4.1 PTSD symptom decreases immediately after the treatment but with high heterogeneity between studies
Effects of PTSD symptoms immediately after the intervention were assessed in all of the included studies. The results are shown in Figure 5. It indicates a high effect size of EEG neurofeedback (Cohen’s d = -1.28, CI: -1.76 ~ -0.80, p < 0.0001) with its upper confidence interval still in the negative boundary. However, there is a high heterogeneity among the studies (τ² = 0.4488, I² = 71.57%, p < 0.01).
4.2 Placebo effect not found
There were three studies [38, 42, 43] that had a sham control, hence these were separately analyzed to test whether a placebo effect existed. The result is shown in Figure 6. Even within sham control studies, the result shows a high effect size of EEG neurofeedback (Cohen’s d = -0.63, CI: -1.01 ~ -0.2, p < 0.05). Notably, the results showed a heterogeneity = 0 (τ² = 0, I² = 0%, p > 0.05). This may be attributed to the fact that all these studies used the same neurofeedback protocol (i.e. down-regulation of the alpha wave at the Pz region) and were all done with a similar population (i.e. treatment-resistant PTSD patients in Canada).
4.3 EEG-NF yields long-term effects
There were 3 studies [34, 35, 41] that conducted a 1-month follow-up and 4 studies [38, 39, 42, 43] that conducted a 3-month follow-up. To test for the long-term effects of the EEG neurofeedback, a meta-analysis was conducted for each 1-month and 3-month follow-up and the results are shown in Figure 7. Note that none of the studies conducted both 1-month and 3-month follow-up. One study that conducted follow-up after about 6 weeks of the intervention is included in the analysis within the 1-month group [35]. A 1-month follow-up showed a substantial effect of the EEG-NF (Cohen’s d = -1.72, CI: -3.13 ~ 0.30, p < 0.05) but with a high heterogeneity (𝜏² = 1.3533, I² = 87.17%, p < 0.01). A 3-month follow-up, on the other hand, showed a moderate but still high effect size (Cohen’s d = -0.72, CI: -1.07 ~ -0.37, p < 0.0001) with zero heterogeneity. Now, 3 out of 4 studies [38, 42, 43] included in the 3-month follow-up used the same neurofeedback protocol (i.e. alpha down-regulation at the Pz site) and treated a similar population (i.e. Canadian). However, at least one study [39] used a completely different neurofeedback protocol as well as population (i.e. AmygEFP protocol with an Israel population). This may indicate that the effects of EEG neurofeedback are constant after 3 months from the intervention whatever the population/intervention method was. However, the small sample size ultimately limits the interpretation.
Overall, these support that EEG neurofeedback can improve PTSD symptoms across a long period, up to 3 months.
4.4 Moderators
Given the high heterogeneity across studies, several within-study factors may contribute to this.
4.4.1 Year
Figure 8 shows the result of the meta-regression computed with the year of publication as a moderator. The amount of residual heterogeneity indicates the amount of heterogeneity existing among the studies when the effect of the moderator is subtracted. The year of publication could not explain approximately 55% of the heterogeneity (p < 0.05). Although the residual heterogeneity remains significant, it accounted for 45 % of the heterogeneity and the test for the moderator also reveals it to be statistically significant (p < 0.01). This indicates that it does impact the outcome. Studies show a reduced effect size as it gets more recent.
4.4.2 # of session
Figure 9 shows the meta-regression computed with the total number of EEG-neurofeedback sessions as a moderator. A total number of EEG-neurofeedback sessions could not explain approximately 71% of the heterogeneity (p < 0.01). Consistently, the effect of the moderator is not statistically significant (p = 0.1940). This indicates that the total number of sessions of EEG neurofeedback does not affect the outcome.
4.4.3 # of weeks, duration of session
Figure 10 shows the meta-regression done with the total number of weeks spent on the intervention and Figure 11 shows the meta-regression done with the duration of each session as a moderator. For both cases, there is still a significant amount of heterogeneity left (i.e. 80% and 73%) and the effect of the moderator is not statistically significant (p > 0.05). Thus these indicate that the total number of weeks for the intervention and the duration of the session do not affect the outcome.
4.4.4 Simultaneous Treatment
The effect of having a simultaneous treatment with the EEG-neurofeedback was evaluated. Three out of eleven studies were conducted with an experimental group taking no other treatment (psychotherapy, medication) except the EEG neurofeedback. The control group took no medical treatment at all. The remaining eight studies allowed their participants to continue their ongoing treatments for PTSD. Among those eight, three didn’t allow any other therapeutic intervention for their participants but allowed psychotropic medication. Another required all their participants, including both the experimental and control group, to take trauma counseling simultaneously as the EEG neurofeedback. The last four did not restrict any treatment for their participant to receive outside the experiment. All of these eight studies prohibited participants from changing their treatment during the duration of the experiment. Figure 12 shows the result of the meta-regression done with the moderator being whether a study allowed simultaneous treatment or not. Whether a study allowed simultaneous treatment or not could not explain approximately 49% of the heterogeneity (p < 0.05). Although the residual heterogeneity remains significant, it accounted for 51% of the heterogeneity and the test for the moderator also reveals it to be statistically significant (p < 0.0001). Notably, the result showed a bigger estimated effect size for the studies that did not allow a simultaneous treatment (Cohen’s d = -2.3377 for no-sim treatment; Cohen’s d = -0.949 for yes-sim treatment).
4.4.5 Targeting Pz
Five out of eleven studies targeted the region Pz, a region just on top of the midline parietal lobe, for their EEG neurofeedback protocol. Figure 13 shows the result of the meta-regression done with this factor as a moderator. Whether it targeted the Pz region or not could not explain approximately 65% of the heterogeneity (p < 0.01). Although the residual heterogeneity remains significant, it accounted for 35% of the heterogeneity and the test for the moderator also reveals it to be statistically significant (p < 0.0001). The result showed a bigger estimated effect size for the studies that did not target Pz (Cohen’s d = -0.8562 for Pz protocol; Cohen’s d = -1.4983 for non-Pz protocol). This also opposes the current trend (i.e. to target the Pz region).
4.4.6 Pz Targeted down-regulation of Alpha Wave
Four out of eleven studies used an EEG-neurofeedback protocol for decreasing the alpha wave in the Pz electrode site, a very common protocol in the current literature. Figure 14 shows the result of the meta-regression done with this protocol as a moderator. 45% of heterogeneity was explained with statistical significance(p < 0.0001). The result showed that despite its frequent utilization, the effect size was bigger for the studies that did not utilize that protocol (Cohen’s d = -0.7339, p < 0.05 for Pz alpha decrease protocol; Cohen’s d = -1.4879, p < 0.0001 for other protocols). However, again, there seems to be a statistically significant residual heterogeneity (τ² = 0.4488, I² = 54.94%, p < 0.05).
4.4.7 Increasing Alpha Wave
Figure 15 shows the meta-regression computed with the neurofeedback protocol, specifically whether its protocol aimed for an increase in alpha wave or not, as a moderator. Whether it aimed for an increase in alpha wave or not could not explain approximately 50% of the heterogeneity (p < 0.05). Although the residual heterogeneity remains statistically significant, it accounted for 50 % of the heterogeneity and the test for the moderator also reveals it to be statistically significant (p < 0.0001). Studies that aim for an alpha up-regulation yield about twice as much bigger effect size (Cohen’s d =-1.7090 for up-alpha protocol; Cohen’s d = -0.8262 for non-up-alpha protocol). Interestingly, this opposes the current trend (i.e. to downregulate alpha).
4.4.8 Feedback Modality
Finally, the feedback modality was evaluated as a moderator. Out of eleven studies, feedback modality was visual only, auditory only, visual + auditory (VA), or visual + auditory + tactile(VAT). The results are shown in Figure 16. Interestingly, feedback modality explained a remarkable 92% of heterogeneity. In detail, the estimated effect size for each modality is from the biggest, VAT = -3.3643 (p < 0.001), Auditory = -2.0186 (p < 0.0001), Visual = -1.1756 (p < 0.05), VA = -0.8258(p < 0.0001). The VAT seems to yield the highest effect while the visual-only feedback yields the least with auditory being the second next and VA being the third. However, the effect size of the VAT is based on only a single study with a small sample size (n = 10). Looking upon the other modalities, contrary to an intuition that more modality would lead to a better treatment outcome, a single modality (i.e. auditory or visual) yields a higher effect size compared to a combined modality (i.e. visual + auditory).