The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users

Mobile Telecommunications and Health Research Programme The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users G J Rubin, A J Cleare, and S Wessely RUM 17

Project title: The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users Project reference: RUM 17 Project Director: Project Monitor: Professor Simon Wessely Dr Z J Sienkiewicz Project start date: 1 April 2003 Project end date: 30 June 2006 Final report date: 05 October 2006 Date approved by Monitor: 20 December 2006 Date approved by Chairman: 05 March 2012

The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users Dr G James Rubin 1,2, Dr Anthony J Cleare 2, Professor Simon Wessely 1 1 Executive Summary The health effects most commonly associated with mobile phone use are subjective symptoms such as headaches, dizziness, fatigue and tingling sensations. As yet, however, no generally accepted biological mechanism has been found which might explain how mobile phone signals could cause these symptoms. Whether a subgroup of people exists who are particularly sensitive to mobile phone signals is also uncertain, although some people do report experiencing symptoms almost every time they use a mobile phone. This apparent sensitivity to mobiles represents a subcategory within a more general condition called electrosensitivity or ES. People who report having ES typically describe negative symptoms as a result of exposure to any from a wide range of electromagnetic devices, including computer monitors, televisions, microwave ovens and overhead powerlines. The causes of ES are the subject of considerable debate. Most previous studies which have exposed people with ES to electromagnetic fields under double-blind conditions have not been able to find any link between the presence of these fields and the reporting of symptoms. As such, it has been suggested that the symptoms experienced by people with ES may be triggered more by psychological than by electromagnetic factors. In this study we tested 71 people who reported often developing headaches as a result of exposure to GSM mobile phones signals and 60 non-sensitive control volunteers. Each volunteer was asked to take part in three testing sessions, involving exposure to 50 minutes of a pulsing GSM signal, exposure to 50 minutes of a non-pulsing signal and exposure to 50 minutes with no active signal present. The order that these conditions occurred in was determined at random for each volunteer and the study was doubleblind with neither volunteers nor researchers being told which sessions were which. During each session, volunteers were asked to complete questionnaires about any adverse symptoms that they experienced and to provide blood and saliva samples which we tested for various hormone levels. Although volunteers did experience negative sensations during the experiment, these were just as likely to occur in the inactive testing sessions as in the GSM sessions or the non-pulsing sessions. In other words, we found no evidence that it was the mobile phone signals that were causing these sensations. Similarly we found no robust evidence that GSM signals affected hormone levels or that resting hormone levels were different in our sensitive and control groups. We did observe several psychological differences between our sensitive and control volunteers, however. In particular, volunteers who reported having ES tended to be more concerned about the health effects of a wide range of environmental issues, had significantly higher levels of depression and were more likely to use mobile phones predominately to make work related calls. Sections of General Hospital Psychiatry 1 and Neurobiology of Mood Disorders 2, Department of Psychological Medicine, Institute of Psychiatry, King s College London, Weston Education Centre, Cutcombe Road, London SE5 9RJ.

The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users Given these findings and given that our inactive testing sessions were sufficient to trigger severe symptoms in some sensitive participants, it appears that psychological factors may play an important role in this condition. 2 Aims and Objectives The main purpose of our research was to address two issues identified as priority areas for additional research by the Stewart Report 1 : Do normal users experience unpleasant symptoms as a result of mobile phone use (para 5.261)? Are some individuals particularly sensitive to the possible health effects of mobile phones (para 5.259)? We also tested several additional hypotheses relating to physiological and psychological factors that may be related to apparent sensitivity to mobile phones. In particular, we tested whether any differences existed between people who report being sensitive to mobile phones and non-sensitive control participants with regard to baseline neuroendocrine function, neuroendocrine function following exposure to a mobile phone signal, critical flicker fusion threshold, resting pulse and psychological variables such as emotional distress and worries about the health effects of modern life. 3 Participants Dr GJ Rubin (Lecturer) 1,2 Mr G Hahn (Senior Research Nurse) 1,2 Professor B Everitt (Professor Emeritus of Biostatistics) 3 Dr A Papadopoulos (Principal Clinical Scientist) 2 Dr AJ Cleare (Reader in Affective Disorders) 1,2 Professor S Wessely (Professor of Epidemiological and Liaison Psychiatry) 1 1. King s College London, Institute of Psychiatry, Department of Psychological Medicine, Section of General Hospital Psychiatry, Weston Education Centre, London SE5 9RJ 2. King s College London, Institute of Psychiatry, Department of Psychological Medicine, Section of Neurobiology of Mood Disorders, 103 Denmark Hill, London SE5 8AZ 3. King s College London, Institute of Psychiatry, Department of Biostatistics and Computing, London SE5 8AZ 4 Achievements The recent uptake of mobile phones has been accompanied by some concern about possible health risks 1. In the general population, the health effects most often attributed to mobile phone use are nonspecific symptoms. Excluding sensations of mild warmth, the most commonly reported symptoms are headache, burning, dizziness, fatigue and tingling 2. Mechanisms to explain these phenomena remain speculative, and although it has been suggested that the pulsing nature of global system for mobile communication (GSM) signals may be partly to blame 3, experiments that have exposed healthy adults to GSM signals under blind conditions have not found any significant effects on the reporting of symptoms 4. Whether a subgroup of people who are more sensitive to GSM exists remains unclear. Of particular interest are people who report symptoms almost every time they use a mobile phone 5. This phenomenon falls within the broader category of electrosensitivity (ES), a medically unexplained condition in which nonspecific symptoms are reported after perceived exposure to any of a wide range of electrical devices, including mobile phones, visual display units, and power lines. The prevalence of self reported ES in the United Kingdom is unknown, but community studies in Sweden and California put the figure at between 1.5% and 3% 6;7. Within this condition, at least one subdivision has been suggested between people with 2

3 The effect of mobile phone use on symptoms and neuroendocrine function in normal and hypersensitive users discrete problems relating to a specific electromagnetic device and those who report a more complex illness involving multiple symptoms associated with several electromagnetic stimuli 8. The second group tend to have more severe symptoms 9, a worse prognosis 8 and a different psychological profile 10 than the first. Provocation studies that have exposed people who report ES to electromagnetic fields under blind conditions have so far failed to provide any good evidence linking the presence of such fields to the reported severity of symptoms 11. Several authors have therefore suggested that psychological mechanisms may be more relevant in causing the condition 12. In particular, it is possible that technostress is one factor which initially leads someone to develop negative symptoms while using their mobile phone 13. Technostress can be defined as the stress associated with having to interact with a new technology which may be difficult to use and which also brings with it a perceived increase in workload. Other factors, such as having a high degree of concern about the health effects of modern life 14;15, being exposed to media stories about the possible adverse health effects of mobile phones 16 and learning that precautionary measures are recommended in relation to mobile phone use 17 may then lead people to believe that these symptoms are the result of exposure to the mobile phone signal. Once this association has been made, a vicious circle may develop in which each new exposure to a mobile phone is accompanied by increased anxiety and increased expectation of further symptoms, factors which are both well known to predict increased symptom reporting across a range of conditions 18. Certain physiological mechanisms may also be relevant in either triggering or perpetuating ES. For instance, it has been reported that individuals who suffer from ES tend to exhibit higher basal heart rates and lower heart rate variability than healthy individuals 19 suggesting that an abnormality in the autonomic nervous system may play some role in the pathogenesis of the disorder. There has also been some suggestion that patients with ES may show generalised hyperresponsiveness to external stimuli, as demonstrated by a heightened ability to discriminate flicker from constant light (the Critical Flicker Fusion (CFF) test) 19. Furthermore, given the similarities in symptom profiles between ES and chronic fatigue syndrome, a condition in which abnormally low levels of cortisol are linked to symptom severity 20, it is possible that altered neuroendocrine functioning is another important factor in ES. Although research examining the effects of mobile phone exposure on neuroendocrine function has so far failed to identify any consistent effects 21, as yet there has been no research into whether such effects are more apparent in people who report sensitivity to mobile phone signals. In this study, we first tested whether people with self reported sensitivity to GSM would experience greater headache severity after double blind exposure to a GSM signal than to a sham signal. Secondary outcomes included other symptoms and ability to discriminate GSM from sham signals. We also tested whether a pulsing signal resulted in greater reporting of symptoms than a non-pulsing signal. The same study was also used to test whether participants with ES were significantly different to control participants with regard to several variables proposed in the psychological model of the condition, namely emotional distress, modern health worries and association of mobile phones with occupational stress. We also tested whether differences in baseline CFF threshold, pulse or neuroendocrine function could be observed between the groups, and whether neuroendocrine function was significantly affected by GSM exposure. Methods In order to assess the effects of mobile phone exposure on symptoms and neuroendocrine function we used a within participants study in which we exposed people who reported adverse reactions to mobile phone signals (sensitive group) or who did not report any such effects (control group) to three conditions: a signal mimicking that produced by a 900

4 The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users MHz GSM mobile phone, an unpulsed continuous wave signal, and a sham exposure with no active signal present. Our Clinical Trials Unit determined the order in which these conditions occurred for each participant on enrolment by using a computerised random numbers generator and counter-balancing within blocks of six consecutive participants. Exposures were double blind that is, neither participants nor researchers were told which type of exposure was present in which testing session. The controls for our exposure equipment allowed for 256 possible settings, of which 15 had been randomly allocated to each condition. Only the Clinical Trials Unit knew which settings related to which exposure. For the first nine control participants and six sensitive participants (11.5% of all participants), Clinical Trials Unit staff told researchers which setting to use on the morning of each exposure. Given the theoretical possibility that the meaning of a setting might eventually be inferred by observing repeated participants reactions to it, for the remaining sessions Clinical Trials Unit staff entered the codes and then obscured them from the researchers with opaque tape. In order to assess differences between sensitive and control participants with regard to basal neuroendocrine levels, CFF threshold, pulse and psychological factors, we used a between-participants design comparing scores on these variables prior to any experimental exposure between our sensitive and control participants. advertising by interested clinicians and by our funding body, posters in general practitioners surgeries, adverts and articles in the press and specialist health publications, email circulars, and word of mouth. Exposures We generated exposures by using the standard GSM handset system used within the UK Mobile Telecommunications and Health Research programme. The antenna for this headband-mounted system was positioned slightly above and behind the left ear and within a few millimetres of the participant s scalp. Both GSM and continuous wave conditions produced a target specific absorption rate adjacent to the antenna of 1.4 W/kg, with an uncertainty of ±30% 22. For the sham exposure, a continuous wave signal was generated to ensure that the system heated up to the same degree as the active exposures but was diverted to an internal load instead of being transmitted through the antenna; only minimal leakage of this signal occurred (specific absorption rate <0.002 W/kg). Questionnaires We assessed severity of symptoms during exposure by using 100 mm visual analogue scales 23, anchored with the phrases no sensation and worst possible sensation. These scales measured headaches; nausea; fatigue; dizziness; skin itching, tingling, or stinging; sensations of warmth or burning on skin; and eye pain or dryness. Participants To be eligible for the sensitive group, participants had to report often experiencing headache-like symptoms within 20 minutes of using a 900 MHz GSM mobile phone. Participants who did not attribute any symptoms to mobile phone signals were eligible for the control group. We excluded people who were aged under 18 or over 75, were pregnant, had a psychotic illness, were currently using antidepressants, or reported severe symptoms at baseline while in our testing room. We recruited participants through mailshots organised by an ES support group, We collected other data at baseline, consisting of demographics and current or previous mobile phone usage. We also asked participants to record the frequency with which they experienced 11 common symptoms after a mobile phone call (never, 25% of calls, 50% of calls, 75% of calls, every call). We asked participants in the sensitive group about duration of illness and symptoms, how near a mobile phone needed to be before they could detect it, whether they considered themselves to have electrosensitivity or sensitivity to electromagnetic fields, whether they had sought treatment and whether their sensitivity impaired their daily functioning 24.

5 The effect of mobile phone use on symptoms and neuroendocrine function in normal and hypersensitive users Other variables assessed at baseline consisted of reason for using a mobile phone (only for work, mainly for work, for both work and social reasons, mainly for social reasons, only for social reasons); perceived usefulness of mobile phones (I find having access to a mobile phone: not at all useful, slightly useful, moderately useful, extremely useful); general health status (Medical Outcomes Survey 36-item Short Form; SF-36 25 ); non-psychotic psychiatric caseness (defined using a cut-off score of 4 or more on the 12-item General Health Questionnaire; GHQ-12 26 ); total depression score (9-item Patient Health Questionnaire; PHQ-9 27 ); and concern about the health effects of various aspects of modern life categorised under the headings toxic interventions, radiation, tainted food and environmental pollution (Modern Health Worries; MHW 15 ). CFF theshold CFF threshold was assessed using a Flicker Fusion instrument (Model 12021; Lafayette Instrument Company, Indiana USA) and calculated as the mean of six trials, which alternated between ascending trials starting at a flicker rate of 1Hz and descending trials starting at 60Hz. All trials used a luminance setting of 91% and flicker increments or decrements of 1Hz per second. Neuroendocrine measures Plasma neuroendocrine levels were measured by taking blood samples using 9ml Monovette syringes containing EDTA KE preservative or lithium heparin preservative (Sarstedt, Leicester, UK). After collection, these samples were centrifuged immediately at 3,500 revolutions per minute at 4C for 10 minutes. The plasma was then pipetted off and stored at 40C. Saliva samples to assess salivary cortisol were collected by asking participants to chew for one minute on a plain Salivette (Sarstedt, Leicester, UK), which was then stored at -20C until assay. Following the end of testing for a participant, their plasma samples were defrosted and those collected using lithium heparin syringes were assayed for cortisol, growth hormone and prolactin, while those collected using EDTA syringes were assayed for ACTH. Saliva samples were also defrosted and assayed for salivary cortisol. Plasma growth hormone, ACTH, prolactin and cortisol concentrations were measured using "Immulite," an automated immunoassay analyzer (DPC,Glyn Rhonwy, Llanberis, Gwynedd, LL55 4EL, UK; www.dpcweb.com). The analyzer was calibrated according to the manufacturer s instructions and the day to day performance of the assays monitored with commercial control sera of 2-3 different concentrations. Inter and intra-assay coefficients of variation were less than 10% for all assays. The analyser's plasma cortisol assay was specially modified and validated by us before being used to measure the salivary cortisol concentrations. Procedure We sent written information to people who contacted us and screened them for eligibility. We invited those who provided verbal consent to attend our unit for three mornings. We instructed participants not to take recreational drugs for one week before attending; not to drink alcohol for 24 hours beforehand; and not to drink more than one cup of tea or coffee, take painkillers, undertake strenuous physical activity or do anything psychologically stressful on the morning of each visit. Sessions began with a 30 minute adjustment period. At the start of this period we attempted to cannulate the participant. Given our desire to avoid causing discomfort to participants due to cannulation, a maximum of two cannulation attempts were made on each visit. Following cannulation in session one s adjustment period participants completed the various demographic and psychological questionnaires. During the adjustment period in session two, participants completed the CFF threshold assessment. At the end of each adjustment period, we asked participants to complete baseline visual analogue scale measures, we took blood and saliva samples, and we measured the participant s pulse. The exposure equipment was then attached and switched on for 50 minutes. Participants

6 The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users completed further visual analogue scale measures after 5, 15, 30, and 50 minutes. Additional blood samples were taken 30 and 50 minutes after the start of the exposure and additional saliva samples at 15, 30 and 50 minutes. If a participant requested that an exposure be terminated early, visual analogue scales were administered immediately. All participants completed a final set of visual analogue scales and provided a final set of blood samples 30 minutes after the end of each exposure. At this point we asked them to state whether they believed a signal had been present and their confidence about this (100 mm visual analogue scale from complete guess to 100% certain ). At least 24 hours after each session we contacted participants and asked them whether they had experienced any visual analogue scale symptoms in the 24 hours since exposure. We ascertained a score of 0 (no sensation) to 10 (worst possible sensation) for any symptoms that were reported, and we categorised participants scoring 5 or more as having experienced a definite symptom. All testing took place between September 2003 and June 2005 in two rooms within King s College London. The rooms, which were lit by two table lamps, were not shielded against outside electromagnetic fields. moderate that is, an effect size of 0.5. Our calculation showed that to detect this effect as significant at the 5% level and with 80% power we would need 60 participants in each group. In practice, although these assumptions turned out to be reasonable, the nature of our data required us to adopt a different analytical strategy from that originally planned. As such, this calculation should be taken as indicative only. Analyses To analyse symptom severity over time, we used generalised estimating equations 29. This approach was needed to accommodate the extremely positively skewed distribution of each response variable and to allow the inclusion of a suitable correlation structure for the repeated measures of each response. These models also allowed us to take into account differing lengths of exposure for participants who requested that an exposure be terminated early. The specific generalised estimating equations model fitted to each response used log(symptom severity+1) as the dependent variable, a gamma error distribution, and an exchangeable correlation structure. We used robust standard errors to judge the significance or otherwise of the explanatory variables included in the fitted models 29. Sample size calculation We based our sample size calculation on our ability to detect a change in headache severity within the sensitive group after 50 minutes of GSM exposure, using a two way analysis of variance (ANOVA) with one between participants factor (sensitive v control) and one within participants factor (GSM v continuous wave v sham). On the basis of previous studies in healthy and electrosensitive participants 4;28, this analysis assumed that control participants would report a mean headache severity of 10 units in all three experimental conditions whereas sensitive participants would report a mean severity of 11.7 in the sham and continuous wave conditions, with standard deviations of 26.8. In the absence of any pre-existing data, we assumed correlations of r=0.5 between conditions and that any effect of GSM in the sensitive group would be For the neuroendocrine assays, samples collected after a participant had requested that an exposure be terminated early were not included in the analyses. Distributions for the salivary cortisol, plasma cortisol, ACTH and prolactin variables were approximately normal. The distributions for growth hormone scores showed positive skew and were transformed using a ln(x) function prior to analysis. t-tests were used to compare the sensitive and control groups in terms of baseline neuroendocrine levels. For these analyses, we took the mean of all available baseline measures for any given participant as the dependent variable. Repeated measures ANOVA were used to assess the main effects of exposure (GSM vs CW vs Sham) on each hormone and to test for any interaction between exposure and duration (30min vs 50min vs 80min for plasma hormones, and 15min vs 30min vs 50min for salivary cortisol) or group (sensitive vs control). These

7 The effect of mobile phone use on symptoms and neuroendocrine function in normal and hypersensitive users analyses were then re-run to test for any main effects of duration or any interaction between duration and group, but this time collapsing the data across all three exposure conditions and including baseline values in the analyses. Collapsing the data across the conditions improved the power of these analyses by allowing participants with missing items of data for one or two of the exposure conditions to be included. For salivary cortisol we also calculated the area under the curve (AuC) using the trapezoidal method for the baseline, 15, 30 and 50 minute values. AuC provides an indication of the total amount of salivary cortisol produced during the course of each exposure session. The effects of group and exposure on AuC were then tested using ANOVA. Differences between groups in terms of CFF, pulse and psychological parameters were assessed using t-tests, χ 2 -tests and ANOVA where applicable. Given the previous evidence suggesting that there may be two subgroups within ES who exhibit different characteristics 9;10;30;31, for these analyses the sensitive group was subdivided into participants who reported having electrosensitivity (ES) and participants who reported problems with mobile phones only (MO). managerial background (χ2=5.6, p=0.02). Restricting the demographic comparisons to participants who completed all three testing sessions did not alter these results. For sensitive participants who completed the testing, the mean reported delay between beginning a call and onset of symptoms in everyday life was 6.5 (SD 6.5) minutes. For 48 people, symptoms usually resolved within two hours. All but one had been sensitive for at least a year (median 4 (interquartile range 2-5) years). Eighteen people reported that their sensitivity to mobile phones caused definite impairment or worse in at least one aspect of daily functioning, and 15 people reported having sought treatment for their condition. Thirteen people reported being sensitive to mobile phones at distances of one metre or more, and the same number reported having electrosensitivity. Sensitive participants reported headache-like symptoms in a mean of 70.4% of calls. The next most common symptoms were skin warmth or burning (43.8% of calls), difficulty concentrating (30.0%), and dizziness (20.8%). Very few control participants reported any symptoms in relation to mobile phone signals; the highest mean frequency was for skin warmth or burning (2.9%). Ethical approval The South London and Maudsley NHS Trust Research Ethics Committee granted ethical approval for this study. All participants provided informed written consent before being tested. Results We were contacted by 83 potential sensitive participants and 69 potential controls who met the inclusion criteria and provided verbal consent. Of these, 71 sensitive participants and 60 controls attended for their first testing session, and 60 in each group attended all three testing sessions and were included in our main generalised estimating equations analyses (see figure 1). Table 1 shows demographic data for those participants who attended at least one session; the only substantive difference between the groups was a significantly higher proportion of sensitive participants from a professional or

8 The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users Self-assessed for eligibility (n = unknown) CONTROL GROUP Randomised (n=69) SENSITIVE GROUP Randomised (n=83) Withdrew (n=9) Unable to attend appointments (n=7) Withdrew (n=12) Unable to attend appointments (n=7) Withdrew consent (n=1) Withdrew consent (n=1) Other reason (n=1) Other reason (n=4) Completed baseline questionnaires (n=60) Completed baseline questionnaires (n=71) Excluded (n=2) Severe symptoms at baseline (n=2) Took part in first exposure session (n=60) Took part in first exposure session (n=69) Withdrew (n=6) Severe symptoms provocation (n=3) during Reason unknown (n=3) Took part in second exposure session (n=60) Took part in second exposure session (n=63) Withdrew (n=3) Severe symptoms provocation (n=2) during Took part it third exposure session (n=60) Took part in third exposure session (n=60) Reason unknown (n=1) Figure 1: Study Flow Diagram

The effect of mobile phone use on symptoms and neuroendocrine function in normal and hypersensitive users Effects of exposure on symptom reporting Table 2 shows the results of fitting generalised estimating equation models to each response variable. The group time interaction term was not needed in any model, so it does not appear in this table. Fitted models for all response variables showed highly significant effects for time (both linear and quadratic effects) and for baseline severity. We found no convincing evidence of an effect of condition or a condition group effect for any of the symptoms. For headache, burning sensations, skin sensations, and eye pain we found evidence of a main group effect sensitive participants reported greater severity. In terms of the original visual analogue scale units, this group effect for headache severity equated to an increase of 1.0 (95% confidence interval 0.4 to 2.0) unit. Figure 2 shows the median headache severity by group for each exposure condition. We also analysed the number of severe reactions seen in each condition, with a severe reaction defined as a participant requesting that an exposure be terminated early or withdrawing from the study entirely after an exposure. Twenty six such reactions occurred in the sensitive group (9 withdrawals; 17 early terminations), and none occurred in the control group. These reactions were equally distributed between GSM (n=7), continuous wave (n=10), and sham (n=9) conditions (χ2=0.54, P=0.76). Excluding data relating to the four participants whose reasons for withdrawal were not explicitly stated to us (see figure 1) did not affect these results (GSM 5, continuous wave 9, sham 8; χ2=1.2, P=0.55). We obtained next day follow-up results for all three sessions for 41 control participants and 49 sensitive participants. Cochran s Q tests identified no significant differences in the number reporting at least one definite symptom after GSM, continuous wave, or sham exposures in either the control group (GSM 0/41, continuous wave 2/41, sham 4/41; Q=4.0, P=0.14) or the sensitive group (GSM 5/49, continuous wave 8/49, sham 4/49; Q=2.0, P=0.37). Table 3 shows participants assessments of whether a signal was present during provocation. The proportion who believed a signal was present during exposure to GSM (60% of sensitive participants, 58% of controls) was slightly less than for the sham exposure (63% of sensitive participants, 68% of controls). Self reported confidence for these judgments did not differ greatly (table 2). Neuroendocrine results Mean baseline plasma neuroendocrine levels were available for 36 or 37 sensitive participants and 40 or 41 control participants, depending on the hormone. Mean baseline salivary cortisol levels were available for 69 sensitive participants and 59 controls. Table 4 provides the mean levels in each instance. No significant differences were observed between sensitive and control groups with regard to any baseline level (t<0.9, p>0.4). Due to missing data, the repeated measures ANOVAs testing for the effects of exposure on the plasma hormones were based on the results for 10 sensitive participants and between 22 and 24 control participants. The salivary cortisol analyses were based on data for 43 sensitive and 46 control participants. These analyses identified a significant main effect of exposure for plasma cortisol (F(2,62)=5.4, p=0.007). Post hoc testing using least significant difference tests suggested that this effect was due to significantly lower cortisol levels in the continuous wave condition (mean=225) than in the GSM (mean=258, p=0.005) or sham conditions (mean=270, p=0.008). No other main effects or interaction terms involving the exposure variable were found to be significant (see table 4). The ANOVA for salivary cortisol AuC also failed to reveal any significant effects of exposure or any significant exposure by group interaction (table 5). The second set of analyses for the plasma hormones using data collapsed across exposure conditions included data for 39 or 40 control participants and 33 or 34 sensitive participants. The same analyses for salivary cortisol used data for 58 control participants and 65 sensitive participants. Significant main effects of duration were noted for all variables except growth hormone (see table 5), relating to the expected reduction in neuroendocrine levels over time due to the body s natural circadian rhythms. No significant duration by group interactions were apparent for any hormone, however. 9

The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users Table 1 Demographics of participants Variable Control group (n=60) Sensitive group (n=71) 33.5 (10.2) 37.1 (13.2) 0.09 Sex (male:female) 27:33 31:40 0.88 Ethnicity (white:other) 45:15 56:15 0.60 Marital status (single:married/cohabiting:divorced/separated) 39:19:2 38:30:3 0.41 Employment status (in work:unemployed:housewife/husband:student) 30:10:2:18 42:9:3:17 0.71 Socioeconomic status (professional, managerial or intermediate:semiroutine, routine, or student) 31:29 51:20 0.02 Educational level (secondary education or lower:higher education) 18:42 26:45 0.42 Weekly frequency of mobile phone use (<4 times:4-12 times:13+ times)* 8:25:27 17:22:32 0.23 Typical length of call (<5 minutes:5-15 minutes:16+ minutes)* 32:22:6 44:18:9 0.37 *Former mobile users (n=10) based their answers on the last time they regularly used one. P value for differences between groups 10

The effect of mobile phone use on symptoms and neuroendocrine function in normal and hypersensitive users Table 2 Estimated regression coefficients (robust standard error) derived from generalised estimating equation models used to assess effects of group, exposure, duration of exposure, and baseline score on symptom severity Symptom Baseline severity Duration (linear function) Duration (quadratic function) Sensitive v control Sham v GSM CW v GSM Group x (sham v GSM) Group x (CW v GSM) Headache 0.04 (0.008) 0.04 (0.004) 0.0004 (0.0004) 0.7 (0.2) 0.07 (0.1) 0.02 (0.1) 0.08 (0.2) 0.2 (0.2) Nausea 0.02 (0.05) 0.006 (0.001) 0.0002 (0.00004) 0.2 (0.3) 0.06 (0.1) 0.2 (0.1) 0.1 (0.4) 0.3 (0.3) Fatigue 0.04 (0.005) 0.01 (0.002) 0.0003 (0.00005) 0.2 (0.2) 0.08 (0.1) 0.2 (0.1) 0.09 (0.2) 0.2 (0.2) Dizziness 0.05 (0.02) 0.007 (0.001) 0.0003 (0.00005) 0.3 (0.2) 0.2 (0.1) 0.09 (0.1) 0.2 (0.3) 0.01 (0.3) Skin 0.05 (0.01) 0.004 (0.001) 0.0003 (0.00005) 0.5 (0.2) 0.09 (0.1) 0.1 (0.1) 0.1 (0.2) 0.3 (0.2) Burning 0.03 (0.00005) 0.007 (0.001) 0.0007 (0.0004) 0.4 (0.2) 0.05 (0.1) 0.09 (0.09) 0.2 (0.2) 0.4 (0.2) Eye pain 0.05 (0.008) 0.007 (0.001) 0.0003 (0.00004) 0.6 (0.2) 0.04 (0.1) 0.2 (0.1) 0.3 (0.2) 0.08 (0.2) CW=continuous wave; GSM=global system for mobile communication. In each model, the dependent variable used was log(symptom severity +1). 11

The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users 30 25 Severity (0-100) 20 15 10 Sensitive (GSM) Sensitive (CW) Sensitive (Sham) Control (GSM) Control (CW) Control (Sham) 5 0 0 5 15 30 50 80 Time (Minutes) Figure 2 Median headache severity (error bars show interquartile range) during provocation with global system for mobile communication (GSM), continuous wave (CW), and sham exposures for sensitive and control participants. For clarity, the graph does not include data relating to exposures that were terminated early, although these data were included in analyses 12

The effect of mobile phone use on symptoms and neuroendocrine function in normal and hypersensitive users Table 3 Number of participants who believed a signal was present for each experimental condition and mean (SD) confidence (0-100) reported by participants for these signal present assessments Exposure Sensitive participants Completed all three Completed at least one Controls exposures exposure N Confidence N Confidence N Confidence GSM 35/60 36.8 (28.5) 36/60 58.6 (30.8) 41/65 61.2 (31.0) CW 42/60 39.7 (33.0) 41/60 57.7 (27.8) 45/64 57.8 (28.9) Sham 41/60 43.9 (31.9) 38/60 64.4 (31.7) 39/63 64.0 (31.3) CW=continuous wave; GSM=global system for mobile communication. Table 4: Mean baseline neuroendocrine levels (standard deviation) [sample size], collapsed across the three experimental conditions Growth Hormone (ln miu/l) ACTH (pmoles/l) Prolactin (miu/l) Plasma cortisol (nmol/l) Salivary cortisol (nmol/l) Sensitive Participants Control Participants t-test Results -1.0 (1.5) [n=37] -0.9 (1.6) [n=40] t=0.2, df=75, p=0.9 4.1 (1.7) [n=37] 3.7 (1.9) [n=41] t=0.9, df=76, p=0.4 204.0 (119.8) [n=36] 209.2 (117.1) [n=41] t=0.2, df=75, p=0.8 323.7 (158.9) [n=37] 319.8 (131.6) [n=41] t=0.1, df=76, p=0.9 7.7 (3.9) [n=69] 7.8 (3.4) [n=59] t=0.1, df=126, p=0.9 13

The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users Psychological variables Of the 71 sensitive participants who attended at least one testing session, 19 reported having electrosensitivity or sensitivity to electromagnetic fields and were included in our ES group: the other 52 sensitive participants constituted the MO group. No differences were observed with regard to the perceived usefulness of mobile phones (χ 2 =7.2, p=0.13) between these groups and the control group. However, a significant difference was observed in terms of the reasons for using a mobile phone (χ 2 =12.3, p=0.02), with MO (7 / 52; 13%) and ES (4 / the ES group than the MO group with respect to toxic interventions (p<0.001) and radiation (p<0.001), and greater concern in the MO group than the control group with respect to radiation alone (p<0.001). 19; 21%) participants being more likely to use or have used their mobile phone mainly for work compared to controls (2 / 60; 3%). In terms of general health status, there were significant differences between the three groups for every SF-36 scale (F(2,128)>3.8, p<0.05) except changes in health (F(2,128=0.02, p=0.98). These differences were all due to the worse health status in the ES group compared to either the MO or control groups (see table 6, p<0.05). The only exception was for the mental health scale, for which no significant difference was observed between the ES and the control groups (p=0.06). Group scores on the GHQ-12, PHQ-9 and MHW scales are also shown in table 6. There were no significant group differences in the percentages of participants classified as psychiatric cases using the GHQ-12 (χ 2 =2.9, p=0.24). However, there was a significant group difference in PHQ-9 depression scores (F(2,126)=7.5, p=0.001), with ES participants having significantly greater depression than control (p=0.002) or MO participants (p=0.001). There were also significant group differences in terms of the MHW toxic intervention (F(2, 128)=7.4, p<0.001), tainted food (F(2,128)=3.5, p=0.03) and radiation (F(2,128)=33.1, p<0.001) subscales, though not for the environmental pollution scale (F(2,128)=1.4, p=0.26). These differences reflected greater concern in ES participants than control participants with respect to toxic interventions (p<0.001), tainted food (p=0.03) and radiation (p<0.001), greater concern in 14

The effect of mobile phone use on symptoms and neuroendocrine function in normal and hypersensitive users Table 5: ANOVA results for effects of exposure, group and duration on neuroendocrine hormones Hormone Main effect of exposure Exposure x group interaction Exposure x duration interaction Main effect of duration b Duration x group interaction b Plasma Growth Hormone F(2,60)=0.7, p=0.54 a F(2,60)=0.1, p=0.94 a F(4,120)=0.6, p=0.68 a F(3,213)=1.86, p=0.14 F(3,213)=0.8, p=0.52 Plasma ACTH F(2,62)=2.0, p=0.14 a F(2,62)=1.5, p=0.24 a F(4,124)=0.7, p=0.63 a F(3,216)=14.5, p<0.001 F(3,216)=0.3, p=0.85 Plasma Prolactin F(2,64)=0.6, p=0.56 a F(2,64)=0.2, p=0.83 a F(4,128)=1.9, p=0.12 a F(3,210)=52.1, p<0.001 F(3,210)=1.5, p=0.21 Plasma Cortisol F(2,62)=5.4, p=0.007 a F(2,62)=2.0, p=0.15 a F(4, 124)=0.2, p=0.96 a F(3,213)=48.6, p<0.001 F(3,213)=0.3, p=0.80 Salivary Cortisol F(2,172)=0.9, p=0.39 a F(2,172)=0.5, p=0.63 a F(4,344)=1.5, p=0.2 a F(3,363)=56.6, p<0.001 F(3,363)=0.3, p=0.80 Salivary Cortisol Area Under the Curve F(2,170)=2.1, p=0.12 c F(2,170)=0.3, p=0.75 c Not applicable Not applicable Not applicable a Analyses including exposure, group and duration as independent variables. b Analyses including duration and group only as independent variables. c Analyses including exposure and group only as independent variables. 15

The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users Table 6: General health and psychological status of participants in control, mobiles only (MO) and electrosensitive (ES) groups. All values are mean (SD) unless stated Variable (range of scores: meaning of higher score) Control Group (n=60) MO Group (n=52) ES Group (n=19) SF-36 physical functioning (0-100: better health) 95.3 (8.6) 93.4 (16.0) 78.7 (28.9) a,b SF-36 social functioning (0-100: better health) 93.3 (10.6) 93.1 (11.5) 69.3 (31.3) a,b SF-36 role limitations physical (0-100: better health) SF-36 role limitations emotional (0-100: better health) 93.3 (20.5) 95.7 (13.8) 48.7 (43.7) a,b 85.0 (27.0) 86.5 (25.8) 61.4 (42.0) a,b SF-36 mental health (0-100: better health) 76.8 (13.3) 78.7 (12.9) 68.2 (19.8) b SF-36 energy / fatigue (0-100: better health) 69.7 (16.1) 67.5 (16.4) 52.9 (27.4) a,b SF-36 pain (0-100: better health) 86.1 (16.3) 85.5 (14.3) 68.7 (27.6) a,b SF-36 general health perceptions (0-100: better health) 75.6 (18.4) 78.0 (17.6) 56.3 (24.3) a,b SF-36 change in health (0-100: better health) 57.5 (18.0) 57.7 (18.9) 56.6 (28.7) Number of GHQ-12 cases 10 / 60 (17%) 4 / 52 (8%) 4 / 19 (21%) Total depression score on PHQ (0-27: more depression) Modern health worries: toxic interventions (1-5: greater concern) Modern health worries: environmental pollution (1-5: greater concern) Modern health worries: tainted food (1-5: greater concern) Modern health worries: radiation (1-5: greater concern) 2.2 (2.8) 1.9 (2.4) [n=50] 4.8 (3.8) a,b 2.2 (1.0) 2.3 (0.8) 3.2 (1.1) a,b 3.0 (1.0) 2.9 (0.9) 3.3 (1.0) 2.8 (1.1) 2.9 (1.2) 3.6 (1.2) a 2.0 (1.0) 2.7 (0.9) a 4.0 (0.8) a,b a Significantly different to control p<0.05; b significantly different to mobiles only p<0.05 16

The effect of mobile phone use on symptoms and neuroendocrine function in normal and hypersensitive users Table 7: Mean critical flicker fusion thresholds and resting pulse rates for participants in control, mobiles only (MO) and electrosensitive (ES) groups. All values are mean (SD). Variable Critical flicker fusion threshold (Hz) Resting pulse rate Control Group 34.5 (3.1) [n=60] 68.4 (7.9) [n=60] Mobiles Only Group ES Group 36.7 (4.1) 32.8 [n=48] a (9.2) [n=15] b 65.9 (7.5) [n=49] 71.9 (7.1) [n=14] b a Significantly different to control p<0.05; b significantly different to mobiles only participants participants p<0.05 CFF and pulse results Mean pulse rates were recorded for all 60 control participants, 49 MO participants and 14 ES participants. CFF thresholds were recorded for 60 control participants, 48 MO participants and 15 ES participants. Table 7 shows the mean CFF thresholds and resting pulse rates for the three groups. A significant difference was observed for CFF (F(2, 120)=5.3, p=0.006) with MO participants having significantly higher thresholds than ES participants (p=0.005) or control participants (p=0.02). A significant difference was also observed for pulse (F(2, 120)=3.8, p=0.03) with ES participants having significantly faster pulses than MO participants (p=0.01). 5 Analysis of objectives met We successfully met our objective of testing the effects of GSM exposure on symptom reporting in a sample of apparently sensitive individuals and a sample of non-sensitive control participants. We also successfully tested baseline differences between these groups in terms of psychological, neuroendocrine and other physiological parameters. With regard to testing the effects of GSM exposure on neuroendocrine function, difficulties with cannulation limited the number of participants we could include in these analyses. For one neuroendocrine variable, cortisol, we were able to overcome this by analysing saliva samples. For growth hormone, prolactin and ACTH we were unable to do this: larger sample sizes would have been preferable for these variables. 6 Interpretation We found no evidence to indicate that self reported sensitivity to 900 MHz GSM mobile phone signals has a biological basis. Nor did we find any evidence to suggest that the pulsing nature of GSM contributes to these symptoms. These findings are in agreement with the large majority of previous blind or double blind provocation studies for electromagnetic sensitivity, which have found no differences in the severity of symptoms elicited by active or sham exposure to electromagnetic fields 11;30;32-34. Did some inadequacy exist in our methods that might account for these negative findings? If it did, we are unaware of it. The exposure represented a relatively worst case scenario mobile phone call, using a high specific absorption rate and lasting almost eight times longer than the mean call length usually needed to trigger symptoms in our sensitive sample. Interference from participants reactions to extraneous electromagnetic fields is also unlikely: after 30 minutes adjusting to our offices, only two participants reported baseline symptoms that might have masked any effects of exposure, and both were excluded. Finally, as we were able to detect changes in symptom severity over time as highly significant, the sensitivity of our visual analogue scales and our statistical techniques do not seem to have had any shortcomings. That symptom severity did increase during exposure is interesting. These symptoms were not trivial. Indeed, for some they were so severe that exposures had to be stopped early or the participants withdrew from the study. The confidence that sensitive participants had in their ability to discriminate active from sham signals also suggests that they experienced reactions similar to those 17

The Effect of Mobile Phone Use on Symptoms and Neuroendocrine Function in Normal and Hypersensitive Users encountered in real life, a finding also reported in previous provocation studies 11. That apparently realistic symptoms can be induced in provocation experiments, despite no differences being observed between active and sham conditions, suggests that the acute symptoms reported by sensitive people in everyday life may be the result of a nocebo phenomenon. Such phenomena have previously been observed in relation to a wide range of stimuli 35, including headaches induced by providing misleading information about the presence of electrical fields 36. The mechanisms governing nocebo effects need further study but seem to include conscious expectation of symptoms and the presence of negative affect 37;38, factors that are likely to be present whenever people who perceive themselves to be sensitive to mobile phones have to make use of the technology. electromagnetic fields that causes people to become concerned about the health effects of modern life and not vice versa, good quality prospective evidence does exist which demonstrates that such fears can indeed precede and predict the development of nonspecific symptoms 14. Finally, although all other mobile phone usage variables such as typical frequency and length of calls were the same between our groups, participants who believed themselves to be sensitive to mobile phones had a greater tendency to use or to have used mobile phones predominantly for work purposes. This association suggests that an initial association between mobile phone use and occupational stress may well be one factor contributing to the original association between mobile phone use and the onset of non-specific symptoms. Data from the psychological questionnaires that we used also supported a psychological interpretation of ES. Firstly, it is important to note that no differences were observed between the groups on the GHQ-12 measure of psychiatric caseness. This is a well-replicated finding within the ES literature 39-41 and suggests that the condition is not explicable solely in terms of formal psychiatric illness. However, ES participants did have significantly higher levels of depression compared to the other two groups as measured on the PHQ-9, as well as lower mental health scores on the SF-36 scale in comparison to the MO group. Although it is difficult to determine the direction of causality implied by this association, these findings are consistent with higher distress levels being a risk factor for the development of ES. Similarly, the high levels of concern that ES and MO participants had about topics related to radiation and electromagnetic devices and also, in the case of ES participants, about food contamination and toxic interventions is consistent with the theory that pre-existing concerns about the health effects of modernity are risk factors for the development of physical symptoms following exposure to worrying stimuli. While it might be argued that in this case it is the development of a sensitivity to 18 With regard to the CFF measure, while we did find some evidence of hyper-responsiveness to external stimuli in our sensitive participants, this was restricted to the mobiles only subgroup. To our knowledge, only one other study has tested CFF threshold in this specific population but without finding any differences in threshold compared to a healthy control group 30. Other studies including typically more severely affected ES participants have identified increased CFF thresholds 19;31. Why our own ES subgroup did not display this trend is therefore uncertain. Meanwhile the increased resting pulses observed for our ES group, but not for our MO group, does correspond more closely with previous findings showing that ES participants, but not those who are sensitive only to mobile phones, have increased heart rates compared to healthy controls 30;31;42;43. No evidence was found to suggest that individuals who report sensitivity to mobile phone signals have altered neuroendocrine function at rest compared to a non-sensitive control group. Similarly, we found no evidence that self-reported sensitivity to mobile phone signals is associated with altered neuroendocrine function as a result of exposure to either GSM or CW radiofrequency