Report of Partial Findings from the National Toxicology Program Carcinogenesis Studies of Cell Phone Radiofrequency Radiation in Hsd: Sprague Dawley® Sd Rats (Whole Body Exposure)

The U.S. National Toxicology Program (NTP) has carried out extensive rodent toxicology and carcinogenesis studies of radiofrequency radiation (RFR) at frequencies and modulations used in the U.S. telecommunications industry. This report presents partial findings from these studies. The occurrences of two tumor types in male Harlan Sprague Dawley rats exposed to RFR, malignant gliomas in the brain and schwannomas of the heart, were considered of particular interest and are the subject of this report. The findings in this report were reviewed by expert peer reviewers selected by the NTP and National Institutes of Health (NIH). These reviews and responses to comments are included as appendices to this report, and revisions to the current document have incorporated and addressed these comments. When the studies are completed, they will undergo additional peer review before publication in full as part of the NTP's Toxicology and Carcinogenesis Technical Reports Series. No portion of this work has been submitted for publication in a scientific journal. Supplemental information in the form of four additional manuscripts has or will soon be submitted for publication. These manuscripts describe in detail the designs and performance of the RFR exposure system, the dosimetry of RFR exposures in rats and mice, the results to a series of pilot studies establishing the ability of the animals to thermoregulate during RFR exposures, and studies of DNA damage. (1) Capstick M, Kuster N, Kuhn S, Berdinas-Torres V, Wilson P, Ladbury J, Koepke G, McCormick D, Gauger J, and Melnick R. A radio frequency radiation reverberation chamber exposure system for rodents; (2) Yijian G, Capstick M, McCormick D, Gauger J, Horn T, Wilson P, Melnick RL, and Kuster N. Life time dosimetric assessment for mice and rats exposed to cell phone radiation; (3) Wyde ME, Horn TL, Capstick M, Ladbury J, Koepke G, Wilson P, Stout MD, Kuster N, Melnick R, Bucher JR, and McCormick D. Pilot studies of the National Toxicology Program’s cell phone radiofrequency radiation reverberation chamber exposure system; (4) Smith-Roe SL, Wyde ME, Stout MD, Winters J, Hobbs CA, Shepard KG, Green A, Kissling GE, Tice RR, Bucher JR, and Witt KL. Evaluation of the genotoxicity of cell phone radiofrequency radiation in male and female rats and mice following subchronic exposure. SUMMARY The purpose of this communication is to report partial findings from a series of radiofrequency radiation (RFR) cancer studies in rats performed under the auspices of the U.S. National Toxicology Program (NTP).1 This report contains peer-reviewed, neoplastic and hyperplastic findings only in the brain and heart of Hsd:Sprague Dawley® SD® (HSD) rats exposed to RFR starting in utero and continuing throughout their lifetimes. These studies found low incidences of malignant gliomas in the brain and schwannomas in the heart of male rats exposed to RFR of the two types [Code Division Multiple Access (CDMA) and Global System for Mobile Communications (GSM)] currently used in U.S. wireless networks. Potentially preneoplastic lesions were also observed in the brain and heart of male rats exposed to RFR. The review of partial study data in this report has been prompted by several factors. Given the widespread global usage of mobile communications among users of all ages, even a very small increase in the incidence of disease resulting from exposure to RFR could have broad implications for public health. There is a high level of public and media interest regarding the safety of cell phone RFR and the specific results of these NTP studies. Lastly, the tumors in the brain and heart observed at low incidence in male rats exposed to GSM-and CDMA-modulated cell phone RFR in this study are of a type similar to tumors observed in some epidemiology studies of cell phone use. These findings appear to support the International Agency for Research on Cancer (IARC) conclusions regarding the possible carcinogenic potential of RFR.2 It is important to note that this document reviews only the findings from the brain and heart and is not a complete report of all findings from the NTP’s studies. Additional data from these studies in Hsd:Sprague Dawley® SD® (Harlan) rats and similar studies conducted in B6C3F1/N mice are currently under evaluation and will be reported together with the current findings in two forthcoming NTP Technical Reports.

cell phone radiofrequency radiation reverberation chamber exposure system; (4) Smith-Roe SL, 1 Wyde ME, Stout MD, Winters J, Hobbs CA, Shepard KG, Green A, Kissling GE,Tice RR,2 Bucher JR, and Witt KL. Evaluation of the genotoxicity of cell phone radiofrequency radiation in 3 male and female rats and mice following subchronic exposure. Lastly, the tumors in the brain and heart observed at low incidence in male rats exposed to GSM-1 and CDMA-modulated cell phone RFR in this study are of a type similar to tumors observed in 2 some epidemiology studies of cell phone use. These findings appear to support the International It is important to note that this document reviews only the findings from the brain and heart and 7 is not a complete report of all findings from the NTP's studies. Additional data from these 8 studies in Hsd:Sprague Dawley ® SD ® (Harlan) rats and similar studies conducted in B6C3F 1 /N 9 mice are currently under evaluation and will be reported together with the current findings in two 10 forthcoming NTP Technical Reports. communication devices, even a very small increase in the incidence of disease resulting from 1 exposure to the RFR generated by those devices could have broad implications for public health. efforts already underway at that time, the NTP concluded that additional studies were warranted 8 to more clearly define any potential health hazard to the U.S. population. Due to the technical 9 complexity of such studies, NTP staff worked closely with RFR experts from the National 10 Institute of Standards and Technology (NIST). With support from NTP, engineers at NIST 11 evaluated various types of RFR exposure systems and demonstrated the feasibility of using a 12 specially designed exposure system (reverberation chambers), which resolved the inherent 13 limitations identified in existing systems. 14 In general, NTP chronic toxicity/carcinogenicity studies expose laboratory rodents to a test 15 article for up to 2 years and are designed to determine the potential for the agent tested to be 16 hazardous and/or carcinogenic to humans. 3 For cell phone RFR, a program of study was 17 designed to evaluate potential, long-term health effects of whole-body exposures. These studies 18 were conducted in three phases: (1) a series of pilot studies to establish field strengths that do not 19 raise body temperature, (2) 28-day toxicology studies in rodents exposed to various low-level Hsd:Sprague Dawley ® SD ® (Harlan) rats were housed in custom-designed reverberation 8 chambers and exposed to cell phone RFR. Experimentally generated 900 MHz RF fields with 9 either GSM or CDMA modulation were continuously monitored in real-time during all exposure 10 periods via RF sensors located in each exposure chamber that recorded RF field strength (V/m).

11
Animal exposure levels are reported as whole-body specific absorption rate (SAR), a biological 12 measure of exposure based on the deposition of RF energy into an absorbing organism or tissue. 13 SAR is defined as the energy (watts) absorbed per mass of tissue (kilograms). Rats were exposed 14 to GSM-or CDMA-modulated RFR at 900 MHz with whole-body SAR exposures of 0, 1.5, 3, or 15 6 W/kg. RFR field strengths were frequently adjusted based on changes in body weight to 16 maintain desired SAR levels.

18
Exposures to RFR were initiated in utero beginning with the exposure of pregnant dams 19 (approximately 11-14 weeks of age) on Gestation Day (GD) 5 and continuing throughout 20 gestation. After birth, dams and pups were exposed in the same cage through weaning on 21 postnatal day (PND) 21, at which point the dams were removed and exposure of 90 pups per sex 22 per group was continued for up to 106 weeks. Pups remained group-housed from PND 21 until 23 they were individually housed on PND 35. Control and treatment groups were populated with no 24 more than 3 pups per sex per litter. All RF exposures were conducted over a period of 1 approximately 18 hours using a continuous cycle of 10 minutes on (exposed) and 10 minutes off 2 (not exposed), for a total daily exposure time of approximately 9 hours a day, 7 days/week. A 3 single, common group of unexposed animals of each sex served as controls for both RFR 4 modulations. These control rats were housed in identical reverberation chambers with no RF 5 signal generation. Each chamber was maintained on a 12-hour light/dark cycle, within a 6 temperature range of 72 ± 3°F, a humidity range of 50 ± 15%, and with at least 10 air changes 7 per hour. Throughout the studies, all animals were provided ad libitum access to feed and water.

10
In pregnant rats exposed to 900 MHz GSM-or CDMA-modulated RFR, no exposure-related 11 effects were observed on the percent of dams littering, litter size, or sex distribution of pups. 12 Small, exposure-level-dependent reductions (up to 7%) in body weights compared to controls 13 were observed throughout gestation and lactation in dams exposed to GSM-or CDMA-14 modulated RFR. In the offspring, litter weights tended to be lower (up to 9%) in GSM and 15 CDMA RFR-exposed groups compared to controls. Early in the lactation phase, body weights of 16 male and female pups were lower in the GSM-modulated (8%) and  17 RFR groups at 6 W/kg compared to controls. These weight differences in the offspring for both 18 GSM and CDMA exposures tended to lessen (6% and 10%, respectively) as lactation progressed. 19 Throughout the remainder of the chronic study, no RFR exposure-related effects on body 20 weights were observed in male and female rats exposed to RFR, regardless of modulation 21 (Figures 1 and 2). At the end of the 2-year study, survival was lower in the control group of 22 males than in all groups of male rats exposed to GSM-modulated RFR ( Figure 3). Survival was 23 Report Revised on June 23, 2016 also slightly lower in control females than in females exposed to 1.5 or 6 W/kg GSM-modulated 1 RFR. In rats exposed to CDMA-modulated RFR, survival was higher in all groups of exposed 2 males and in the 6 W/kg females compared to controls (Figure 4). A low incidence of malignant gliomas and glial cell hyperplasia was observed in all groups of 2 male rats exposed to GSM-modulated RFR (Table 1). In males exposed to CDMA-modulated 3 RFR, a low incidence of malignant gliomas occurred in rats exposed to 6 W/kg (Table 1). Glial 4 cell hyperplasia was also observed in the 1.5 W/kg and 6 W/kg CDMA-modulated exposure 5 groups. No malignant gliomas or glial cell hyperplasias were observed in controls. There was not 6 a statistically significant difference between the incidences of lesions in exposed male rats 7 compared to control males for any of the GSM-or CDMA-modulated RFR groups. However, 8 there was a statistically significant positive trend in the incidence of malignant glioma (p < 0.05) 9 for CDMA-modulated RFR exposures. In females exposed to GSM-modulated RFR, a malignant glioma was observed in a single rat 21 exposed to 6 W/kg, and glial cell hyperplasia was observed in a single rat exposed to 3 W/kg 22 (Table 2). In females exposed to CDMA-modulated RFR, malignant gliomas were observed in 23 two rats exposed to 1.5 W/kg. Glial cell hyperplasia was observed in one female in each of the 24 CDMA-modulation exposure groups (1.5, 3, and 6 W/kg). There was no glial cell hyperplasia or 25 malignant glioma observed in any of the control females. Detailed descriptions of the malignant 1 gliomas and glial cell hyperplasias are presented in Appendix C.  Heart 10 Cardiac schwannomas were observed in male rats in all exposed groups of both GSM-and 11 CDMA-modulated RFR, while none were observed in controls (Table 3). For both modulations 12 (GSM and CDMA), there was a significant positive trend in the incidence of schwannomas of 13 the heart with respect to exposure SAR. Additionally, the incidence of schwannomas in the 6 14 W/kg males was significantly higher in CDMA-modulated RFR-exposed males compared to 15 controls. The incidence of schwannomas in the 6 W/kg GSM-modulated RFR-exposed males 16 was higher, but not statistically significant (p = 0.052) compared to controls. Schwann cell 17 hyperplasia of the heart was also observed in three males exposed to 6 W/kg CDMA-modulated 18 RFR. In the GSM-modulation exposure groups, a single incidence of Schwann cell hyperplasia 19 was observed in a 1.5 W/kg male.  In females, schwannomas of the heart were also observed at 3 W/kg GSM-modulated RFR and 12 1.5 and 6 W/kg CDMA-modulated RFR. Schwann cell hyperplasia was observed in one female 13 in each of the CDMA-modulation exposure groups (1.5, 3, and 6 W/kg).
14 15 Schwann cells are present in the peripheral nervous system and are distributed throughout the 22 whole body, not just in the heart. Therefore, organs other than the heart were examined for 23 schwannomas and Schwann cell hyperplasia. Several occurrences of schwannomas were 24 observed in the head, neck, and other sites throughout the body of control and GSM and CDMA 25 RFR-exposed male rats. In contrast to the significant increase in the incidence of schwannomas 1 in the heart of exposed males, the incidence of schwannomas observed in other tissue sites of 2 exposed males (GSM and CDMA modulations) was not significantly different than in controls 3 (Table 5). Additionally, Schwann cell hyperplasia was not observed in any tissues other than the 4 heart. The combined incidence of schwannomas from all sites was generally higher in GSM-and 5 CDMA-modulated RFR exposed males, but not significantly different than in controls. The

6
Schwann cell response to RFR appears to be specific to the heart of male rats.  In female rats, there was no statistically significant or apparent exposure-related effect on the 18 incidence of schwannomas in the heart or the combined incidence in the heart or other sites 19 (Table 6).  The two tumor types, which are the focus of this report, are malignant gliomas of the brain and 9 schwannomas of the heart. Glial cells are a collection of specialized, non-neuronal, support cells 10 whose functions include maintenance of homeostasis, formation of myelin, and providing 11 support and protection for neurons of the peripheral nervous system (PNS) and the central In the heart, exposure to GSM or CDMA modulations of RFR in male rats resulted in a 19 statistically significant, positive trend in the incidence of schwannomas. There was also a 20 statistically significant, pairwise increase at the highest CDMA exposure level tested compared 21 to controls. Schwann cell hyperplasias also occurred at the highest exposure level of CDMA-22 modulated RFR. The intracardiac schwannomas in male rats were not observed in animals from 1 the same litter. Schwann cell hyperplasia in the heart may progress to cardiac schwannomas. No 2 Schwann cell hyperplasias or schwannomas of the heart were observed in the single, common 3 control group of male rats. The historical control rate of schwannomas of the heart in male 4 Harlan Sprague Dawley rats is 1.30% (7/539) and ranges from 0-6% for individual NTP studies 5 ( Table D2, Appendix D). The 5.5-6.6% observed in the 6 W/kg GSM-and CDMA-modulated 6 RFR groups exceeds the historical incidence, and approaches or exceeds the highest rate 7 observed in a single study (6%). The increase in the incidence of schwannomas in the heart of 8 male rats in this study is likely the result of whole-body exposures to GSM-or CDMA-9 modulated RFR.

11
In the brain, there was a significant, positive trend in the incidences of malignant gliomas in 12 males exposed to CDMA-modulated RFR, and a low incidence was observed in males at all 13 exposure levels of GSM-modulated RFR that was not statistically different than in control males.
14 The male rats in which gliomas were observed were not from the same litter. Glial cell 15 hyperplasia, a preneoplastic lesion distinctly different from gliosis, was also observed at low 16 incidences in rats exposed to either GSM or CDMA modulation. of the GSM-modulation groups and in the 6 W/kg CDMA-modulated group only slightly 1 exceeds the mean historical control rate and falls within the observed range. The survival of the control group of male rats in the current study (28%) was relatively low 4 compared to other recent NTP studies in Hsd:Sprague Dawley ® SD ® (Harlan) rats (average 47%, 5 range 24-72%). If malignant gliomas or schwannomas are late-developing tumors, the absence of 6 these lesions in control males in the current study could conceivably be related to the shorter 7 longevity of control rats in this study. Appendix E lists the time on study for each animal with a 8 malignant glioma or heart schwannoma. Most of the gliomas were observed in animals that died 9 late in the study, or at the terminal sacrifice. However, a relatively high number of the heart 10 schwannomas in exposed groups were observed by 90 weeks into the study, a time when 11 approximately 60 of the 90 control male rats remained alive and at risk for developing a tumor. between RFR exposure and the neoplastic lesions in the heart than in the brain. No biologically 18 significant effects were observed in the brain or heart of female rats regardless of modulation.

21
The results reported here are limited to select findings of concern in the brain and heart and do The Poly-k test Portier and Bailer, 1989;Piegorsch and Bailer, 1997) 5 was used to assess neoplasm prevalence. This test is a survival-adjusted quantal-response 6 procedure that modifies the Cochran-Armitage linear trend test to take survival differences into 7 account. More specifically, this method modifies the denominator in the quantal estimate of 8 lesion incidence to approximate more closely the total number of animal years at risk. For 9 analysis of lesion incidence at a given site, each animal is assigned a risk weight. This value is 10 one if the animal had a lesion at that site or if it survived until terminal sacrifice; if the animal 11 died prior to terminal sacrifice and did not have a lesion at that site, its risk weight is the fraction 12 of the entire study time that it survived, raised to the kth power. This method yields a lesion 13 prevalence rate that depends only upon the choice of a shape parameter, k, for a Weibull hazard 14 function describing cumulative lesion incidence over time . A further 15 advantage of the Poly-k method is that it does not require lesion lethality assumptions.

17
Unless otherwise specified, the NTP uses a value of k=3 in the analysis of site-specific lesions 18 . Bailer and Portier (1988) showed that the Poly-3 test gives valid results if 19 the true value of k is anywhere in the range from 1 to 5. In addition, Portier et al. (1986) modeled 20 a collection of relatively common tumors observed in control animals from two-year NTP rodent 21 carcinogenicity studies, showing that the Weibull distribution with values of k ranging between 1 22 and 5 was a reasonable fit to tumor incidence in most cases. In cases of early tumor onset or late 23 tumor onset, however, k=3 may not be the optimal choice. Tumors with early onset would 24 require a value of k much less than 3, while tumors with late onset would require a value of k 25 much greater than 3. In the current studies, malignant brain gliomas occurred only in animals 26 surviving more than 88% of the length of the study. For these brain tumors, a Weibull 27 distribution with k=6 is a better fit to survival time than with k=3 (Portier, 1986). Malignant 28 schwannomas of the heart occurred in animals surviving at least 65% of the length of the study; a 29 Weibull distribution with k=3 adequately fits these heart tumor incidences. Therefore, poly-6 30 tests were used for analyses of brain tumors and poly-3 tests were used for schwannomas. Variation introduced by the use of risk weights, which reflect differential mortality, was 2 accommodated by adjusting the variance of the Poly-k statistic as recommended by Bieler and 3 Williams (1993) and a continuity correction modified from Thomas et al. (1977) was applied. Poly-k tests were used in the analysis of lesion incidence, and reported P values are one sided. Body weights and litter weights were compared to the control group using analysis of variance 10 and Dunnett's test (1955). The probability of survival was estimated by the product-limit 11 procedure of Kaplan and Meier (1958). Statistical analyses for possible exposure-related effects 12 on survival used Cox's (1972) method for testing two groups for equality and Tarone's (1975) 13 life table test to identify exposure-related trends. Survival analysis p-values are two-sided.  Piegorsch, W.W., and Bailer, A.J. (1997). Statistics for Environmental Biology and Toxicology, 33 Section 6.3.2. Chapman and Hall, London.

4
Portier, C.J., and Bailer, A.J. (1989). Testing for increased carcinogenicity using a 5 survivaladjusted quantal response test. Fundam. Appl. Toxicol. 12, 731-737. The three papers make the argument that in studies that have low power to detect an effect, a 8 significant finding (p value <0.05) is more likely to be a false positive than a true positive. In 9 some cases this may be correct, but for rare tumors, as observed in the current study, it is very 10 unlikely that the significant findings are false positives. One reason for this is that the actual 11 significance level of the tests is not 0.05; it is much less, as illustrated below. Another reason is 12 the introduction and use of rates of true prevalence of effects, which we consider first.
This expression involves R, which is the pre-study odds that the tested effect is a true effect.

19
That is, As illustrated below in the statistical examples, R is a major modifying factor governing whether 21 a result with a significant p value is appropriately considered a true or false positive. The

22
selection of an appropriate R, or expected odds of an effect, in this case a carcinogenic effect*, 23 permits the introduction of bias in the interpretation of the p values in this report on the NTP 1 cancer studies of radiofrequency radiation. 2 For example, R could be pre-assigned as the expected odds of a positive cancer finding at any 3 site in male or female rats or mice (as in scenario 1 below where R = 1.2), or as the odds of a 4 positive cancer finding in only the male rat (as in scenario 2 below where R = 0.54). In these two 5 cases, the chances of our findings being true positives (PPV) are very high (>94%), despite the 6 low power of the current study to detect such an effect.

7
R could alternatively be pre-assigned as the odds of seeing these specific tumor types occur only 8 in the brain and/or heart of male rats. In this case, because gliomas and schwannomas have only

24
To illustrate, suppose that the background tumor rate is 1.5%, which is similar to the rate of 25 schwannomas in the hearts of male rats in the NTP historical control database (1.3%), and that 26 there are two groups: Control and Treated, with n = 90 animals per group. Further suppose that 27 the null hypothesis that tumor rates are the same in the two groups, H 0 , is tested against the 28 alternative hypothesis that the Treated group has a higher rate, H a , using a one-sided Fisher's 29 exact test. We reject the null hypothesis if p is less than 0.05. The actual significance level of this 1 test, α, is the probability of rejecting the null hypothesis when is it actually true. In other words, • 0 tumors in the Control group and 5 or more in the Treated group, or 4 • 1 tumor in the Control group and 7 or more in the Treated group, or 5 • 2 tumors in the Control group and 8 or more in the Treated group, or 6 • 3 tumors in the Control group and 10 or more in the Treated group, or The probability of making a Type I error (false positive decision, rejecting H 0 when it is true) is: Thus, when the tumor rate is low and the decision rule is to reject H 0 when the one-sided Fisher's 3 exact p-value is less than 0.05, the actual false positive rate, α, is 0.0032.

4
This significance level can be used to calculate the probability that a significant result is a true 5 positive, Positive Predictive Value (PPV). Following the Button et al. (2013) paper's notation: concluded that the test article was carcinogenic*. The pre-study odds of a carcinogenic 10 effect, R, is 326/(595-326) = 1.2119. Thus, the probability that a significant test 11 represents a true positive is This says that, under the low power/low tumor rate conditions described above, if a test is 13 significant at the 0.05 level, it almost certainly indicates a real carcinogenic effect. 3) If there is no prior information and it is thought that it is as equally likely that there is a 1 real effect as it is that there is no effect, then R = 1 and PPV = 0.97. 2 Furthermore, the relationship between R and PPV can be rearranged to solve for R, In this low power/low tumor rate situation, R could be as low as 0.28 and the PPV would be at 4 least 90%, or R could be as low as 0.13 and the PPV would be at least 80%.  ______________________________________________________________________________ 20 *The term "carcinogenic" in this case refers to NTP studies in which any group of male or 21 female rats or mice was judged to show "clear" or "some" evidence of carcinogenic activity.

22
Keep in mind that many of these agents were selected for cancer studies based on a suspicion 23 that they would cause cancer. Other agents, such as cell phone RFR, were chosen based more on 24 the sheer numbers of people exposed. The "level of evidence" definitions are indicated below.

25
As one can see, statistical significance is only one of many considerations that go into the study 26 interpretation. 27 We have not assigned a specific level of evidence to the NTP RFR study, as it is not complete.
1 Rather, we evaluated the partial study findings and concluded that the tumors highlighted are 2 "likely" related to the RFR exposure.    For studies showing multiple chemical-related neoplastic effects that if considered 41 individually would be assigned to different levels of evidence categories, the following 42 convention has been adopted to convey completely the study results. In a study with clear 1 evidence of carcinogenic activity at some tissue sites, other responses that alone might be 2 deemed some evidence are indicated as "were also related" to chemical exposure. In 3 studies with clear or some evidence of carcinogenic activity, other responses that alone 4 might be termed equivocal evidence are indicated as "may have been" related to chemical 5 exposure. Pathology data presented in this report on cell phone RFR were subjected to a rigorous peer 3 review process. The primary goal of the NTP peer-review process is to reach consensus 4 agreement on treatment-related findings, confirm the diagnosis of all neoplasms, and confirm 5 any unusual lesions. At study termination, a complete necropsy and histopathology evaluation 6 was conducted on every animal. The initial pathology examination was performed by a 7 veterinary pathologist, who recorded all neoplastic and nonneoplastic lesions. This examination 8 identified several potential treatment-related lesions in target organs of concern (brain and heart), 9 which were chosen for immediate review. 1 The initial findings of glial cell tumors and 10 hyperplasias in the brain and schwannomas, Schwann call hyperplasia, and schwannomas from 11 all sites were subjected to an expedited, multilevel NTP pathology peer-review process. The data 12 were locked 2 prior to receipt of the finalized, study-laboratory reports to ensure that the raw data 13 did not change during the review. The QA review pathologists then met with Dr. Mark Cesta, NTP pathologist for these studies, 24 and Dr. David Malarkey, head of the NTP Pathology Group, to review lesions and select slides 25 for the Pathology Working Group (PWG) reviews. All PWG reviews were conducted blinded 26 with respect to treatment group and only identified the test articles as "test agent A" or "test The reviewing PWG pathologists largely agreed on the diagnostic criteria for the lesions and on 8 the diagnoses of schwannomas in the head and neck, and granular cell lesions in the brain. margins. The neoplastic cells were typically very densely packed with more cells than neuropil.

5
The cells were typically small and had round to oval, hyperchromatic nuclei. Mitoses were 6 infrequent. In some of the neoplasms, invasion of the meninges, areas of necrosis surrounded by 7 palisading neoplastic cells, cuffing of blood vessels, and neuronal satellitosis were observed. The 8 malignant gliomas did not appear to arise from any specific anatomic subsite of the brain.

10
Glial cell hyperplasia consisted of small, proliferative, and poorly demarcated foci of poorly 11 differentiated glial cells that accumulated and invaded into the surrounding parenchyma. In some 12 cases, there was a small amount of perivascular cuffing. The hyperplastic cells appeared 13 morphologically identical to those in the gliomas but were typically less dense with more 14 neuropil than glial cells. There were no necrotic or degenerative elements present, so there was 15 no evidence that the increased number of glial cells was a reaction to brain injury.  The cell types described for the endocardial neoplasms were both present, but in fewer numbers. 26 In both subtypes of schwannomas, there was a minimal amount of cellular pleomorphism. In 27 some larger neoplasms, Antoni type A and B patterns were present.

29
The Schwann cell hyperplasias were similar in appearance to the schwannomas, but were smaller 1 and had less pleomorphism of the cells. In the case of the endocardial Schwann cell hyperplasia, 2 there was no invasion of the myocardium.  Purpose: To provide independent peer review of an initial draft of this partial report. The peer 6 reviewers were blind to the test agents under study. Introductory materials on RFR and details of 7 the methods dealing with the field generation and animal housing were redacted from the version 8 sent to the reviewers. The reviewers were provided a study data package, also blinded to test The reviewers stated that the NTP had performed an adequate and objective peer review 7 of the pathology data, and the statistical approaches used were consistent with other NTP 8 studies. The methods were described as objective and reasonable. The interpretations of 9 the data, including the limitations, were also reasonable and objective. One reviewer 10 found the data on schwannomas of the heart to be more compelling with respect to an 11 association with treatment than the brain gliomas. This reviewer summarized the findings 12 as: The proliferative lesions in the brain are more difficult to interpret because 1) 26 their low incidence that was well within the historical control range, 2) lack of One reviewer suggested that more information be given on the time when tumors were 6 observed (e.g., at terminal necropsy, or early in the study) to help assess the possible impact 7 of the decreased survival times in the control animals on tumor incidence. This reviewer also 8 suggested a discussion of how the survival of control male rats in this study compared to the 9 historical control data. There was also concern that the diagnostic criteria developed by the 10 PWG and used in the current study would impact the historical control incidence rates 11 reported in Table D. The reviewers had the option of agreeing, agreeing in principle, or disagreeing with the draft 21 conclusions. All three reviewers agreed in principle, reiterating issues discussed above. Charge: To peer review the draft report, statistical analyses, and pathology data and comment on 25 whether the scientific evidence supports NTP's conclusion(s) for the study findings. This is a partial report, a report which is presumably part of a larger set of studies involving 2 species (mice and rats), 2 sexes (male, female), and multiple tissue types, all based on 90-week studies of two different types (GSM and CDMA) of cell phone radiofrequency radiation (RFR). In this partial report, we are given findings regarding brain gliomas and heart schwannomas in male and female Harlan Sprague Dawley rats which were exposed exposed to control or 3 different levels (1.5, 3.0, 6.0) of two types (GSM and CDMA) of RFR. There were 90 rats in each group. Using the poly-3 test with the Bieler-Williams variance adjustment, the authors found a statistically significant increase in the rate of brain gliomas in males exposed to CDMA RFR. Using the poly-6 test, the authors found a statistically significant increase in the rates of heart schwannomas in males exposed to GSM and CDMA. There were no statistically significant differences in rates of gliomas or schwannomas in females; also there was no statistically significant increase in rates of gliomas in males exposed to GSM RFR.

Comments:
1) Why aren't we being told, at least at a high level, of the results of other experiments (i.e., male and female mice, tissues other than heart and brain, tumors other than glioma and schwannoma)? Given the multiple comparisons inherent in this kind of work (see pages 27-30 and Table 13 of the FDA guidance document), there is a high risk of false positive discoveries. In the absence of knowing other findings, we must worry about selective reporting bias. 2) I was able to reproduce the authors' positive P-value findings (see Appendix 1, R code) using the MCPAN R package. However, I'm getting slightly different values for adjusted denominators (also in Appendix 1). 3) I was able to reproduce the authors' findings of longer survival with RFR (see Appendix 1, R code). 4) I have a number of questions about the study design: a. Were control rats selected in utero like the exposed rats were? b. Were pregnant dams assigned to different groups by formal randomization? If not, why not? c. Why were pups in the same litter included? Did the authors take any steps in their analyses to account for the resulting absence of i.i.d? d. The authors state that at most 3 pups were chosen per litter. How were the 3 pups chosen (and the others presumably not used for this experiment)? Were the 3 pups that were chosen selected by formal randomization? If not, why not?
e. Were all analyses based on the intent-to-treat principle? Were there any crossovers? Were all rats accounted for by the end of the experiment and were all rats who started in the experiment included in the final analyses? f. Blinding: The authors state that "All PWG reviewer were conducted blinded with respect to treatment group," but in the very next phrase write "only identifying the test articles as 'test agent A' or 'test agent B.'" Why was this information (test agent A or B) given? The blinding was not complete. 5) Sample size: a. Did the authors perform a prospective (that is before initiation of the work) sample size calculation? If so, what were the prior assumptions? In other words, why did the authors choose to study 90 rats in each group and why did they set the maximum duration to 90 weeks (instead of 104 weeks)? b. I used a publicly available simulation package 1 to calculate the study power for male rats based on the following (see Appendix 2, power calculation simulation studies): i. Control tumor rate of ~1.5%.
ii. Risk ratio 2.5 in the group receiving the highest dose iii. 2-sided Alpha = 0.005 (based on Table 13 of the FDA guidance document). Note this low alpha of 0.005 for poly-k trend tests is recommended to minimize the risk of false positive discoveries. iv. Sample size of 90 for each group with one planned sacrifice. v. Low lethality with lethality parameters set according to study duration and Weibull shape parameter (see Table 3 of Moon et al 1 ). When I re-ran the simulations using intermediate lethality, results were not materially changed. vi. Study duration 90 weeks vii. 5000 simulations viii. Note -I used dose levels of 0,1,2, and 4 because I was unable to adjust these on the web site (despite trying 3 different browsers). c. Based on these inputs, the recommendations in Table 13 of the FDA guidance document, and a sample size of 90 rats in each group, I find very low power (<5%, see Appendix 2). Even allowing for a risk ratio of 5.0 (a level that is clinically unlikely), the power for 2-sided alpha=0.005, k=3 and low lethality is only ~14% (see Appendix 2). d. The low power implies that there is a high risk of false positive findings 2 , especially since the epidemiological literature questions the purported association between cell phone exposure and cancer. 3 6) Summary: I am unable to accept the authors' conclusions: a. We need to know all other findings of these experiments (mice, other tumor types) given the risk of false positive findings and reporting bias. It would be helpful to have a copy of the authors' statistical code. b. We need to know whether randomization was employed to assign dams to specific groups (control and intervention).
1. Scientific criticisms: a. Please comment on whether the information presented in the draft report, including presentation of data in any tables, is clearly and objectively presented. Please suggest any improvements.
Overall, the information included in the report is presented in a comprehensive and accurate manner. Specifically, the experimental design and conditions are sufficiently documented and the choice of statistical approaches is explained; the results are well organized and necessary details are provided.
Nevertheless, a few additions could be suggested: (1) Appendix tables for all poly-k tests performed could be added. I believe this would enhance the presentation of the adjusted rates and the strength of the statistical evidence. As a possible example I prepared the below table using the R package MCPAN and its poly3test() function. (2) In the portion of the text describing poly-k test results, p-values are given for significant pairwise comparisons; I would also give the p-values estimated for the significant trends (maximum test).
(3) Information could be included regarding the software or programming environment used for the computations.
(4) In the portion of the text describing differences in survival at the end of the study between control and RFR-exposed animals (page 5 §2) the compared characteristic is not named (median survival, TSAC?) and also no numerical values of the estimates or the range of differences are given. I would add numbers in the text or an Appendix Appropriate statistical design and methods were applied in accord with the FDA/NTP guidelines for conducting long-term rodent carcinogenicity studies and analyses. The results and limiting issues were objectively discussed. The critical issue of shorter survival in the male control group was addressed with regard to the percentage of animals surviving to terminal sacrifice in historical control data (avg. 47%, range 24% to 72%) and the possible impact of the observed age of tumor occurrence on the statistical inference.
I believe detailed information about animal selection and randomization procedures should be given so that the potential for allocation bias could be judged. As shown in the figure below, the lower survival rate to terminal sacrifice (28%) in the male control is accompanied by the higher rate of moribund sacrifice (49%); in the male group exposed to CDMA with 6 W/kg, a higher rate of natural death was observed (46%).
It has been reported that insufficient randomization can lead to differences in survival rates. As an example, in a carcinogenicity study on aspartame it was suggested that lack of randomization to different rooms may have possibly been the cause of low survival rates (27%) in the control female group due to a high background infection rate (EFSA, 2006;Magnuson, B., Williams, G.M., 2008).
A statement of the required statistical significance level should be added. FDA guidance suggests the use of significance levels of 0.025 and 0.005 for tests for positive trends in incidence rates of rare tumors and common tumors, respectively; for testing pairwise differences in tumor incidence the use of significance levels of 0.05 and 0.01 is recommended for rare and common tumors, respectively. If power calculations to determine the required sample size were performed, the results should also be included.
3. The scientific evidence supports NTP's conclusion(s) for the study findings: The NTP's overall draft conclusion was as follows: "Under the conditions of these studies, the observed hyperplastic lesions and neoplasms outlined in this partial report are considered likely the result of exposures to test article A and test article B. The findings in the heart were statistically stronger than the findings in the brain." In my view, the results support the conclusion of likely carcinogenic effect of the RFR-exposure on Schwannoma heart lesions in male Harlan Sprague Dawley rats.
Possible carcinogenic effects in the brain are marginal and are not sufficiently supported by statistical evidence in the male Harlan Sprague Dawley rats.
In the female Harlan Sprague Dawley rats very few lesions were observed in either site and statistical significance was not reached at all.
Analysis of National Toxicology Program (NTP) study evaluating risk in rat lifetime exposure to GSM or CDMA RFR.

Notes:
The NTP study document acknowledges several study limitations [page 10, discussion section]. Potential limitations should prominently factor into considerations regarding the context of the findings, as well as their interpretation and application.
Working list of limitations potentially impacting NTP study interpretations • Difficulty in achieving diagnostic consensus in lesions classifications of rare, unusual, and incompletely understood lesion association • Document appears to indicate that the second Pathology Working Group (PWG) empaneled to review and obtain lesion classification consensus, following the inability of the initial PWG to do so, may have reviewed different lesions sets • No record of clinical disease manifestations due to lesions involving heart and brain [note lesions in heart and brain are mutually exclusive; affected rats have either one or the other and do not appear to have the involvement of both organs together (appendix E)] • Lesions, including malignancies, do not appear to materially shorten lifespan, except for a subgroup of rats (less than 1/3 of affected rats) with malignant Schwannomas in heart • Lack of shortened lifespan as a consequence of malignancy for the majority of affected rats contrasts with shortened lifespan of male control rats for which there is absence of attributable cause of death. The survival of the control group of male rats in the current study (28%) was relatively low compared to other recent NTP studies (avg 47%, range 24 to 72%). Creates greater reliance on statistical controlling for survival disparities and reliance on historical controls • Reliance on historical controls made up of rats of different genetic strain background, held under different environmental conditions • Absence of data on incidence of more frequently expected tumor occurrences in rats (background lesions) Documenting the nature of the brain and cardiac lesions observed in RFR exposed rats and placing them into test article exposure-related context, in contrast to potential for their occurring spontaneously, are important and challenging goals. The NTP study limitations make the interpretation of reasonable risk more complicated. NTP acknowledgements of study limitations appear factored into one of NTP's reviewer's study conclusion, i.e., findings represent "some evidence" for a test article effect in statistically significant trend for Schwannomas; an opinion which is coupled with a conclusion for "equivocal evidence" of an effect in relation to malignant gliomas of the brain [NTP Appendix F, Reviewer Comments].
The summation from Appendix F reviewers regarding existence of test article effect is less than conclusive. The NTP study documents a series of cytoproliferative changes in heart and brain. The nature of some of the changes is challenging diagnostically and appears to be incompletely understood. These findings are presented in the absence of complete analysis of the entire consequences of the study effects. For example, no potential significance for test article effect context is given to any of granular cell proliferative lesions of the brain, a finding mentioned only as a contrast to what was less well understood pathologically (NTP Appendix C, Pathology). It is noteworthy that the lesion types analyzed in the NTP RFR study under review are uncommon historically in rats, in the organs discussed. Furthermore, the malignancies of neuroglia appear to be paired with the occurrence of poorly understood changes involving neuroglial cell hyperplasias in the central and peripheral nervous systems. Little information can be gleaned from the literature about the nature and significance of these latter proliferative changes, interpreted by NTP as nonneoplastic and noninflammation-reactive neuroglial cell in nature. Although unclear in the NTP study document, it is plausible that the particular lesion constellation, along with the relative novelty of some lesions, contributed to the lack of consensus regarding the nature of the lesions on the part of the initial PWG study pathologists. Concern raised by one of the reviewers (Appendix F, Reviewer Comments) regarding how this difficulty in ability to classify lesions might impact comparisons to historical control lesion incidence data (NTP Table D) is certainly principled.
The extraordinary PWG process, presumably posed by the difficult diagnostic interpretations, has the potential to influence the reliance on historical controls. In this regard, study limitations concerning determination of whether or not there is a test article effect include the substantially poor survival of male rats in the control group. The survival of the control group of male rats in the study under review (28%) was relatively low compared to other recent NTP studies (avg 47%, range 24 to 72%). This apparently led to greater statistical construction to account for the impact of study matched controls, and created increased reliance upon historical data of rare tumor incidences in control animals taken from other chronic carcinogenicity studies. NTP acknowledges a limitation in using the historical incident data and a small study match control group due to poor survivability. There are potential sources of variability when using historical controls of different rat strains and fluctuating study conditions (environment, vehicle, route of exposure, etc.), as is the case here. It seems less than clear what appropriate background lesion incidence is, as NTP indicates some data involve other strains of rats. The range of lesion incidence in historical controls could mean that the true incidence of some lesions varies considerably and might be considered rare or more common depending upon the incidence rate.
The guidance manual on Statistical Aspects of the Design, Analysis and Interpretation of Chronic Rodent Carcinogenicity Studies of Pharmaceuticals by the FDA provided for this review discusses applying comparisons using historical control lesion incidences at some length [beginning page 27, line 996]. Considering lesions as being rare or more common appears to influence selection of the level of statistical significance for comparisons. It appears that analysis for significant differences in tumor incidence between the control and the dose groups for these NTP studies has been established at the 0.05 level (NTP Tables 1, 3,5). Interpretations of trend tests may be influenced by the choice of decision rule applied. Such choices can result in about twice as large overall false positive error as that associated with control-high pairwise comparison tests [page 28, line 1012-1026]. The FDA guidance manual [page 31, line 1136] highlights concern regarding reliance upon historical control incidence data, stating that using historical control data in the interpretation of statistical test results is not very satisfactory because the range of historical control rates is usually too wide. This is especially true in situations in which the historical tumor rates of most studies used are clustered together, but a few other studies give rates far away from the cluster. When the range of historical control data is simply calculated as the difference between the maximum and the minimum of the historical control rates, the range does not consider the shape of the distribution of the rates. These circumstances may impose some limitations on optimal risk assessment designs.
Somewhat paradoxically then, NTP study limitations including that imposed due to reliance upon less than optimal historical control lesion incidence data for much of the comparisons between treated and untreated rats, is confronted by existence of a difficult to classify and incompletely understood lesion constellation interpreted to include neuroglial cell hyperplasia. Notwithstanding, this confounding proliferative lesion occurring in the context along with malignancies of apparently similar histogeneses, sustains a level of concern for a rare injury mechanism related to test article effect. Additional information about the study together with an assessment of the statistical analyses may enhance the value of this analysis.