SAR distributions from mobile phones
Identification of phone classes
Analysis showed some clustering of phone types related to the combination of phone position and shape.31
The position in which phones are held varies between individuals and, for the same individual, during use and with different phones. Classifications of SAR distributions using phone position therefore cannot be used in epidemiological studies.
Frequency is known to be inversely related to penetration depth of radiation.34
Phones were therefore classified by band as follows: 800–900, 1500 and 1800–1900 MHz. No measurement and limited simulations were available on 450 MHz phones, and only 14 subjects in the five countries reported ever using them; they were therefore included with 800–900 MHz phones.
Average spatial SAR distribution for each phone class
From each phone measured, 3D SAR distributions were estimated in the Gridmaster and, for each phone class, the average SAR distribution was derived. shows average SAR distributions in the brain at 800–900 and 1800 MHz.
Surface and axial views of 1 cm3 SAR (W/kg) distributions in the GRIDMASTER brain at 800-900 MHz and 1800 MHz respectively, for a phone held on the right side (the colour scale used in all views is the same)
Most of the SAR in the brain (97%–99% depending on frequency band) appeared to be absorbed in the hemisphere on the side where the phone is used, mainly (50%–60%) in the temporal lobe. The average relative SAR decreased rapidly with depth, particularly at higher frequencies (). The SAR distribution appeared similar across phone models, between older and newer phones, and between phones with different antenna types and positions.27
Output power levels
Influence of APC
In GSM phones, before APC was introduced in the early 1990s, phones worked to maximum power. The SMP study conducted between 2001 and 2005 with four GSM models, and over 63 000 calls made by 516 subjects in 12 countries, showed that, on average, APC reduced power to around 50% of the maximum power levels in both 900 and 1800 MHz frequency bands.28
Results from this and other studies35–40
are broadly consistent in the period 1999–2005 for GSM networks.
For any historical dose index, however, extrapolation in time is necessary. GSM networks were introduced in early to mid-1990s in Europe and more recently in North America. Earlier, first-generation analogue networks had no or very limited APC capability (K Hansson Mild, personal communication, 2007), so phones were nearly always operating at maximum power. Since the start of GSM systems, improved network coverage may have reduced power levels, but increased indoor use of phones, more frequent use of handovers and higher traffic density may have increased them. Thus, there are uncertainties about backward extrapolation and we decided, for GSM and other second-generation networks, to assume a factor of 1 for APC up to its introduction and of 0.5 afterwards.
For Code Division Multiple Access (CDMA) networks, information about power control is scant, except that systems are very efficient. In a limited number of measurements in Canada, the average power was 13–15 times less than peak; an APC factor of 0.067 was therefore chosen for CDMA networks, compatible with recently published data comparing CDMA and GSM exposure levels.41
Influence of network operator
The SMP study indicated important differences between operators.28
Within a frequency band, operator was the most important factor explaining differences in average power levels between users. The study, however, was only a snapshot in time and place, and network optimisation factors vary considerably over time and place. SMP volunteers moreover did not form a geographically representative sample of the study population in most countries, and no information was available for some networks in study regions. For these reasons, though this adds uncertainty, the dose algorithm could not include a factor relating to operator.
Influence of phone use environment Urban/rural use
A small (13%–14%) difference in average power levels was found in the SMP study between subjects using phones mainly in rural versus urban environments.28
This was mainly driven by Sweden, where volunteer selection covered sparsely populated areas. Few rural users were included in SMP studies elsewhere. Base-station data from Sweden36
showed greater urban/rural differences (factor of 1.6–1.9 depending on time of day), but data were for 1 week and one operator only.
An urban/rural factor was therefore included in the dose algorithm only where the study covered a large proportion of the country (Israel and New Zealand): based on SMP and base-station studies, average output power was estimated to be 30% greater in rural users than in urban users for second-generation phones.
Use while moving or stationary
No difference in average power levels was found in the SMP study between users using phones mainly while moving in a vehicle and those reporting mainly stationary use.28
This was initially surprising because networks must hand-over a moving phone from one base-station to another, and with each handover, the power returns to a high level before APC lowers output to optimum level. Handovers, however, occur frequently in stationary situations because networks need to distribute traffic equitably between base-stations, and stationary phone calls may require extra power in built-up areas because of shielding, thus explaining the lack of an effect. The SMP study, moreover, may not have been sufficiently sensitive to detect differences: information available concerned average use (the only information collectable in epidemiological studies) rather than use circumstance of each individual call. Based on the results of the SMP study, it was decided not to differentiate moving and stationary use in the dose algorithm.
A difference was found between output power inside and outside buildings in a small study.37
A question was added to the SMP study, but no difference was found.28
As no information is available in the Interphone questionnaire on use inside buildings, this variable was not included in the algorithm. This decision is unlikely to bias dose estimates because quality of indoor communications improved over time, leading to gradual increase of indoor use from a low level in the earliest years. However, this is a source of uncertainty.
Influence of DTX
Use of DTX could not be measured with the SMPs. Data published in 200035
suggest that DTX reduced average power by about 30% once it was enabled in a network. As little information was available from network operators about dates of DTX enablement and because operators which provided information reported it occurred early in GSM networks, it was assumed all GSM networks in the study were DTX enabled from 1994 to 1995.
Use of phone-specific SAR values
The dosimetry database of SARmax for phones in use before and during the Interphone Study included 1233 values. Analysis showed large variation in SAR, with some SARmax measured below 0.01 W/kg (current limit of detection) possibly due to measurement errors. Analyses by time period (online appendix figure 3) provided no statistical evidence of a trend over time for a given communication system. The methods and standards used to measure SARmax have evolved over time, so SARmax values are not directly comparable. Analyses (not shown) of SARmax values for more recent models in the database, restricted to measurements in reputable laboratories, showed differences as high as a factor 3 for specific phone models.
For these reasons and because efficiency of phones may partially compensate or exaggerate differences, phone-specific SARmax values were not used in the algorithm: instead, the median of available SARmax measurements was assigned to phones in a class. For this, SARmax measured with averaging volumes of 1 or 10 g were converted to the SAR of the 1 cm3 Gridmaster cell with the highest average SAR for the phone class. Conversion factors were derived from the database of SARmax (1 g) and SARmax (10 g) in the French and Japanese measurements. Single ratios were used for all frequency bands for SARmax (1 g) and for SARmax (10 g) as analyses showed no significant difference across bands (not shown).
Information on phone use from study subjects
Information on duration and number of calls was reported by Interphone Study subjects by period of use for each phone they had used. Information on laterality of use, call environment and use of hands-free devices was also collected. For time periods for which a subject reported use of hands-free devices, amount of use was reduced by 100%, 75%, 50% or 25%, respectively, if devices were used always or almost always, more than half, about half or less than half of the time. If the subject reported a preferred side of use, 90% of use was assigned to that side of the head and 10% to the other. Otherwise, 50% was assigned to each side of the head. Sensitivity analyses were also conducted in which reported laterality was not used at all, and 50% of use was assigned to each side for all subjects.
Construction of the dose algorithm and application to Interphone Study subjects
Based on the above, an algorithm was developed for RF dose from mobile phones in terms of cumulative specific energy (CSE) (in joules per kilogram) absorbed at a given location in the brain (l
), for a given frequency band (f
) (450, 800–900, 1500, 1800–1900 MHz) and a specific communication system (s
) (GSM, CDMA, etc). Location can be anywhere within the Gridmaster cells, and the quantity estimated is the average within the volume of the relevant cells. The algorithm estimates the CSE (CSEl,f,s
) absorbed at that location for a given study subject as follows:
i denotes month of use within a period defined by operator and phone,
X l,c proportion of the SARmax received at location l, for a phone in class c
Averagec average over all phones in class c of the argument in parentheses
SARx SARmax (1 or 10 g) for phone x in class c from the SARmax measurements of phones.
Conv1/10g conversion factor from SARmax measured for averaging volumes of 1 or 10 g to SARmax in the Gridmaster cell with highest average SAR.
Ti reported average call time in month i from answers to the questionnaire
Hi modifier for reported use of hands-free devices in month i from the questionnaire (see above)
Yi,u/r effect of exposure circumstances modifying output power (only urban/rural use—see above)
Pi,f,s o proportion of traffic in frequency f and communication system s for operator o at time i based on information from network operators questionnaires. Proportion is 1 if the phone operated in a single band and system. When dual bands/systems were enabled, the relative proportions of traffic were taken to increase linearly from 1.0/0.0 at time of introduction of the second frequency band/system to p/(1−p)—the proportions stated at time of response by the operator—and to remain constant thereafter.
Oi,f,s,o effect of other modifiers as appropriate (APC and DTX, see above).
The total CSE absorbed at a given location (l
for a particular study subject was obtained by summing over all combinations of communication systems and frequencies he/she used. Risk analyses can be conducted both using TCSEl
as the explanatory variable (assuming that carrier frequencies and their modulation by different communication systems do not modify possible biological effects) and investigating separately the effect of CSE in different frequencies and communication systems.42
Based on this algorithm, other metrics can be developed including the time-weighted average of the absorbed power (TWA-SAR), obtained by dividing TCSE by total duration of calls, as well as TCSE and TWA-SAR in different time windows before case diagnosis to explore the hypotheses that risk, if any, may be related to dose rate rather than cumulative dose and that doses in different time windows may have different effects.
Distribution of doses
and show the resulting distribution of CSE at tumour location by communication system and frequency among glioma study subjects from the five countries (results for meningioma study subjects are similar—not shown). Substantial differences were seen between technologies, with CSE being highest on average for Advanced Mobile Phone System (AMPS) and lowest for GSM-1800 and CDMA systems.
Results of dose estimation for glioma study subjects in five Interphone countries: estimated TCSE absorbed at the estimated centre of the tumour in joules per kilogram, by communication system, up to 1 year before reference date
Figure 3 Distribution of estimated CSE (in J/kg) at the origin of the tumour by mobile phone communication system and frequency band. (Glioma study subjects 5 Interphone countries. For controls, location of tumour is taken to be the location of the tumour of the (more ...)
The distribution of TCSE at the tumour location by cumulative call time, cumulative number of calls and time since start of use of mobile phones is shown in overall (see online appendix table 2 for specific frequencies and communications systems).
Table 2 Distribution of deciles of TCSE absorbed at the location of the tumour (in joules per kilogram) by cumulative call time, cumulative number of calls and time since start of mobile phone use. All frequencies and communication systems. (Glioma study subjects (more ...)
While substantial agreement (34%) exists between categorisation of subjects by TCSE and cumulative call time (κ 0.68), there is non-negligible misclassification: 37%, 21% and 8% of subjects, respectively, have 1, 2 or more deciles of difference. Because higher frequencies penetrate less in the brain, the agreement is substantially lower for GSM-1800 (with 5% having complete agreement and 86% more than 2 deciles difference) than for lower frequencies (online appendix table 2). Results are similar for cumulative number of calls though κ statistics, and agreement are lower. TCSE is also related to time since start of use ( and online appendix table 2), with most subjects with high TCSE being long-term users and most short-term users having low TCSE.
In analysis of variance including the main factors in the dose algorithm, only cumulative call time and tumour location were statistically significant predictors (predicting, respectively, 43% and 13% of the variability) for TCSE overall. Results were similar for analogue systems and 800–900 MHz frequency bands. For CDMA 1900, APC was statistically significantly associated with CSE (p=0.015).