LL1-04 — Ch4 Benzene: an historical perspective on the American and European occupational setting#

Late lessons from early warnings: the precautionary principle 1896–2000 (EEA Environmental Issue Report No 22, 2001). Chapter 4, report pages 38–51 (PDF pages 38–51; PDF and printed page numbers coincide). Main text pp. 38–46; Table 4.1 p. 47; references pp. 47–51.

Reading record. I read the whole text extract in order, through the final page marker (PDF/report p. 51). I rendered PDF pages 38–47 and checked them visually against the extract. The extract is faithful. The layout is a clean two-column text with one glossary box (p. 38) and one table (Table 4.1, p. 47, “Source: EEA”). The 2001 report has no panels or commentaries by other authors, so everything in the main text is Infante’s own voice. The one exception is Table 4.1, which is attributed to the EEA (the editors), not to Infante. For standpoint and context I also consulted, in the same PDF, the Introduction (pp. 11–12), Infante’s author biography (p. 197) and the editors’ uses of benzene in Chapter 16 “Twelve late lessons” (pp. 171–182). These are marked as such below and are not treated as part of this chapter’s argument.

Authors and standpoint#

Author. Peter F. Infante, sole author (p. 38).

Stated position (p. 197, author biographies). Health Standards Programs, Occupational Safety and Health Administration (OSHA), US Department of Labor, Washington DC. His credentials are given as D.D.S. and Dr.P.H., with a doctorate in public health from the University of Michigan’s Department of Epidemiology. He is a Fellow of the American College of Epidemiology. For 25 years he had “evaluated and regulated toxic substances found in US workplaces for the Department of Labor”. He had more than 100 publications and had served on panels for the US National Cancer Institute (NCI), the President’s Cancer Panel, the congressional Office of Technology Assessment and the National Academy of Sciences. He had also been an expert consultant to the WTO in the EU asbestos-ban dispute.

Editors’ disclosure (p. 12). The editors note that case-study authors “are not without strong views”, being mostly “active participants” in the histories they recount. They say specifically that Infante “did the first cohort epidemiological study of benzene-exposed workers” and had worked for many years in the US health and safety agency to reduce workers’ exposure. Authors were asked to be “as objective as possible” (p. 12) in answering four set questions (listed on p. 11): 1. When was the first credible early warning? 2. What were the main actions and inactions? 3. What were the resulting costs and benefits, and how were they distributed? 4. What lessons follow?

Authors wrote pro bono and were asked to keep chapters brief (p. 12). The editors also chose benzene deliberately as a chapter on the “North American handling” of an issue relevant to Europe, written by a North American scientist (p. 11).

How the standpoint shows in the text. - Participant-historian. Infante is lead author of the pivotal 1977 Pliofilm cohort study (p. 40). He co-wrote the 1988 estimate of deaths caused by regulatory delay (with DiStasio, p. 41), wrote the 1997 risk assessment based on the Chinese data (p. 41), and co-wrote the 1990 gasoline paper (p. 45). He co-authored the 1978 consumer-products paper (Young et al., 1978, p. 40; he is a co-author per the reference list, p. 51, though the text does not say so). He was a named party to the 1977 exchange with industry consultants in the Lancet (Infante et al., 1977b, p. 43). Eleven of the roughly 83 references have him as author or co-author. The chapter is therefore partly a first-person account of disputes in which he was a principal protagonist. - Regulator writing against the constraints on his own agency. He is a serving OSHA official. He reports the Supreme Court’s “Benzene Decision” fairly, including its hedges, but says it “has had a major impact on OSHA’s ability to control exposures” (p. 40). He sharply criticises the risk-assessment regime OSHA has operated under since (pp. 41, 44), and he calls OSHA’s own 1987 limit inadequate (p. 41). - Explicit first-person opinion, flagged as such: “From my personal experience” (p. 43); “In my opinion, it is part of a new anti public health approach…” (p. 43); “In my opinion, the protracted argument…” (p. 45). - Evident stance. Strongly pro-worker-protection and pro-regulation, hostile to industry consultants (who are named individually), and critical of consensus standard-setters (the ACGIH). His epistemic position is roughly this: human data on benzene have long been sufficient, and demands for more certainty are a delaying tactic. The tone is often polemical. Examples: “clearly fictitious” (p. 44); “incomprehensible” (p. 46); “make a mockery of public health education” (p. 46). - Panels/commentaries. None. (LL1 has no panels.) No industry or regulator response is printed alongside the chapter. Opposing positions appear only as Infante summarises and rebuts them.

Section-by-section notes#

4.1 Early warnings (p. 38)#

• First warnings, 1897. Santessen [sic; usually spelled Santesson] reported aplastic anaemia among young women making bicycle tyres in Sweden. LeNoir and Claude reported haemorrhaging in a young man in a French dry-cleaning operation. Since then benzene “has been known to be a powerful bone marrow poison” (p. 38). Similar reports “increased dramatically” through the first half of the 20th century.
• Internal inconsistency: on p. 42 and p. 46 the same Santessen 1897 cases are described as among women “manufacturing raincoats” and in the “raincoat manufacturing industry”. The p. 46 argument about Sweden depends on the 1897 cases being in “this same industry” as a 1940s raincoat plant. Both are rubber-goods trades using benzene as a rubber solvent (Helmer’s 1944 title refers to “the rubber industry”, p. 48), so the broad point survives. The specific “same industry” claim needs checking against Santesson (1897).
• Expansion of use.
• First major use as a solvent in rubber came in 1910–14.
• Production was “stimulated greatly” by First World War demand for toluene for explosives (p. 38).
• After the war, use spread to artificial leather, rubber goods, glue, hats, rotogravure printing, paint, adhesives, coatings, dry cleaning, car manufacturing and tin-can assembly. Benzene also became a starting material for organic synthesis and a component of petroleum products and motor fuel blends (p. 38).
• Harm tracked use. Expansion “was accompanied by a vast increase” in aplastic anaemia, generally called “benzene poisoning” (p. 38). Some workers were diagnosed within weeks of starting work, and some died within months (Hogan and Schrader, 1923). Exposures were “mostly between 200 ppm … and 1 000 ppm” (p. 38).
• Greenburg et al. (1926). A survey of 12 US plants found:
• 32% of workers had abnormally low white cell counts (<5,500/cc) and 12% had counts below 4,000/cc;
• exposures were 90 ppm and higher.
• Greenburg recommended medical removal if a worker developed clinical symptoms or if blood values fell by 25% or more (p. 38).
• This is the first quantitative, population-level warning in the chapter. Structurally it is the strongest early evidence: a multi-plant prevalence survey linked to measured exposure levels.
• Glossary box (p. 38). Defines leukaemia (acute/chronic; myelogenous, lymphatic, monocytic), aplastic anaemia, “benzene poisoning” (“usually refers to aplastic anaemia”), “hypersusceptibility”, haemopathy and multiple myeloma (“a cancer of the lymphatic system”, a loose definition).

4.1.1 First report of benzene-induced leukaemia (p. 38)#

• 1928. Dolore [sic; usually Delore] and Borgomano published the first case of benzene-induced leukaemia: acute lymphatic leukaemia in a pharmaceutical worker in a job “considered dangerous because of high benzene exposure levels”. Another worker at the plant died of aplastic anaemia. The authors thought some earlier aplastic anaemia cases “may also have been leukaemia” (p. 38). My gloss: this was an early signal that the “poisoning” (aplastic anaemia) and leukaemia outcomes were linked, and that some leukaemia may have been hidden under the “poisoning” label. The chapter does not draw this out.
• The company’s response was not to reduce exposure: it chose “to rotate the workers out of the specific job every month” (p. 38). This spreads exposure across the workforce instead of reducing it.

4.2 Actions and inactions#

4.2.1 Exposure recommendations (p. 39)#

• Substitution recommended. By 1939 “the vast number of benzene poisonings among workers in all parts of the world” had led several investigators to recommend replacing benzene with other solvents (Greenburg, 1926; Erf and Rhoads, 1939; Mallory et al., 1939) (p. 39).
• 1939. Hunter and Mallory et al. reported 89 cases of “poisoning” and 3 of leukaemia across occupations. Two of the poisonings were associated with levels of <25 ppm and 10 ppm respectively (p. 39). Much later argument (pp. 39, 43; Table 4.1, p. 47) rests on these two cases, which the chapter treats as showing a “10 ppm level already associated with aplastic anaemia” by 1939.
• The limits that followed (p. 39):

• The API 1948 document. Because of evidence of “hypersusceptibility”, the API itself concluded that “the only absolutely safe level from exposure to benzene was zero” (p. 39). The basis was that some workers developed bone-marrow depression while working alongside colleagues whose blood counts were normal. In other words, variation between individuals meant no exposure level could be assumed safe for everyone. The API “then proceeded to recommend a limit of ‘50 ppm or less’” (p. 39). The trade body’s own review stated the no-safe-level conclusion, then recommended a limit above the levels already linked to poisoning.

4.2.2 Disregard for recommendations (p. 39)#

• Continued harm despite the limits. Aplastic anaemia and central nervous system (CNS) toxicity continued to be reported in the 1940s–50s: headache, nausea, giddiness, staggering gait, paralysis and unconsciousness, including 13 deaths in the UK (Browning, 1965) (p. 39).
• Exposure inferred, not measured. Such CNS effects “are thought to be associated with” 3,000–20,000 ppm (Flury, 1928). The chapter describes this as “200 to 800 times higher” than the recommended limits and “2 000 times higher” than the 10 ppm level linked to aplastic anaemia in 1939 (p. 39).
• Arithmetic check: the multipliers use the top of the range (20,000 ppm ÷ 100 = 200; ÷ 25 = 800; ÷ 10 = 2,000). At the bottom of the range (3,000 ppm) the ratios would be about 30–120× the limits and 300× 10 ppm. The exposure levels for the UK deaths are inferred from the symptoms, not measured. The qualitative point (exposures far above the limits) stands; the precise multipliers are rhetorical upper bounds. The conclusions (p. 46) then go further, to “most likely more than 200 to 800 times”, which the arithmetic does not support.
• Global pattern. In the 1950s–60s there was “an obvious lack of precaution for workers exposed to benzene … in many parts of the world”, including France, Italy, Russia, Turkey, the UK and the US. The evidence is published case reports of blood diseases at exposures well above levels known to be toxic (p. 39).
• Italy.
• Vigliani and Saita (1964): acute leukaemia risk among workers “heavily exposed” in rotogravure and shoe industries (Milan and Pavia) was at least 20 times that of the general population.
• Vigliani (1976): over 200 cases of benzene haemopathy, including 34 acute leukaemias, were treated at the Milan and Pavia institutes from 1942 to 1965. Exposures were mostly 200–500 ppm, “with occasional peaks above these levels” (p. 39).
• France. Goguel et al. (1967): 44 cases of benzene-induced leukaemia, mostly chronic forms, in the Paris region from 1950 to 1965 (p. 39). (The reference title on p. 48 says “a study of 50 cases”; the text says 44, presumably 44 leukaemias among 50 cases. Not explained.)
• Turkey. Aksoy is quoted: benzene-containing glue adhesives were “extremely practical and cheaper in the market”, so the shoe and slipper industry switched to them from its customary petroleum-based adhesives in 1961 (p. 39).
• Exposures were 150–650 ppm (Aksoy, 1978).
• “By the mid-1970s, an epidemic of aplastic anaemia and leukaemia among Turkish shoe workers began to unfold” (p. 39).
• (Note that Aksoy et al. 1971 already reports haematological effects in 217 workers, p. 47.)
• Summary claim (p. 39). Most of those in these European reports who died of leukaemia and other benzene-related blood diseases “were exposed to levels shown decades earlier to cause benzene poisoning”. IARC (1974) is cited as a review of the case reports.

4.2.3 Epidemiological evidence for leukaemia (pp. 39–40)#

• Early 1970s, University of North Carolina. A series of studies found significant excesses of leukaemia, mostly chronic forms, among rubber workers exposed to “presumably low” benzene levels (McMichael et al., 1975). Exposure came from contaminated rubber solvents (petroleum naphtha, toluene, mineral spirits) containing roughly 1–5% benzene in the 1940s and about 0.5% (naphtha) in the 1970s (p. 40).
• 1977, Infante et al. This was the first cohort study of workers exposed specifically to benzene, among makers of a rubberised food wrap, “Pliofilm”. It found a 5- to 10-fold leukaemia risk at exposures “generally considered within the various limits recommended over the time period 1940–1971, that is a 10 ppm time-weighted average to a maximum limit of 100 ppm” (p. 40).
• Before 1977, causation rested on clinical evidence. Benzene was considered a cause of leukaemia “based not upon epidemiological studies, but rather upon case reports of leukaemia and the clinical observation” that benzene-induced aplastic anaemia and other blood diseases could progress to acute leukaemia (p. 40). This is an important admission for any fairness assessment. Before 1977 the quantitative leukaemia risk at the then-permitted levels had not been shown epidemiologically.
• Internal tension: the flat statement that pre-1977 causation rested “not upon epidemiological studies” sits awkwardly with material the chapter itself reports just before it: Vigliani and Saita’s (1964) comparison showing at least a 20-fold acute leukaemia risk among heavily exposed workers, and the University of North Carolina studies (McMichael et al., 1975) (pp. 39–40). The best reading is that Infante means no benzene-specific cohort study existed, and none at within-limit exposures. The first of these is how Table 4.1 describes his 1977 study.
• Narrative gap: the chapter never explains when or how the limit moved from 25 ppm (ACGIH 1957) to the “old limit of 10 ppm” referred to on p. 40.

4.2.4 US attempt to control occupational exposure (p. 40)#

• 1977 emergency standard. Based on Infante et al. (1977a) and the world literature, OSHA issued an emergency temporary standard (ETS) that would have cut the limit to 1 ppm (8-hour TWA). This followed OSHA’s then policy that carcinogen exposure be lowered “to the lowest feasible limit”, with feasibility covering both technology and economics (p. 40).
• Challenge and appeals.
• The ETS was stayed in 1977 after a challenge in the US Court of Appeals by the API. The API “in essence argued that there was no increased risk of developing leukaemia” below the old 10 ppm limit (p. 40).
• OSHA then proposed a permanent standard, took comments and held a hearing. It issued a final standard (1978) at 1 ppm.
• The API challenged it on the same grounds; the Court of Appeals vacated it; the case went to the Supreme Court (p. 40).
• 1978 consumer withdrawal. “In an unrelated development”, benzene was voluntarily withdrawn from US consumer products (1978). This followed a showing that home use of a paint stripper could produce up to 200 ppm in a short time (Young et al., 1978, p. 40). Infante is a co-author of that paper (p. 51), which the text does not mention.

4.2.5 The US Benzene Decision and dose-response analyses (pp. 40–41)#

• The decision. In July 1980 the Supreme Court issued the Benzene Decision (cited as IUD 1980, “AFL-CIO vs. American Petroleum Institute, 1980, 448 US 601”, p. 49. The standard citation is Industrial Union Department, AFL-CIO v. American Petroleum Institute, 448 U.S. 607, decided 2 July 1980; confirmed from outside sources, so “601” is a slip). Infante’s headline judgement: the decision “has had a major impact on OSHA’s ability to control exposures to benzene and other toxic substances in the workplace” (p. 40).
• Before issuing any permanent health standard, the Secretary of Labor must make a threshold finding that a workplace is unsafe, in that “significant risks are present and can be eliminated, or lessened, by a change in practices” (p. 40).
• Significant risk should, “if at all possible”, rest on the best available evidence “through such means as quantitative risk assessments” (p. 40).
• The court’s hedges, reported fairly.
• The decision “recognised the uncertainties involved” (p. 40).
• The requirement “is not a mathematical straitjacket”.
• It is OSHA’s responsibility to decide what counts as significant, “based largely on policy considerations”.
• The court’s one concrete illustration: a one-in-a-million cancer risk from drinking chlorinated water “clearly could not be considered significant”. A one-in-1,000 fatality risk from regularly breathing gasoline vapour that is 2% benzene is one a reasonable person “might well consider … significant” (p. 40).
• (Background, not from the chapter: the lead opinion, by Justice Stevens, was a plurality opinion, which the chapter does not mention. The plurality placed the burden on the agency to show significant risk.)
• OSHA’s practice since then. OSHA has treated a 45-year occupational lifetime risk of 1 extra case of cancer, “or other material impairment of health consequence”, per 1,000 workers as “significant” (pp. 40–41). It “has not yet considered” the non-significant end of the range, because every standard issued since the decision (except perhaps formaldehyde) left an estimated residual risk above 1 per 1,000 (p. 41; Infante, 1995b).
• Infante’s inversion of the court’s phrase. “A straightjacket, however, appropriately describes the risk analyses that OSHA currently engages in” (p. 41). His argument (p. 41):
• The pre-1980 delays have been “superseded” by further detailed risk analyses.
• These are “encumbered” by analyses of cancer mechanism. Because no substance’s exact carcinogenic mechanism is known, “speculation and argument about various unproven hypotheses” consume time.
• They are also encumbered by disputes over which animal species (mouse, hamster, rat) is most appropriate when human data are lacking.
• Because OSHA “is required to review and comment on all possible cancer mechanisms, appropriateness of species, etc.”, the process “has created additional years of delay”.
• “Instead of reasonable precautions being promulgated by government and employers, years pass as analyses are performed to determine the dose response between exposure and risk of disease.”
• The analyses may also incorporate “speculative mechanistic data that have not been validated”.
• Concessions. “While on the surface such analyses may seem appropriate…” (p. 41). He accepts that risk analysis has a surface legitimacy; his complaint is that it is open-ended and gets captured by procedure.

4.2.6 Cost in lives of the prolonged regulatory process (p. 41)#

• The 1987 standard. “Eleven years after” the ETS, OSHA issued a new standard with a 1 ppm TWA (OSHA, 1987). The chronology is 1977 to 1987, which is ten years, and the same page also says “10 years” and “10-year delay”, so the text is inconsistent.
• The limit rested on feasibility, not risk. The limit was “based on ‘economic feasibility’, not elimination of ‘significant risk’”, because an occupational lifetime risk of 10 extra leukaemia deaths per 1,000 workers “was associated with the 1 ppm limit” (p. 41). The chapter does not say whose estimate this is; presumably it is OSHA’s 1987 risk assessment. My arithmetic: this is ten times OSHA’s own significance benchmark of 1 per 1,000.
• Competing, lower estimates.
• Crump (1994) used the NIOSH (National Institute for Occupational Safety and Health) cohort as updated by Paxton et al. (1994), restricted to acute myelogenous and monocytic leukaemia deaths. He estimated 0.02 to 5.1 excess deaths per 1,000, depending on exposure estimates and model (p. 41).
• Infante discounts these because they “are based on a follow-up period of the NIOSH cohort that results in selection bias for the reasons described below”. This refers forward to p. 42 on the company’s removal of pre-1950 records.
• Analytical note: two choices drive these lower estimates: the narrower endpoint (acute myelogenous and monocytic leukaemia only) and the exposure reconstruction. Infante’s rebuttal addresses only follow-up/selection. He does not show in this chapter that Crump or Paxton used the pre-1950 start date he criticises in Wong (1995).
• Cost of the delay (Infante and DiStasio, 1988). Using OSHA’s final risk assessment and estimates of extra exposure across the US workforce during the 10-year delay, the paper estimated that 198 extra leukaemia deaths and 77 extra multiple myeloma deaths “will eventually develop”, “deaths that could have been prevented” (p. 41).
• The figure is a model projection by the chapter author, not a count of observed deaths.
• It excludes other benzene-caused blood diseases because “there were no dose-response data available” for them (p. 41). So its coverage of diseases is incomplete by construction, which biases it low. It is not a lower bound overall: the figure depends on OSHA’s risk model, and the chapter itself reports lower competing risk estimates (Crump, 1994, above).
• Table 4.1 (EEA) turns this into “This 10-year delay caused more than 200 deaths” (p. 47). That wording drops the projection framing (“will eventually develop”) and states a model estimate as historical fact. “More than 200” is consistent only with the combined 275 (198 + 77); the leukaemia figure alone is 198.

4.2.7 Expansion of lymphohaematopoietic diseases (pp. 41–42)#

• Non-Hodgkin’s lymphoma (NHL). The delay estimate also left out NHL, which had since been associated with occupational exposure “to very low levels of benzene” (Hayes et al., 1997) (p. 41).
• Infante’s 1997 risk assessment. Based on “the 1996 Hayes et al. study results”, it suggests 54 extra deaths from leukaemia/lymphoma per 1,000 workers “exposed over an occupational lifetime (45 years)”. The exposure level is not stated; the next sentence, on “the inadequacy of the 1 ppm exposure limit”, implies 1 ppm. That is “54 times greater than a level considered significant by OSHA” (p. 41).
• Evidence-quality note: Infante (1997) is a four-page contribution to Toxicology Forum meeting proceedings (reference, p. 49), not a peer-reviewed paper. The chapter’s most striking risk figure therefore rests on the author’s own conference-level analysis. “The 1997 Hayes et al. study results and dose-response analyses based upon those data clearly demonstrate the inadequacy of the 1 ppm exposure limit for benzene based on cancer risk alone” (p. 41).
• Two different Hayes papers: the reference list has Hayes et al. 1996 (“Mortality among benzene-exposed workers in China”) and Hayes et al. 1997 (“Benzene and the dose-related incidence of hematologic neoplasms in China”) (p. 48). The death-based 54/1,000 estimate is tied to the 1996 mortality paper and the “inadequacy” claim to the 1997 incidence paper. This is not a date slip, though the text does not explain the difference.
• Mitigating facts, acknowledged (pp. 41–42):
• “most occupational settings in the United States are able to achieve” 0.2–0.3 ppm or lower;
• the US standard includes exposure monitoring, medical surveillance and hazard training, “all of which should presumably further reduce the risk”.
• Direct observation, not modelling (p. 42). Hayes et al. (1997), plus Hayes personal communication (1999), show significantly elevated relative risks for all lymphohaematopoietic cancers combined, and for acute non-lymphocytic leukaemia plus myelodysplastic syndrome combined. This was in workers exposed to a constant average of 1.2 ppm for 5.5 years (6.7 ppm-years as printed; 1.2 × 5.5 is 6.6, a rounding matter), far below the 45 ppm-years a 1 ppm limit allows over a working life. Some individuals in benzene cohorts who died of leukaemia or lymphoma had estimated exposures of 0.5–2 ppm for only 1–2 years or less (Infante, 1992) (p. 42). Individual case exposures of this kind are illustrative, not causal evidence.
• Evidence-quality note: one key low-dose figure rests on an unpublished personal communication.
• The expanding spectrum (p. 42). By the 1990s, animal toxicology had shown multiple-site carcinogenicity. Further epidemiology and case reports had extended benzene’s carcinogenicity to:
• all major leukaemias in aggregate;
• acute myelogenous leukaemia (AML) and variants;
• acute lymphatic leukaemia (ALL);
• chronic lymphatic leukaemia (CLL);
• chronic myelogenous leukaemia (CML);
• hairy cell leukaemia;
• myelodysplastic syndrome (MDS);
• myeloproliferative disorders;
• NHL, “including multiple myeloma” (MM).

Each has citations. Some rest on single case reports or short items. For hairy cell leukaemia, Aksoy (1987) reports three cases (of chronic lymphoid and hairy cell leukaemia combined), and Flandrin and Collado (1987) is a two-page item asking whether male predominance in hairy cell leukaemia relates to occupational exposures. For ALL, one of the two citations is a population-based case-control study of childhood leukaemia in Shanghai (Shu et al., 1988), and AML citations include a 1965 reference book on industrial solvents (Browning) (reference titles, pp. 47–50). CLL rests on a single citation, McMichael et al. (1975), a study of rubber workers exposed to solvents in which benzene was a contaminant (p. 40). MDS rests on a single study (Hayes et al., 1997). The list mixes strong and weak evidence without grading it. - The Wong reanalysis and selection bias (p. 42). An “updated” analysis (Infante’s scare quotes) of the NIOSH cohort by Wong (1995) found the multiple myeloma excess no longer significant. Wong had moved the start of follow-up from 1950 to 1940. That “introduced selection bias … because the company removed records for individuals who died at one of the study locations for several of the years prior to 1950”. Analyses starting before 1950 “cannot be relied upon” for multiple myeloma “or any other cause of death” (p. 42). - Mechanism of note: the regulated firm’s handling of records shapes which analyses are valid. A reanalysis choice that looks neutral (a longer follow-up) can therefore produce a null result.

4.3 Discussion (p. 42)#

• Concern without adequate reduction. Occupational health showed concern “at times”, especially in the early decades of worker surveys. Yet “benzene exposure levels were not reduced to levels commensurate with the toxicity data available at the time”, and the epidemic of poisonings and leukaemia continued “through the first six decades of the 20th century” (p. 42). The reasons, he says, are “apparent”.

4.3.1 Some reasons for lack of precaution (pp. 42–45)#

Four headings are offered as causes.

(a) Lack of knowledge (p. 42). - The lack of precaution “has been attributed in part” to ignorance of toxicity in the first four decades. - Example: after Santessen’s 1897 report of four cases of aplastic anaemia in Sweden (here “women manufacturing raincoats”; the reference title reads “Vier Todesfalle”, four deaths, p. 50), exposures were not lowered enough. Helmer (1944) reported 60 cases of benzene poisoning (58 women) at a single raincoat factory in Sweden in 1940–41. Helmer attributed the epidemic partly to ignorance among management and workers. - Infante reports this as an explanation others gave (“has been attributed in part”; the Swedish case is Helmer’s own attribution), not as his own view. His “even though” framing already signals doubt, and on p. 46 he rejects the Swedish ignorance claim as “difficult to accept” and “incomprehensible”.

(b) Cost of solvents (pp. 42–43). - Greenburg (1926), Erf and Rhoads (1939) and Mallory et al. (1939) had all recommended substitution. Yet worldwide consumption kept growing after the Second World War. - Two stated reasons: - in the synthetic rubber industry, benzene “was such a good rubber solvent” (p. 42); - per Aksoy (1977), it was cheaper than other solvents in Turkish shoe and slipper manufacture (p. 43). - “Economic consequences” led to other solvents being replaced with benzene in Turkey “as late as 1961”, causing high exposures and the epidemic of leukaemia, preleukaemia, pancytopenia and other blood diseases. “Economic considerations in the 1960s … further contributed” (p. 43).

(c) Consensus recommendations and corporate influence (p. 43). - He repeats the mismatch: poisonings were reported at 25 and 10 ppm in 1939, against the ACGIH’s 100 ppm in 1946 and 35 ppm in 1948. - “From my personal experience over the years”: consensus organisations “usually base their recommended exposure levels on what is easily achievable in the workplace”. Exposure and toxicity data “are reviewed, but are not translated into health-based exposure-limit recommendations” (p. 43). - Castleman and Ziem (1988) investigated the ACGIH and concluded that its threshold limit values (TLVs) “were based heavily on corporate influence” (p. 43). - Infante’s own conclusion is hedged: consensus recommendations “were inadequate and corporate influence may have played a role” in them and in the resulting disease (p. 43). - Castleman and Ziem (1988) called for an international effort “to develop scientifically based guidelines to replace TLVs in a climate of openness and without manipulation by invested interests”. This “has not been achieved” (Castleman and Ziem, 1994) (p. 43).

(d) “Anti public health attitude (call for scientific certainty)” (pp. 43–45). This is the longest subsection, and the argumentative core of the chapter.

• Periodisation (p. 43). “In the 1970s, benzene manufacturers and users began a new approach” to communicating benzene toxicity to the public, workers and plant managers. This “contributed to a continuation of overexposure”. Manufacturers “began to hire consultants to downplay the importance of the scientific observations … and to introduce unresolvable arguments about dose-response analyses”, which delayed regulation.
• Changed economic motive (p. 43). Economics was “again” given priority over public health. This time the concern was the cost of lowering exposures (OSHA, 1987) “and perhaps” the rising cost of litigation and liability. The era “fostered the development of arguments that seek to minimise or misrepresent study results”.
• Opinion, flagged (p. 43). “In my opinion”, this is “a new anti public health approach that calls for scientific certainty in terms of causality for every specific lymphohaematopoietic disease related to benzene by exposure level”. As a result, workers worldwide “may not be afforded the appropriate protection to the extent feasible”, and many who fall ill “may receive little or no compensation”.
• Example 1: the 1977 OSHA hearings (pp. 43–44).
• “It was argued” (passive voice; the next sentence attributes the cluster argument to “consultants to the industry”) that the NIOSH 5–10-fold leukaemia finding was “meaningless in terms of public health intervention”.
• One argument (“Although not persuasive”) was that the study had simply found a random leukaemia cluster, since leukaemia clusters occur in time and space (Tabershaw and Lamm, 1977; rebuttal in Infante et al., 1977b).
• It was also argued that benzene “could not cause leukaemia in workers because there was no evidence that benzene caused cancer in experimental animals” (Olson, 1977; the text does not state Olson’s affiliation).
• Infante calls this argument “clearly fictitious given the overwhelming evidence of carcinogenicity provided by the study of humans” (pp. 43–44).
• Animal carcinogenicity followed “shortly thereafter” (Maltoni and Scarnato, 1979; NTP, 1986) (p. 44). Maltoni was about two years later; the NTP bioassay nine years later, which stretches “shortly”.
• Fairness note: the absence of animal evidence in 1977 was factually true. What Infante calls “fictitious” is the inference that human evidence was insufficient without it.
• Example 2: the 1980s rulemaking under the Benzene Decision (p. 44).
• Attention focused on OSHA’s and its consultants’ dose-response analyses.
• “Most of the argument addressed the benzene exposure assessment”, with the parties supplying new exposure estimates for periods with no exposure data. “These ‘educated’ guesses by the various parties involved in the rule-making could not be confirmed.”
• The result was multiple new dose-response analyses and a protracted debate over which exposure assessment was right, a question “the resolution of which can never be determined with scientific certainty”. There was also argument over which dose-response model to use.
• Key concession and claim: “While nobody would object to debate on these issues”, protracted continuation left workers and the public “unnecessarily exposed” to levels that a shorter process could have reduced (p. 44).
• Aphorism (citing his own 1987 paper): “Studying a subject to death often results in the death of those we are trying to protect” (p. 44).
• Example 3: the NCI/CAPM Chinese studies, 1990s (pp. 44–45).
• The US NCI and the Chinese Academy of Preventive Medicine (CAPM) published studies (Hayes et al., 1996, 1997; Dosemeci et al., 1996). They showed a dose-response for leukaemia, lymphoma, MDS and aplastic anaemia, and high relative risks at “very low average benzene exposures, such as around 1 ppm” (p. 44).
• Because governments in Europe, the US and elsewhere may use them to estimate general-population risk from low-level exposure, they “have broad implications for public health intervention” (p. 44).
• “Consultants to the chemical industry” published critiques of both the health findings and the exposure estimates (Wong, 1998, 1999; Budinsky et al., 1999). These, “in the opinion of some”, misrepresent the data (Hayes et al., 1998; Hayes pers. comm.) (p. 44). The attribution is hedged and relies partly on personal communication.
• Example 4: Wong’s broader positions, taken apart point by point (pp. 44–45).
• Claim: Wong (1995), using NIOSH cohort data, concluded that benzene causes only AML, with a threshold of 370–530 ppm-years (p. 44).
  • Rebuttal: Wong’s own industry-wide study (Wong, 1987a, b) reported a significant dose-response for leukaemia at cumulative exposures from <15 to >60 ppm-years. In that study “none of the leukaemia deaths were from acute myelogenous leukaemia”. His own data therefore contradict both the AML-only claim and the threshold (p. 44). This is a strong argument from internal consistency, if accurately reported. (The Wong 1987 study was industry-wide among chemical workers; its sponsor is not named in the chapter.)
• Claim: no association with multiple myeloma in the NIOSH study, because no dose-response could be found in four cases (p. 44).
  • Rebuttal: the inability to see a dose-response with four deaths “is essentially meaningless” for lack of statistical power. Rinsky et al. (1987, NIOSH) had already shown a significant multiple myeloma excess in the cohort and linked it to low-level exposure (pp. 44–45).
• Claims: Wong also argues benzene is not associated with multiple myeloma generally, nor with NHL (Wong, 1998). These conclusions “appear to be at odds with the views of others” (Goldstein and Shalat, 2000; Goldstein, 1990; Rinsky et al., 1987; DeCoufle et al., 1983; Infante, 1995a; Savitz and Andrews, 1996, 1997; Hayes et al., 1998, 2001) (p. 45).
• Bergsagel et al. (1999, in Blood; Wong is a co-author) is mentioned among “surprising views” (p. 44), but its content is not discussed.
• Closing argument (p. 45).
• The arguments over NCI/CAPM “seem reminiscent” of the post-1977 delay.
• “It will be unfortunate if more precaution is taken with the use of data from the NCI/CAPM study than with the protection of populations exposed to levels of benzene that can be reduced with technology currently available.” This turns the language of precaution around: caution about the data versus caution about the people.
• “Scientific certainty is difficult to achieve, but stressing the uncertainty does not do justice to the data on benzene exposure and related diseases.”
• Historically, precaution “in accordance with the scientific data available at the time would have eliminated much needless suffering and death”.
• “In my opinion” the protracted dose-response argument and “denial” about the diseases “most likely associated” with benzene are counterproductive.
• He grants that the continuing debate “may be interesting from an academic viewpoint”, but says it “also raises the question of whether it may be more a reflection of economic concerns and potential liability on the part of companies than a concern for the public’s health” (p. 45). The motive attribution is framed as a question.

4.3.2 Benzene in gasoline a continuing hazard (p. 45)#

• Low awareness and benzene content. Most consumers and many medical personnel do not know gasoline contains benzene. US gasoline has averaged about 1.5% benzene for two decades and may reach 5% by volume (Infante et al., 1990). Petrol “in most European countries” has historically contained more, a trend that “apparently” persisted through 1994 (Deschamps, 1995), though content “has supposedly been reduced more recently” (p. 45).
• Evidence cited on gasoline exposure (p. 45):
• leukaemia (Schwartz, 1987; Jakobsson et al., 1993; Infante et al., 1990);
• other blood diseases (Infante et al., 1990; Lumley et al., 1990; Naizi and Fleming, 1989);
• chromosomal defects (Lumley et al., 1990; Hogstedt et al., 1991);
• other genetic damage (Nilsson et al., 1996).
• These are mixed-quality designs: a proportionate mortality study, case reports, small cross-sectional biomarker studies. The chapter says “Not surprisingly” but does not grade them.
• Information failure. Pumps “do not provide adequate information” on benzene cancers, and gasoline material safety data sheets (MSDS) omit the chromosomal and genetic damage evidence (p. 45).
• Consequence of the “lack of candour” (p. 45):
• mechanics and highway maintenance workers use gasoline as a solvent for cleaning parts (Infante, 1993);
• consumers use it as a solvent and “fail to take the necessary precautions” when using it in lawn mowers, trimmers, saws and so on.
• Nigeria. Roadside vendors of repackaged gasoline had 26% neutropenia vs 2–10% in controls (Naizi and Fleming, 1989) (p. 45).
• Case reports. The hazard has been recognised since at least 1928 (Askey: aplastic anaemia in a US worker exposed to gasoline). Later cases include myelofibrosis in a Swedish petrol station attendant (Tondel et al., 1995) and aplastic anaemia in a US roofer who used petrol to clean seams (Infante et al., 1990) (p. 45).
• Institutional failure. “Despite the overwhelming literature on the hazards of benzene in petrol”, public health and safety officials, “including those employed by industry, have yet to come to grips with the task of adequately informing workers and consumers” (p. 45). “Overwhelming” is stronger than the mixed studies just cited would support.

4.4 Conclusions and lessons for the future (p. 46)#

In order: 1. Failure to act on knowledge. Knowledge of toxicity and “the failure to take precautions … over the past century is cause for concern”. Inaction by “consensus organisations and governments alike throw into question the ability of these organisations to protect the health of the public” (p. 46). 2. “In the case of benzene exposure in the workplace, the precautionary principle is not relevant” (p. 46). Infante does not explain the sentence. Read in context (it is followed at once by the ignored substitution advice), it most plausibly means the hazard was known. On that reading the problem was failure to act on established evidence, not a problem of acting under uncertainty. This is a significant framing move inside a report about precaution. - Nuance: the chapter’s own vocabulary blurs this line. It repeatedly describes the failure as a “lack of precaution” (pp. 39, 42). It says “taking precaution … in accordance with the scientific data available at the time” would have saved lives (p. 45). And it faults the absence of “precautionary measures to reduce exposure levels below those known, or reasonably anticipated, to cause blood diseases” (p. 46), which is partly anticipatory. So “precaution” in the ordinary sense of protective action is central to the chapter; only the formal principle is set aside. 3. Ignored substitution advice. Substitution recommendations made “in the United States and the United Kingdom in the 1920s” went “unheeded for decades” even though high percentages of surveyed workers had blood disorders (p. 46). The text elsewhere documents only a US 1926 recommendation (Greenburg) and 1939 US sources. No UK 1920s source is cited. 4. Late, unverified consumer withdrawal. Benzene was withdrawn from US consumer products only in 1978, voluntarily by manufacturers, “and it has never been adequately validated” (p. 46). 5. Sweden. The “claimed ignorance” in the 1940s Swedish raincoat industry is “difficult to accept”. Sixty cases occurred in a single factory, and case reports in “this same industry” had been published in 1897. Unawareness “in such a small industry in a country known for its humanitarian concerns is incomprehensible” (p. 46). This is rhetorical, not evidential. See also the bicycle-tyre/raincoat inconsistency above. 6. UK deaths. The 13 UK neurotoxic deaths occurred at levels “most likely more than 200 to 800 times” the recommended limits. “Clearly, this situation could, and should, have been avoided had there been any serious concern for worker health” (p. 46). (The multipliers are upper bounds on p. 39’s own arithmetic, so “more than” overstates; see 4.2.2 above.) 7. Turkey 1961. The substitution is “difficult to understand” given what was known. Aksoy says it happened because benzene was cheaper. “Data on the costs … are not available, but it is unlikely that the difference could have been more than a few cents a gallon” (an explicit conjecture). The resulting epidemic “had to have been very costly”: disease, health care costs, lost wages. “This is simply a case, as in other instances, of the cost of production being more important to the manufacturers than the cost of human life” (p. 46). 8. Consensus limits and producer influence. - Despite “numerous case reports (in the thousands)”, precautionary measures were not taken. Consensus limits (ACGIH) “were based on those that were easily achievable in the workplace”. - According to Castleman and Ziem, these stemmed from the participation of corporate-employed scientists on the TLV Committee. - The one explicitly labelled lesson: “consensus organisations … should maintain distance from the producers of the chemicals and their ‘consultants’ when evaluating evidence” (p. 46). 9. Gasoline warning labels. Affixing pump warning labels listing the cancers and other diseases known or likely to be caused by benzene “may serve to reduce unnecessary benzene exposure” to mechanics, attendants, highway workers and consumers, including home solvent use. Failure to inform the public “is to repeat our failures of the 20th century in the 21st century and to make a mockery of public health education” (p. 46).

Table 4.1 “Benzene: early warnings and actions” (p. 47; Source: EEA)#

This is an editorial compilation. Its entries: 1897 Santessen (bone marrow poison); 1926 Greenburg (low white counts); 1928 first leukaemia case; 1939 substitution recommended “but this was not implemented”; 1946 ACGIH 100 ppm “even though” poisonings at 25 and 10 ppm; 1947 50 ppm; 1948 35 ppm; 1948 API “only absolutely safe level is zero, but recommends 50 ppm or less”; 1957 ACGIH 25 ppm; 1950s–60s “Obvious lack of precaution … with fatal consequences”; 1977 Infante et al. first cohort study; 1977 Department of Labor seeks 1 ppm, challenged by API; 1978 voluntary consumer withdrawal; 1980 Benzene Decision “severely limiting regulatory actions”; 1987 1 ppm standard (“This 10-year delay caused more than 200 deaths in the United States”); 1996 studies showing disease from 1 ppm; 2001 petrol contains benzene, giving public exposure.

Editorial compression to note: - The 1926 substitution recommendation is folded into 1939. - “Caused more than 200 deaths” converts a projection (198 leukaemia deaths plus 77 myeloma deaths, “will eventually develop”) into a completed historical fact. The number itself is a conservative rounding of the 275 total. - “Severely limiting regulatory actions” is stronger than the court’s own hedges as the chapter reports them (“not a mathematical straitjacket”, p. 40). It is close, though, to Infante’s own verdict that the decision “has had a major impact on OSHA’s ability to control exposures” (p. 40) and to his “straightjacket” (p. 41). - The 1996 date stands in for Hayes et al. 1996/1997.

4.5 References (pp. 47–51): observations on the evidence base#

• There are about 83 entries. Primary documents are used for the core chronology:
• ACGIH Transactions for 1946, 1948 and 1957;
• the API 1948 Toxicological Review;
• OSHA Federal Register notices from 1977 (×2), 1978 and 1987;
• the Supreme Court decision;
• Olson’s 1977 testimony (OSHA docket exhibit).

This gives the limit-versus-evidence chronology a solid documentary basis. - Eleven entries are by Infante as author or co-author, including the four most consequential quantitative claims: the 1977 cohort, the 1988 delay-deaths estimate, the 1997 risk assessment and the 1990 gasoline paper. - Opposing literature is cited, mainly so that it can be rebutted: Wong (1995, 1996, 1998, 1999), Budinsky et al., Bergsagel et al., Tabershaw and Lamm, Olson, Crump, Paxton et al. Wong’s earlier industry-wide study (1987a, b) is cited the other way, as supporting evidence for all leukaemias and CML (p. 42) and against Wong’s own later claims (p. 44). - Some key evidence is unpublished (“Hayes, pers. comm., 1999”; “Hayes, pers. comm.”) or “in press” (Hayes et al., 2001). - Name variants suggest editing slips: Santessen/Santesson; Dolore/Delore; Schrader (text) vs Shrader (references); Zeim (p. 46) vs Ziem; Naizi (possibly Niazi). OSHA is expanded as “Occupational Health and Safety Administration” (p. 40); the agency is the Occupational Safety and Health Administration, as the author biography has it (p. 197). The Supreme Court citation is given as 448 US 601; the correct citation is 448 U.S. 607 (confirmed from outside sources).

Case timeline#

The column headed Strength gives my rating of the warning or action as evidence, at the time.

Lags (derived from the chapter’s dates): - First credible warning of bone-marrow toxicity to the first dated recommended limit: 1897 → 1946, about 49 years. Caveat: p. 40 refers to “various limits recommended over the time period 1940–1971”, which implies some limit existed from 1940 (about 43 years). The chapter dates no limit before the ACGIH’s in 1946. Using Greenburg’s 1926 survey as the first strong quantitative warning: 20 years to a limit, and the limit set (100 ppm) was above the 90 ppm at which Greenburg had found the effect. - Substitution first recommended (1926) → broad substitution: never achieved in the period covered. Benzene use expanded after 1945, and in Turkey the substitution ran the other way in 1961. - First leukaemia case report (1928) → a limit set explicitly with cancer in view (1977 ETS; 1987 final standard): 49 years to the attempt, 59 years to the enforceable 1 ppm limit. - First benzene-specific cohort (1977) → final 1 ppm standard (1987): about 10 years, delayed mainly by litigation and the post-1980 risk-assessment requirement. - Low-dose evidence (≈1 ppm; 1996–97) → revision of the US 1 ppm limit: none as of 2001. The chapter argues the limit is already inadequate. - Gasoline hazard first reported (1928) → adequate consumer information: not achieved as of 2001 in the author’s view.

The authors’ own lessons and conclusions#

Lessons Infante derives from his evidence: 1. Knowledge was available; action did not follow. Exposure levels “were not reduced to levels commensurate with the toxicity data available at the time” (p. 42). Precaution “in accordance with the scientific data available at the time would have eliminated much needless suffering and death” (p. 45). This is supported by the limit-versus-evidence chronology (pp. 38–39). 2. This was not a precautionary-principle case. In the workplace, “the precautionary principle is not relevant” (p. 46). My gloss, since the author does not explain the sentence or use the word “prevention”: the failure was one of prevention in the face of known harm. His own text still speaks of “lack of precaution” and “precautionary measures” (pp. 42, 45, 46); see 4.4 item 2. 3. Consensus standard-setters anchored limits to achievability and were subject to corporate influence (pp. 43, 46). This rests on the numerical mismatch, one external study (Castleman and Ziem, 1988) and personal experience. In the text itself he hedges (“may have played a role”, p. 43). In the conclusions he reports Castleman and Ziem’s finding (corporate scientists on the TLV Committee) without that hedge and draws a lesson from it “if not already obvious” (p. 46). 4. Economic motives drove overexposure. First the cheapness and quality of benzene as a solvent (pp. 42–43, 46). Later the cost of lowering exposures and of liability (p. 43). 5. Protracted dose-response argument delayed protection and cost lives. The estimate is 198 + 77 projected deaths from the 1977–87 delay (p. 41). “Studying a subject to death…” (p. 44). 6. The post-1980 risk-assessment regime has become a “straightjacket” that adds years of delay through mechanism and species arguments (p. 41). 7. The current 1 ppm limit is inadequate on cancer risk alone (p. 41). 8. Industry-consultant critiques misrepresent data (hedged as “in the opinion of some”, p. 44) and repeat the post-1977 pattern (p. 45).

Recommendations and advocacy (normative prescriptions): - “Consensus organisations … should maintain distance from the producers of the chemicals and their ‘consultants’” (p. 46). This is the only item labelled “one of the lessons”. - Warning labels on gasoline pumps listing the cancers and other diseases known or likely to be caused by benzene (p. 46), and fuller MSDS disclosure (implied, p. 45). - Implied: act on the NCI/CAPM low-dose data rather than wait for certainty (p. 45). Put “reasonable precautions” in place of open-ended dose-response analysis (p. 41). - Implied: independent, open, internationally developed exposure guidelines (endorsing Castleman and Ziem, p. 43).

Normative judgments beyond the evidence (advocacy register): - “the cost of production being more important to the manufacturers than the cost of human life” (p. 46); - the “anti public health approach” (p. 43); - the Swedish “humanitarian concerns” argument (p. 46); - “make a mockery of public health education” (p. 46).

Editors’ use of this chapter in Chapter 16 (not Infante’s text; for cross-reference): - acute effects preceding chronic ones, “sometimes by many decades”, though the editors add that this “should not necessarily be taken as a general rule” (p. 171); - benzene data “not collated”, no systematic monitoring (p. 172); - petrol as a neglected major exposure route (p. 175); - undue influence of interested parties; the animal-carcinogenicity objection; “demonstrably erroneous claims (based, for example, on basic statistical errors)” about low-dose effects (p. 179); - short-term horizons (p. 180); - the “evidently anti-precautionary ‘straightjacket’” imposed by the Supreme Court as an example of “paralysis by analysis” (p. 181); - the differing US regulatory sub-culture (p. 182); - the editors’ own balancing caveat: whether a need for more information “risks ‘paralysis by analysis’ or is merely part of a ‘prudent and careful evaluation’” depends on each party’s assessment of “the likely pros and cons as they impinge upon them” (p. 182).

Mechanisms and dynamics#

1. How warnings arose#

• Clinical medicine first. The warnings came first from physicians seeing individual workers (from 1897; the first leukaemia case in 1928), then also from industrial hygiene surveys (1926, 1939), then from epidemiology (1970s–90s), and only last from animal bioassays (1979, 1986) (pp. 38–40, 44). The human evidence came before the laboratory evidence. This reverses the order the 1977 objection assumed, which treated animal data as a precondition (pp. 43–44).
• Acute signals preceded chronic ones. Visible, fast outcomes (CNS toxicity, aplastic anaemia within weeks, p. 38) came decades before the chronic outcome (leukaemia) was quantified (1977). Clinicians suspected the link early (1928: aplastic anaemia cases “may also have been leukaemia”, p. 38). The editors note the acute-before-chronic pattern across several cases, but caution that it “should not necessarily be taken as a general rule” (p. 171).
• The disease spectrum kept widening as more and better studies arrived (p. 42): aplastic anaemia → acute leukaemia → chronic leukaemias → MDS, NHL, multiple myeloma. The “safe” level kept falling as well: effects at 200–1,000 ppm (early poisoning reports; the chapter does not date this range to a single source), ≥90 ppm (1926), 10–25 ppm (1939), within 10–100 ppm limits (1977), around 1 ppm (1997) (pp. 38–42). Limits were set above effect levels already reported (100 ppm in 1946, against 1926 and 1939 findings). Each limit was later found to sit above newly observed effect levels (1977, 1997).
• Caveat (mine): the sequence mixes endpoints (blood counts and aplastic anaemia early; leukaemia, lymphoma and MDS later) and study designs of rising sensitivity (case reports, then surveys, then cohorts). Part of the falling trend therefore reflects better detection of smaller risks, not only new hazard knowledge.

2. How warnings were ignored, then contested#

The chapter describes two regimes. - Neglect regime (c. 1900–1960s). Warnings were not so much disputed as not acted on: - substitution advice was ignored; - limits were set above documented effect levels; - a firm rotated workers instead of reducing exposure (one reported instance, 1928); - ignorance was claimed (pp. 38–39, 42–43).

The implied driver is economic convenience: a cheap, excellent solvent. - Contestation regime (1970s–2000). “In the 1970s, benzene manufacturers and users began a new approach” (p. 43): - consultants were hired; - evidence was reinterpreted (“random cluster”, p. 43); - new evidential preconditions were imposed (animal data, pp. 43–44); - exposure reconstructions were contested (p. 44); - disease-by-disease causal specificity was demanded (p. 43); - reanalyses produced nulls (p. 42, 44–45); - standards were litigated (p. 40).

The chapter links the timing to rising stakes: “the cost of lowering exposures” (citing OSHA, 1987) “and perhaps” the rising cost of litigation and liability (p. 43). Connecting the cost concern to OSHA’s move towards 1 ppm is my inference. Read this way, active contestation grew as the evidence strengthened and the cost of acting rose, and not while the evidence was weak.

3. Law, courts and the standard of proof#

• Litigation as the decisive venue. The API’s legal challenges stayed the 1977 ETS and vacated the 1978 standard (p. 40). The scientific dispute was settled, or deferred, in court.
• The Benzene Decision moved the burden onto the regulator. (“Burden” is my word; the chapter says the Secretary of Labor “is required to make a threshold finding”. Outside sources confirm that the plurality put the burden on the agency.) OSHA had to make a threshold showing of “significant risk”, quantified where possible (p. 40). In effect (my paraphrase of the chapter’s pre- and post-1980 contrast, pp. 40–41; Infante does not put it this way), this changed the regulatory question from “is this a carcinogen, and what is the lowest feasible level?” (OSHA’s pre-1980 policy, p. 40) to “demonstrate a quantified, significant risk at current levels”. That in turn made every link in the risk model (mechanism, species, exposure reconstruction, model form) a contestable step (p. 41, 44).
• Stated standards versus operative standards. Formally the court asked only for the “best available evidence” and disclaimed a “mathematical straitjacket” (p. 40). In practice, Infante says, the obligation to respond to every mechanistic and species argument produced one (p. 41). The formal standard of proof was moderate; the procedural standard, in effect, became very high.
• Feasibility trumped risk in the end. The 1987 limit was set on economic feasibility despite a residual risk ten times OSHA’s own significance benchmark (p. 41). My inference: the risk-quantification apparatus delayed action without determining its level.

4. How uncertainty was framed#

• Demands for certainty in the plural. Causal certainty was demanded “for every specific lymphohaematopoietic disease … by exposure level” (p. 43). Splitting the question into disease × dose cells guarantees that some cells lack statistical power (e.g., four multiple myeloma cases, p. 44). Underpowered cells can then be presented as “no evidence”.
• Absence of evidence treated as evidence of absence. Examples: the four-case myeloma dose-response (p. 44); “no evidence that benzene caused cancer in experimental animals” (p. 43).
• Irreducible uncertainty. Exposures in periods with no measurements can “never be determined with scientific certainty” (p. 44). Demanding their resolution before acting is effectively a demand for indefinite delay.
• Infante’s counter-frame. “stressing the uncertainty does not do justice to the data” (p. 45), and his inversion of precaution: more caution is being applied to using the data than to protecting people (p. 45).
• Endpoint definition as a lever. Crump’s lower estimates were restricted to AML/monocytic leukaemia (p. 41). Wong argued for AML only (p. 44). Narrowing the outcome definition lowers the estimated risk. The chapter does not state this mechanism in general terms, but its examples show it.

5. Control of data and reanalysis#

• Record removal by the company. The company removed death records for certain pre-1950 years at one plant. A reanalysis that extended follow-up back to 1940 therefore built in selection bias (p. 42). The regulated party’s custody of records shaped what could be validly inferred.
• Selective use of one’s own data. Infante says Wong’s 1995 claims ignored Wong’s own 1987 dose-response findings (p. 44).
• Competing exposure reconstructions from “the various parties” became the main arena of dispute (p. 44). Parties could each produce plausible-looking but unverifiable estimates.

6. Consensus bodies, conflicts of interest and mental models#

• ACGIH mental model: “achievable = acceptable”. Limits track “what is easily achievable in the workplace”. Data are reviewed but not translated into health-based limits (p. 43). This is a feasibility-anchored mindset that treats existing practice as the reference point.
• Participation by corporate scientists on the TLV Committee, per Castleman and Ziem (p. 46).
• API mental model (1948). It accepted the science (no safe level, because of variable susceptibility) yet recommended a limit (≤50 ppm) that tolerated known harm (p. 39). This separates knowing from standard-setting, and it happened inside industry.
• Employer mental model (1928). Rotation: the problem is treated as individual dose accumulation to be spread around, not exposure to be removed (p. 38).
• Management in Sweden: claimed ignorance (p. 42). Infante disputes this (p. 46).
• Industry consultants’ model (1970s–90s): thresholds (no risk below 10 ppm, p. 40; 370–530 ppm-years, p. 44), a narrow disease spectrum (AML only), and a requirement for animal confirmation.
• Court’s model: risk management as a matter of “significance”. Trivial risks should not be regulated, and quantification is the discipline that tells them apart (p. 40). The chapter reports this fairly but does not engage with its rationale: prioritising among many hazards and avoiding costly regulation of trivial risks.
• Infante’s own model: lowest feasible exposure (OSHA’s pre-1980 policy, which he reports on p. 40 and implicitly favours over the post-1980 regime, preferring “reasonable precautions” to open-ended analysis, p. 41); human evidence is sufficient (p. 44); delay is itself a quantifiable harm (p. 41). A no-safe-threshold view is implied (he rejects the proposed thresholds, pp. 40, 44) but never stated as such.

7. Economics: costs, benefits and who bears them#

• Benefits to producers and users. Benzene was cheap and an excellent rubber solvent (pp. 42–43); lowering exposures has costs (p. 43); liability can be avoided (p. 43).
• Costs to workers. Disease, death, health care, lost wages (p. 46), and “little or no compensation” (p. 43). The groups named are:
• young women in Swedish rubber-goods manufacture (58 of 60 cases were women, p. 42);
• Turkish shoe and slipper workers (p. 39);
• Italian rotogravure and shoe workers (p. 39);
• US Pliofilm workers (p. 40);
• mechanics, highway workers and petrol attendants (p. 45);
• informal roadside fuel vendors in Nigeria (p. 45);
• consumers (pp. 40, 45).
• Asymmetry the chapter asserts without quantifying. A saving of perhaps “a few cents a gallon” (explicit conjecture) set against an epidemic of fatal disease (p. 46). The chapter gives no numbers for the benefits of benzene use or the compliance costs of the 1 ppm standard.
• Delay costs are locked in. Deaths “will eventually develop” from exposures during the delay (p. 41). My inference, since the chapter does not discuss latency: harms from delay cannot be reversed once the exposure has happened, and they appear only years later.
• Accounting asymmetry. Harms without dose-response data (the other blood diseases) are left out of the quantified cost of delay (p. 41). Formal cost estimates therefore systematically understate harm.

8. Substitutes, feasibility and innovation#

• Substitutes were available and recommended from 1926 (pp. 39, 42, 46). The chapter asserts they were “known to be less toxic to bone marrow” (p. 46) but does not name them or assess their own hazards. Among the solvents McMichael’s cohort used (petroleum naphtha, toluene, mineral spirits, p. 40), some were contaminated with benzene.
• Reverse substitution. In Turkey in 1961 a known hazard was introduced on cost grounds (p. 39). This shows that hazard knowledge in one place does not stop adoption in another where the price signal dominates.
• Feasibility demonstrated after the fact.
• Consumer products were withdrawn voluntarily after high home exposures were shown (p. 40). The chapter gives no timing that would show this was prompt, and Infante frames it as late (“not withdrawn … until 1978”, p. 46).
• Most US workplaces achieved 0.2–0.3 ppm, below the 1 ppm limit (p. 41).

The chapter implies that lower exposures were technically attainable. My inference is that the limit then lagged behind achieved practice, so even the achievability anchor stopped moving the limit downward. The chapter says nothing about innovation induced by regulation.

9. Lock-in and pervasiveness#

• Embedding across uses. Benzene spread into a wide range of industrial uses and into petroleum products and motor fuel (p. 38). By 2001 the chapter presents gasoline (US average about 1.5% benzene, up to 5%) as a continuing and largely unaddressed route of worker and public exposure (p. 45). It does not claim this is the largest route; the editors’ version is that controlled routes “are not necessarily the routes of greatest exposure” (p. 175).
• Mismatch between control and exposure routes. Occupational control improved while the diffuse route, through a ubiquitous product, stayed unlabelled and largely unaddressed (p. 45). The editors turn this into a general point: controlled routes “are not necessarily the routes of greatest exposure” (p. 175).
• The chapter does not analyse why benzene remained in fuel (e.g., its role in fuel properties). Chapter 10 on MTBE, p. 111, mentions benzene’s octane role; that is outside this section.

10. Information and communication#

• “Lack of candour” in hazard communication. Pumps give no adequate information and MSDS omit genotoxicity (p. 45). The result is risky everyday practices: gasoline used as a solvent at work and at home (p. 45).
• Naming frames the hazard. Calling aplastic anaemia “benzene poisoning” (p. 38) framed the hazard as acute poisoning, which by implication can be managed by limiting heavy exposure, rather than as a carcinogenic, no-threshold hazard. This is my inference; the chapter only records the terminology.
• Framing language on each side.
• Industry consultants: “random cluster” (p. 43).
• Infante: “anti public health” (p. 43); scare quotes around “risk assessment” (p. 41), “updated” (Wong’s reanalysis, p. 42), “consultants” (p. 46) and “educated” in “‘educated’ guesses”, a sceptical label he applies to the exposure estimates of “the various parties involved in the rule-making”, not to industry alone (p. 44); “straightjacket” turned back on the court’s own disclaimer (p. 41).

11. Institutional behaviour#

• Governments and consensus bodies failed to act for decades (p. 46). OSHA acted in 1977 (an emergency standard at 1 ppm) but was held back by courts and procedure (pp. 40–41). The public health community, including industry safety staff, failed on gasoline communication (p. 45). Voluntary industry action on consumer products came, in Infante’s framing, late (1978) and was never “adequately validated” (pp. 40, 46).
• Short horizons (editors, p. 180) and paralysis by analysis (editors, p. 181) are the editors’ generalisations.

12. Transnational dynamics#

• Uneven protection across places. Harm recurred in country after country (Sweden, Italy, France, UK, Russia, Turkey, US) over decades. Published knowledge did not reliably protect workers elsewhere (p. 39).
• Evidence travelled; protection did not. Chinese cohort data are, in Infante’s account, the key new low-dose evidence. They “may be used in the future by governments in Europe, the United States and other countries” to estimate general-population risk (p. 44). This is a prospective claim, not a record of use. Informal-economy workers (Nigeria) appear as a highly exposed, unprotected group (p. 45).
• Thin European regulatory side. Despite the chapter title, it contains no account of European regulatory responses (e.g., national or EU occupational limits). The “European setting” is represented by harm reports and petrol composition only. This partly reflects the editors’ brief: they commissioned the chapter as one of three focused “primarily on the North American handling” of issues relevant to Europe (p. 11).

Transferable insights (technology-neutral)#

1. Knowing is not acting: the binding constraint was institutional, not epistemic. Credible, repeated, published warnings (clinical from 1897, quantitative from 1926) did not produce protective limits for decades. When limits came, they were set above documented effect levels. Evidence: pp. 38–39, 42, 46; Table 4.1 p. 47. Strength: strong. The chronology rests on primary documents (ACGIH transactions, the API review, surveys). The causal attribution for why is weaker; see 2, 3 and 6.

2. Standard-setting anchored to what is “easily achievable” builds current practice into the definition of acceptable risk. Participation by producer-affiliated experts can reinforce this. Evidence: ACGIH 100 ppm (1946) against poisonings at 10–25 ppm (1939) and effects at ≥90 ppm (1926), p. 39, 43; Castleman and Ziem (1988), p. 43, 46; Infante’s personal experience, p. 43. Strength: moderate. The numerical mismatch is documented. The mechanism (achievability anchoring, corporate influence) rests on one external study plus participant testimony, and the chapter itself hedges (“may have played a role”, p. 43).

3. Those who produce a hazard can hold, and even state, knowledge that runs ahead of the standards they recommend. Evidence: API 1948: “only absolutely safe level … was zero” yet recommending ≤50 ppm (p. 39). Strength: strong as a documented instance; suggestive as a general pattern (single example in this section).

4. Administrative work-arounds that spread a hazard around can stand in for reducing it. Evidence: monthly job rotation in response to high exposures (1928), p. 38. Strength: suggestive. A single instance, reported briefly.

5. Cost advantages can drive adoption of a known hazard in new places long after the hazard is established elsewhere. Knowledge of harm does not travel with the product. Evidence: Turkey 1961 switch to benzene because it was cheaper, followed by an epidemic (pp. 39, 43, 46); continued global growth in use after 1945 despite advice to substitute (p. 42). Strength: moderate. The switch and the epidemic are documented via Aksoy. The cost comparison is explicitly conjectural (p. 46).

6. Contestation grows as evidence strengthens and the stakes of acting rise, and it shifts from neglect to active challenge. The tactics described are: - alternative explanations (chance clustering); - moving evidential preconditions (animal confirmation); - demands for certainty disease by disease and dose by dose; - reanalyses built on choices that produce nulls.

Evidence: pp. 43–45 (1977 hearings; 1980s rulemaking; 1990s NCI/CAPM critiques; Wong reanalyses); editors, p. 179. Strength: moderate. Specific episodes are documented with citations, and some rebuttals are strong internal-consistency arguments (Wong’s own 1987 data; four-case power). But the account comes from a protagonist in these disputes, the opposing case is presented only in summary, and motives are attributed “in my opinion” (pp. 43, 45).

7. Requiring comprehensive quantification before action can itself cause delay, especially when every alternative mechanism or model must be answered. The delay has its own health cost. Evidence: Benzene Decision and after (pp. 40–41, 44); the ~10-year 1977–87 gap (p. 41); projected 198 + 77 deaths (p. 41); editors’ “paralysis by analysis” (p. 181). Strength: moderate. The legal requirement and the timeline are documented. That the requirement caused the delay is plausible but untested: there is no comparison with a counterfactual or with other standards. The death figure is the author’s own projection. The editors themselves concede that whether a call for more information is “paralysis by analysis” or “prudent and careful evaluation” depends on how the pros and cons fall on each party (p. 182).

8. Some uncertainties cannot be reduced, so demanding they be resolved before action means delaying indefinitely. Evidence: retrospective exposure estimates for periods without data: “educated guesses … could not be confirmed”; the question “can never be determined with scientific certainty” (p. 44). Strength: moderate–strong. The logic is sound and the instance is documented. The general claim that such disputes are only delay is the author’s inference.

9. Whoever keeps the records shapes what the evidence can show. Analytic choices that look neutral can produce null results. Evidence: company removal of pre-1950 death records; the follow-up start date moved to 1940 introduces selection bias (p. 42); endpoint narrowing lowers risk estimates (pp. 41, 44; my inference from the chapter’s examples, not a point it states). Strength: moderate. A specific, checkable claim, but asserted without the underlying documentation in this chapter. Its consequences for Crump and Paxton are asserted, not shown.

10. What cannot be quantified drops out of the accounting, so formal estimates of the cost of delay are biased low in their coverage of harms. Evidence: the delay-deaths estimate excluded other known benzene blood diseases for lack of dose-response data (p. 41); NHL was excluded too (p. 41). Strength: strong for the coverage point, which the chapter states explicitly. Its size is unknown. The estimate as a whole is not shown to be an underestimate: it rests on OSHA’s risk model, and the chapter reports lower competing estimates (Crump, 1994: 0.02–5.1 per 1,000 for AML and monocytic leukaemia only, presented as an alternative to the 10 per 1,000 associated with 1 ppm; p. 41). Model uncertainty cuts both ways.

11. The scope and level of harm tend to be revised in one direction over time: more outcomes, at lower doses. Each era’s “safe” level has sat above the next era’s observed effect level. Evidence: 200–1,000 ppm → ≥90 → 10–25 → within 10–100 ppm limits → ≈1 ppm (pp. 38–42); outcomes widening from aplastic anaemia to multiple leukaemias, MDS, NHL and myeloma (p. 42); editors on acute→chronic, with their caution that it is not “a general rule” (p. 171). Strength: strong for the falling-threshold trajectory within this case, as a description. The trajectory mixes endpoints and increasingly sensitive study designs, so it partly measures better detection rather than only new hazard. Moderate for the disease-spectrum expansion, where several endpoints (myeloma, NHL, CLL) were genuinely contested in 2001 and the evidence cited varies from cohorts to single case reports. Generalising beyond this case is suggestive.

12. Diffuse exposure through a ubiquitous product can persist, poorly communicated, after the concentrated occupational route has been controlled. Evidence: benzene in gasoline (1.5–5%); no adequate warnings; MSDS omissions; gasoline used as a solvent (p. 45); editors, p. 175. Strength: moderate. The information failure is documented. The health burden at consumer exposure levels is not quantified, and the studies cited are mixed in design and quality.

13. Harms fall on those with least information and power, and benefits accrue to producers. Compensation often fails. Evidence: women rubber-goods workers (p. 42); Turkish shoe workers (p. 39); informal fuel vendors (p. 45); “little or no compensation” (p. 43); “cost of production” versus “cost of human life” (p. 46). Strength: suggestive. The pattern is asserted and illustrated, not analysed. There are no data on compensation, and the benefits side is not quantified.

14. Voluntary industry action can come late and go unverified. Evidence: benzene “was not withdrawn from consumer products in the United States until 1978”, voluntarily by manufacturers, “and it has never been adequately validated” (p. 46). It followed a demonstration of high home exposures from paint stripper (Young et al., 1978, p. 40), which Infante co-authored. The chapter gives no timing that would show the withdrawal was prompt. Strength: suggestive. A single instance, stated without supporting detail.

15. Some of the “late lessons” are failures of prevention in the face of known harm, not failures of precaution under uncertainty. The two call for different remedies. Evidence: “the precautionary principle is not relevant” to workplace benzene (p. 46). The sentence fits the pre-1970s neglect era best. Two later episodes in the chapter are genuinely about uncertainty. The 1977 emergency standard cut the limit to 1 ppm under a “lowest feasible” policy for carcinogens, before any quantified leukaemia risk below 10 ppm existed and against the API’s argument that there was none (p. 40). The 2001 low-dose and disease-spectrum dispute is the second (pp. 44–45). Strength: moderate. The author’s sentence is explicit, but it is a single unexplained line. His own vocabulary (“lack of precaution”, pp. 39, 42; “precautionary measures” against levels “reasonably anticipated” to cause disease, p. 46) blurs the line he draws. The prevention/precaution distinction is my conceptual framing, prompted by the text. It still matters for how any analysis based on these reports is structured.

16. Formal limits can come to lag behind practice: industry achieves more than the rules require, while the rules stay put. Evidence: most US workplaces at 0.2–0.3 ppm against the 1 ppm limit (p. 41); 1 ppm set on feasibility with a residual risk of 10/1,000 (p. 41). Strength: suggestive. One data point. Partly my inference from the chapter’s figures.

Limitations, contestation and bias check#

Where the section is advocacy rather than analysis - Motive attribution: - “anti public health approach” (p. 43); - “cost of production being more important to the manufacturers than the cost of human life” (p. 46); - “denial” (p. 45); - “incomprehensible” given Sweden’s “humanitarian concerns” (p. 46); - “make a mockery” (p. 46).

Some are flagged as opinion (“In my opinion”, pp. 43, 45) or hedged (“raises the question”, p. 45). Others are stated flatly in the conclusions (p. 46). - Opponents speak only through the author. No industry, consultant or court-side rationale is presented in its own terms. The strongest case for quantitative risk assessment is never stated or engaged: that it sets regulatory priorities across thousands of substances and avoids expensive regulation of trivial risks, which was the court’s concern (p. 40). He grants only that such analyses “may seem appropriate” on the surface (p. 41) and that “nobody would object to debate” (p. 44). - Participant-author problem. The chapter’s most consequential numbers are the author’s own work: the 1977 cohort, the 1988 delay-deaths estimate, the 1997 risk assessment, the gasoline evidence. He was a party to the disputes he adjudicates (Infante et al., 1977b, vs Tabershaw and Lamm). The editors disclose this (p. 12), but the chapter does not flag it at the points where it matters. He does not mention, for example, that he co-authored Young et al. (1978), p. 40.

Where the evidence is thin - The pre-1977 low-dose claim (“10 ppm level already associated with aplastic anaemia” by 1939) rests on two cases among 89, with 1930s exposure measurement (p. 39). The chapter’s argument about the 1946–57 limits is much better supported by Greenburg’s 1926 survey (effects at ≥90 ppm, against a 100 ppm limit in 1946) than by the 10 ppm figure it repeatedly stresses. - The “200 to 800 times” multipliers use the top of an inferred exposure range (pp. 39, 46). - The cost of the Turkish substitution (“a few cents a gallon”) is explicitly conjecture (p. 46). - “Numerous case reports (in the thousands)” (p. 46) is never counted in the chapter. - Low-dose evidence partly unpublished (Hayes, pers. comm., pp. 42, 44). The headline 54-per-1,000 risk figure comes from the author’s own Toxicology Forum proceedings paper (Infante, 1997; p. 41, reference p. 49). - The disease-spectrum list (p. 42) mixes robust cohort findings with single case reports and letters without grading them. - The gasoline health evidence (p. 45) is a heterogeneous mix of small studies and case reports. “Not surprisingly” and “overwhelming literature” do more work than the evidence. - The discounting of Crump and Paxton (p. 41) relies on a selection-bias argument set out for a different analysis (Wong, 1995) and not shown to apply to them.

Internal inconsistencies and errors - Santessen 1897 cases in bicycle tyres (p. 38) versus raincoats (pp. 42, 46). The Sweden “same industry” argument depends on the latter. - “Eleven years” versus “10-year delay” (p. 41). - Pre-1977 leukaemia causation “not upon epidemiological studies” (p. 40), after citing Vigliani and Saita’s 20-fold risk (1964) and the UNC studies (pp. 39–40). - “200 to 800 times” (p. 39) becomes “more than 200 to 800 times” (p. 46). - OSHA expanded as “Occupational Health and Safety Administration” (p. 40). - “United States and the United Kingdom in the 1920s” substitution recommendations (p. 46): no UK source is given. - Goguel “44 cases” versus reference title “50 cases” (pp. 39, 48). - Name spellings (Santessen, Dolore, Schrader/Shrader, Zeim/Ziem, Naizi). - Supreme Court citation (448 US 601; correctly 448 U.S. 607). - Table 4.1 (EEA) hardens the projection “will eventually develop” into “caused more than 200 deaths” (p. 47).

Counter-arguments acknowledged or ignored - Acknowledged: - before 1977 leukaemia causation rested on case reports and clinical inference, not epidemiology (p. 40); - the court disclaimed a “mathematical straitjacket” and left significance to policy (p. 40); - US workplaces achieve 0.2–0.3 ppm, and the standard’s ancillary provisions reduce risk (pp. 41–42); - debate on exposure and model issues is legitimate in itself (p. 44); - lower risk estimates exist (Crump, 0.02–5.1/1,000, p. 41). - Ignored or absent: - benzene’s economic and industrial value; - OSHA’s compliance-cost estimates for the 1 ppm standard; - the rationale for QRA as a prioritisation tool; - any scientific case for sublinear dose-response or thresholds (dismissed as “speculative mechanistic data”, p. 41); - the role of unions and litigants in bringing the Supreme Court case (the case name, “AFL-CIO vs. American Petroleum Institute”, signals union involvement, but the text does not discuss it); - NIOSH’s own recommendations; - the ACGIH’s later movements (Infante 1992 in the references concerns a “0.1 ppm ACGIH proposed TLV”, p. 49, but the text never mentions it); - the transition to a 10 ppm limit (p. 40); - European regulatory action, entirely absent despite the chapter title and the report’s transatlantic aim (pp. 11–12). The editors did commission the chapter as focused “primarily on the North American handling” of the issue (p. 11), so this is partly by design.

Hindsight bias - The editors’ own test. Case-study conclusions “should be based on ‘the spirit of the times’ and not on the luxury of hindsight” (p. 11). The judgements below apply that test. - For the 1920s–1960s neglect era, the charge that limits ignored contemporaneous evidence is well founded on the chapter’s own sources (1926 survey, 1939 series, API 1948). This is not hindsight: the evidence was published and in some cases held by industry itself. - For leukaemia specifically, the quantitative risk at 10–25 ppm was not established before 1977 (p. 40). Judging the 1946–57 limits on cancer grounds would involve some hindsight. Infante’s case against those limits rests mainly on the aplastic anaemia evidence, which is appropriate. - For the 1977–87 dispute, the chapter treats the post-1977 arguments as essentially bad faith. Some points (the exposure reconstruction for the Pliofilm cohort, the choice of dose-response model) were genuine scientific uncertainties with real consequences for the estimates (0.02–5.1 vs 10 per 1,000, p. 41). The claim that the delay was unnecessary is a value judgement about how much uncertainty should delay action, not a finding.

How the pro-precaution framing shapes the conclusions - All the report’s case studies are “false negatives”. The editors say they sought “false positive” examples, including by inviting industry representatives to submit them, but “no suitable examples emerged” (p. 12). Benzene is a case where harm was real and severe, so it cannot test whether the same interventions (acting on case reports, rejecting QRA) would over-regulate elsewhere. - Infante himself says precaution is “not relevant” to the workplace case (p. 46), because the harm was known. This cuts against using benzene as a precautionary-principle exemplar. Yet his 2001 argument about low doses and the disease spectrum is a precautionary argument: act on the NCI/CAPM data despite contested exposure estimates (p. 45). So, in character, was the 1977 emergency standard he supports: 1 ppm as the “lowest feasible limit” for a carcinogen, set when the API could still argue there was no risk below 10 ppm (p. 40). Both kinds of argument are present, and they should be kept apart.

Fair in the other direction - Strong documentary spine. The core chronology uses primary sources (ACGIH transactions, API 1948, Federal Register notices, the Supreme Court decision, docket testimony). - Several rebuttals are specific and checkable rather than rhetorical: Wong’s own 1987 dose-response and absence of AML; the four-case power problem; the pre-1950 record removal (pp. 42, 44–45). - He uses hedges where the evidence is indirect: “may have played a role” (p. 43); “in the opinion of some” (p. 44); “raises the question” (p. 45); “presumably” (pp. 40, 41). - He flags his own method’s limits: the delay-death estimate excludes other diseases (p. 41). He also acknowledges that achieved exposures are lower than the limit (p. 41). - No dissenting panel exists. The absence of any industry or regulator response is a structural feature of the 2001 report, not the author’s choice.

Notable quotes#

1. “The company’s means of dealing with the high exposure levels that caused these blood diseases was to rotate the workers out of the specific job every month.” (p. 38)
2. The API “concluded that the only absolutely safe level from exposure to benzene was zero” … then recommended “50 ppm or less”. (p. 39)
3. “consensus organisations usually base their recommended exposure levels on what is easily achievable in the workplace.” (p. 43)
4. “a new anti public health approach that calls for scientific certainty in terms of causality for every specific lymphohaematopoietic disease related to benzene by exposure level.” (p. 43, flagged “In my opinion”)
5. “A straightjacket, however, appropriately describes the risk analyses that OSHA currently engages in prior to proposing any regulatory action.” (p. 41)
6. “These ‘educated’ guesses by the various parties involved in the rule-making could not be confirmed.” (p. 44)
7. “Studying a subject to death often results in the death of those we are trying to protect.” (p. 44)
8. “It will be unfortunate if more precaution is taken with the use of data from the NCI/CAPM study than with the protection of populations exposed…” (p. 45)
9. “In the case of benzene exposure in the workplace, the precautionary principle is not relevant.” (p. 46)
10. Consensus organisations “should maintain distance from the producers of the chemicals and their ‘consultants’ when evaluating evidence for the diseases of concern.” (p. 46)

Open questions#

About the text and its evidence (answerable from primary sources): 1. Were Santesson’s 1897 cases in bicycle-tyre or raincoat manufacture, or both? Does the p. 46 “same industry” argument hold? 2. What exactly did the ACGIH record as the basis for the 1946 100 ppm value (ACGIH Transactions 1946, p. 40)? Does Castleman and Ziem (1988) document corporate participation for benzene specifically? 3. How and when did the operative limit fall from 25 ppm (1957) to the 10 ppm TWA referred to on p. 40 (likely via ANSI/ACGIH and OSHA’s 1971 adoption of consensus standards, to verify)? 4. The pre-1950 record removal (p. 42): what is the documentary source? Did Crump (1994) or Paxton et al. (1994) rely on follow-up affected by it, as p. 41 implies? 5. The 1978 voluntary consumer withdrawal: what regulatory pressure accompanied it (e.g., a Consumer Product Safety Commission proposal) and was it ever verified? 6. Why is European regulatory action (national limits, EC/EU occupational and fuel measures) absent from a chapter titled “American and European”? (The editors’ brief was “primarily … North American handling”, p. 11, but the title promises more.) What would a European regulatory timeline show about transatlantic differences, the report’s stated aim (pp. 11–12)? 7. Olson (1977): in what capacity did he testify?

Counterfactual and interpretive: 8. Would the 1987 limit have been lower, or earlier, without the Benzene Decision? Or did QRA mainly change the justification, since the final limit rested on feasibility anyway (p. 41)? 9. How should one separate legitimate scientific dispute (exposure reconstruction, model choice) from strategic delay, without relying on a protagonist’s characterisation? What independent markers exist: funding source, reproducibility, whether the objections were later vindicated? 10. How large was the burden from gasoline-route exposure in 2001, compared with occupational exposure? Would pump labels have changed behaviour?

Pointers for the hindsight strand (from my background knowledge, NOT from the report; all to be verified against primary sources): - IARC’s later evaluations (Monographs Vol. 100F, 2012; Vol. 120, 2018) are generally reported to find sufficient evidence for AML/ANLL and limited evidence or “positive associations” for several other lymphohaematopoietic cancers (e.g., NHL, CLL, multiple myeloma, ALL). If so, this is partial support for p. 42’s broad spectrum claim, with AML firmly established and the rest less so. - Low-dose evidence after 2001: NCI-affiliated studies are reported to have found haematotoxicity in Chinese workers below 1 ppm (Lan et al., 2004, Science). Pooled petroleum-industry cohorts were later reported to show MDS risk at low cumulative exposures. The exposure-assessment disputes over the Chinese cohort reportedly continued. - Limits after 2001: - US OSHA PEL: understood to remain 1 ppm (29 CFR 1910.1028). If true, the p. 41 “inadequacy” claim was never acted on. - EU binding OEL: understood to be 1 ppm (Directive 97/42/EC), lowered by Directive (EU) 2022/431 to 0.5 ppm (2024) and 0.2 ppm (2026). - ACGIH TLV: understood to be 0.5 ppm since the late 1990s, with much lower values proposed in the 2020s. - Fuel benzene limits: - EU: 1% v/v from 2000 (Directive 98/70/EC), consistent with p. 45’s “supposedly … reduced”. - US: EPA’s 2007 mobile-source air toxics rule set a 0.62% annual average from 2011.

These were reductions in content, not the warning-label route Infante recommended. California Proposition 65 pump signage (from the late 1980s) bears on the p. 45 claim that pumps lack “adequate” information. - The pace of OSHA health standards after 1980, relevant to the “straightjacket” claim. - The author’s later career: worth checking only to contextualise his later publications, not this 2001 text.

Audit log#

Independent audit against the full extract (report pp. 38–51) and the PDF (pp. 11–12, 111, 171–182, 197 for the editors’ and biographical material the notes cite). Every quotation in the notes was re-checked verbatim. The changes below were made in the notes (N) and digest (D).

1. N, standpoint: reference count corrected from “roughly 75–80, about a dozen his own” to about 83, 11 with Infante as author or co-author.
2. N, standpoint: “sharply criticises the Supreme Court’s Benzene Decision” corrected. He reports the decision fairly and says it had a “major impact” (p. 40); his sharp criticism targets the post-1980 risk-assessment regime (p. 41).
3. N, standpoint: “wrote the 1988 estimate … 1990 gasoline analysis” corrected to co-wrote (with DiStasio; with Schwartz and Cahill).
4. N, standpoint: the four set questions are listed on p. 11; “as objective as possible” is on p. 12.
5. N, 4.1.1: “acute and chronic outcomes lie on a continuum” rewritten as a flagged gloss (aplastic anaemia and leukaemia linked).
6. N, 4.2.1: page references for the 10 ppm argument corrected from pp. 39, 43, 46 to pp. 39, 43 and Table 4.1 p. 47 (p. 46 does not use it).
7. N, 4.2.2: Vigliani peaks restored to “occasional”; added that p. 46 escalates to “more than 200 to 800 times”, beyond p. 39’s own upper-bound arithmetic.
8. N, 4.2.3: added the internal tension between “not upon epidemiological studies” (p. 40) and the Vigliani and Saita and UNC evidence reported on pp. 39–40.
9. N, 4.2.5: added Infante’s “major impact on OSHA’s ability to control exposures” (p. 40), previously omitted; “review and respond to” corrected to the source’s “review and comment on”.
10. N, 4.2.6: the 10/1,000 figure is no longer called “OSHA’s own estimate”; the chapter does not name its source. “Dismisses” softened to “discounts”, with the verbatim quote.
11. N, 4.2.6 and Table 4.1: the claim that Table 4.1 “drops the 77 myeloma deaths” was wrong. “More than 200” only fits the combined 275, since 198 < 200. The projection-as-fact point is retained.
12. N, 4.2.6: “by construction a lower bound” corrected. It is incomplete in disease coverage, but it depends on OSHA’s risk model, and the chapter reports lower competing estimates.
13. N, 4.2.7: “54/1,000 at the 1 ppm limit” corrected. The level is implied, not stated. Added that Infante (1997) is a four-page meeting-proceedings paper.
14. N, 4.2.7: hairy cell characterisation corrected to match the reference titles; the ALL citation to a childhood-leukaemia case-control study and the AML citation to a 1965 reference book added.
15. N, 4.3.1(a): added “four cases” (p. 42) and the reference title’s “four deaths” (p. 50) for Santessen.
16. N, 4.3.1(b): “synthetic rubber industry” context restored.
17. N, 4.3.1(d): the “meaningless” argument is passive-voice in the source; the attribution to consultants is now flagged. NTP 1986 is nine years after 1977, so “shortly thereafter” stretches; noted.
18. N, closing argument: truncated quotation restored to verbatim (“…the data on benzene exposure and related diseases”).
19. N, 4.3.2: “European petrol” corrected to “most European countries”; “take no precautions” corrected to “fail to take the necessary precautions”. Added the “overwhelming literature” overstatement.
20. N, 4.4 item 2 and lessons item 2: “precautionary principle is not relevant” is now marked as unexplained. Added that the chapter’s own “lack of precaution” and “precautionary measures … reasonably anticipated” language blurs the prevention/precaution line; the “prevention” label is flagged as the annotator’s.
21. N, Table 4.1: “severely limiting” is now set against Infante’s own “major impact” wording, not only the court’s hedges.
22. N, references: added the “Occupational Health and Safety Administration” name slip (p. 40).
23. N, timeline: the 200–1,000 ppm range is no longer attributed to Hogan and Schrader. Olson’s affiliation is marked as not stated. The NCI/CAPM “strong design (large cohort)” rating is replaced by what the chapter says (“well-conducted”, no design details). The 54/1,000 row is qualified.
24. N, lags: the first-limit lag is caveated, since p. 40 implies limits from 1940 (about 43 years, not 49).
25. N, recommendations: garbled “Replace ‘reasonable precautions’ for” corrected.
26. N, editors’ Ch 16: p. 171 now quoted as “sometimes by many decades”, with the editors’ caveat that it is not “a general rule”; mechanism 1 no longer says the editors “generalise” it.
27. N, mechanisms 1: “each generation’s limit above the next generation’s effect level” refined. Added the caveat that the falling trend mixes endpoints and study sensitivity.
28. N, mechanisms 2: “firms rotated workers” corrected to a single reported firm (1928).
29. N, mechanisms 3: the “changed regulatory question” framing and “delayed action without determining its level” are now flagged as the annotator’s paraphrase and inference.
30. N, mechanisms 6: “no safe threshold” is marked as implied, not stated, in Infante’s model.
31. N, mechanisms 7: the latency claim is marked as annotator inference (the chapter does not discuss latency).
32. N, mechanisms 8, 11 and insight 14: the 1978 consumer withdrawal is no longer called “quick” or “fast”. The chapter gives no timing, and Infante frames it as late (“not … until 1978”, p. 46). “OSHA acted boldly” is de-editorialised.
33. N, mechanisms 9: “main population exposure route is gasoline” corrected; the chapter does not claim it is the largest route.
34. N, mechanisms 10: “‘educated’ guesses” moved from industry framing to Infante’s own sceptical label, which he applies to all rule-making parties (p. 44).
35. N, mechanisms 12: the Chinese data are no longer “gathered under high exposures” or said to have “became key evidence for Western standard-setting”. The chapter says they “may be used in the future”. The European omission is tempered by the editors’ “North American handling” brief (p. 11).
36. N, insights 10, 11, 15: insight 10 narrowed to a coverage point, with model uncertainty noted. Insight 11 caveated for endpoints and design sensitivity. Insight 15 downgraded from strong to moderate.
37. N, limitations: added the proceedings-paper source for the 54/1,000 figure, the “overwhelming literature” overstatement and three internal inconsistencies. “False negatives by design” corrected: the editors sought false positives but found none suitable (p. 12). The European-regulation omission is tempered.
38. D: author-disclosure sentence corrected; Italian figures reworded; the Benzene Decision no longer “required a quantified” risk (hedges added); the 1987 residual risk specified (leukaemia, working lifetime); the 1988 projection attributed to Infante and DiStasio using OSHA’s model; the 54/1,000 level marked as implied; gasoline percentages attributed to the US.
39. D: lessons “Prevention, not precaution” and “Convenience-led limits” reworded. Mechanisms “Industry knew but did not act” and “reintroduced” corrected. Insight 9 narrowed. Insight 10 caveated. Insight 12 downgraded to moderate. Crump’s competing estimate added to caveats. Europe, hindsight and case-selection caveats corrected.
40. Checked and found accurate (no change): all chronology figures (ppm limits, dates, 32%/12%, 89/3 cases, 13 UK deaths, 20×, 200+/34, 44 cases, 150–650 ppm, 5–10×, 198/77, 0.02–5.1, 54/1,000, 1.2 ppm for 5.5 yr, 6.7 ppm-yr as printed (strictly 6.6; rounding), 26% vs 2–10%). Also the multiplier arithmetic, the bio (p. 197), the editors’ disclosure (p. 12), the Ch 16 citations (pp. 172, 175, 179–182), the MTBE octane reference (p. 111), and the absence of any contemporary-technology references.
41. D: the 1926 recommendations are attributed to Greenburg; “educated guesses” is attributed to Infante as a label for all parties; insight 4 reworded (“bring … into new places”, not “reintroduce”).

Second audit pass#

A second independent audit re-read the full extract (report pp. 38–51) and re-checked the editors’ and biographical pages (PDF pp. 11–12, 111, 171–182, 197) using a text extraction from the PDF. Table 4.1 was checked against a fresh PDF text extraction; this pass could not render pages to check them visually. One outside source was used, to confirm the Supreme Court citation. The first pass’s corrections held up. The changes below were made in the notes (N) and digest (D).

42. N, 4.2.5 and references and limitations: the Supreme Court citation is confirmed as 448 U.S. 607, decided 2 July 1980 (outside source), so “448 US 601” (p. 49) is a slip. The earlier “to verify” / “likely 607” wording is replaced. The plurality point is now given as outside background.
43. N, 4.2.5: added the court’s recognition of “the uncertainties involved” (p. 40), which had been left out of the hedges. “Never had to define” the non-significant end is replaced by the source’s “has not yet considered”. The significance benchmark now quotes “or other material impairment of health consequence”.
44. N, 4.2.7 and limitations: the claim that the text “mixes” or “uses interchangeably” Hayes 1996 and 1997 is withdrawn. These are two distinct papers (1996 mortality; 1997 incidence, p. 48), tied to different claims. The “1997 Hayes … clearly demonstrate” sentence is now quoted verbatim.
45. N, 4.2.7 spectrum: added that CLL rests on one citation (McMichael et al., 1975: solvent-exposed rubber workers, benzene as contaminant) and MDS on one study (Hayes et al., 1997).
46. N, 4.2.7 and mechanisms 10: Infante’s scare quotes around “updated” (p. 42) and “risk assessment” (p. 41) are added to the framing-language record.
47. N, 4.3.1(a): corrected “He offers it here as a stated cause”. Infante reports lack of knowledge as others’ attribution (Helmer’s for Sweden), and his “even though” framing already signals doubt.
48. N, 4.3.1(d) closing argument: added his concession that the debate “may be interesting from an academic viewpoint” (p. 45).
49. N, references: corrected “Wong ×6 … cited so that it can be rebutted”. Wong 1987a/b are cited as supporting evidence (p. 42) and against Wong’s later claims; only Wong 1995, 1996, 1998 and 1999 are opposing.
50. N, timeline: 1978 row notes that the chapter does not date the vacatur.
51. N, lessons item 3: “less so in the conclusions” is made precise. On p. 46 he reports Castleman and Ziem’s finding without the hedge and calls the lesson obvious.
52. N, editors’ Ch 16 and insight 7: added the editors’ balancing caveat (p. 182) that “paralysis by analysis” versus “prudent and careful evaluation” depends on how the pros and cons fall on each party.
53. N, mechanisms 1: the clinical-then-surveys ordering is adjusted, since the 1926 survey predates the 1928 case. The 200–1,000 ppm range is no longer dated “1923”, consistent with the timeline caveat.
54. N, mechanisms 2: the link between “cost of lowering exposures” and OSHA’s move to 1 ppm is flagged as annotator inference; the source wording is quoted.
55. N, mechanisms 3: “burden” is flagged as the annotator’s word. The chapter’s “threshold finding” wording is quoted, with outside confirmation that the plurality put the burden on the agency.
56. N, mechanisms 6: “which he endorses” (OSHA’s lowest-feasible policy, p. 40) is softened to “implicitly favours”. He reports the policy and prefers “reasonable precautions” (p. 41), but never states an endorsement.
57. N, insight 9: “manufacture null results” is softened to “produce”. Endpoint narrowing is flagged as annotator inference.
58. N, insight 11: fixed broken bold formatting.
59. N, insight 15 and “pro-precaution framing”: added that the 1977 emergency standard (1 ppm as “lowest feasible” when the API argued no risk below 10 ppm, p. 40) was itself a decision under uncertainty. “Not relevant” fits the pre-1970s neglect era best.
60. N, hindsight: added the editors’ instruction to judge by “the spirit of the times” and not “the luxury of hindsight” (p. 11).
61. D: ACGIH “set” changed to “recommended”. “Within the permitted limits” is replaced by the source’s “generally considered within” the 1940–71 recommended limits, and the pre-1977 absence of epidemiology is added. The 54/1,000 figure is tied to the 1996 Hayes mortality study. The p. 46 unhedged claim is attributed to Castleman and Ziem. “Burden” is flagged as the annotator’s word. The editors’ p. 182 caveat is added to insight 6. The case-level claim “precaution … only to the 2001 dispute” is corrected to include the 1977–87 low-dose dispute. The 448 U.S. 607 citation slip is added to the inconsistencies.
62. Re-checked and found accurate (no change): reference count (83) and Infante’s 11 entries; the bio (p. 197); the editors’ disclosure, four questions, pro bono and false-positive statements (pp. 11–12); Ch 16 quotations (pp. 171, 172, 175, 179, 180, 181); the MTBE octane reference (p. 111); all quotations in “Notable quotes”; and the absence of any contemporary-technology references.