Article publication date
October 2025
ARPANSA review date
November 2025
Summary
This commentary discusses the recently published set of systematic reviews that were commissioned by the World Health Organization (WHO) to evaluate various health endpoints in relation to radiofrequency electromagnetic field (RF-EMF) exposure. The commentary presents criticisms of some aspects of each review with a particular focus on those relating to human observational studies (SR1 A & B, SR3A & B, SR5), oxidative stress (SR9) and non-specific symptoms (SR7, SR8). The commentary touches on study selection criteria, purported authorship bias, the use of meta-analysis and other direct criticisms of included studies. The conduct of risk of bias (RoB) and certainty of evidence (CoE) assessments were also the subject of criticisms as well as some additional semantic commentary around phrasing used in GRADE CoE statements.
Published In
Melnick et al. Environmental Health
Link to study
https://pubmed.ncbi.nlm.nih.gov/41034851/
Commentary by ARPANSA
Cancer in human observational studies (SR1 A and B)
SR1 A and B remain the most comprehensive and highest-quality systematic reviews on human observational studies investigating RF-EMF and cancer; they conclude that RF-EMF does not cause cancer with varying degrees of certainty for different cancers based on the availability of evidence. The primary criticisms of SR1 by Melnick et al., are due to the inclusion of the Danish cohort study which bases exposure on mobile phone subscription. The exposure assessment in the Danish cohort study has been validated, showing a clear distinction between subscribers and non-subscribers, with only 16% of non-subscribers actually using a mobile phone (Schuz & Johansen, 2007), meaning its exposure assessment can be rated very well and far better than previous studies. The conclusions of SR1 on brain cancer, in particular, are supported by the Cosmos study, (Freychting et al., 2024), which is the most comprehensive observational study to date, that includes a large cohort of participants and assesses exposure via questionnaires on mobile phone use and operator data. The Cosmos study found no association between mobile use and brain cancer, but this result was not included in SR1 because it was published after the cut-off date for inclusion. Melnick et al., also criticised SR1 for how it assessed exposure comparisons saying it was too basic and only used exposure metrics like“ever versus never mobile phone use”. However, SR1 goes beyond simple exposure proxies by analysing duration of use and dose-related metrics such as cumulative call time and number of calls.
Cancer in experimental animals (SR2)
In contrast to their criticisms of the other systematic reviews, the commentary praised the systematic review on the effect of RF-EMF on cancer in experimental animals despite the significant flaws in the narrative synthesis of SR2. An extensive critique of SR2 can be found on the ARPANSA website.
Adverse reproductive outcomes (human observational studies) (SR-3 A and B)
The reviews (SR3A and SR3B) provide state-of-the art methods on investigating whether RF-EMF exposure is related to reproductive outcomes in human populations. Melnick et al, comments that the use of surrogate measures of exposure (hours of mobile phone use), do not provide reliable information on exposure to the genitalia or the developing foetus. However, this is a valid and widely accepted approach in epidemiological studies when objective exposure data are unavailable (Teschke, 2003). Another criticism of Melnick et al. is that evidence on the effects of RF-RMF on female reproductive outcomes was made without consideration of the extensive literature on oxidative stress due to RF-EMF exposures. Although several experimental studies suggest that RF-EMF exposure may induce oxidative stress, the validity of these findings is undermined by heterogeneity, and other methodological limitations that have been described in the systematic review on oxidative stress (SR9) (Meyer et al., 2024).
Male fertility in both in-vivo and in-vitro studies (SR4)
The main criticism for SR4 by Melnick et al. was that they believe that for some of the outcomes where effects were found, the certainty of the evidence should not have been downgraded. They argue that heterogeneity in the included papers does not impact the certainty of results. This is counter to standard systematic review methodology, such as outlined in the GRADE assessment, which typically considers methodological heterogeneity as a source of inconsistency that reduces confidence in the overall body of evidence. By dismissing these variations as inconsequential, the authors risk overlooking important sources of bias or effect modification that could influence the interpretation of results. Melnick et al., suggests that the conclusions for SR4 should be changed because of some negative results, particularly those reported for laboratory animals and human sperm in vitro. However, the only negative outcome with a high level of certainty in the evidence, had an average exposure across the studies of 23.87 W/kg. This average is over 28,000 times the public limit for whole body average exposure set in the ARPANSA safety standard (RPS S-1)
Effects on cognition in human observational studies (SR5)
Melnick et al. had various criticisms of SR5, one of the mains ones being that a study by Grigoriev et al. (2018) should not have been excluded. However, this study was excluded due to significant methodological flaws, particularly, the lack of a description of exposure assessment, follow-up protocol and participant details. For instance, the study fails to account for mobile phone use changing between the age of 7 to 17 years, which is critical given that previous research (Thomas et al., 2010; Bhatt et al., 2017) consistently shows increased mobile phone use as children age. Therefore, its exclusion from SR5 is well-founded. Regarding the exclusion of other studies, namely, cross-sectional studies were excluded as they are unable to establish causality, and this reason is clearly described in SR5 and its protocol.
Effects on cognitive performance in human experimental studies (SR6)
Melnick et al.’s main criticism of SR6 is the use of the neuropsychological assessment classification system for cognitive domains (Lezak et al., 2012) and suggest that the incorrectly cited Cattell-Horn-Carrol (CHC) taxonomy should have been used. However, the cognitive domains used for grouping the systematic review outcomes are in fact based on CHC taxonomy and Lezak et al. (2012) only describes ways these different domains can be tested for. This comment by the authors is therefore invalid. Melnick et al., further criticise the heterogeneity of the included studies stating this would reduce the ability of the meta-analyses to detect a small effect. Statistical power is an issue that was discussed in the systematic review and remains an issue in the literature, particularly in singular one off studies. However, the pooling data for use in a meta-analysis, as done in this systematic review, is how the lack of statistical power in the literature can be alleviated.
Symptoms (human observational studies) (SR-7)
Melnick et al. purport that the health outcomes (tinnitus, migraines, and sleep disturbances) included in SR7 should not have been evaluated. This is contrasting to literature which clearly shows individuals report both of these as short-term and long-term effects (Medic et al., 2017; Lipton et sl., 2001; Zeleznik et al., 2024). The authors also criticised one of the key included studies, the COSMOS study (Auvinen et al., 2019), allegedly citing its inappropriate study methodology, such as exclusions and follow-up time. The COSMOS study excluded the subjects with a history of tinnitus or weekly headaches at baseline to avoid potential reverse causation. This is a common practice in epidemiological studies when the goal is to test the relationships between environmental exposures and health outcomes (Rezende et al., 2022). This approach helps isolate the effect of preexisting conditions and those that align with outcomes after exposure (Poorolajal, 2025). The four-year follow-up period in the COSMOS longitudinal cohort study is a strong design choice for evaluating health symptoms (Kamal et al., 2025). The methodological approaches adopted in the COSMOS study, represent the most robust design and are well supported by evidence.
Human experimental non-specific symptoms (SR-8)
With respect to the systematic review on non-specific symptoms in human experimental studies (SR-8) Melnick et al. assert that, because studies on people without idiopathic environmental intolerance attributed to EMF (IEI-EMF, also known as electromagnetic hypersensitivity or EHS) were included in the review, it cannot effectively review studies on people with IEI-EMF. However, in SR8, the analyses were subdivided by IEI-EMF and non IEI-EMF populations. The commentary also questions why human provocation studies using EMF frequencies outside of the RF section of electromagnetic spectrum are excluded from the review and suggests that their exclusion prevents consideration of the results of studies using RF-EMF exposure. SR-8’s protocol, and the entire set of systematic reviews more broadly, clearly define the exposure that is to be studied and so a large departure of the review from this paradigm would be inappropriate. The inclusion criteria of SR-8 adequately cover the topic under investigation (RF-EMF).
Oxidative Stress (SR-9)
The primary criticism Melnick et al. ascribe to the systematic review on the effect of RF-EMF on biomarkers of oxidative stress (SR-9) is the inclusion criteria. Melnick et al. take issue with the fact that numerous studies were excluded from SR-9 for using an unreliable method of outcome assessment and that studies were excluded for not adequately characterising their exposure system. However, both reasons are valid causes for exclusion regardless of the number of studies that fall under that category. The secondary criticism Melnick et al. have of SR-9 is the subdivision of the meta-analyses into biomarker and biological system pairs which they assert may dilute an overall effect. However, combining outcomes across different organs and markers would create a lack of specificity thus reducing the usability of the results for directing further research and drawing conclusions relevant to human health outcomes. Combining biomarkers and biological systems into a net category only enables vague discussions of miscellaneous oxidative stress.
ARPANSA has written brief evaluations of each systematic review and published them as part of our radiation literature survey program. They are available here: SR1A & 1B, SR2, SR3A & B, SR4, SR5, SR6, SR7, SR8, SR9.
