LL2-10 digest — Ch10 Bisphenol A: contested science, divergent safety evaluations#

Late lessons from early warnings II (EEA 2013), pp. 215–239 (main text pp. 215–229). Authors: Andreas Gies (German Federal Environment Agency, UBA) and Ana M. Soto (Tufts). No panels. The summary’s reference to a panel on animal carcinogen testing (p. 215) is a copy error from Ch8 (Panel 8.2, pp. 194–196).

Core story#

• Origins. BPA’s oestrogenic activity was found in the 1930s by the team that later produced DES. Unsuitable as a medicine, it became the building block of polycarbonate (1957) and epoxy resins. Production reached 3.8 Mt a year by 2006 (p. 216).
• Exposure. BPA leaches from polycarbonate, can linings, thermal paper, dental sealants and medical devices, so exposure is near-universal (p. 216).
• Rediscovery by accident. In 1993 Stanford endocrinologists traced an oestrogenic contaminant to their autoclaved polycarbonate labware. No screening programme had flagged BPA (p. 217).
• The evidence split. From 1997 academic studies reported low-dose developmental effects in rodents; Table 10.1 lists 49 at ≤50 μg/kg bw/day (pp. 219, 227–228). EFSA and the FDA relied on a few industry-sponsored GLP guideline studies (Tyl et al., 2002, 2008) that reported none. The chapter says low-dose differences in the rat study were dismissed by its authors, and an “independent” study (Ryan et al., 2010) also found no low-dose effects. EFSA and the EU risk assessment report excluded the academic studies for few doses, small samples, weak statistics or inconsistency (pp. 220–221).
• Divergent verdicts on largely shared evidence:
• EFSA set a TDI of 50 μg/kg (p. 221);
• by the authors’ own arithmetic, the Nordic minority’s studies imply 0.13 μg/kg and the NTP’s data 0.008 μg/kg (p. 222);
• Canada listed BPA as toxic on evidence it called “limited” (p. 222);
• the FDA’s Science Board called the FDA’s margins “inadequate”; the FDA voiced “some concern” but kept its number (pp. 222–223);
• ANSES found effects below reference doses (p. 223);
• EFSA reaffirmed safety in 2010 (p. 223).
• Internal dose. Measured free BPA in maternal blood is 4–6 ng/ml; model estimates are 0.1–10 pg/ml (p. 224).
• Influence (pp. 225, 228). A consultancy with tobacco-industry history claimed credit for EU classification as Category 3 rather than Category 2. Nine of 21 EFSA panel members declared industry-linked ties. Funding tracked outcomes: 94 of 104 government-funded studies found harm, versus 0 of 11 industry-funded.
• Markets moved first. Pressure from consumers and politicians forced product withdrawals while agencies still said “safe” (p. 225). The EU banned BPA baby bottles from 2011 (p. 226).

Authors’ lessons and recommendations#

• Lessons:
• a “same old story” of deployment before understanding (p. 226);
• monotonic dose-response fails for hormone-like agents, and “the time makes the poison” (pp. 218–219);
• BPA is not simply a “weak oestrogen” (p. 229);
• GLP compliance “does not indicate that good science has been performed” (Box 10.1, p. 220);
• excluding all non-guideline studies “cannot be defensible” (p. 223).
• Recommendations:
• restart the EU assessment, led by the high-impact researchers (p. 226);
• “Until final decisions are made”, end BPA uses involving close contact via food or the environment (p. 226);
• an industry-financed, government-managed testing fund;
• academic results not to be outweighed by contract-lab results;
• updated guidelines and stronger adviser independence (p. 229).

Main mechanisms#

• Paradigm defence: findings dismissed as implausible under old assumptions (p. 220).
• Evidence-selection rules as the battleground (p. 223, fn 2).
• Ossified test endpoints (pp. 222–223, 229).
• Latency defeating epidemiology (p. 219).
• Hidden stocks and unknown uses (p. 216).
• Structural funding conflicts and procedural advocacy (pp. 225, 229).
• Markets outrunning regulators (p. 225).

Transferable insights (technology-neutral)#

1. Known hazardous properties may not follow a substance into new uses (pp. 216–217, 220). Moderate.
2. Formal screening can miss what serendipitous research finds (p. 217). Moderate.
3. Methods that assume monotonic scaling can yield artefactual “no effect” results (pp. 218–221). Moderate.
4. Evidence-eligibility rules can move “safe” levels by orders of magnitude on the same evidence (pp. 221–223). Strong for the pattern; the extreme ratios are the authors’ own calculations.
5. Process assurance is not scientific adequacy (p. 220). Moderate.
6. Validation lag discounts newer, more sensitive measures (pp. 222, 229). Moderate.
7. With latent, transient exposure, epidemiology struggles to detect harm, so demanding human proof risks waiting until harm is irreversible (p. 219). Strong for the latency mechanism, given the DES precedent. Moderate for the normative corollary, which is an inference beyond the chapter.
8. Funding source correlates with findings (p. 228). Moderate: small, protagonist-compiled sample.
9. Advocacy targets classification chokepoints (p. 225). Suggestive: self-reported.
10. Market and public pressure can outpace assessments (p. 225). Moderate.
11. The most exposed (neonates) are the least visible (p. 224). Moderate.
12. Institutions sharing evidence reach divergent verdicts (pp. 222–223). Strong for the divergence. Moderate for the causes: only assessment stage and study criteria are evidenced (fn 2, p. 223).

Main caveats#

• Protagonist authors, no dissenting panel. Soto co-authored many sources, including the Chapel Hill statement, which the chapter lists among the assessments while noting it “is not a risk assessment in the classical sense”. The UBA view cited is Gies’s own agency’s.
• Framing outruns the evidence. The report calls BPA a “false negative” (p. 10), yet the chapter concedes that cross-sectional associations “are not a proof of causation” (p. 226). Its other human associations are listed without that caveat.
• “Overlap” holds only at the margins. Animal effect doses overlap human intake only where the lowest-dose studies (0.2–2.5 μg/kg) meet upper-percentile or worst-case intakes (up to 11–13 μg/kg in young children). Typical adult intakes are lower (pp. 224–225, 227).
• No precaution standard. The summary promises to address “challenges for applying the precautionary principle” (p. 215), but the chapter never says what weight of evidence should trigger action.
• Overstatements: “far-reaching agreement” (p. 219) versus acknowledged dispute (p. 218); “irrefutable” (p. 220); “over 800” toxic studies (p. 223) versus “at least 46” (p. 221).
• Selective handling of counter-evidence: failed replications are unengaged (Ashby 1999 is even listed as positive); Teeguarden et al. 2011 (free BPA undetectable in serum) is cited but undiscussed; the chapter’s own contamination warning (p. 216) is not applied to the high serum values.
• Errors: the Tyl AGD finding differs between text (male increase) and table (female decrease), and Tyl reported a female increase; the cited letter concerns ovary weights. Braun 2009 is a cohort, not cross-sectional. Figure 10.1’s inverted-U caption fits only two of its four panels: (a) has its largest effect at the highest dose and (b) peaks at the lowest (p. 218).
• Silent on benefits, alternatives and substitution risk.
• Hindsight. EFSA’s 2023 TDI of 0.2 ng/kg bw/day (verified) moved strongly in the direction the chapter urged. But CLARITY-BPA reproduced the guideline-versus-academic split, and the BfR opposed EFSA, so the conflict did not resolve.