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

Late lessons from early warnings (EEA, 2001), Ch. 4, pp. 38–51. Sole author: Peter F. Infante, OSHA. He led the first benzene cohort study and wrote or co-wrote several of the key estimates. The editors disclose his cohort study and agency role (p. 12), but the chapter never flags his authorship where it cites his own work. There are no panels. Table 4.1 (p. 47) was compiled by the EEA, not Infante.

Core story#

Early warnings (1897–1939). - Benzene was reported as a bone-marrow poison in 1897 (p. 38). Harm grew as its uses spread after the First World War. - A 1926 survey of 12 US plants found low white cell counts in 32% of workers at 90 ppm and above. Greenburg recommended medical removal and substitution (pp. 38–39). - The first leukaemia case was reported in 1928. The employer rotated workers monthly rather than cut exposure (p. 38). - In 1939 there were calls to substitute. Two poisonings had been recorded at 10–25 ppm (p. 39).

Limits above the evidence (1946–1957). The ACGIH (US industrial hygienists’ body) recommended 100 ppm in 1946, falling to 25 ppm by 1957. In 1948 the American Petroleum Institute called zero “the only absolutely safe level” but recommended “50 ppm or less” (p. 39).

Harm continued (1940s–1970s). - 13 UK neurotoxic deaths. - More than 200 cases of benzene blood disease treated in Milan and Pavia, including 34 acute leukaemias. - 44 leukaemias around Paris. - Turkey’s shoe industry switched to benzene in 1961 because it was cheaper, and an epidemic followed (pp. 39, 42–43).

The US standard (1977–1987). - In 1977 Infante’s cohort study found a 5–10-fold leukaemia risk at levels “generally considered within” the limits recommended in 1940–71 (p. 40). Until then, leukaemia causation rested on case reports and clinical observation, not epidemiology (p. 40). - OSHA’s 1 ppm standards of 1977–78 were stayed or vacated after industry challenges (p. 40). - The Supreme Court’s 1980 Benzene Decision required a threshold finding of “significant risk”, quantified “if at all possible”, though “not a mathematical straitjacket” (p. 40). - A 1 ppm limit came in 1987. It was set on economic feasibility, with a residual risk of 10 extra leukaemia deaths per 1,000 workers over a working lifetime (p. 41). - Infante and DiStasio (1988), using OSHA’s risk assessment, project 198 leukaemia and 77 myeloma deaths from the delay (p. 41). Table 4.1 hardens this into “caused more than 200 deaths” (p. 47).

Low-dose evidence (1990s). - US–Chinese cohort studies showed elevated risks at around 1 ppm (pp. 41–42, 44). - Infante’s own 1997 proceedings paper, based on the 1996 Hayes mortality study, estimates 54 excess leukaemia/lymphoma deaths per 1,000 workers over 45 years. The level is implied, not stated, to be 1 ppm (p. 41). - Industry consultants disputed the findings. Infante rebuts them point by point (pp. 44–45).

Gasoline. US gasoline averages about 1.5% benzene, up to 5%, and most European petrol historically more. There are no adequate warnings at the pump or in MSDS (p. 45).

Authors’ lessons#

• Principle not relevant. In the workplace “the precautionary principle is not relevant” (p. 46). The sentence is unexplained; in context it means the harm was known and ignored, i.e. a failure of prevention (my gloss). He still calls the failure a “lack of precaution” throughout (pp. 39, 42, 45–46).
• Achievability-led limits. Consensus limits reflected what was “easily achievable”. Corporate influence “may have played a role” (p. 43); the conclusions report Castleman and Ziem’s finding of corporate scientists on the TLV Committee without that hedge (p. 46).
• Cost over lives. The cost of production was put ahead of human life (p. 46).
• Certainty as delay. Demands for scientific certainty and protracted dose-response argument delayed protection (pp. 43–45).
• Risk assessment as a “straightjacket” (p. 41).
• 1 ppm is inadequate (p. 41).
• Recommendations:
• consensus bodies should keep their distance from producers and their “consultants” (p. 46);
• gasoline pumps should carry warning labels (p. 46).

Mechanisms#

• Industry’s knowledge ran ahead of its standard. The API’s 1948 review said zero was the only safe level, then recommended “50 ppm or less” (p. 39).
• Limits followed what was achievable (p. 43).
• From neglect to contestation. Neglect gave way, from the 1970s, to active contestation (pp. 42–45):
• a “random cluster” explanation;
• a demand for animal data despite human evidence;
• demands for certainty disease by disease;
• reanalyses that produced nulls.
• The burden of proof moved to the regulator through litigation and the court’s test, which required OSHA to make a “threshold finding” of significant risk (p. 40). “Burden” is my word, not the chapter’s.
• Some uncertainty was irreducible. Past exposures could only be reconstructed by what Infante calls the “‘educated’ guesses” of all parties to the rule-making (p. 44).
• Record custody shaped the evidence. Company removal of records biased one reanalysis (p. 42).
• The cost of delay was undercounted in coverage. Diseases without dose-response data were left out (p. 41).
• A known hazard was adopted anew where it was cheaper (Turkey, 1961; pp. 39, 46).
• The harm fell unequally (pp. 42–43, 45).

Transferable insights#

1. Knowing is not acting. Strong (pp. 38–39, 46).
2. “Achievable” standards build current practice into what counts as acceptable. Moderate: the mismatch is documented; the account of influence rests on one study and the author’s own experience (pp. 39, 43).
3. Producers may state knowledge that runs ahead of the standards they recommend. Strong as an instance; suggestive as a pattern (p. 39).
4. Cost advantages can bring a known hazard into new places. Moderate (pp. 39, 46).
5. Contestation grows as evidence strengthens and stakes rise. Moderate, because the account comes from a protagonist (pp. 43–45).
6. Requiring full quantification before action causes costly delay. Moderate: plausible, but the causal link is untested (pp. 40–41, 44). The editors concede that whether more analysis is “paralysis by analysis” or “prudent and careful evaluation” depends on how the pros and cons fall on each party (p. 182).
7. Irreducible uncertainty plus a demand for certainty means indefinite delay. Moderate–strong (p. 44).
8. Who holds the records, and how analyses are set up, shapes the evidence. Moderate (pp. 41–42).
9. Harms that cannot be quantified drop out, so cost-of-delay estimates are biased low in coverage. Strong for the coverage point (p. 41). This does not show the overall estimate is too low: it rests on OSHA’s risk model, and lower competing estimates exist (Crump, 1994, p. 41).
10. “Safe” levels fall and the list of outcomes grows over time. Strong for thresholds as a description of this case, though the sequence mixes endpoints and ever more sensitive study designs. Moderate for the disease spectrum (pp. 38–42).
11. Diffuse exposure through everyday products outlasts workplace controls. Moderate (p. 45).
12. Prevention failures and precaution failures are different problems. Moderate. It rests on one unexplained sentence (p. 46), and the chapter’s own “lack of precaution” language blurs the line (pp. 42, 46). The distinction is my framing, prompted by the text.

Caveats#

• Protagonist-author. Infante’s own studies supply the key numbers. Opponents appear only through his rebuttals, and he attributes motives to them (pp. 43, 45–46).
• Thin evidence:
• the pre-1977 low-dose claim rests on two cases (p. 39);
• the “200–800×” figures use the top of an exposure range inferred from symptoms (p. 39);
• the Turkish cost gap is conjecture (p. 46);
• some data are personal communications (p. 42), and the 54-per-1,000 figure comes from the author’s own meeting-proceedings paper (p. 41);
• lower competing risk estimates (Crump, 1994: 0.02–5.1 per 1,000, AML and monocytic only) are discounted for a selection bias that is asserted, not shown, to apply (p. 41).
• Inconsistencies:
• Santesson’s 1897 cases are described as bicycle tyres on p. 38 but raincoats on pp. 42 and 46;
• the delay is given as both “eleven” and “10” years (p. 41);
• 1920s UK substitution advice is mentioned without a source (p. 46);
• the Supreme Court decision is cited as 448 US 601 (p. 49); the correct citation is 448 U.S. 607.
• Little on Europe. European regulation is absent despite the chapter’s title. This is partly by design: the editors commissioned it for the “North American handling” of the issue (p. 11).
• Hindsight. Leukaemia risk at the permitted levels was not shown epidemiologically before 1977 (p. 40), so the neglect charge rests best on the aplastic anaemia evidence.
• Case selection. All the report’s cases are “false negatives”. The editors sought “false positive” examples, including from industry, but none emerged that they judged suitable (p. 12).
• What benzene supports. Mainly “act on known harm”, especially for the pre-1970s era. Precaution under uncertainty applies to the low-dose disputes after 1977: the 1977 standard cut the limit to 1 ppm when the API still argued there was no risk below 10 ppm (p. 40), and by 2001 the argument had moved to about 1 ppm and below (pp. 44–45).