This study took place in two distinct periods, during the relaxation of lockdown conditions after the first and second waves of COVID-19 in Belgium, respectively, while the roll-out of 5G in Belgium was launched in April 2020, which led to numerous protests among the concerned citizens.

The survey was developed under.Net (C#) with storage in an internal Structured Query Language (SQL) server database. In the first period (P1), the survey was published online in French (June 2020) and in Dutch (July 2020) and was available online for 3 weeks. This period corresponds to the relaxation of the lockdown measures following the first COVID-19 wave. In the second period (P2), it was published in French (February 2021) and in Dutch (March 2021) and was available online for 4 weeks. This second period corresponds to the post-COVID-19 lockdown period, although less strict, linked to the second wave of COVID-19 (Table 1).

Respondents who did not wish to complete the survey digitally could request a paper format to be filled in by hand and sent by post. The final anonymized dataset includes one hand-filled form in P1 and two in P2.

The survey included the following domains (see details in Supplementary material S1).

Demography includes questions related to age, gender, municipality, and employment (possibly temporarily interrupted). The degree of urbanization (DEGURBA) was assigned based on the municipality (28). The classification identifies three zones (level 1): cities, towns and suburbs, and rural areas.

Health status: Respondents were asked to evaluate their health on a 5-point Likert scale (from very good to very poor—Health_status) and to report the frequency over time of symptoms common in the IEI-EMF (migraine, insomnia, fatigue, memory problems, heart palpitations, joint pain, digestion problems, itching, depressed mood, and irritability) in four categories (from never to every day). The symptom score was calculated by averaging the answers related to symptoms (SymptomScore).

Risk Perception, by way of the MHW scale (8, 20): The scale was translated into French and Dutch and used to assess how concerned respondents perceive the impact of various aspects of modern life on their health. Translations were independently proofread by two members of the research team in both languages. The scale initially consists of 24 items, with scores ranging from 0 (not at all concerned) to 4 (extremely concerned). Two items were added to consider new issues: risk perceptions linked to “5G antennas” and “COVID-like viruses,” while due to technical issues, the item on genetically modified food was removed (Supplementary material S2). MHW scores were calculated by summing the answers. This resulted in an overall MHW score (25 items), a specific score for items related to radiation (4 items: mobile phones, 2G-4G antennas, 5G antennas, high voltage powerlines (HVPL)—EMF_worries score) and an MHW score excluding radiation items (21 items—noEMF_worries score), ranging from 0 to 100, 0 to 16 and 0 to 84, respectively. The four items related to EMF sources were also considered separately, each ranging from 0 to 4 (mobile phone-worries, 2G-4G antennas-worries, 5G antennas-worries, and HVPL-worries).

Exposure perception: The level of exposure perception to the various agents of the MHW scale was evaluated by the following question: “Are you very exposed to this agent?” (yes/no). Exposure scores were calculated, assigning 1 to “yes” and 0 to “no” answers, resulting in two exposure scores ranging from 0 to 4 and 0 to 21 for items related to EMF (EMF_exposure) or not (noEMF_exposure), respectively. The perception of exposure to the four items on EMF sources was also considered one by one, ranging from 0 to 1.

EMF sensitivity: The respondents’ perceptions of their sensitivity to EMF were examined using five categories (from not sensitive to hypersensitive).

Exposure avoidance strategies: These strategies could be used to reduce exposure (Supplementary material S3). A score was calculated: for each strategy, 1 or 2 points were assigned if the strategies had been in place for less or more than 1 month, respectively, to give more weight to those strategies adopted over a longer period of time. An avoidance score was derived by summing the answers to the 15 questions, ranging from 0 to 30.

Univariate statistics were processed with Stata/SE 15.1. Fisher’s exact test was used to compare the characteristics of the participants between the two periods, as well as to compare the questionnaires included/excluded due to missing information on sensitivity to EMF and MHW. The distribution of results between the different sensitivity categories was also examined using this test.

Comparisons of health status and EMF sensitivity by period were performed by ANOVA. Comparisons related to EMF sensitivity were performed by ANOVA. The distribution of the results among the different categories of sensitivity was explored by the chi-squared test.

Multivariate analyses were conducted to integrate the contributions of the different variables to the varying levels of EMF sensitivity reported by respondents: (1) ordered logistic regression (OLR) with Stata/SE 15.1 and (2) exploratory multivariate analyses using gradient boosting machine (GBM) analysis within the R environment (R version 4.1.2, The R Foundation for Statistical Computing). GBM modeling is a machine learning technique of interest in managing possible non-linear relationships between independent and dependent variables, without requiring explicit model specifications. The boosting approach used in boosted regression trees has its origins within machine learning (46), but subsequent developments in the statistical community reinterpret it as an advanced form of regression (45) [(29), p.803]. As recommended by Elith et al. (29), we were able to fit GBM models with at least 1,000 trees. A Poisson distribution fits the EMF-sensitivity variable. The GBM prediction models were fitted following the gbm.step routine in the gbm package version 2.1.8 and dismo package version 1.3–5. The trees were built with default parameters: a tree complexity of 5, a learning rate of 0.001 and a bag fraction of 0.5. To enable analyses to be replicated, the seed was set at 123.

In both multivariate analyses, the dependent variable was EMF sensitivity, while the independent variables were gender, age, employment, region, and urbanization as generic variables; symptom score and health status as health variables; and MHW, exposure perception and avoidance scores as specific variables, based on our hypotheses.

We used a convenience sampling method. In P1 and P2, 153 and 446 people, respectively, participated in the survey. However, as a number of them did not complete the questions on sensitivity to EMF and the MHW scale, they were excluded from further analysis. Thus 97 participants were included in P1 (57.5% female and 42.5% male) and 285 (52.6% female and 47.4% male) in P2 (see Supplementary material S4).

In P1, no significant differences were found between the general characteristics of the included and excluded participants (Supplementary material S4). In P2, the proportions of women (p = 0.037) were slightly higher in the excluded participants group.

In the included participant group, the respondents’ age category distribution, and sex distribution were similar between the two periods (Supplementary material S4). However, the distribution of living areas (Region variable) differed between the two periods (p < 0.001), with a higher proportion of Walloon residents in P1 and higher proportions of Brussels and Flemish residents in P2 due to the more intensive recruitment in those regions, especially in Flanders. The difference in the degree of urbanization distributions (p = 0.001) follows the regional characteristics, as indicated by the population density in the three regions of 488, 7,511, and 216 inhabitants/km² in Flanders, Brussels and Wallonia, respectively (30). Approximately 70% of the respondents, in both periods, declare themselves professionally active.

Regarding health, EMF sensitivity, symptoms and (Table 2), 26.8% of the P1 respondents and 33.7% of the P2 respondents stated that they were not sensitive to EMF, 46.3% of the P1 respondents and 46% of the P2 respondents were not very sensitive or somewhat sensitive, while 26.8% of the P1 respondents and 19.6% of the P2 respondents stated that they were very sensitive or hypersensitive to EMF. There were no significant differences in the sensitivity distribution across the different categories between P1 and P2.

In P1, a higher proportion of people in the excluded participants group rated their health as very good. In P2, both the proportions of women (p = 0.037) and of the most sensitive individuals (p = 0.028) were slightly higher in the excluded group.

MHW assessment did not reveal an extreme concern, with the highest averages ranging from moderate to very much in P1 for air pollution, pesticides in food, additives in food, pesticide sprays and mobile phone antennas, and in P2 for pesticides in food, air pollution, additives in food, climate change and antibiotics in food (Figure 1). In relation to the COVID-19 pandemic-related variables, concerns are less than moderate.

The MHW and EMF worries showed significantly different values in the respective EMF sensitivity categories in the two periods: respondents who considered themselves more sensitive showed greater concern than those who considered themselves not sensitive or somewhat sensitive, both in terms of the MHW and EMF worries scores (Table 3), as well as in terms of the items related to EMF sources considered one by one (Table 4). Greater differences between the EMF sensitivity categories were observed for mobile phones and 2G-4G antennas, with mean scores ranging from less than 1 for not at all sensitive to 4 or close to 4 for extremely sensitive respondents.

In P1, 44.3% of participants reported adopting strategies to limit their exposure at least once in the past month, while 27.3% did so in P2. Regardless of the period, the three most common strategies to avoid exposure were to eliminate or reduce the use of RF devices, to go to places that people considered to be unexposed to recover and recharge, and to avoid going to exposed places or only when there were fewer people (Figure 2).

The avoidance score was significantly different for the five EMF sensitivity categories in both periods (P1: F (4,92) = 20.04; p < 0.001; P2: F (4,280) = 97.46; p < 0.001) (Figure 3). The greater the reported sensitivity to EMFs, the greater the number of avoidance strategies.

The aim of this study was to collect data to improve knowledge of the characteristics of people who report different levels of sensitivity to EMF by assessing their MHW, perception of exposure and avoidance strategies.

The analysis of MHW items showed, contrary to Baliatsas et al. (20) and Bailer et al. (5), that all EMF sources considered, especially 2G-4G antennas, are among the greatest worries. This could be related to the characteristics of our population, which includes a larger proportion of people who consider themselves to be EMF sensitive, as was intended during recruitment.

We observed differences between P1 and P2 regarding the percentages of participants from different regions (with a larger proportion of Flemish participants in P2) and the level of urbanization (consistent with regional proportions) (Supplementary material S4). However, there were no differences in participant characteristics related to health, sensitivity, and symptoms (Table 2) between the two periods.

Our first hypothesis was based on the positive association between reporting higher levels of EMF sensitivity and higher MHW scores, particularly on the EMF items of the MHW scale. The results of this study confirmed this hypothesis.

Considering our second hypothesis, we observed more frequent avoidance strategies for those reporting greater EMF sensitivity.

Finally, regarding our third hypothesis that the adoption of exposure avoidance strategies would lead to lower exposure perceptions of EMF sources, our results suggest that this hypothesis should be rejected. Indeed, despite the more frequent avoidance strategies of those reporting sensitivity, there is no indication of a reduction in the exposure perception among the most EMF sensitive individuals or in relation to the increased EMF sensitivity. This was confirmed in the multivariate analysis, where EMF exposure did not account for EMF exposure in the model. The hypothesis of lower exposure perception is thus not confirmed. However, despite the apparent lack of effect on their perception of being exposed, individuals express setting up exposure avoidance strategies. It is recognized that avoidance strategies can lead to social exclusion, work incapacity and financial difficulties (27), which need to be considered (11).

The use of GBM models provided complementary insights to the ordered logistic regression analyses. For example, GBM analyses highlighted that avoidance strategies accounted for the largest share of variance in EMF sensitivity, particularly in P2, where they represented over 50% of the model’s explanatory power. This finding underscores the central role of behavioural responses in perceived EMF sensitivity. Additionally, the non-linear relationships identified by the GBM models—for instance, the plateau effect observed for the symptom score and avoidance strategies—provided deeper insights into how these factors influence sensitivity at different levels.

Our results showed that the most sensitive people perceived themselves to be more exposed than did the other participants. This can be interpreted in two directions.

First, the exposure perception of highly sensitive people could be influenced by their avoidance strategies. To reduce their symptoms and to make their avoidance strategies as effective as possible, they are forced to look for - and in some ways find - what they consider to be the potential origin of these symptoms. By focusing on them, highly sensitive people may become aware of the wide diversity of EMF sources in their surroundings. This could explain why these avoidance strategies, as they are implemented, are not sufficient to reduce their exposure perception. Second, they use avoidance strategies, but these strategies are not effective. Based on Dieudonné (31) and Van den Bergh et al. (32), several potential explanations are possible:

The sample in this study could be a point of concern. Indeed, particularly in P1, the number of respondents, notably the proportion of Flemish citizens, is quite low due to limited access to contact persons or institutions likely to relay survey information during this period. This may limit the generalizability of the findings and their applicability as clinical guidance. Future research should aim for larger and more diverse participant pools to strengthen the reliability of the conclusions.

Moreover, in contrast to previous surveys dealing with MHW and IEI-EMF (20, 24), which included a larger number of participants, the recruitment of individuals was directed primarily towards IEI-EMF people. As a result, there was a greater proportion of people with this profile in this study, which was relevant for the analysis of their behavioral characteristics. Furthermore, this study does not follow a longitudinal design; therefore, it does not allow us to identify a genuine evolution of the sensitivity or of its characteristics, as Traini et al. (36), Martens et al. (37) or Röösli et al. (38).

Our analysis revealed regional differences between the P1 and P2 populations, particularly in terms of levels of urbanization. These differences may influence perceptions of EMF exposure, as urban areas often have visible EMF infrastructure, such as transformers, electricity cables, and mobile phone masts. In fact, ordered logistic regressions showed a significant impact of living in Flanders on the rating of EMF sensitivity in P2, with a higher proportion of Flemish inhabitants. However, this was not linked to the degree of urbanization, as evaluated here based on postal codes. Moreover, GBM analyses did not confirm this. One of the limitations of our study is the lack of data on the visibility of infrastructure, which could affect participants’ perceptions. Future research should include questions on the proximity and visibility of these infrastructures in order to better understand their impact on perception and behavior.

Another limitation that could also be mentioned is the classification of respondents as more or less sensitive to EMFs based on a single question. Indeed, Szemerszky et al. (39) proposed complementing the sensitivity assessment commonly used in surveys with additional questions considering the ratio of symptoms and the impact of sensitivity on people’s lives. More specificity in the definition of EMF sensitive people could have provided more accurate information for comparisons between groups. Finally, multivariate analyses by GBM applied for exploratory purposes, which are still not widely used in this type of analysis, revealed their interests, but adjustments could improve their performance.

We have observed that risk perception and avoidance strategies are important variables in defining the level of sensitivity. However, avoidance strategies have been shown to be ineffective in reducing perceived exposure to EMF sources, while often leading to substantial financial costs and significant social and professional consequences (40). These behaviors, which reflect a response to a perceived threat, can be analysed through the prism of Protection Motivation Theory (PMT) (41), which provides a framework for understanding how individuals assess threats and adapt their protective behaviors. Indeed, four elements are expected to play a role in how individuals are driven to react in a protective way towards a perceived threat, such as NIR and its possible health implications: the perceived severity of the hazard and its likelihood of occurring decide on their “threat appraisal,” while their sense of being able to cope with the threat, the so-called “coping appraisal,” is influenced by both their response efficacy (the belief that the threat can be mitigated) and their perceived self-efficacy (the belief in their own ability to take action to mitigate the threat).

On the basis of the PMT, we could hypothesize that self-efficacy and response effectiveness scores are low in these individuals, due to the ubiquity of EMFs (difficult to avoid completely) and the perceived lack of control over exposure. This ineffectiveness of avoidance strategies raises questions about their relevance in reducing perceived exposure and associated symptoms. Our results highlight the need for empirical verification of weak coping appraisals in IEI-CEM individuals to better tailor interventions. It would also be relevant to explore other approaches, such as cognitive-behavioral interventions, to reinforce the feeling of control and self-efficacy (42). Confirming this hypothesis would highlight the need to critically evaluate avoidance strategies and their psychological impact, which could inform more effective interventions for highly sensitive individuals.

Beyond these questions, there is a need to focus on the resources and capacity to improve the quality of life of IEI-EMF people. Whatever the reasons for the ineffectiveness - or very relative effectiveness - of their avoidance strategies to feel less exposed, as described in Section 4.2, their suffering requires effective care to reduce the impact of symptoms on their daily lives. Therefore, despite the limited evidence to date, cognitive behavioral therapies (CBTs) could be an interesting research prospect for developing therapeutic tools (43). Nevertheless, other strategies should also be tested on the basis of new IEI-EMF models, for example, the comprehensive model explaining the onset of symptoms and their link with environmental agents developed by Van den Bergh et al. (32, 44). It may offer new perspectives for helping people with IEI-EMF to cope with their symptoms, but further work is needed to test their validity.