cancers diagnosed and included in the study. These rare
ENVIRONMENTAL CLAIMS JOURNAL © 67
cancers are included in the report category ‘all other types, in which 44 cancer cases were reported, compared with an expected number of 39. The report found “higher than the expected number of cases of sinus cancers and other cancers of the nasal cavity,’ but this excess “was not statistically significant over the 20-year investigation period.” This ambiguous statement suggests the possibility that these cancers indeed might have been elevated significantly over a shorter observation period, but this is speculation in the absence of data.
A second type of rare cancer quantified by NYS DOH was carcinoid tumors (‘carcinoids’), which are neuroendocrine tumors having diverse origins. Six cases were identified, with just four expected. This 50-percent excess likewise did not rise to statistical significance (at the 95-percent confidence level), according to NYS DOH.
Local Village of Hoosick Falls physician Marcus Martinez noted (orally) that an unusually high number of cancers that he had seen among Village residents are rare. Their rarity, however, was attributable to their unusually early age at onset and/or their unusual aggressiveness rather than to their tissue of origin. This variety of rare cancers was not investigated in the NYS DOH study.
Anecdotal evidence
An important source of information about cancer clustering is anecdotal, but NYS DOH excluded such evidence from consideration. Anecdotal evidence includes residents’ experiences with pet dogs, pet cats, and cattle. Other anecdotal evidence involves people whose cancer cases were not reported in the NYS cancer registry, but are known from oral and other communications. For example a case of unusually aggressive pancreatic cancer that killed a man at an unusually young age, well before an age at which pancreatic cancers tend to appear, might be classifiable as a ‘rare’ cancer.
The first person to raise the alarm about cancer clustering in the Village was Michael Hickey, in part based upon a test of his residential drinking water, confirming the presence of PFOA. The alarm also was triggered by the death from cancer of his father and of several long-time co-workers of his father at the Saint Gobain Performance Plastics Company. Each co-worker contracted and died of cancer at an unusually early age. This evidence is factual and, as evidence goes, it is credible. It also is ‘anecdotal’
Excluding anecdotal evidence of this kind might have resulted in inclusion of the same cancer cases in the study but, even if that occurred, the anecdotal cases would have been ‘diluted’ statistically into the full, much larger study population. Such dilution would have tended to obscure the anecdotal association of the affected individuals with one another occupationally, and with PFOA occupationally and residentially. NYS DOH (2017) did not explain its decision to exclude anecdotal evidence and, in particular, to exclude from its analysis the possibility that these specific cancer deaths might have been related, or PFOA-related, rather than being merely random as implied by NYS DOH’s cancer cluster report.
68 © RAMICHAELS
Protecting science vs. protecting public health
Tension between the mandate of protecting the body of scientific knowledge vs. protecting the human body does not reduce the onus on public health professionals to do both simultaneously. Classifying a cancer cluster as real, for example, might be required for public health professionals to justify taking protective measures such as health monitoring to safeguard life and health, and funding such measures. Protective measures should be taken if a cluster probably is real, not just when it is 95-percent certain to be real.
The conflict between conservatism in the interest of academic science vs. public health protection has not always been resolved in favor of the latter. Investigating brain cancers in Los Alamos, New Mexico, Kulldorff et al. (1998), using a widely accepted statistical program called SaTScan, found that perceived clusters actually were statistical flukes: “The community was informed that such a finding could easily have resulted from random fluctuation in the incidence of a rare disease within a small population ... With adjustment for age, sex, and race, the most likely cluster is in the Albuquerque-Santa Fe area during 1985 through 1989... With a P value of 0.074 [92.6 percent confidence level], the cluster is not statistically significant” (Kulldorff et al. 1998, pages 1378-9).
Notwithstanding the negative finding made above, all effects have causes and, specifically, all cancer cases have causes. The suspicious brain cancer cluster in Los Alamos actually might have been real. It might have been caused by an environmental stressor, such as radioactivity of recent vintage, or radioactive residues dating back to the era of the Manhattan Project at Los Alamos. To protect science and public health simultaneously, public health professionals can and should analyze cancer incidence data to identify not only statistically unambiguous cancer clusters, but also candidate clusters whose confirmation might be possible with further investigation.
If incidence data fail to reach the 95-percent confidence criterion, one might ask what would justify further study as proposed here. A partial answer is to ask a question in response: what justified conducting a health study in the first place? One answer to both questions is: anecdotal evidence creating the perception of a cancer cluster. Given the power of some anecdotal observations, the premise seems inappropriate that investigation should cease if a wider study fails to reveal a wider effect.
A wider effect might suggest a worst-case scenario. In the case of the Village of Hoosick Falls, for example, a co-worker cancer cluster, even if anecdotal, might constitute an early warning of a developing scenario of contamination emanating from their workplace to the environment. The substance, in this example PFOA, might have affected employees more intensely because of their longer and more intense exposure, and might later affect the surrounding community more intensely if not removed from pathways of potential exposure.
Thus, evidence that does not attain the 95-percent statistical significance criterion typical of academic science still potentially may be important. This may count as an ‘inconvenient truth. Accordingly, the American Statistical Association recently issued a statement to combat pervasive misunderstanding in the scientific, business,
ENVIRONMENTAL CLAIMS JOURNAL @) 69
and public policy communities of the 95-percent confidence level and its routine, often inappropriate application (Goodman 2016; Wasserstein and Lazar 2016).°
If a cancer cluster (or vaccine side effect, or other public health effect) fails to attain 95-percent statistical significance, what should public health professionals do? They can and should refrain from concluding that a cluster has been found with a scientific standard of certainty. However, they also can and should evaluate what if any further probative studies might be justified. Such further studies should be designed to decide the possible reality of the candidate cluster(s), and the possible causality of their association with a stressor such as, here, PFOA.
Overall conclusion
An apparent cluster of cancer deaths was observed among co-workers, precipitating discovery in Village of Hoosick Falls drinking water of PFOA to which the coworkers might have been exposed at its source, occupationally. This contamination and anecdotal cluster of cancers in turn precipitated a NYS DOH study of possible “unusual elevations” of incidence rates for multiple cancer types among Hoosick area residents. NYS DOH’s study design applied the stringent academic standard of 95-percent confidence for accepting cancer clusters as real, and found multiple elevations but none except lung cancer reaching that stringent 95-percent confidence criterion. NYS DOH dismissed the lung cancer elevation, unconvincingly attributing it to smoking rather than PFOA. NYS DOH failed otherwise to investigate possible associations of cancer incidence elevations specifically with PFOA exposure as indicated by blood serum levels, or with PFOA exposure pathways as indicated by occupational and residential exposure analyses. To protect science and public health simultaneously, public health professionals can and should analyze data to identify statistically unambiguous cancer clusters, as well as candidate clusters whose confirmation might be possible with further investigation. Further types of investigation are addressed here, such as considering reliable anecdotal evidence and analyzing biological mechanisms. Considering these additional information sources together might strengthen or weaken the hypothesis of causal PFOA association with cancer incidence elevations, thereby guiding risk management efforts.
About the author
Robert A. Michaels, PhD, CEP (ram@ramtrac.com) is President of Schenectadybased RAM TRAC Corporation, and a toxicologist specializing in assessment and management of risks to public health potentially posed by environmental contaminants. He has served numerous corporate clients, the U.S. Congressional Office
5 Wasserstein and Lazar (2016): “Scientific conclusions and business or policy decisions should not be based only on whether a p-value passes a specific threshold. Practices that reduce data analysis or scientific inference to mechanical “bright-line” rules (such as “p < 0.05") for justifying scientific claims or conclusions can lead to erroneous beliefs and poor decision making. A conclusion does not immediately become “true” on one side of the divide and “false” on the other. Researchers should bring many contextual factors into play to derive scientific inferences, including the design of a study, the quality of the measurements, the external evidence for the phenomenon under study, and the validity of assumptions that underlie the data analysis. Pragmatic considerations often require binary, “yes-no” decisions, but this does not mean that p-values alone can ensure that a decision is correct or incorrect. The widespread use of “statistical significance” (generally interpreted as “p < 0.05”) as a license for making a claim of a scientific finding (or implied truth) leads to considerable distortion of the scientific process.”
70 ©) RA.MICHAELS
of Technology Assessment, and public interest organizations such as the Natural Resources Defense Council (NRDC). Dr. Michaels chaired the State of Maine Scientific Advisory Panel, and for 20 years chaired the Certification Review Board of the Academy of Board Certified Environmental Professionals (ABCEP). He has been Secretary of the NFPA Committee on Classification and Properties of Hazardous Chemicals, Board Member of the National Association of Environmental Professionals, and Member of the Editorial Advisory Boards of Springer-Verlag and Cambridge University Press journals. He earned his doctorate at SUNY at Stony Brook in 1979, and in 2004 was awarded ABCEP’s Kramer Medal recognizing his professional contributions.
References
Census Reporter. 2017. Hoosick Falls, NY. https://censusreporter.org/profiles/16000US3635474hoosick-falls-ny/, 9 pages, access date 3 June
Goodman, S. N. 2016. Aligning statistical and scientific reasoning. Science 352:1180-1. doi:10.1126/science.aaf5406.
HealthData.org. 2017. County profile: Rensselaer county, New York. http://www.healthdata.org/ sites/default/files/files/county_profiles/US/2015/County_Report_Rensselaer_County_Ne w_York.pdf, 9 pages, access date 8 June
Kulldorff, M., W. F. Athas, E. J. Feuer, B. A. Miller, and C. R. Key. 1998. Evaluating cluster alarms: A space-time scan statistic and brain cancer in Los Alamos, New Mexico. American Journal of Public Health 88 (9):1377-80. doi:10.2105/AJPH.88.9.1377.
Michaels, R. A. 2016. Perfluoroalkyl compounds (PFCs) in the Village of Hoosick Falls, Rensselaer County, New York: Health risks and successive approximation toward enforceable national regulation. Albany, New York State Bar Association, NYSBA New York Environmental Lawyer 36 (2):88-111, Fall
Michaels, R. A. 2017a. Lessons learned from legacy contaminants of emerging concern: Perfluoroalkyl compounds (PFCs) in the Village of Hoosick Falls, Rensselaer County, New York. Environmental Claims Journal 29 (1):4—48. doi:10.1080/10406026.2016.1270666
Michaels, R. A. 2017b. Confidence in courts: A delicate balance. Science 357 (6353):764, Letters. doi:10.1126/science.aao3967.
NYS DOH. 2016a. Hoosick Falls and Town of Hoosick Questions and answers about PFOA blood testing program. Albany, New York: New York State Department of Health. http://www. villageofhoosickfalls.com/Water/Documents/HealthMonitoringFactSheet-06022016.pdf,
6 pages
NYS DOH. 2016b. PFOA biomonitoring (blood sampling) program. Albany: New York State Dept. of Health. http://www.health.ny.gov/environmental/investigations/hoosick/docs/pfoa_blo od_sampling_q_and_a_8-1-16.pdf, 5 pages
NYS DOH. 2016c. PFOA in drinking water in the village of hoosick falls and town of hoosick. Albany: NYS Department of Health. https://www.health.ny.gov/environmental/investiga tions/hoosick/. access date 16 February
NYS DOH. 2017. Cancer incidence investigation 1995-2014, village of hoosick falls, Rensselaer county, New York. Albany: New York State Department of Health. http://www.health.ny. gov/hoosick, 18 pages including Appendices
Wasserstein, R. L., and N. A. Lazar. 2016. The ASA’s statement on p-values: Context, process, and purpose. The American Statistician 70 (2):129-33. doi:10.1080/00031305.2016.1154108.
Publication citation
Michaels, RA. Protecting science and public health in communities – PFOA and possible cancer clusters in the Village of Hoosick Falls, Rensselaer County, New York. Routledge, Taylor & Francis Group, Environmental Claims Journal, 30(1):58-70, January-March 2018, doi: 10.1080/10406026.2017.1422660, (https://doi.org/10.1080/10406026.2017.1422660), published online 31 January 2018;