LL2-21 digest — Ch21 Mobile phones and brain tumour risk: early warnings, early actions?#

Late lessons from early warnings II (EEA 2013), pp. 509–529 (PDF 511–531). Authors: Lennart Hardell and Michael Carlberg (the Swedish “Hardell group”), and David Gee (EEA). There are no panels or responses, so opposing views appear only as the authors characterise them.

Core story#

The chapter tells how IARC came to classify radiofrequency fields as a Group 2B “possible” carcinogen in May 2011. That decision rested mainly on two sets of case-control studies: the Hardell group’s Swedish studies and the IARC-coordinated Interphone study (pp. 509, 520). The chapter then argues that industry, the media and governments failed to respond adequately. It also argues that some scientists and official bodies misread null or inconclusive studies as reassurance.

Written by protagonists, it is both history and advocacy: - Hardell defends his own studies. - Gee’s EEA narrates its own 2007/2009 warnings (Box 21.2, p. 515). - The EEA’s withdrawal from the IARC meeting is reported in a footnote (p. 520, fn 15). Gee was its invited representative.

Key evidence#

• Hardell pooled results (1997–2003 cases): glioma, ipsilateral use, >10 years latency: mobile OR 2.9 (1.8–4.7), cordless OR 3.8 (1.8–8.1). Acoustic neuroma, wireless >10 years: OR 2.2, but cordless >10 years OR 1.0 (0.3–2.9). No consistent meningioma pattern (p. 513; Table 21.1, p. 516).
• Hardell, first use before age 20: glioma OR 3.1, acoustic neuroma OR 5.0, with wide CIs. One printed CI is impossible (p. 514).
• Interphone (~EUR 20m, 5.5m from industry): a four-year delay and a three-way split among investigators. Top-decile glioma (≥1640 h) OR 1.40 (1.03–1.89). Saracci and Samet called the compromise conclusion “elegant and oracular” (p. 517). Headlines contradicted each other on the same day, and co-authors gave conflicting public readings (p. 518).
• Studies attacked for misclassifying exposure: Interphone and CEFALO (children) counted cordless users as “unexposed” (pp. 514, 519). The Danish cohort did the same for corporate and post-1995 users; IARC set it aside for this reason (pp. 520, 522).
• Post-classification framing: industry statements (“not a definite nor a probable” carcinogen) and WHO’s “no adverse health effects … established” (pp. 520–521).
• Epilogue: an Italian Supreme Court compensation ruling in October 2012 (pp. 524–525).

Authors’ lessons and recommendations#

Lessons: long-term use raises glioma and acoustic neuroma risk, including from cordless phones and in adolescents (p. 524). “Non-positive” studies are wrongly read as “negative” (p. 511). Incidence data are “of limited value” (p. 523). Ambiguous consensus lets “not causal” be read as “no link” (p. 517). The IARC classification had no “significant impact” on governments (p. 524).

Recommendations: reduce head exposure through texting, hands-free use and better design, especially for young people. Reconsider standards built on the thermal paradigm. Label phones. Fund independent research through a levy on phones and masts. Publish divergent expert views side by side. Compensate heavy users who develop these tumours. Keep monitoring national tumour trends, but do not use them to dismiss case-control findings. (Box 21.2, p. 515; fn 12, p. 518; pp. 523–525.)

Main mechanisms#

• Latency versus the speed of adoption: fewer than 10% of Interphone cases had 10+ years of use (pp. 512, 517).
• Exposure misclassification pushes results toward “no effect” (pp. 514, 519, 522).
• Experts’ prior commitments and the thermal paradigm (pp. 513, 515, 520).
• Consensus-by-ambiguity in large consortia (pp. 517–518).
• Funding and independence disputes: Interphone, the Danish cohort, ICNIRP, and Ahlbom’s removal as IARC chair (pp. 517, 519–521).
• Strategic reframing of an authoritative hazard judgement (pp. 520–521).
• Media reliance on press releases (p. 524).
• Behaviour and technology change as an unplanned route to precaution (fn 9, p. 515).

Transferable insights (technology-neutral)#

1. Studies that find nothing are not evidence of safety, especially when exposure is short relative to latency (pp. 511, 512, 514). Strong in principle.
2. Early studies of new, long-latency exposures are built to miss harm; evidence arrives after exposure is universal (pp. 510, 512, 517). Strong in principle. The chapter accepts early positive results without asking why risk would appear so soon.
3. Defining the “unexposed” group badly (leaving out substitutes, using administrative proxies) hides risk (pp. 514, 520, 522). Strong for the Danish cohort; moderate otherwise.
4. Divided expert consortia produce ambiguous conclusions that interested parties then capture; structured dissent is an alternative (pp. 517–518). Moderate.
5. An authoritative classification does not settle a dispute; each party reframes it (pp. 520–521). Moderate–strong as description; the tobacco-style “doubt manufacture” charge is asserted.
6. Funding structures and evaluators’ prior commitments shape credibility; independence must be structural and applied to every side (pp. 513, 515, 517, 519–521). Moderate.
7. When precaution is cheap, a lower evidence threshold is proportionate. This was shared across the scientific divide (pp. 515, 518, 520). Moderate–strong as decision logic.
8. A dominant mechanistic model sets the burden of proof (pp. 515, 520, 524). Suggestive.
9. Deciding which evidence streams count is itself a governance act (pp. 519–520, 523). Moderate.
10. Courts and compensation systems can act on individual cases before science settles (p. 525). Suggestive.
11. Participants scrutinise evidence harder when it cuts against them; the chapter does this itself (pp. 514, 517, 523). Moderate.

Main caveats#

• Self-review with no counter-panel. Critics are heard only through the authors, often via partisan secondary sources (pp. 518, 521).
• Interphone under-reported. The chapter omits Interphone’s overall glioma OR of 0.81, its sub-1 ORs in nine of ten call-time deciles, and the “implausible values” of reported use among the heaviest users. It also omits the UICC funding firewall for EUR 3.5m of the 5.5m industry money. (All verified in IARC Press Release 200, 2010.) One point cuts the chapter’s way: Interphone itself said the sub-1 ORs “possibly reflect participation bias”, a downward bias.
• Little et al. 2012 summarised one-sidedly. That paper found Hardell’s estimates incompatible with US incidence but Interphone’s compatible. The chapter’s quotation ends with an ellipsis just before the Interphone clause (p. 523).
• CEFALO summarised one-sidedly. The chapter omits CEFALO’s findings of no relation to amount of use and no excess in the most-exposed brain areas (p. 514; Aydin et al. 2011 abstract).
• Weak arguments. Laterality results could reflect recall bias. The claim that different results by tumour type exclude bias does not follow (p. 524). Box 21.3’s reading of 2B as up to “nearly one in two probability” contradicts IARC’s own statement that the terms have “no quantitative significance” (p. 522).
• Asymmetric reasoning. Latency is used to discount null studies, while short-latency positive ones are accepted without asking why risk would appear so early. Rising Danish incidence is presented without the caveats applied to flat Swedish and US trends (pp. 512, 514, 523).
• Later evidence has mostly gone against the chapter’s specific claims. MOBI-Kids 2022 found no causal association in young people. The Million Women Study 2022 found glioma RR 0.89. Nordic incidence to 2016 rules out risk ratios of 1.08 or more at a 10-year lag. A WHO-commissioned 2024 review found “moderate certainty” of no increased risk, including for cordless phones and children. CERENAT 2014 (heavy users’ glioma OR 2.89) and NTP 2018 rat findings keep a heavy-use and plausibility question open. IARC listed RF for re-evaluation in 2019 and 2024; as of this check, the 2B classification stands.
• What holds up: the chapter’s institutional and communication observations are stronger than its epidemiological conclusions.