In view of the widespread and increasing use of GSM mobile phones, potential health risks have been a key focus of concern. Over the past years, several studies have investigated the effects of electromagnetic fields (EMFs) emitted by mobile phones on cognitive functioning. It is still unclear, however, whether there are effects and, if so, which functions of the central nervous system may be influenced. Although a number of studies have been conducted to examine this issue, results are inconsistent. Some research suggests that acute exposure to mobile phones affects cognitive performance. Lee et al recorded a mild facilitating effect on attention.1 Preece et al found a small decrease in reaction time in choice reaction task.2 Tests conducted by Smythe and Costall reflected facilitated memory performance, but only in male subjects.3 Koivisto et al found that response times in simple reaction task and vigilance tests increased.4 Cognitive time to complete an arithmetic task, however, decreased. In a second study, Koivisto et al revealed that response times in some working memory tasks decreased.5 Edelstyn and Oldershaw demonstrated a facilitatory effect on attention.6 Eliyahu et al investigated spatial recognition, verbal recognition and spatial compatibility tasks and reported a slowdown of response time.7 Curcio et al also showed an improvement of the performance in simple and choice reaction tasks.8 In contrast, Keetley et al showed an impairment in simple and choice reaction tasks, while performance on trail making was improved.9

Other studies, however have shown no effects. Haarala et al10 11 tried to replicate the findings of Koivisto et al4 5 but were unable to find any significant difference on cognitive function. Besset et al found no effects in the following neuropsychological categories: information processing, attention capacity, memory function and executive function.12 Russo et al demonstrated no change in performance in attention tasks.13 Hladky et al found no impact in memory load and attention.14 Recent findings from Haarala et al indicate that electromagnetic exposure (differentiated in continuous wave and pulse modulated EMFs) have no effect on human cognitive functions.15

The current paper uses a meta-analysis to examine whether cognitive functions are affected by acute exposure to EMFs emitted by GSM mobile phones.

METHODS

PubMed was used for a literature search on original articles published before 28 February 2007 using the phrases “mobile phone” or “cellular phone” and “cognitive”. The search revealed 40 articles. In addition, relevant articles were identified by personal reference lists. We decided to investigate all abstracts from these articles. Then, the full text articles of all relevant human studies examining the question of neuropsychological effects of mobile phones were assessed for inclusion in the meta-analyses.

Nineteen studies published between 1999 and January 2007 were taken into consideration: Russo et al;13 Besset et al;12 Haarala et al;10 Haarala et al;11 Lee et al;1 Edelstyn and Oldershaw;6 Koivisto et al;4 Koivisto et al;5 Elihayu et al;7 Smythe and Costall;3 Preece et al;2 Maier et al;18 Krause et al;19 Hladky et al,14 and Lee et al;16 Keetley et al;9 Curcio et al;8 Wilen et al;17 and Haarala et al.15

These studies investigated the effect of GSM mobile phones on cognitive functions using test procedures for recording change in performance: information processing; reaction time; attention; memory and executive functions. The minimum criteria for inclusion in our meta-analysis encompassed the following:

1. Either: treatment group (mobile phone switched on) and control group (mobile phone switched off) with baseline measures; or repeated measurement of subjects with mobile phone switched on and switched off.

2. Means and standard deviation of the dependent variables are documented for both groups or both time points or test statistics (t or F values).

3. Single-blind or double-blind experimental study design was applied.

4. Exposure by a GSM mobile phone specified at a range of ∼900 MHz to ∼1800 MHz under two conditions: on versus off.

5. Enrolled subjects were considered healthy.

6. The study must include at least one neuropsychological test that is used in another study, which fulfils requirements 1–4.

Nine studies did not fulfil at least one of these minimum requirements: Hladky et al,14 Preece et al,2 Smythe and Costall,3 Eliayhu et al,7 Lee et al, 16 Wilen et al,17 Haarala et al,15 Maier et al,18 and Krause et al.19

Ten studies fulfilled all the requirements mentioned above. Table 1 gives an overview of the design, sample size and sociodemographic variables, showing that three studies used a double-blind method; five single-blind; five used a repeated design while three others used independent variables, which were matched via education/intelligence or subjects were randomly allocated to either an experimental or control group. In total 555 subjects were tested; 470 subjects were exposed to GSM cellular phones and 421 were tested under sham condition. As the studies of Haarala et al10 and Haarala et al11 described different cognitive tests administered to the same samples, they will be treated as one study in the following.

The minimum requirement for the inclusion of a single test procedure in the analysis was that test results were provided from at least two different studies. The remaining 10 studies provided data in order to analyse the results of 10 tests comprising a total of 29 psychological parameters. The following provides a brief outline of the tests included in the meta-analysis:

• Simple Reaction Task (SRT): measured subjects’ reaction time to a stimulus (compare Haarala et al10).

• Choice Reaction Task (CRT): measured reaction time to complex stimuli. Subjects were shown a number and had to press the corresponding button (compare Haarala et al10).

• Vigilance (VIG): single capital letters were presented and subjects were to press a button when one of the target letters appeared (compare Russo et al13).

• Subtraction (SUB): subjects had to solve subtraction exercises, in which a one-digit number was to be subtracted from the number nine. Correct answers and reaction times were recorded (compare Russo et al,13 Haarala et al10).

• Sentence Verification (VER): measured attention. Statements, including numbers, were presented to subjects in the form of x<y, x>y, x = y. Subjects had to respond by pressing “yes” or “no”. Correct answers and reaction time were recorded (compare Haarala et al10).

• N-back test: measured short-term or working memory. Stimuli (capital letters) were presented, and subjects had to respond to target letters by pressing “yes” and to non-target letters by pressing “no”. There are four memory load conditions: 0-back, 1-back, 2-back and 3-back. Under these conditions the target letter was any letter presented 0, 1, 2, 3 trials back. Correct answers and reaction times were recorded (compare Haarala et al11).

• Trail Making Test: measured visual-conceptual and visual motor tracking skills. Subjects are asked to draw a continuous line connecting 25 circled digits/letters (compare Keetley et al9).

• Digit span forward and digit span backward: measured attentional capacity. Seven pairs of random number sequences were read aloud. Subjects were asked to repeat the same sequence in either the same or reverse order (compare Edelstyn and Oldershaw6).

• Spatial span forward and spatial span backward: measured attentional capacity. Increasingly larger amounts of information were presented. Subjects had to estimate spatial span using white cubes fastened in a random order to a board. In each trial the blocks were tapped in a prearranged sequence. Subjects were asked to repeat this sequence in either the same or reverse order (compare Edelstyn and Oldershaw6).

The meta-analysis was done as a group comparison between exposed and non-exposed subjects. The results of all studies were first transformed into their respective effect sizes, indicating whether there exists an effect of mobile phones, and then these effect sizes were transformed into the meta-analytic delta-measure (Δ). The analysis begin with the assumption that there is a common population effect in the different studies and that one single effect size could be determined for the different studies. When the assumed homogeneity for single effect sizes was not given (the variance of population effects was unequal zero, tested via χ² test), a random effects model was applied; otherwise computations were according to the fixed effects model. A correction of the estimators of the effect sizes based on the reliabilities was not possible because of missing documentation from the cognitive tests in the studies. As a result of missing documentation on the correlations, the effect sizes of repeated measures were calculated according to the formulae of independent samples, taking into account a reduction of the power.

RESULTS

Table 2 summarises the results of the meta-analyses. For all eligible parameters (except for vigilance (time) and digit span backward) the effect sizes of the studies can be seen as homogeneous and the mean effect size (Δ̄) describes the population effect. For vigilance and digit span backward the single effect sizes of the included studies are not homogenous and Δ̄ describes the mean effect size of the studies. The confidence interval depicts the range of the population effect based on the variance of the population effects.

Significant effect sizes were found in two tests: reaction time in the subtraction task was facilitated by exposure—that is, subjects reacted faster. In the N-back test under condition 0-back target reaction time was also facilitated. However, under condition 2-back target, reaction time was impaired from exposure. The errors in condition 2-back non-target are higher—that is, subjects make more errors under exposure. No significant effect sizes were found in all the other tests. All effect sizes are shown in figure 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Meta-analytic effect sizes (Δ̄) and their confidence intervals (diamonds represent fixed effects models, rectangles random effects models and the size of the squares indicates the number of subjects included). SRT, Simple Reaction Task; CRT, Choice Reaction Task; VIG, Vigilance; SUB, Subtraction; VER, Sentence Verification.

DISCUSSION

The meta-analysis confirms the assumption that cognitive performance measured by the subtraction task is mildly facilitated from EMF exposure. This effect appears only in reaction time, not accuracy. The significant effects concerning the N-back test for working memory show discrepant effect sizes: under condition 0-back target response time is lower under exposure, while under condition 2-back target response time increased. The number of errors under condition 2-back for non-targets appears to be higher under exposure. At the other levels of the N-back test no significant effect sizes were detected.

No effects were found in the following tasks: Simple and Choice Reaction Task (SRT, CRT), Vigilance (VIG), Sentence Verification (VER), all parameters of the N-back test except the three ones mentioned above; digit span forwards and spatial span forwards/backwards.

What conclusions can be drawn from this analysis? The results of the meta-analysis show that the original spectrum of cognitive abilities, which might be influenced by EMF emitted by mobile phones, can be reduced to two areas, which are measured by two tests: short-term memory (N-back test) and cognitive time needed in a mental arithmetic task (subtraction task). Furthermore, the effects seem to be so small that implications for human performance in everyday life can be practically ruled out.

However, several problems arise in the interpretation of our results. Firstly, the results of the N-back test seem to be paradoxical. On a low level of difficulty, there is a facilitatory effect, but on a higher one there is an inhibitory effect. From a scientific point of view, this conflicts with conventional patterns of dose-response relation. Another anomaly is that there are no discernable effects on the levels between 0-back target and 2-back target as well as on the higher level 3-back target. One explanation for this paradoxical result could be that Koivisto et al,5 whose study is included in the analysis, only used a single-blind study design. In such a design, non-optimal administration of the exposure source as well as the test procedure may influence the results.

Secondly, the problem of single-blind study designs also occurs in the subtraction task. The single-blind study of Koivisto et al4 shows the highest and per se the only significant effect size. Russo et al,13 Haarala et al10 11 and Curcio et al8 showed no significant effect size in the subtraction task.

Thirdly, the biological mechanism influencing the test outcomes is unknown. There are only vague speculations that potential negative effects may be due to a rise in brain tissue temperature, but empirical evidence is lacking.2 The absence of a biological mechanism complicates interpretation of the results.

What are the limitations of this study? In general, the administered meta-analysis is a very useful technique to objectively aggregate single results of a number of separate studies. It not only gives a conclusive overview about a certain field of research, but also overcomes the problem of reduced power due to small sample sizes by an enhanced pooled sample size, and therefore leads to more accurate estimations of the effect size. However, to conduct a conclusive meta-analysis the requirements already mentioned in the statistical section must be fulfilled. Moreover, the main problem was the fact that many instruments were applied in only one study and therefore could not be included in the meta-analysis—for example, spatial item recognition task (FACE) used by Eliyahu et al,7 word-recall task used by Smythe and Costall3 or tests from the Cognitive Drug Research (CDR) used by Preece et al.2 From a scientific point of view, it is not efficacious to apply new and non-standardised test procedures. Moreover, some studies omitted certain statistics—for example, Hladky et al14 and Smythe and Costall3 did not include means and standard deviations in their results. As meta-analyses will be used more frequently in the future, a statistical standard should be set, such as documenting effect sizes or correlations in the case of repeated measurements.

CONCLUSION

Summarising the current analysis, the following can be concluded: there may be two cognitive areas that are mildly affected by exposure to EMF by GSM mobile phones. Firstly, human attention measured by the subtraction task may be mildly affected in regard to decreased reaction time. Secondly, working memory measured by the N-back test seems to be affected. However, as the latter result cannot be explained by conventional patterns of scientific dose-response relation, it should be judged with caution. As both effects are quite small, the implication in everyday life can virtually be ruled out.

Further research should focus on attention and working memory using the same tests included in this meta-analysis. Basis requirements such as a double-blind study design should also be considered. A final conclusion—if possible—can only be reached on a biological level once it is clarified how electromagnetic fields influence cognitive functions.