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

Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013), Part C “Emerging issues”. Report pages 509–529 (chapter text pp. 509–525; references pp. 525–529). PDF pages 511–531.

Read in full from the text extract, and checked against the rendered PDF for pp. 509, 514, 515, 516, 517, 520 and 522: the abstract box, the children section, Box 21.2, Tables 21.1–21.2, the Interphone section, footnote 15 and Box 21.3. The tables extracted cleanly. The PDF confirms that the one numerical oddity (p. 514) is in the printed original and was not introduced by extraction.

Authors and standpoint#

Authors: Lennart Hardell, Michael Carlberg and David Gee (p. 509). The chapter gives no institutional affiliations beyond these: - Hardell (“LH”) leads the “Hardell group” whose four Swedish case-control studies are the chapter’s main evidence. His interest in the topic began with a Swedish committee on the cancer risks of extremely low frequency (ELF) fields from power lines (p. 510). The chapter describes the group’s investigators as “oncologists who have been trained in cancer epidemiology” (p. 513). Carlberg is a co-author on most of the Hardell group papers cited (reference list, pp. 527–529). - Gee was at the EEA. The reference list gives his title as “Senior Adviser, Science, Policy and Emerging issues” (p. 526, EEA 2011a). (Outside the chapter, not stated in it: Gee was a lead editor of the Late lessons project.) Footnote 15 (p. 520) says IARC invited Gee to join its 2011 Working Group “as ‘a representative of your organization, rather than as an observer’”. The EEA then withdrew a few days before the meeting. - Funding of the chapter: grants from two Swedish cancer charities, Cancer- och Allergifonden and Cancerhjälpen (p. 509, fn 1).

Standpoint. This is an advocacy chapter written by participants, not a detached case history. All three authors are protagonists in the story they tell: - Hardell’s group produced the principal positive studies. The chapter defends them against every critique it discusses and critiques competing studies in detail. - Gee’s EEA issued the 2007 and 2009 “early warnings” that the chapter reproduces approvingly in Box 21.2 (p. 515). The EEA was also invited by the BioInitiative group to contribute a chapter (p. 513), and withdrew from the IARC meeting, citing delays in publishing the full Interphone results and Ahlbom’s “intellectual bias” (p. 520, fn 15).

The stance is plainly pro-precaution. The chapter argues that the evidence already justified exposure reduction, and that industry, some scientists, the media and governments failed to respond properly.

Panels / commentaries: None. Many 2013 chapters include panels or industry and regulator responses. This one has only the authors’ text and three boxes (Box 21.1 on epidemiological concepts, Box 21.2 on the EEA warnings, Box 21.3 on IARC classifications), all written from the authors’ side. Opposing views appear only as the authors quote and characterise them: Ahlbom, Feychting, Boice, McLaughlin, Rowley, Milligan, Swerdlow/ICNIRP, WHO, and industry trade bodies. No dissenting voice is given its own space. That matters for how much weight the chapter’s characterisations of its opponents can bear.

Section-by-section notes#

Abstract box (p. 509)#

• IARC classified radiofrequency electromagnetic fields (RF-EMF) from mobile phones and similar devices as Group 2B (“possible” human carcinogen) in 2011. It had given ELF magnetic fields from power lines the same classification nine years earlier.
• The decision rested “principally” on two sets of case-control studies, Interphone and the Hardell group. The chapter calls them “complementary and generally mutually supportive”.
• National incidence data are dismissed as of “limited value”, because other risk factors may change and the mechanism is unclear.
• Three charges: “mobile phone industry inertia in considering the various studies and taking the IARC carcinogenic classification into account”; failings from the media in giving the public “robust and consistent information”; and no “significant impact” of IARC’s classification on governments’ sense of responsibility. Plus a recommendation: precautionary action “now” to reduce head exposures would limit the size and seriousness of any risk that may exist, and may reduce other possible harms not considered in the chapter.
• Epilogue preview: “Evidence is increasing that workers with heavy long-term use of wireless phones who develop glioma or acoustic neuroma should be compensated”. It cites the Italian Supreme Court ruling of 12 October 2012.
• Wording to note: the abstract says benefits “need to be accompanied by consideration of the possibility of widespread harms” (p. 509). The concluding remarks say benefits “need not to be accompanied by the possibility of widespread harms” (p. 524). Read literally they say different things. P. 509: benefits must be weighed with possible harms. P. 524: benefits can be had without widespread harms, i.e. precaution need not give up the benefits. This may be a drafting slip. Either way the thrust is the same: the benefits are no reason to ignore possible harms. (The abstract also calls the Italian tumour a “neuroma”; the Epilogue calls it a “neurinoma”.)

21.1 Introduction (p. 510)#

• Restates the IARC 2B decision of 31 May 2011 and the parallel with the 2002 ELF classification.
• Poses the organising question: why were the case-control studies initiated?

21.2 The Hardell group studies 1999–2011 (p. 510)#

• Technology context. Sweden, along with Israel, was an early mass adopter:
• NMT analogue phones: NMT 450 from 1981 to 31 December 2007; NMT 900 from 1986 to 2000.
• GSM digital (900/1800 MHz) from 1991; it now “dominates the market”.
• 3G/UMTS (1900/2100 MHz) in Sweden from 2003. 4G (800/2600 MHz) and TETRA (380–400 MHz) are “being established”.
• DECT/desktop cordless phones from 1988: analogue at 800–900 MHz at first, digital at 1900 MHz since the early 1990s.
• Mobile phones are now used more than landlines in Sweden. “The real increase in use and exposures … has been since the end of the 1990’s” (p. 510).
• The brain is “the main target organ” during handheld use (citing Cardis et al. 2008).
• Origins of the research question:
• LH served on a Swedish committee that concluded there was an increased childhood leukaemia risk near power lines (Hardell et al. 1995). IARC classified ELF fields as 2B in 2002.
• A literature review suggested increased brain tumour risk in the electronics industry, which prompted a case-control study.
• Media attention to a US lawsuit alleging that mobile phone use caused a fatal brain tumour (LA Times headline “Suit Over Cellular Radiation Raises Hazard Questions”, via Carlo and Schram 2001) led the group “to add questions on mobile phone use” to the first of four linked studies. So the mobile-phone questions were added partly in response to litigation and media coverage. They were added to a brain tumour study prompted by an occupational (electronics-industry) signal and by the ELF experience. The chapter cites no earlier biological or epidemiological signal specific to mobile phones. It does not date the lawsuit.
• Aim: explicitly “not to give a thorough review” or to deal with other RF effects (p. 510).

21.3 First Hardell study, 1999 (pp. 510–511)#

• Mailed questionnaire: 209 cases (90%) and 425 controls (91%) responded.
• No overall association.
• Analogue (NMT) use with >10 years latency: OR 1.20 (95% CI 0.56–2.59), not significant.
• Tumours in temporal, occipital or temporoparietal areas with ipsilateral exposure: OR 2.42 (0.97–6.05).
• The authors themselves note “low numbers” and mixed histopathology: “no firm conclusions could be drawn” (p. 511).
• Cordless phones were not included.
• The “non-positive vs negative” point. A 2001 NEJM editorial (Trichopoulos and Adami) on the null US study by Inskip et al. said it “allays fears raised by alarmist reports”. The chapter responds that this “goes far beyond what was scientifically defensible” (p. 511): among 782 patients, only 22 had 5+ years of use and no longer latencies were presented. This is the chapter’s first statement of a recurring theme, that “a ‘non–positive’ study is often assumed to be a ‘negative’ study when in fact the data do not support this assumption” (p. 511).

21.4 Second and third Hardell studies, 2002–2006 (p. 511)#

• Design:
• Two larger studies with cases diagnosed 1 Jan 1997–30 Jun 2000 and 1 Jul 2000–31 Dec 2003, using an identical questionnaire, so the results could be pooled.
• Cases were registry-reported and histopathologically verified, aged 20–80. Controls were drawn from the Swedish Population Registry.
• Cordless (DECT) phones were included, “an exposure which most other studies ignore” (p. 511). Fn 4 says Interphone asked about cordless phones in some countries, but that information “has never been properly analysed or published”.
• Exposure assessment:
• Self-administered questionnaire, supplemented by phone.
• Ear used was asked three ways (questionnaire, phone call, follow-up letter), with “very good agreement”.
• Tumour location came from CT/MRI.
• Ipsilateral was defined as >50% of use on the tumour side.
• Cumulative hours were built from first and last year of use and minutes per day.
• Car use with an external antenna and handsfree use were disregarded. Minimum latency was one year.
• Fn 2: exposure is highest in the temporal lobe (review of 110 phone models, Cardis et al. 2008).

Box 21.1 Some concepts and tools (p. 512)#

• Defines OR, SIR, CI and latency.
• Warns that absent “statistical significance” can be “a weak guide to the strength of evidence”. Fn 5 cites Bradford Hill (1965): “grasp the shadow and loose [sic] the substance” (p. 512), and cross-refers to Chapter 26.
• Latency claim: for solid tumours such as brain cancers, latency “can be from 15–45 years on average”. So “any study of cancer has to be at least as long as the average latent period … before there will be any clear evidence of a cancer risk” (p. 512).
• Fn 6: it took “almost 50 years” to be sure that “the atomic bomb dropped on Japan in 1945 also caused brain cancers”. The source cited is on meningiomas in Nagasaki survivors (Shibata et al. 1994); meningiomas are mostly benign. Fn 6 also cites a sentinel case series of cancer patients with occupational EMF exposure (Stein et al. 2011) for the latency claim.
• Analytical note: the latency claim cuts both ways (see Limitations). If average latency is 15–45 years, the studies of 1997–2003 cases (with Swedish use starting in 1981) mostly capture latencies under 20 years. Strong associations at “>10 years” would then imply either a promotional effect or an unusually short latency. The chapter uses the long latency to discount null studies, but accepts short-latency positive findings, including CEFALO’s result at >2.8 years (p. 514). This asymmetry is partly defensible: a short study can reveal an early effect but cannot rule out a later one. But the chapter does not discuss what mechanism would produce early excess risk. It does not consider whether early positive results might reflect bias. (It mentions initiation, promotion and progression only in the incidence-data context, p. 523.)

21.5 Fourth Hardell study, 2010: deceased cases (pp. 512–513)#

• Trigger: a review commissioned by the Swedish Radiation Protection Agency (Boice and McLaughlin 2002) suggested that excluding deceased cases was a source of bias. The chapter says: “The scientific reason for this suggestion was not given” (p. 512).
• Design: 464 deceased malignant brain tumour cases (mostly grade IV astrocytoma). Two control groups came from the Death Registry: 464 people who died of other cancers and 463 who died of non-cancer causes. Next of kin answered the questionnaires.
• Response: 346 cases (75%), 343 cancer controls (74%), 276 other-cause controls (60%).
• Results:
• Mobile phones, >10 years latency: OR 2.4 (1.4–4.1).

• 2000 cumulative hours: OR 3.4 (1.6–7.1).

• Cordless: no clear association, although >2000 h gave OR 1.7 (0.8–3.4) (p. 513).
• Conclusion: the Boice and McLaughlin critique was “scientifically unfounded” (p. 513).
• Note: Excluding deceased cases in a study of a highly lethal tumour is a standard selection-bias concern. Boice and McLaughlin’s rationale may simply not have been spelled out. The chapter’s verdict rests on evidence: the deceased-case study reproduced the earlier results. But calling the critique “scientifically unfounded” sits oddly beside the group’s decision to run a whole new study to answer it, which suggests the question was worth asking. (External context, not drawn in the chapter: Boice and McLaughlin were at the International Epidemiology Institute (IEI). The chapter later says IEI co-designed the Danish cohort and that “The source of money for the IEI has not been disclosed” (p. 521). The reference list shows Boice and McLaughlin as co-authors of the Danish cohort papers (Johansen 2001; Schüz 2006) and of the Danish Interphone arm (Christensen 2005), and Boice as co-author of the CEFALO editorial (pp. 526–529). The chapter makes no explicit connection between these.)

21.6 Some Swedish responses to the Hardell studies (p. 513)#

• Ahlbom and Feychting (1999) wrote to the journal suggesting that selection bias of cases might explain the high response rate. The chapter calls this “unfounded and easy to rebut”, attributing the high response rates to the oncologist investigators (p. 513).
• Prior commitment of an evaluator. Anders Ahlbom, later in the Swedish part of Interphone, had stated “even before the study started” in an opinion letter (Adami et al. 2001) that an association was “biologically bizarre” (p. 513). The reference list gives that piece’s title as “Opinion — Experts who talk rubbish” (p. 525). The chapter points out the irony that Ahlbom’s own work (Feychting and Ahlbom 1993) supported an association between power-line magnetic fields and childhood leukaemia, which “would also have to be regarded as biologically bizarre”. (The phrase “biologically bizarre” is not in quotation marks in the chapter, so it may paraphrase the letter. The letter had five authors, led by Adami, who also co-wrote the 2001 NEJM editorial criticised on p. 511. The chronology is also loose: the letter is dated 2001, but Interphone fieldwork ran “between 2000 and 2004” (p. 517), so “even before the study started” may be imprecise.)
• Maria Feychting (Swedish Interphone) asked whether questions were put the same way to cases and controls (Björkstén 2006). The chapter replies that they were in Hardell’s studies, but that Interphone used different methods for cases, for example bedside interviews of cases only.
• BioInitiative Report (2007). The Hardell studies and other evidence inspired a group of scientists to produce it. The chapter says it had “considerable impact” in alerting people to “a small but growing minority of experts who did not agree with the WHO EMF Project statements and other reports that there was no evidence of risk (e.g. of SCENIHR 2007)” (p. 513).
• EEA’s own entry. The BioInitiative group invited the EEA to contribute a chapter on the relevance of the 14 Late lessons case studies. “Having considered the published evidence, the EEA decided it was timely to issue a guarded early warning” in September 2007 (p. 513; see Box 21.2).

21.7 Pooled analysis of the Hardell studies (p. 513; Table 21.1 p. 516)#

• The studies of cases diagnosed 1997–2003 were pooled, with the deceased glioma study added. Latency categories were >1–5, >5–10 and >10 years.
• Glioma: increased risk overall, highest at >10 years, and higher still ipsilaterally: mobile OR 2.9 (1.8–4.7); cordless OR 3.8 (1.8–8.1).
• Meningioma (n = 916): “no consistent pattern”. Ipsilateral >10 years: mobile 1.6 (0.9–2.9), cordless 3.0 (1.3–7.2).
• Acoustic neuroma (n = 243): wireless OR 2.2 (1.3–3.7) at >10 years. Ipsilateral use gave higher risks than contralateral.
• Table 21.1 (p. 516):
• Glioma, all latencies >1 year: wireless 1.3 (1.1–1.5), mobile 1.3 (1.1–1.6), cordless 1.3 (1.1–1.6).
• Glioma >10 years: wireless 2.1 (1.6–2.8), mobile 2.5 (1.8–3.3), cordless 1.7 (1.1–2.6).
• Meningioma: no statistically significant OR except cordless ipsilateral >10 years (3.0). Wireless >10 years 1.4 (0.97–2.0); all totals 1.0–1.1.
• Acoustic neuroma, mobile: ipsilateral >10 years 3.0 (1.4–6.2); >10 years 2.6 (1.5–4.6); total 1.7 (1.2–2.3).
• Acoustic neuroma, cordless: ipsilateral >10 years 2.3 (0.6–8.8), not significant; >10 years 1.0 (0.3–2.9); total 1.5 (1.04–2.0). So there was no long-latency excess for cordless phones and acoustic neuroma. That qualifies the concluding claim that cordless use “increases the risk when properly assessed” (p. 524). The claim is carried mainly by glioma.
• Controls: 2438 for glioma (living and deceased), 2162 for meningioma and acoustic neuroma.
• Adjusted for age, gender, socioeconomic code and year of diagnosis, plus vital status for glioma. Ipsilateral analyses used living subjects only.

21.8 Risks to children (p. 514; Table 21.2 p. 516)#

• Mechanistic argument: children’s brains absorb more RF because of smaller heads, thinner skulls and higher tissue conductivity (Cardis 2008; Christ 2010; Gandhi 2012). The developing brain is more sensitive to toxins (Kheifets 2005) and keeps developing until about age 20. Children also have a longer lifetime in which to develop tumours. Together these leave “children at a higher risk than adults” (p. 514). This is stated as a conclusion rather than as a hypothesis.
• Hardell data by age at first use:
• Glioma, first use <20: OR 3.1 (1.4–6.7). A similar pattern for cordless phones (data not shown).
• Acoustic neuroma, first use <20: OR 5.0 (1.5–16). No conclusion possible for cordless phones (only one case with first use under 20).
• Ipsilateral, youngest group: glioma OR 4.4 (1.3–15); acoustic neuroma OR 6.8 “(95 % CI = 1.4–3.4)”. The printed CI cannot be right, since it excludes the point estimate. This was confirmed on the rendered PDF page and is probably a typo for 1.4–34. The very wide intervals reflect small numbers.
• No age pattern for meningioma.
• Table 21.2 (p. 516), glioma: 20–49 → 1.4 (1.1–1.7); ≥50 → 1.3 (1.01–1.6). Meningioma <20: 1.9 (0.6–5.6).
• The text defines the age groups by “first use of a wireless phone”. The ORs it reports and Table 21.2 are for mobile phones only; cordless results are “data not shown”. The text cites Hardell and Carlberg 2009, but the table’s sources are Hardell et al. 2006b, 2006c, 2010 and 2011a.
• Other studies: “very few” apart from CEFALO (Aydin et al. 2011) and the ongoing EU Mobikids study (fn 8).
• CEFALO critique (Denmark, Sweden, Norway and Switzerland; critique drawn from Söderqvist et al. 2011, Hardell’s group):
• The study authors reported they “did not observe that regular use of a mobile phone increased the risk”. A Boice and Tarone editorial said it showed no increased risk, and a Karolinska Institute press release called the results “reassuring”. The chapter: “the statements go far beyond what the study really showed” (p. 514).
• Specific flaws alleged:
  • Cordless use was assessed only “in the first 3 years of use”, “a most peculiar definition”.
  • Wireless use was never analysed as a combined category.
  • Cordless users were placed in the “unexposed” group for mobile analyses, and vice versa, “thereby potentially concealing an increased risk”.
• Data the chapter reads as supportive:
  • Regular users: OR 1.36 (0.92–2.02), non-significant, rising somewhat with duration of subscription and of calls. Only latency of 5 years or more was presented, “with very few cases”.
  • Operator-recorded use (62 cases, 101 controls), time since first subscription >2.8 years: OR 2.15 (1.07–4.29), trend p = 0.001.
• Omitted by the chapter (verified against the CEFALO abstract, Aydin et al. 2011, via Europe PMC): in the operator-data subset, risk was related to time since subscription “but not to amount of use”. “No increased risk of brain tumors was observed for brain areas receiving the highest amount of exposure.” The CEFALO authors concluded that “the absence of an exposure-response relationship” in either respect “argues against a causal association”. The chapter reports the time-since-subscription trend as “further support of a true association” but does not mention these two null dose and location results.
• Verdict: “a moderately increased risk, in spite of low exposure, short latency period and limitations”. CEFALO “cannot be used as reassuring evidence against an association” (p. 514, emphasis in original). Had it been available to IARC, “it would have provided additional evidence to support” 2B.

Box 21.2 The EEA early warnings, 2007–2011 (p. 515)#

• 2007: “Appropriate, precautionary and proportionate actions taken now to avoid plausible and potentially serious threats to health from EMF are likely to be seen as prudent and wise from future perspectives” (EEA 2007).
• 2009 update: “The evidence for a head tumour risk from mobile phones, although still very limited, and much contested, is, unfortunately, stronger than two years ago”. It was “now strong enough, using the precautionary principle” to justify four steps: 1. Governments, industry and the public should take “all reasonable measures to reduce exposures”, “particularly … children and young adults who seem to be most at risk”. Means: texting, hands-free sets, and “phones of an improved design which could generate less radiation and make it convenient to use hands free sets”. 2. Reconsider the scientific basis of exposure standards, which have “serious limitations such as reliance on the contested thermal effects paradigm; and simplistic assumptions about the complexities of radio frequency exposures”. 3. Provide effective labelling and warnings. Special Eurobarometer (fieldwork Oct/Nov 2006): 80% of EU citizens do not feel informed about the protection framework, and 65% are not satisfied with the information they get on EMF health risks. 4. Fund “urgently needed research”, including industry grants and “possibly a small levy on the purchase and or use of mobile phones”. This is modelled on what the EEA “think[s] the US pioneered” with a 1970s research levy in the rubber industry. The funds would be used “by independent bodies”.
• 2011: the warning was updated in evidence to the Council of Europe hearing (February 2011).
• Fn 9: the EEA “has since noted, with some relief, what appears to be an increased use of hands free devices, particularly in the younger generation, due to enhanced applications” (p. 515). This is an important aside: the recommended exposure reduction was partly happening through changes in technology and use, not through regulation.
• Fn 10: the EEA has noted “increasing evidence of ‘funding bias’” (results linked to funding source) across pharmaceuticals, tobacco, lead, asbestos, BPA and EMF, and also in cost-benefit analysis and transport cost estimates. No specific EMF funding-bias study is cited.

21.9 The Interphone study 2000–2010: disagreements and delays (pp. 517–518)#

• Set-up: an international collaboration under IARC guidance, initiated on the recommendation of expert groups (McKinlay 1997 for the European Commission; Cardis et al. 2007). Sixteen centres in 13 countries, fieldwork 2000–2004.
• Cost: “nearly EUR 20 million of which industry contibuted 5.5 million” (p. 517). By contrast, the Hardell studies cost ~EUR 410,000, funded by the Swedish Work Environment Fund, two cancer charities, Telia (the Swedish telecom operator), Fondkistan and the Örebro University Hospital Cancer Fund (fn 11, p. 517). The chapter discloses Telia funding of its own studies but never discusses it, while it scrutinises industry funding of Interphone at length.
• Divergence and delay:
• Country analyses differed. Some were positive; some showed “decreased risks, i.e. seemingly a ‘protective’ effect of the radiation”.
• There was “a 4 year delay” between the country results and the overall publication.
• Three factions, as the chapter “reportedly” describes them: one saw a positive link (especially in the 10+ year group); one saw an artefact of design; a third “could agree to neither position” (p. 517). The chapter gives no source for this account.
• IARC Director Christopher Wild “brokered sufficient agreement” to publish in May 2010.
• Results as the chapter reports them:
• No association with meningioma overall.
• Glioma in the highest exposure group (≥1640 h, “about half an hour of use per day for ten years”): OR 1.40 (1.03–1.89).
• Ipsilateral: 1.96 (1.22–3.16). Temporal lobe: 1.87 (1.09–3.22).
• The compromise text. Two sentences point different ways. First: “There were suggestions of an increased risk of glioma … at the highest exposure levels”. Then: “biases and errors limit the strength of the conclusions … and prevent a causal [our emphasis] interpretation” (p. 517).
• Continuum of evidence. The chapter argues that Interphone failed to explain how evidence strength runs from “scientific suspicion of risk” through “strong association” and “reasonable certainty” to “causality” (Bradford Hill 1965). As a result, “the media and the public could assume that ‘not causal’ meant ‘no link’” (p. 517).
• Saracci and Samet’s editorial (IJE 2010) called the conclusion “both elegant and oracular … (which) tolerates diametrically opposite readings” (p. 517). They gave methodological reasons for likely underestimation, including short latency: “less than 10 % of the Interphone cases had more than 10 years exposure”. An italicised passage, apparently quoting them: “None of the today’s established carcinogens, including tobacco, could have been firmly identified as increasing risk in the first 10 years or so since first exposure” (p. 517).
• Media consequences (pp. 517–518). On 17 May 2010 the Daily Telegraph reported a risk and BBC News reported no risk. “This conflicting media reporting pattern was widely repeated elsewhere” (p. 518).
• Fn 12 (p. 518): the EEA “had anticipated this confusion”. It had proposed to IARC that the conflicting interpretations be published side by side in the same article; “this suggestion was not adopted”.
• Co-authors’ divergent public statements (p. 518, via Microwave News, i-sis.org.uk and YouTube):
• Cardis (coordinator): “Overall…the results show a real effect”.
• Armstrong: “some indication of an increased risk of gliomas, but I cannot say this with certainty”.
• Sadetzki: not strong enough for a causal interpretation but “sufficient to support precautionary policies”.
• Feychting: over ten years’ use “shows no increased risk”.
• Ahlbom (on Chinese television): “nothing in these data or in previous data, really, to indicate that there is any risk”.
• Later Interphone outputs (p. 518):
• Cardis et al. 2011 (five countries): risk rose with estimated total cumulative RF energy (J/kg) absorbed at the estimated tumour centre, counting exposure more than 7 years before diagnosis. Top quintile OR 1.91 (1.05–3.47). This was “for the first time” linking absorbed dose, not just its proxy, to tumour induction, and it “received very little media attention”.
• Larjavaara et al. 2011, a “similar study based on less sound methods”, appeared to contradict it.
• Acoustic neuroma (Interphone 2011): ipsilateral use started >10 years before, with >1640 h, OR 3.74 (1.58–8.83).
• Parotid gland: the total results “have never been published”. IARC “terminated” Interphone (fn 14: officially completed February 2012). Only the Swedish and Israeli results are out. Subgroups suggest increased risk, but other studies do not show a consistent pattern, and long-term data are scarce.

21.10 Reviews and discussions of Hardell and Interphone (pp. 518–520)#

• Interphone methodological problems as the chapter lists them:
• Differential exposure misclassification from observational and recall bias, “which would tend to underestimate the risk”. This direction is asserted without a specific citation. Differential recall by cases more usually inflates ORs. The downward pressure in Interphone is more plausibly from participation (selection) bias, which the Interphone group itself invoked (see Limitations).
• Low participation: in some countries ~50% of cases and ~40% of controls, against Hardell’s 90% (malignant), 88% (benign) and 89% (controls) (p. 518). The chapter itself notes the figures are not like-for-like: Interphone counted deceased cases in its participation rates, whereas Hardell handled deceased cases in a separate sub-study (pp. 518–519).
• About 40% of cases interviewed in hospital. Interviewers always knew case or control status. Cordless phones not properly assessed (p. 519).
• Bedside interviews were “a major disadvantage and … ethically questionable”. Patients may be post-surgical, uninformed about their diagnosis or sedated. In the Danish arm, patients scored lower due to aphasia and paralysis (Christensen et al. 2005) (p. 519). By contrast, Hardell cases got a postal questionnaire about 2 months after diagnosis and later phone interviews “blinded as to case or control status”.
• Hardell’s recall-bias checks (Hardell 2002): wireless use was similar whether or not subjects reported a previous cancer or had a relative help them. Exposure changes after the phone interviews did not differ between cases and controls (p. 519).
• Myung et al. 2009 meta-analysis: the Hardell studies were of “higher quality” than Interphone (p. 519). The chapter adds that Myung missed cordless phones. Cordless emissions are “of the same magnitude” as digital mobiles, and cordless calls tend to be longer. Classing cordless use as “unexposed” biases estimates against a risk.
• Industry comment (Rowley and Milligan 2010, GSMA/MMF): they claimed Interphone was independent of industry influence. The chapter rebuts:
• Industry provided EUR 5.5 million plus extra national funding.
• The Interphone protocol (IARC 2001) allowed industry “observers or consultants”.
• It also required that “representatives of industry and other concerned organisations… shall be informed shortly (maximum of seven days) before publication” (p. 519).
• Rowley and Milligan’s charges of selection, information and recall bias and “unusually high” participation in Hardell’s studies are called “ad hoc statements … not substantiated”. “A high participation rate is a pre-requiste for high quality in case-control studies” (p. 519).
• More favourable analyses of Hardell: Kundi 2009, Myung 2009, Mead 2009, Cardis and Sadetzki 2011, Levis et al. 2011. (The reference list dates Levis as 2001, a reference error.) IARC relied mainly on Hardell and Interphone (p. 519).
• Cardis and Sadetzki 2011, both Interphone scientists, are “particulary interesting”. Their conclusions: “It is not possible to evaluate the magnitude and direction of the different possible biases on the study results and to estimate the net effect of mobile phones on the risk of brain tumours. The overall balance of the above mentioned arguments, however, suggest the existence of a possible association”. And: “Simple and low cost measures, such as the use of text messages, handsfree kits and/or the loudspeaker mode of the phone could substantially reduce exposure to the brain from mobile phones. Therefore, until definitive scientific answers are available, the adoption of such precautions, particularly among young people, is advisable” (Cardis and Sadetzki 2011, p. 170; quoted pp. 519–520).

21.11 IARC evaluation 2011 (p. 520)#

• The meeting: 24–31 May 2011, Lyon, covering all RF sources. It included all published Hardell studies and the overall Interphone results (2010, 2011; Cardis 2011).
• The chapter claims “The results on glioma are similar in the Hardell group and Interphone studies if the same inclusion and exclusion criteria are used (Hardell et al., 2011b)”, contrary to “widespread claims” that they differed. The source is the group’s own re-analysis.
• Process: 30 scientists (per the chapter; IARC PR 208 says “31 scientists from 14 countries”, see below; the difference may reflect a late change of membership and is unresolved here) in four subgroups (animal, epidemiology, exposure, mechanistic). Pre-meeting drafts were agreed “sentence by sentence” in plenary.
• Conclusion: “limited evidence in humans” based on positive associations for glioma and acoustic neuroma from Interphone and Hardell. The Danish cohort was set aside for “considerable misclassification in exposure assessment” (Baan et al. 2011). The final vote was a “very large majority” for 2B, “based also on occupational studies” (p. 520).
• Fn 15 (p. 520), a key governance detail:
• IARC invited Gee as an EEA “representative”.
• A few days before the meeting, the EEA withdrew “because of further delays in publishing the full Interphone results and because of the intellectual bias of Ahlbom who was then the Chair of the epidemiology group”.
• The day before the meeting, IARC removed Ahlbom from the chair “as a result of a reported conflict of interest”, and the meeting received part of the unpublished Interphone data. By then it was “too late for the EEA to then participate”.

21.12 Responses to the IARC conclusion (pp. 520–521)#

• “Even the authoritative IARC evaluation has been interpreted very differently by different groups” (p. 520).
• WHO EMF fact sheet (June 2011): “To date, no adverse health effects have been established as being caused by mobile phone use”, and “Tissue heating is the principal mechanism of interaction”. The chapter says this ignores non-thermal effects (citing Giuliani and Soffritti 2010) (p. 520).
• Industry statements:
• Michael Milligan (MMF): “it is significant that IARC has concluded that RF electromagnetic fields are not a definite nor a probable human carcinogen”.
• Jack Rowley (GSMA): “a hazard is possible but not likely” (p. 520).
• Patrick Frostell (Federation of Finnish Technology Industries): in line with the “dominant interpretation … neither carcinogenic to humans nor probably carcinogenic to humans” (p. 521).
• Dariusz Leszczynski (STUK, Finland; member of the IARC panel): the classification caused “a stir of pro and contra opinions”, and the only outcome is “confusion … by this flurry of contradictory opinions and spin-statements” (p. 521, from his blog).
• The Economist (Babbage blog): the “brouhaha” is “monumentally irrelevant compared with all the other things there are to worry about” (p. 521).
• Microwave News on ICNIRP: “a self-perpetuating group that declines to disclose its finances”, whose epidemiology committee “has only welcomed the like-minded”. Its former chair Ahlbom has said risks are “nonexistent”. Maria Blettner, “the lone dissenting voice in the final vote”, and Ahlbom both worked on Interphone (p. 521).
• “Perhaps even IARC has contributed to this confusion by seeming to agree with the largely non-positive but much criticized Danish cohort study” (p. 521, citing only the Microwave News home page). The chapter does not explain this. It sits awkwardly with its own report that the IARC Working Group drew “No conclusions” from the Danish cohort (p. 520).
• Tobacco analogy: “No doubt the IARC decision started a world-wide spinning machine perhaps similar to the one launched by the tobacco industry” over passive smoking in the 1990s (Ong and Glantz 2000). “Sowing confusion and ‘manufacturing doubt’ is a well known strategy” (Michaels 2008; McGarity and Wagner 2008; Oreskes and Conway 2010) (p. 521). Fn 16 describes Philip Morris’s three-pronged scientific, communications and government strategy against IARC’s second-hand smoke study. Philip Morris planned to spend USD 2 million in one year alone “and up to USD 4 million on research”, against IARC’s USD 2 million over ten years (p. 521). No direct evidence of a coordinated campaign on RF is offered; the parallel is inferred from the divergent press statements.

21.13 Further studies since IARC (pp. 521–523)#

• Larjavaara et al. 2011 (Nordic Interphone, tumour location): uses a “different, less clear method”. Only 42 cases had >10 years of use, and there was no analysis of the highest-exposed, longest-duration group. “Much less informative and less sophisticated” than Cardis 2011 (p. 521).
• Danish subscriber cohort (Johansen 2001; Schüz 2006, 2011; Frei 2011):
• Set up by IEI (Rockville, USA) and the Danish Cancer Society, funded by two Danish operators (TeleDenmark Mobil, Sonafon), IEI and the Cancer Society. IEI’s funding sources were “not disclosed” (p. 521).
• 723,421 subscribers identified (1982–1995), but the initial cohort had only 58% of them (p. 522).
• Box 21.3 IARC and its classifications (p. 522):
• Explains hazard (“an agent that is capable of causing cancer under some circumstances”) versus risk.
• Counts (IARC 2012): Group 1 = 108; 2A = 64; 2B = 272; 3 = 508; 4 = 1.
• Definitions: 2B means “limited evidence of cancer in humans and less than sufficient evidence in animals” and “chance, bias or confounding cannot be ruled out with reasonable confidence”.
• Contains an interpretive gloss not taken from IARC: “Some 2B agents will be at the lower end of the probability range, others will be close to the nearly one in two probability”. This probability reading is the authors’ own, and it contradicts the source the box cites. The IARC 2006 Preamble (verified; p. 22) states: “The terms probably carcinogenic and possibly carcinogenic have no quantitative significance and are used simply as descriptors of different levels of evidence”. The box itself also stresses the hazard/risk distinction.
• It criticises journalists for “loosely lumping together” 2B agents such as dry cleaning fumes and coffee with mobile phones. “Each agent needs to be considered on its own evidence.”
• Small inconsistency: 272 agents, then “the 271 in Group 2B”.
• Danish cohort misclassification, per the chapter, acknowledged by the cohort authors themselves (Frei et al. 2011):
• Subscribers who did not use phones counted as “exposed”. Non-subscriber users counted as “unexposed”.
• 200,507 corporate subscribers (likely heavy users) counted as “unexposed”.
• Post-1995 subscribers (“over 80 % of the Danish population”) counted as “unexposed”.
• Cordless users counted as “unexposed” (p. 522).
• Also: no laterality analysis and “complete absence of actual exposure data” (p. 523).
• The chapter cites Ahlbom et al. 2007 (Ahlbom, Feychting, Cardis, Elliott) among the critiques of the cohort (p. 523). So researchers the chapter elsewhere treats as opponents had also criticised the Danish cohort. The camps were less neatly divided than the narrative suggests.
• Hence the authors’ conclusion of “no increased risks … providing little evidence for a causal association” is “not soundly based” (p. 523).

21.14 Monitoring national tumour trends (p. 523)#

• The argument addressed: some researchers (Aydin 2011; Ahlbom and Feychting 2011; Deltour 2012; Little 2012) use incidence data to test the case-control associations.
• Chapter’s critique:
• Ahlbom and Feychting (2011) cited Swedish Cancer Registry data (no overall increase since the 1990s) in support of the Danish cohort, rather than the more relevant Danish registry.
• The Swedish registry’s completeness for CNS tumours, especially high-grade glioma, “has been seriously questioned” (Bergenheim 2007; Barlow 2009). Sweden was ~40% of the Deltour 2012 population.
• Danish counter-evidence offered:
• NORDCAN 2000–2009: brain and CNS tumours rising per year by 2.7% in men (1.1–4.3) and 2.9% in women (0.7–5.2).
• Sundhedsstyrelsen: age-standardised incidence up 40% in men and 29% in women, 2001–2010.
• A Danish Cancer Society news release: “an almost 4-fold increase” in glioblastoma over 10 years; “these incidence data are not generally available” (p. 523).
• Little et al. 2012 (US SEER):
• The chapter says the method used LA men aged 60–64 as the baseline, found only in the web appendix, “Thereby numerous assumptions were made”.
• The conclusion that Swedish-study risks “are not consistent with observed incidence trends” “goes far beyond scientific evidence”. The chapter’s quotation ends with an ellipsis (“US population data…”) just before Little’s qualifying clause, “although the US data could be consistent with the modest excess risks in the Interphone study” (verified against the abstract; see Limitations).
• The chapter notes Little’s own data showed high-grade glioma rising 0.64%/yr (0.33–0.95), 1992–2008, “a result not commented further” (p. 523).
• General point: “Much care is needed when using descriptive data … to dismiss results from analytical epidemiology”. Other risk factors may change. Cancer involves initiation, promotion and progression. “As the mechanism for RF-EMF carcinogenesis is unclear it supports the view that descriptive data on brain tumour incidence is of limited value” (p. 523).
• The section title (“Need for monitoring long term trends …”) implies a recommendation: keep monitoring national trends, but do not use them to dismiss analytical results.
• Note the partial asymmetry: the chapter does not explicitly claim that rising Danish incidence shows causation, and its “limited value” verdict is stated generally. But it presents the rising Danish figures, including an unpublished “4-fold” glioblastoma figure, without the caveats it applies to flat Swedish and US trends: changes in diagnosis and registration, or other risk factors.

21.15 Concluding remarks (pp. 523–524)#

• Four claims attributed to “the telecommunications industry and others” (p. 524): 1. ICNIRP limits adequately protect against cancer. 2. Children are no more sensitive than adults. 3. There are no biologically significant non-thermal effects. 4. If there are, there are no acceptable mechanisms.

Against these the chapter cites the 400-page Ramazzini Institute/ICEMS review (Giuliani and Soffritti 2010), summarised by the EEA for the Council of Europe (2011). - Scientific conclusions: - Hardell and Interphone “show an increased risk for glioma and acoustic neuroma associated with long term mobile phone use”. - “Also use of cordless phones increases the risk when properly assessed and analysed.” - Risk is highest ipsilaterally. “Adolescents seem to be at higher risk than adults.” No consistent meningioma pattern (p. 524). - Anti-bias argument: different results for different tumour types “strongly argues against systematic bias as an explanation of the findings. In that case the results would have been similar regardless of tumour type” (p. 524). This is a questionable inference; see Limitations. - ICNIRP critique: Swerdlow et al. (ICNIRP epidemiology committee), online 1 July 2011, concluded “the trend in the accumulating evidence is increasingly against the hypotheses that mobile phone use can cause brain tumors in adults”. The chapter charges “unfounded attacks on individual researchers as exemplified in this article, a pattern that repeats similar experiences in the asbestos, lead and tobacco histories”. It also charges that results are questioned by “obscure methods and citing single results out of context” (p. 524). - Media: “There is a lack of investigating journalists”; most rely on press releases. Limitations are “rarely in the abstract which is most often all that is read”. Without accurate reporting, the public cannot “make their own judgements about how precautionary they should be” (p. 524). - Governments: “It is remarkable that the IARC carcinogenic classification does not seem to have had any significant impact on governments’ perceptions of their responsibilities … especially given the ease with which exposures can be reduced (i.e. texting, handsfree devices and better phone design)” (p. 524). - Research: “Independent research … urgently needed”, given global exposure of “over 5 billion people and many other species”. This includes bees and birds “whose navigation systems are possibly being affected” and breeding of wild birds (Balmori 2005, 2009; Sharma and Kumar 2010; Everaert and Bauwens 2007). Funding could come from “levies on phones and masts but used independent from their influence” (p. 524). - Closing: “Precautionary actions now to reduce head exposures … would limit the size and seriousness of any brain tumour risk that may exist” and “may also help to reduce the other possible harms” (p. 524).

21.16 Epilogue (pp. 524–525)#

• The case: the Italian Supreme Court affirmed an earlier ruling that INAIL (the Insurance Body for Work) must compensate a businessman. He had used mobile and cordless phones for 12 years, 5–6 hours per day, “preferably on the same side” as a neurinoma of the trigeminal (Gasser’s) ganglion, the same tumour type as acoustic neuroma but on the 5th cranial nerve. The tumour is benign but disabling.
• The chapter’s reading: “The Italian case fulfils the criteria for a causal association; more than 10 years use …, high cumulative exposure on the same side …, and a tumour type that would be predicted based on previous research” (p. 525). “No further appeal … is possible.”
• Note: this treats a legal finding about one individual as meeting causal criteria. The chapter does not discuss legal standards of proof, or how an individual attribution is inferred from population-level ORs.

Case timeline#

Lag between first credible warning and effective action. What counts as “credible” is itself contested here: - The first Hardell study (1999) was weak by its authors’ own admission. - The pooled Hardell results (2006) and the EEA warning (2007) are the chapter’s implicit “early warning”. - The IARC 2B classification (2011) is the first authoritative hazard judgement. - On the chapter’s account, the IARC classification “does not seem to have had any significant impact on governments’ perceptions of their responsibilities” (p. 524). That was 5–6 years after the EEA warning and 1–2 years after IARC. The chapter does not survey national measures or advice, so this is a characterisation, not a documented finding of no government action.

The only “actions” the chapter records are: - voluntary behaviour change (more hands-free use and texting, “due to enhanced applications”, fn 9 p. 515); - advisory statements (the EEA’s warnings; Cardis and Sadetzki’s advice); - a single court ruling (p. 525); - continued research. No harms are quantified. The chapter makes no estimate of attributable cases, costs or burden, consistent with its framing as an “emerging issue” where harm is unestablished.

The authors’ own lessons and conclusions#

Lessons derived from their evidence (empirical and interpretive): 1. Hardell and Interphone together show increased risk of glioma and acoustic neuroma with long-term use. The risk is highest ipsilaterally, adolescents appear at higher risk, and there is no consistent pattern for meningioma (p. 524). 2. Cordless phones also increase risk “when properly assessed and analysed” (p. 524). Studies that treat cordless users as unexposed bias results toward the null (pp. 514, 519, 522). 3. Null or “reassuring” studies (Inskip 2001, CEFALO 2011, the Danish cohort, incidence-trend studies) are methodologically weak and have been over-interpreted as evidence of no risk (pp. 511, 514, 523). 4. Results that differ by tumour type argue against systematic bias (p. 524). 5. The Interphone consortium’s compromise wording, and its failure to explain the continuum of evidence strength, let media and public read “not causal” as “no link” (p. 517). 6. Even an authoritative hazard classification is reinterpreted by interested parties. This is framed as a “spinning machine” comparable to tobacco-industry tactics (pp. 520–521). 7. Descriptive incidence data are of limited value for testing the association (pp. 509, 523). 8. Governments did not change their view of their responsibilities after IARC (p. 524). Media failed to provide robust information (pp. 509, 524). 9. Statistical non-significance is “a weak guide” to the strength of evidence, and causation sits at the far end of a continuum of evidence strength (Box 21.1, p. 512; p. 517). 10. Hazard categories should not be “lumped”: “Each agent needs to be considered on its own evidence” (Box 21.3, p. 522). 11. “A high participation rate is a pre-requiste for high quality in case-control studies” (p. 519).

Recommendations and advocacy: - Reduce head exposure now, especially for children and young adults, through texting, hands-free use and better phone design (Box 21.2 p. 515; pp. 509, 524). - Reconsider exposure standards based on the thermal paradigm (p. 515). - Effective labelling and warnings (p. 515). - Independent research, funded partly by industry levies on phones and masts but insulated from industry influence (pp. 515, 524). - Publish divergent expert interpretations side by side (fn 12, p. 518; proposed to IARC, not adopted). - Compensation for heavy long-term users with glioma or acoustic neuroma (pp. 509, 525). - Research on ecological effects (bees, birds) (p. 524). - Continue monitoring long-term national trends in nervous-system tumours (21.14 heading, p. 523), without using descriptive data to dismiss analytical results.

The chapter does not recommend restricting the technology. It explicitly affirms that “the benefits of mobile telecommunications are many” (pp. 509, 524). Its precaution is behavioural and design-based exposure reduction, plus information and research, which it presents as low-cost.

Mechanisms and dynamics#

1. How the warning arose. It came from three sources: an adjacent hazard (ELF fields and childhood leukaemia; brain tumours in the electronics industry), a piggy-backed research question added to an existing study design, and US litigation with its media coverage (p. 510). The signal was weak and exploratory at first (p. 511). It grew mainly within one research group through repeated, pooled analyses using the same methods (pp. 511–513). The notes’ reading: replication by independent groups was partial at best, since Interphone’s positive findings appear only in the highest-exposure subgroups (p. 517). The chapter’s own view is that the two sets of studies are “complementary and generally mutually supportive” (pp. 509–510), and that their glioma results are “similar” when the same inclusion criteria are used (p. 520, citing the Hardell group’s own re-analysis).

2. Epistemic contest over study design and which evidence counts. The dispute is less about data than about which designs and biases dominate: - Case-control studies are vulnerable to recall and selection bias. The Hardell group argues its methods minimise these (pp. 511, 519). Critics argue the opposite (pp. 513, 519). - Cohorts based on subscription records avoid recall bias but, as IARC put it, involve “considerable misclassification in exposure assessment” (pp. 520, 522–523). - Incidence trends can test for large population effects (the notes’ gloss). The chapter’s objections are different: they are confounded by changes in other risk factors, registry quality is poor, and the mechanism is unclear (p. 523).

Each side treats the design flaws of the other side’s preferred evidence as disqualifying. IARC’s decision that “No conclusions could be drawn” from the Danish cohort (p. 520) is a concrete instance of an authoritative body adjudicating which evidence counts.

3. Misclassification as a systematic force toward “no effect”. A recurring mechanism is that ignoring a substitute that delivers the same exposure (cordless phones), or defining exposure by an administrative proxy (subscription status), dilutes any real effect (pp. 511, 514, 519, 522). The chapter’s point is general: how the “unexposed” reference group is defined can conceal risk.

4. Burden and standard of proof: the “non-positive ≠ negative” error. On the chapter’s account, several authoritative voices treated inconclusive studies as exonerating: - the NEJM editorial (p. 511); - the CEFALO editorial and the Karolinska press release (p. 514); - the Danish cohort authors (p. 523); - arguably the WHO fact sheet (p. 520), though its “no adverse health effects have been established” is literally accurate. The chapter’s main complaint there is the thermal framing.

The chapter calls this “a common misconception” and says the statements go “far beyond” what the data showed. The notes’ gloss: in burden-of-proof terms, the absence of statistically significant findings in short-latency, low-exposure studies was presented as evidence of safety (Box 21.1, p. 512; p. 517).

5. Latency and timing. Mass exposure began only in the late 1990s (p. 510). Brain tumour latencies may be decades (Box 21.1, p. 512). Fewer than 10% of Interphone cases had more than 10 years of use (p. 517). Epidemiology therefore lags exposure by design. By the time studies can be informative, exposure is universal (“over 5 billion people”, p. 524) and the technology has changed generations several times (p. 510). This rapid turnover of exposure regimes (NMT → GSM → 3G → 4G, TETRA, DECT) means many studies describe exposure conditions that have since changed. The chapter notes the generations (p. 510) but does not draw this lesson explicitly.

6. Mental models and priors of experts. The chapter highlights: - Ahlbom’s pre-study view that an association was “biologically bizarre” (p. 513). - The dominance of the “thermal effects paradigm” in standards and WHO statements (pp. 515, 520, 524). - The claims the chapter attributes to “the telecommunications industry and others”: that there are “no biologically significant effects from non-thermal levels of EMF” and that, if there are, “there are no acceptable mechanisms of action that could explain these effects” (p. 524).

The dynamic: an established mechanistic model sets the prior. Epidemiological anomalies are then attributed to bias rather than treated as signals. The chapter uses the same mechanistic uncertainty in reverse, arguing that because the mechanism is unclear, flat incidence data cannot refute risk (p. 523).

7. Consortium consensus and institutional ambiguity. Interphone’s internal three-way split produced a 4-year delay, publication only after the IARC Director intervened, and a compromise conclusion that Saracci and Samet called “elegant and oracular”, one that “tolerates diametrically opposite readings” (p. 517). Co-authors then publicly contradicted one another (p. 518). The EEA’s proposal to publish dissent explicitly was rejected (fn 12, p. 518). Consensus-seeking norms in large collaborations can suppress rather than clarify disagreement, and ambiguity then transfers interpretive power to whoever frames the result most effectively.

8. Conflicts of interest, funding and independence. The chapter documents several: - Industry co-funding of Interphone (EUR 5.5m of ~20m) and protocol clauses on industry observers and pre-publication notice (pp. 517, 519). - Telecom-operator funding of the Danish cohort and undisclosed IEI funding (p. 521). - ICNIRP’s undisclosed finances and like-minded membership (p. 521, quoting Microwave News). - Ahlbom’s removal from the IARC chair for a “reported conflict of interest” (fn 15, p. 520). - A general “funding bias” claim (fn 10, p. 515).

Its proposed remedy is a structural firewall: levy-funded, independently administered research (pp. 515, 524). But the chapter applies this scrutiny one way. Its own studies received Telia money (fn 11, p. 517), and the authors’ advocacy roles (the EEA warnings; the EEA’s invitation to contribute to the BioInitiative Report) are not treated as potential interests.

9. Communication and media. Press releases shaped coverage (“reassuring”, p. 514), editorials made overreaching claims (pp. 511, 514), and headlines diverged on the same day (p. 518). Journalists read abstracts and releases, not papers (p. 524). A panel member described public “confusion” (p. 521). Box 21.3 notes the “lumping” of 2B agents such as coffee, which invites trivialising comparisons (p. 522). Strategic ambiguity plus a thin media layer leaves the public unable to form “their own judgements about how precautionary they should be” (p. 524). That last phrase shows the chapter’s model of precaution: partly individual choice by informed users, not only state regulation.

10. Reframing of an authoritative finding by interested parties. Industry bodies described 2B in largely accurate but selective terms: “not a definite nor a probable” carcinogen; “possible but not likely”. FFTI went further, calling it “in line with the dominant interpretation” that RF is “neither carcinogenic … nor probably carcinogenic” (pp. 520–521). Shortly afterwards (June 2011) the WHO EMF fact sheet reaffirmed “no adverse health effects … established” (p. 520). The chapter reads this as doubt manufacture on the tobacco pattern (p. 521), an analogy asserted rather than demonstrated for this case.

11. Credibility contests and personalisation. The dispute is heavily personalised on both sides: - An opinion piece titled “Experts who talk rubbish” (reference, p. 525). - Accusations of “intellectual bias” (fn 15, p. 520). - Named rebuttals of individual critics as “unfounded”, “ad hoc” or “easy to rebut” (pp. 513, 519). - Complaints of “unfounded attacks on individual researchers” (p. 524).

Credibility, independence and motive become proxies for evidence quality. That makes resolution harder and pulls the fight toward venues such as courts and the media.

12. Institutional inertia. Governments did not treat IARC’s classification as creating new responsibilities (p. 524). Standards stayed anchored to thermal effects (p. 515). WHO messaging stayed reassuring (p. 520). The chapter offers no detailed analysis of why. The implied explanations are industry influence, the prevailing paradigm, and ambiguous science that allows inaction.

13. Costs, benefits and who bears risk. The benefits are “many” and widely shared (pp. 509, 524). The risk, if real, falls on heavy long-term users, children and adolescents (pp. 514–515) and occupational heavy users (pp. 509, 525). Precaution is framed as cheap and borne mostly by users (texting, hands-free) or manufacturers (design) (pp. 515, 520, 524). A research levy would move research costs to industry and consumers (p. 515). The chapter does not quantify costs on either side.

14. Alternative venues. Where the scientific and regulatory process stalls, law offers another route. The Italian workers’ compensation system and courts acted on the evidence of harm in an individual case (pp. 509, 525).

15. Technology and design as a precautionary channel. Hands-free use rose “due to enhanced applications” (fn 9, p. 515). Better phone design was suggested as a means to lower exposure and make hands-free convenient (p. 515). Exposure reduction can come through product evolution and changes in use rather than regulation. The chapter notes this with “relief” but does not develop it.

16. Lay and civil-society knowledge. The BioInitiative Report is presented as giving voice to “a small but growing minority of experts” (p. 513). Eurobarometer data show public dissatisfaction with information (p. 515). Lay or local knowledge in the sense used in other chapters (workers, communities) plays almost no role. This is an expert-versus-expert dispute.

Transferable insights (technology-neutral)#

1. A study that fails to find an effect is not evidence of no effect, especially when exposure is short relative to plausible latency. Authoritative bodies repeatedly present such studies as reassuring. Evidence: NEJM editorial on Inskip (p. 511); CEFALO editorial and Karolinska “reassuring” release (p. 514); Danish cohort conclusion (p. 523); Box 21.1 on statistical significance (p. 512). Strength: strong as a methodological principle, since it is standard epidemiological reasoning (Bradford Hill). Moderate as a description of this case, where the chapter’s examples are real but chosen by a partisan author.

2. When a hazard’s effects take years to appear and exposure is new, early studies are structurally unable to rule out harm. Evidence arrives after exposure is universal. Evidence: late-1990s growth in exposure (p. 510); 15–45 year latency claim (p. 512); <10% of Interphone cases with 10+ years of use (p. 517); A-bomb brain tumours took ~50 years (fn 6, p. 512); 5 billion exposed (p. 524). Strength: strong in principle. The chapter accepts short-latency positive results. That is partly defensible (early studies can reveal but not exclude harm), but it leaves unexamined whether such early positives are plausible or biased. The specific 15–45 year figure is asserted.

3. How the “unexposed” comparison group is defined can hide a risk. Leaving out substitutes that deliver the same exposure, or using administrative proxies for exposure, pushes results toward “no effect”. Evidence: cordless phones classed as unexposed in CEFALO and Interphone (pp. 514, 519); Danish cohort misclassification, acknowledged by its own authors and by IARC (pp. 520, 522). Strength: strong for the Danish cohort (independently acknowledged). Moderate for the cordless-phone argument, where the hazard of the substitute is itself contested.

4. Expert consortia that must reach consensus on divided results can produce deliberately ambiguous conclusions. Ambiguity then hands interpretation to whoever frames it best. Explicitly publishing dissent is an alternative. Evidence: 4-year delay, three factions, conclusion brokered by the IARC Director (p. 517); “elegant and oracular” (Saracci and Samet, p. 517); opposite headlines on the same day (p. 518); co-authors’ contradictory statements (p. 518); EEA’s side-by-side proposal rejected (fn 12, p. 518). Strength: moderate. Well documented from sources independent of the authors (the IJE editorial, co-authors’ quotes), but a single case.

5. Communicating evidence strength as a binary (causal / not causal) invites the misreading “no link”. Graded vocabularies for strength of evidence are needed, and the difference between a hazard classification and a risk estimate must be explained. Evidence: Bradford Hill continuum (p. 517); Box 21.3 on hazard versus risk and “lumping” of 2B agents (p. 522); confusion after IARC (p. 521). Strength: moderate. Plausible and illustrated. But the chapter itself glosses 2B as reaching up to “nearly one in two probability” (p. 522), which contradicts IARC’s own statement that the terms have “no quantitative significance”. That shows how graded categories can themselves be over-read.

6. An authoritative hazard judgement does not settle a contested question. Interested parties reframe it, each accurately but selectively, and official bodies may keep issuing reassurance alongside it. Evidence: MMF, GSMA and FFTI statements (pp. 520–521); WHO fact sheet (p. 520); Leszczynski on “confusion” (p. 521); The Economist (p. 521). Strength: moderate–strong for divergent reinterpretation (the quotes are documented). Asserted for the claim that this amounts to coordinated doubt manufacture on the tobacco pattern (p. 521); no RF-specific evidence of coordination is offered.

7. Who funds, organises and staffs research and evaluation shapes its credibility and possibly its results. Structural independence (firewalled or levy-based funding, disclosure, recusal) matters, and scrutiny must apply to all sides. Evidence: Interphone industry funding and protocol clauses (pp. 517, 519); Danish cohort and IEI funding (p. 521); ICNIRP finances (p. 521); Ahlbom’s removal (fn 15, p. 520); funding-bias claim (fn 10, p. 515); levy proposals (pp. 515, 524); Hardell’s own Telia funding (fn 11, p. 517). Strength: moderate. The chapter asserts funding bias for RF without citing RF-specific evidence. External evidence exists for experimental (laboratory) RF studies (Huss et al. 2007, see below), not for the epidemiology at issue here. The chapter’s own one-sided scrutiny illustrates the need for symmetry. Material context it omits: most of Interphone’s industry money passed through a UICC “firewall” (see Limitations).

8. Evaluators who have publicly pre-committed to a conclusion undermine the perceived legitimacy of assessments. Institutions may act on this only at the last moment. Evidence: Ahlbom’s “biologically bizarre” statement before Interphone began (p. 513); his removal as IARC chair the day before the meeting and the EEA’s withdrawal (fn 15, p. 520). Strength: moderate for the documented episode; suggestive as a general pattern from this chapter alone.

9. A dominant mechanistic model can set the burden of proof. If effects are expected only through a known mechanism, standards and official statements anchored to it discount anomalous observations as bias. Evidence: “contested thermal effects paradigm” (p. 515); WHO’s “tissue heating is the principal mechanism” (p. 520); industry’s “no acceptable mechanisms” (p. 524); Ahlbom’s “biologically bizarre” (p. 513). Strength: suggestive. The dynamic is plausible and visible in the quotes, but the non-thermal evidence the chapter relies on (Giuliani and Soffritti; BioInitiative) is itself heavily contested.

10. When exposure reduction is cheap and easy, a lower evidence threshold for precaution is proportionate. That judgement can be shared across the scientific divide. Evidence: EEA recommendations (p. 515); Cardis and Sadetzki, both Interphone scientists, advising “simple and low cost measures” (p. 520); Sadetzki: “sufficient to support precautionary policies” (p. 518). (IARC’s Director made the same call in 2011, though the chapter does not quote it; see below.) Strength: moderate–strong as decision logic, since it was endorsed by people who disagreed about causation. The chapter gives no quantified cost-benefit analysis.

11. Exposure can shift through changes in design and use faster than through regulation. Precaution may be achieved, or made moot, by technological evolution, which also makes the epidemiology harder to interpret. Evidence: fn 9 (p. 515) on more hands-free use “due to enhanced applications”; “phones of an improved design” (p. 515); successive technology generations (p. 510). Strength: suggestive. Noted only in passing in the chapter.

12. Vulnerable subgroups may differ in both dose and sensitivity, and are the least studied. Precaution arguments focus on them early. Evidence: children’s higher absorption and developing brains (p. 514); Hardell’s age-at-first-use ORs (p. 514, Table 21.2 p. 516); “very few” child studies (p. 514). Strength: moderate for the dosimetry and data-gap point. Weak/contested for the specific risk estimates, which rest on few cases and very wide CIs.

13. Parties to a controversy disagree about which evidence counts (analytic vs descriptive, case-control vs cohort) as much as about the data. Deciding admissible evidence is a governance act. Evidence: IARC’s exclusion of the Danish cohort (p. 520); dispute over incidence data (p. 523); Myung vs Rowley and Milligan over study quality (p. 519). Strength: moderate. Clearly shown in the text.

14. When science and regulation stall, compensation and courts can recognise harm in individual cases on different standards. This creates a parallel track of accountability. Evidence: Italian Supreme Court and INAIL (pp. 509, 524–525). Strength: suggestive. Single case, and the chapter overstates its causal significance.

15. Personalised credibility contests (attacking or defending individuals’ motives and competence) substitute for resolving evidence questions and entrench polarisation. Evidence: “biologically bizarre” and “Experts who talk rubbish” (pp. 513, 525); “intellectual bias” (p. 520); “unfounded attacks on individual researchers” (p. 524); repeated “unfounded / ad hoc / easy to rebut” (pp. 513, 519). Strength: moderate as an observation that the chapter itself exemplifies. The chapter does not frame it as a lesson.

16. Media dependence on press releases and abstracts amplifies the framing chosen by study authors and institutions. Limitations buried in full texts do not reach the public. Evidence: pp. 514, 518, 524. Strength: suggestive / asserted. Illustrated with examples but no systematic evidence.

17. Small relative risks spread across near-universal exposure can matter at population scale, which motivates precaution even when individual risk is small or uncertain. Evidence: “over 5 billion people” (p. 524); “limit the size and seriousness of any brain tumour risk that may exist” (p. 524). Strength: asserted. Standard public-health logic, but the chapter does not quantify the potential burden.

18. Participants in a controversy tend to scrutinise evidence harder when it cuts against them. A precautionary argument is more credible when its evidentiary standards are applied symmetrically. Evidence (from the chapter’s own practice):

    • CEFALO’s positive result at >2.8 years accepted and its null dose and location results left out (p. 514), while null studies are discounted for short latency (pp. 511, 512, 517);
    • rising Danish incidence presented without the caveats applied to flat Swedish and US incidence (p. 523);
    • Little et al.’s conclusion quoted with its Interphone-compatible clause cut off (p. 523);
    • Interphone industry funding scrutinised while Telia funding goes unremarked (pp. 517, 519).

    Strength: moderate. Demonstrated within the text, though some of the latency asymmetry is logically defensible (see Box 21.1 note). A caution for using this chapter as a lens.

Limitations, contestation and bias check#

Nature of the chapter#

• Self-review by protagonists. The lead author assesses his own group’s studies and rebuts every critique of them. The third author narrates and endorses his own institution’s warnings (Box 21.2). No panel or response gives opposing scientists, regulators or industry their own voice, unlike many other 2013 chapters. Readers see critics only through the authors’ summaries, often via partisan secondary sources (Microwave News, i-sis.org.uk, blogs, YouTube; pp. 518, 521).
• An “emerging issue”, not a harm case. Unlike the historical Late lessons cases (asbestos, tobacco, lead), harm here was not established in 2013 and still is not. The volume’s framing, and the chapter’s own analogies to tobacco, asbestos and lead (pp. 521, 524), invite readers to project a known-harm narrative onto an unresolved question. Hindsight bias does not apply as it does elsewhere, but foresight bias (assuming the story will end like the others) does.

Selective reporting of evidence (checked against primary sources)#

• Interphone’s overall findings are under-reported. IARC’s 17 May 2010 press release (PR No. 200, read in full for this note) quotes the Interphone summary:
• “A reduced odds ratio … related to ever having been a regular mobile phone user was seen for glioma [OR 0.81; 95% CI 0.70-0.94]”.
• “ORs were < 1.0 for all deciles of lifetime number of phone calls and nine deciles of cumulative call time”.
• In the top decile “there are implausible values of reported use”.
• “No elevated OR was observed ≥ 10 years after first phone use” (glioma OR 0.98).

The chapter mentions “protective” results only at country level (p. 517). It does not report the sub-unity overall ORs, the null ≥10-year result, or the implausible-use caveat. Those facts underpinned the Interphone group’s joint refusal of a causal interpretation (the chapter itself says the investigators split three ways). They show a pattern (ORs below 1 almost everywhere, above 1 only in the top decile) that is consistent with participation and recall bias as well as with a threshold effect. - In the chapter’s favour: the Interphone summary itself said the reduced ORs “possibly reflect participation bias or other methodological limitations” (PR 200). That is a downward bias, consistent with the chapter’s (and Saracci and Samet’s) argument that Interphone likely underestimated risk. - The funding firewall is omitted. The same press release (re-checked in this audit) gives the total cost as EUR 19.2 million, of which 5.5 million came from industry sources. Of that, 3.5 million came from MMF and GSMA “through a firewall mechanism provided by the UICC”. “Most of the rest” came indirectly to centres from operators and manufacturers, for example “through taxes and fees collected by government agencies”. “Only 0.5 million € (2.5%)” came directly from industry, in Canada and France, “under contracts which preserved the independence of the study”. The chapter presents the 5.5m and the protocol clauses (pp. 517, 519) without the firewall. That does not settle independence, but it is material context. - Little et al. 2012 is summarised one-sidedly. Its abstract (verified via Europe PMC) found that predicted glioma rates under the Swedish (Hardell) risk estimates “should have been at least 40% higher than observed rates in 2008”. By contrast, “the US data could be consistent with the modest excess risks in the Interphone study”. The chapter reports only the criticism of Hardell’s compatibility (p. 523). Its quotation of Little’s conclusion ends with an ellipsis just before the clause “although the US data could be consistent with the modest excess risks in the Interphone study” (abstract re-checked in this audit). Little compared published, not harmonised, estimates, so this does not directly test the chapter’s claim that the two sets of glioma results are “similar” under the same inclusion criteria (p. 520). But it sits uneasily with that claim. - CEFALO’s null exposure-response results are omitted. The CEFALO abstract (Aydin et al. 2011, verified via Europe PMC) reports risk related to time since subscription “but not to amount of use”, and no increased risk “for brain areas receiving the highest amount of exposure”. The chapter cites the time-since-subscription trend as “further support of a true association” (p. 514) and omits both null results. - IARC’s own precautionary advice is not quoted. IARC PR No. 208 (31 May 2011) has Director Wild saying: “it is important to take pragmatic measures to reduce exposure such as hands-free devices or texting”. This would have supported the chapter’s argument. The same release says the Working Group comprised “31 scientists from 14 countries” (the chapter says 30, p. 520). It also says several in-press Interphone papers “were made available to the working group shortly before it was due to convene” (re-checked in this audit). Fn 15 describes this as part of the unpublished data being given to the meeting.

Methodological arguments that are one-sided or weak#

• Laterality. Higher ipsilateral ORs are treated as supporting causation (pp. 511, 513, 517, 524). The well-known alternative is not discussed: cases may over-attribute use to the tumour side (recall bias). In Interphone this showed up as raised ipsilateral and depressed contralateral ORs. Hardell’s triple-check of ear use (p. 511) addresses consistency of reporting, not its accuracy relative to the truth.
• “Different results for different tumour types … strongly argues against systematic bias” (p. 524). This does not follow. Recall and participation biases can differ by tumour type: malignant glioma patients have more cognitive impairment and higher mortality, and benign tumours have different diagnostic pathways. A tumour-specific pattern is necessary but not sufficient to exclude bias.
• Asymmetric latency reasoning. Box 21.1 says clear evidence needs studies at least as long as a 15–45 year latency (p. 512). Yet the chapter treats positive findings at >10 years, and CEFALO’s OR at >2.8 years (p. 514), as support, while null studies are discounted for the same short latencies. Part of this asymmetry is logically defensible, since short follow-up can reveal an effect but not exclude one. What is missing is any discussion of why excess risk would appear so early, or whether early positives reflect bias.
• Asymmetric use of incidence data. Rising Danish incidence and a “4-fold” glioblastoma news release (not publicly available data, p. 523) are presented without the caveats (diagnostic and registration change, other risk factors) that the chapter applies to flat Swedish and US trends. The chapter does not, however, explicitly claim the Danish rises show causation.
• Direction of bias asserted. The chapter says recall and observational bias in Interphone “would tend to underestimate the risk” (p. 518) without specific support. Differential recall by cases more commonly inflates ORs.
• Participation rates. Hardell’s ~90% rates are presented as proof of quality (pp. 513, 518, 519). Critics’ concern that they are unusually high is answered only by assertion: the oncologist-investigator explanation (p. 513), and the claim that a high rate is “a pre-requiste for high quality” (p. 519).
• Deceased-case critique. Called “scientifically unfounded” and lacking a stated reason (p. 512), though selection bias from excluding a fast-fatal tumour’s deceased cases is a standard concern that the group’s own fourth study was designed to address.
• Epilogue overreach. A court’s compensation decision in one case is said to “fulfil the criteria for a causal association” (p. 525). The chapter does not distinguish legal from scientific causation. Per Microwave News’s report (a secondary source), the plaintiff’s expert witness was Angelo Gino Levis, co-author of Levis et al. 2011, which the chapter cites among the “more favourable” analyses (p. 519). The court reportedly preferred Hardell’s work because it was independent of industry funding.

Source quality#

• Heavy reliance on advocacy and secondary sources: Microwave News (pp. 518, 521, 525), i-sis.org.uk (p. 518), blogs (p. 521), YouTube (p. 518), a non-public Danish news release (p. 523).
• The non-thermal effects claim rests on a single ICEMS/Ramazzini monograph (p. 524).
• Ecological claims (bees, birds, p. 524) rest on a handful of small field and behavioural studies.
• The BioInitiative Report (p. 513) is presented approvingly, as the voice of “a small but growing minority of experts” with “considerable impact”. [background, not re-verified] It was criticised by several health and radiation-protection bodies at the time; the chapter does not mention that.

Errors and inconsistencies (minor but relevant to reliability)#

• Acoustic neuroma CI “6.8 (95 % CI = 1.4–3.4)” (p. 514), an impossible interval confirmed in the PDF.
• Cross-reference to “Section 20.9” for Interphone (fn 4, p. 511); it is 21.9.
• 272 vs 271 Group 2B agents (p. 522).
• “Levis et al., 2001” in the references vs 2011 in the text (pp. 510, 519, 528).
• Working Group size 30 (p. 520) vs IARC PR 208’s 31. The discrepancy is unexplained and may reflect a late membership change, so it is not necessarily a chapter error.
• “need to be accompanied” (p. 509) vs “need not to be accompanied” (p. 524).
• Box 21.3’s “nearly one in two probability” gloss on 2B (p. 522) is the authors’ interpretation. It contradicts the IARC 2006 Preamble, which the box itself cites and which says the terms have “no quantitative significance”. It also sits awkwardly with the box’s own emphasis on hazard, not risk.
• Ahlbom’s opinion letter is said to predate the start of the Swedish Interphone study (p. 513), but it is dated 2001, and Interphone fieldwork ran 2000–2004 (p. 517). A possible chronology slip.
• “Perhaps even IARC has contributed to this confusion by seeming to agree with … the Danish cohort” (p. 521) is unexplained, given that IARC drew “No conclusions” from that cohort (p. 520).
• Fn 6 supports a claim about “brain cancers” with a paper on meningiomas, which are mostly benign (p. 512).
• Spelling variants in the text: “Guiliani and Soffriti” (p. 520) vs Giuliani and Soffritti; “Ong and Glanz” (p. 521) vs Glantz; “Dariuz” (p. 521) for Dariusz; “Hazleton” (p. 523) vs Hazelton.

Fair points in the chapter’s favour#

• The “non-positive ≠ negative” critique is sound, and several of its targets (the NEJM editorial, the “reassuring” CEFALO messaging) did overreach relative to their data.
• The Danish cohort’s exposure misclassification is real, and was acknowledged by IARC and by the cohort’s own authors (pp. 520, 522).
• The “oracular” critique of Interphone comes from independent senior epidemiologists (Saracci and Samet), not only from the authors.
• IARC’s 2B classification by a “very large majority” (p. 520) vindicated the chapter’s minimum claim that the human evidence was “limited”, not absent.
• The low-cost precaution argument was shared by Interphone leaders (Cardis, Sadetzki) and by IARC’s Director.
• Industry trade-body statements did frame 2B in the most reassuring terms available (pp. 520–521).
• The chapter is transparent about its own funding, including Telia (fn 11), and about the EEA’s involvement (Box 21.2; fn 15).
• Interphone’s own summary attributed its sub-unity ORs possibly to participation bias (PR 200), a downward bias. That supports the chapter’s argument that Interphone likely underestimated any risk.
• The chapter reports its own acoustic neuroma and meningioma results that fail to support a cordless or age effect (Table 21.1; “No clear age dependent pattern … for meningioma”, p. 514), and says outright that there is “no consistent pattern” for meningioma (p. 524).
• External evidence supports the general funding-bias concern for experimental RF research. Huss et al. 2007 (Environ Health Perspect; verified via Europe PMC) found that studies funded exclusively by industry were least likely to report effects: OR 0.11 (0.02–0.78) compared with publicly or charity-funded studies. They concluded that “interpretation of results … should take sponsorship into account”. The chapter does not cite it. It strengthens footnote 10 for laboratory studies, but does not speak directly to the epidemiology at issue here.

What later evidence says (post-2013)#

Items marked [verified] were checked against primary abstracts or documents in this session (Europe PMC, IARC and NTP documents). Items marked [background] come from general knowledge and were not re-verified here.

• IARC re-evaluation pending.
• [verified] The IARC Advisory Group on priorities for 2020–2024 (report, 2019) recommended RF as “High priority (and ready for evaluation within 5 years)”, advised for the latter half of the period. It noted “the evidence remains mixed” and cited Coureau 2014, Carlberg and Hardell 2015, NTP 2018 and Falcioni 2018.
• [verified] The 2025–2029 priorities poster (IARC 2024) again lists “Non-ionizing radiation (radiofrequency)*”.
• I could not confirm whether a re-evaluation meeting has been scheduled or held. As far as verified, the 2011 2B classification still stands.
• Largest synthesis. [verified] Karipidis et al. 2024 (Environment International), “commissioned and partially funded by the World Health Organization”, reviewed 63 aetiological articles (1994–2022). They found “moderate certainty evidence” that mobile phone RF exposure “likely does not increase the risk” of glioma, meningioma, acoustic neuroma, pituitary or salivary gland tumours in adults, or of paediatric brain tumours. Null findings extended to cordless phones and occupational exposure. There was no increase in pooled relative risks with time since starting use.
• Note: co-authors include Maria Blettner (the lone dissenter in the IARC vote per p. 521) and Susanna Lagorio (a critic of the Italian court decision, per Microwave News). The contest over who evaluates therefore continues.
• [verified] A 2025 commentary (Frank, Front Public Health) argues most primary studies “probably underestimate” causation. The debate is not closed.
• Children. [verified] MOBI-Kids (Castaño-Vinyals et al. 2022, Environ Int; 899 cases aged 10–24, 1,910 controls, 14 countries; co-authors include Cardis, Sadetzki and Kundi, all treated favourably in the chapter) found “no evidence of a causal association between wireless phone use and brain tumours in young people”. Odds ratios fell with time since start of use, attributed partly to recall and prodromal-symptom bias. This directly tests, and does not support, the chapter’s claims on adolescents (pp. 514, 524).
• Cohort. [verified] Million Women Study update (Schüz et al. 2022, JNCI; 776,156 women, 14 years): glioma RR 0.89 (0.80–0.99). Acoustic neuroma and meningioma were not significantly elevated, even though an earlier 2013 report had found increased acoustic neuroma risk with long-term use (per the IARC 2019 AG report). The authors conclude that use “under usual conditions does not increase brain tumor incidence”.
• Case-control support.
• [verified] CERENAT (France; Coureau et al. 2014, Occup Environ Med): no association for regular use overall, but heaviest users (≥896 h) had glioma OR 2.89 (1.41–5.93) and meningioma OR 2.57 (1.02–6.44). This supports a heavy-use signal.
• [verified] The Hardell group’s pooled 1997–2003 and 2007–2009 data (Hardell and Carlberg 2015, Pathophysiology) reported glioma OR 3.0 (1.7–5.2) at >25 years latency.
• [verified] Hardell and Carlberg 2013 (Rev Environ Health) argued that glioma and acoustic neuroma should be considered caused by RF, i.e. Group 1.
• Incidence trends.
• [verified] Nordic males 1979–2016 (Deltour et al. 2022, WHO-funded): trends “incompatible with risk ratios of 1.08 or higher” at a 10-year lag, and “no changes … consistent with a substantial risk”.
• [verified] Australia 1982–2013 (Karipidis et al. 2018, BMJ Open): “no increase in any brain tumour histological type or glioma location that can be attributed to mobile phones”.
• These studies became central to the weight-of-evidence case against large risks, contrary to the chapter’s view that incidence data are “of limited value” (p. 523).
• [background] Some analyses reported rising glioblastoma incidence in England (e.g., Philips et al. 2018). Others attribute such rises at least partly to changes in diagnosis and registration. This remains contested.
• Animal evidence (not covered in the chapter, which excludes it from scope). [verified] The US National Toxicology Program (2018) found “clear evidence” of malignant heart schwannomas and “some evidence” of malignant gliomas in male rats exposed whole-body to 900 MHz RF at high levels. NTP states the findings “should not be directly extrapolated to human cell phone usage”. Schwannoma is the same cell type as acoustic neuroma, so this is partial support for biological plausibility, but the exposures were high and whole-body.
• Standards. [background] ICNIRP’s 2020 RF guidelines retained a thermal basis and judged non-thermal evidence insufficient. The chapter’s call to reconsider the thermal paradigm (p. 515) was not taken up by ICNIRP.
• Law.
• [background] Further Italian decisions reportedly awarded compensation for acoustic neuroma (e.g., Ivrea 2017, upheld on appeal in Turin in 2020).
• [background] In US litigation, courts excluded plaintiffs’ general-causation experts in Newman v. Motorola (D. Md. 2002, 218 F. Supp. 2d 769; a CourtListener search confirms the opinion mentions Hardell, but the text was not retrieved). The chapter does not mention US litigation or Hardell’s role in it.
• Research levy. [background] France has reportedly funded an RF research programme through ANSES from a tax on network operators since about 2011. This is a partial real-world instance of the Box 21.2 levy proposal. Not verified here.
• Precaution via use patterns. Voice calls held to the head declined as a share of use while data, texting and hands-free/speaker use grew. This is the mechanism flagged in fn 9 (p. 515). The IARC 2010 press release already noted “lower emissions, on average, from newer technology phones, and the increasing use of texting and hands-free”. Exposure reduction largely happened through technology and behaviour. That also complicates any later detection of risk.

Net hindsight assessment (as of 2026). The chapter’s institutional and communicative observations have held up well as descriptions of how the controversy unfolded: ambiguity, divergent reinterpretation, funding disputes, “non-positive” read as “negative”, and personalised conflict. Its central empirical claims have mostly weakened: magnitude of risk, excess risk in adolescents, risk from cordless phones, and the irrelevance of incidence data. A possible small excess among the heaviest long-term users is not excluded. CERENAT and the Interphone top decile are consistent with one, and IARC still lists RF as a priority for re-evaluation. The low-cost-precaution argument was reasonable on the 2011 evidence and was endorsed by IARC’s Director. Its value in hindsight lies in the decision logic, not in vindication of harm.

Notable quotes#

1. “a ‘non–positive’ study is often assumed to be a ‘negative’ study when in fact the data do not support this assumption” (p. 511)
2. “any study of cancer has to be at least as long as the average latent period for the tumour being studied before there will be any clear evidence of a cancer risk” (Box 21.1, p. 512)
3. An association was “biologically bizarre” (Ahlbom, as reported, p. 513)
4. “Appropriate, precautionary and proportionate actions taken now … are likely to be seen as prudent and wise from future perspectives” (EEA 2007, Box 21.2, p. 515)
5. The Interphone conclusion was “both elegant and oracular…(which) tolerates diametrically opposite readings” (Saracci and Samet, quoted p. 517)
6. “This meant that the media and the public could assume that ‘not causal’ meant ‘no link’” (p. 517)
7. “it is significant that IARC has concluded that RF electromagnetic fields are not a definite nor a probable human carcinogen” (Milligan, MMF, quoted p. 520)
8. “No doubt the IARC decision started a world-wide spinning machine perhaps similar to the one launched by the tobacco industry” (p. 521)
9. “It is remarkable that the IARC carcinogenic classification does not seem to have had any significant impact on governments’ perceptions of their responsibilities” (p. 524)
10. “Without accurate and reliable reporting in the media the public do not get a robust and consistent information on potential health risks to make their own judgements about how precautionary they should be” (p. 524)

Open questions#

1. Weighing conflicting evidence streams. When case-control associations conflict with population incidence trends, what decision rule is defensible? Incidence data can rule out large risks but not small or late ones. The chapter rejects incidence data; later syntheses lean on them heavily.
2. Thresholds for low-cost precaution. If the cost of precaution is near zero (texting, hands-free), what evidence threshold should trigger official advice, and whose responsibility is it? IARC’s Director, Interphone leaders and the EEA all endorsed exposure reduction in 2011, yet governments largely did not act (p. 524). Why?
3. Consortium dissent. Could consortia be required, or encouraged, to publish structured dissent, as the EEA proposed (fn 12, p. 518)? Would that reduce or amplify public confusion?
4. Symmetric scrutiny of interests. How should assessments handle the full range of interests (industry funding, advocacy commitments, expert-witness roles, institutional reputations) symmetrically, rather than only those of the other side?
5. Fast-changing exposures. How can evidence be generated when exposure regimes change faster than epidemiology can characterise them (p. 510)? Prospective exposure monitoring? Industry data-sharing obligations?
6. Who acts when the science is stuck. What did the Italian compensation line of cases actually rest on scientifically, and how do legal venues interact with unresolved scientific assessments?
7. Precaution by design and market. Did the move to hands-free use and lower-emission designs (fn 9, p. 515) happen because of precautionary messaging, or independently? What does that imply for “precaution by design” versus regulation?
8. The pending IARC re-evaluation. If IARC re-evaluates RF, will the NTP and Ramazzini animal findings together with the heavy-user case-control signals outweigh the null cohort, child and incidence evidence? How will the dispute over who sits on the panel (fn 15) play out again?
9. Case selection in Late lessons. What does including an unresolved, and on later evidence probably low-magnitude, hazard alongside historical harms imply about the volume’s case selection? Would a chapter on a “false positive” or “unresolved” warning teach different lessons?
10. The “funding bias” claim. Footnote 10 asserts it across fields, but the chapter offers no RF-specific evidence. How robust is the RF-specific evidence (e.g., Huss 2007 for experimental studies), and does it extend to epidemiology?

Audit log#

Independent audit against the full text extract (PDF pp. 511–531), with rendered PDF checks of pp. 514, 516 and 517. Outside checks: IARC PR 200 and PR 208, the IARC 2006 Preamble, and the CEFALO and Little et al. 2012 abstracts. All numbers, dates and quotations not listed below were checked and found accurate.

1. Authors: marked “Gee was a lead editor of Late lessons” as outside knowledge, not stated in the chapter.
2. Standpoint: “contributed a chapter to the BioInitiative Report” corrected to “was invited to contribute” (the chapter says only that the EEA was invited, p. 513).
3. Standpoint: “withdrew … in protest” replaced with the reasons fn 15 actually gives.
4. Standpoint: “every published critique” narrowed to “every critique it discusses”.
5. Abstract: “Four charges” corrected to three charges plus a recommendation, with the fuller wording.
6. Abstract: refined the note on “need to” vs “need not to”, which say literally different things; added the neuroma/neurinoma variance.
7. 21.2: removed the overstatement “not a biological signal”. The mobile questions were added partly because of litigation and media coverage, to a study prompted by an occupational signal and the ELF experience.
8. Box 21.1: corrected fn 6 wording (“the atomic bomb”, meningiomas mostly benign) and added the omitted Stein et al. citation.
9. Box 21.1 analytic note: softened the latency-asymmetry charge. Short studies can reveal but not exclude effects; what is missing is any discussion of early-onset mechanism or bias.
10. 21.5: reworded the IEI note. It was presented as a link the chapter “implied”; it is now labelled outside context, cross-checked against the reference list, and the chapter is not credited with a link it does not draw.
11. 21.6: noted that “biologically bizarre” is unquoted in the source, that Adami led the letter and co-wrote the NEJM editorial, and that the 2001 date conflicts with “before the study started”.
12. 21.7: added omitted Table 21.1 values, notably cordless phone and acoustic neuroma >10 years OR 1.0 (0.3–2.9), which qualifies the “cordless increases risk” conclusion.
13. 21.8: replaced the overstated “inconsistency” between text and Table 21.2 with an accurate description; noted the source mismatch.
14. 21.8 CEFALO: added the chapter’s own “very few cases” caveat, and the CEFALO null dose and location results the chapter omits (verified abstract).
15. 21.9: added the chapter’s “reportedly” hedge on the three factions; corrected Cardis 2011 exposure window from “≥7 years” to “>7 years” and restored “J/kg at the estimated tumour centre”.
16. 21.10: flagged the unsupported claim that recall bias “would tend to underestimate” risk; added the chapter’s own caveat that the participation-rate comparison is not like-for-like.
17. 21.10: replaced the spliced Cardis and Sadetzki quotations (which produced ungrammatical composites) with the verbatim text.
18. 21.11: softened “30 vs 31” from a chapter error to an unexplained discrepancy.
19. 21.12: added the full “Perhaps even IARC …” quote and its tension with p. 520; added fn 16’s “up to USD 4 million on research”.
20. 21.13 Box 21.3: added that the IARC 2006 Preamble, which the box cites, says 2A/2B have “no quantitative significance” (verified), contradicting the “one in two probability” gloss.
21. 21.13: added that Ahlbom et al. 2007 (chapter-cited) had also criticised the Danish cohort, so the camps were not neatly divided.
22. 21.14: noted the chapter’s ellipsis cutting Little et al.’s Interphone-compatible clause; softened the incidence “asymmetry” to a partial one; added the implied monitoring recommendation.
23. Timeline: 2007 row no longer calls the EEA warning the “first institutional” warning (not claimed in source); ~2006 row flagged as inferred; late-May and post-IARC rows re-dated (Frei is not June–July; EEA withdrawal preceded 23 May).
24. Lag paragraph: “no effective government action” replaced with the chapter’s actual wording, and noted that the chapter does not survey national measures; advisory statements added to recorded actions.
25. Authors’ lessons: added three omitted lessons (significance a weak guide; each agent on its own evidence; participation rate and quality). Recommendations: added national trend monitoring.
26. Mechanisms 1: added the chapter’s own view (Hardell and Interphone “mutually supportive”, “similar”), set against the notes’ replication reading.
27. Mechanisms 2: “misclassify exposure massively” changed to IARC’s “considerable”; “exclude” changed to “No conclusions could be drawn”; the incidence-data gloss separated from the chapter’s actual reasons.
28. Mechanisms 4: WHO’s inclusion hedged; “burden of proof” labelled as the notes’ gloss rather than the chapter’s framing.
29. Mechanisms 5, 6, 7, 8, 10: softened “technology that no longer exists”; corrected the paraphrase of the industry mechanism claim; attributed “elegant and oracular” to Saracci and Samet; BioInitiative wording; “simultaneously” changed to “shortly afterwards” and FFTI distinguished.
30. Insights 2, 5, 7, 18: latency asymmetry softened; IARC “no quantitative significance” added; Huss 2007 limited to experimental studies and the firewall noted; the evidence list for symmetric scrutiny updated.
31. Limitations (Interphone): “most Interphone authors declined” corrected; added Interphone’s own attribution of sub-1 ORs to participation bias (cuts in the chapter’s favour); the firewall paragraph re-verified and expanded.
32. Limitations (Little): softened “cuts against” (Little compared unharmonised estimates); noted the ellipsis. Added the CEFALO omission.
33. Limitations (methods): the latency and incidence asymmetries rebalanced; added the asserted direction of bias; participation-rate critique made fairer (the chapter does answer, by assertion).
34. Source quality: “presented as authoritative” changed to “presented approvingly”; agency criticism marked [background].
35. Errors list: 30/31 softened; added Preamble contradiction, Ahlbom chronology, IARC/Danish-cohort tension, fn 6 meningioma point, spelling variants.
36. Fair points: added Interphone participation-bias acknowledgement, and the chapter printing its own non-supportive table values; Huss limited to experimental studies.
37. Digest: corrected “Gee’s withdrawal” to the EEA’s withdrawal; rephrased who “failed to respond”; the “same day” claim limited to headlines; added cordless acoustic neuroma OR 1.0; added Interphone to cordless misclassification; softened latency asymmetry; added CEFALO omission, Little ellipsis, Preamble point, participation-bias counterpoint and the monitoring recommendation.
38. Strand check: no mentions of contemporary technologies or companies outside the report were found. Telia, MMF, GSMA, FFTI, Philip Morris and the operators named are all in the chapter.