This newsletter was first published in NeuroConnections, the joint publication of the AAPB and ISNR, Winter 2011
This year’s ISNR meeting seemed to have more invited speakers who were comfortable talking neurofeedback. In the past, one had the feeling that some presenters were there mainly to collect their speaking fees, and were not really prepared to engage with us on our core assumptions. There is a rising tide in the neurosciences that is lifting all boats, even ours. The conversation is shifting toward a language of networks, of structural and functional connectivity, as the key issue in psychopathology, and toward neuromodulation as a strategy for functional recovery. And there we are, having occupied that space already for some decades.
What has allowed this shift to occur—and it is really just getting underway—is the fact that the discussion is taking place on the rather dry ground of neurophysiology as opposed to our historical battleground of clinical claims. The latter exposes us to a surround of cat’s claws that inhibit progress. The contrast between these two frontiers was dramatically on view one day at the conference, when in quick succession we had a presentation by Nick Lofthouse on plans at Ohio State for yet another ‘definitive’ study of neurofeedback for ADHD, a presentation by Andrew Hill on physiological changes with neurofeedback, and a presentation by Martijn Arns on using physiological measures to judge outcomes for ADHD and depression.
Andrew Hill’s study, which tracked Event-Related Potential (ERP) data along with Event-Related Spectral Perturbations (ERSP) through training with some of our favorite old protocols for ADHD, for comparison with sham training, demonstrated that physiological measures can readily distinguish active training from sham. Functional changes with training were tracked with the Lateralized Attention Network Test (LANT) of Michael Posner. The results of functional testing were nothing to write home about, but then the training only involved five sessions and the trainees were normally functioning at the outset. Significantly, the physiological measures were able to distinguish between the different protocols being used, which were our favorite old C3beta and C4SMR, along with C3SMR and sham.
With respect to the lukewarm results of the functional testing, I cannot resist interrupting the narrative to note that the whole point of having several protocols to choose from was to match the protocol to the client. Research designs invariably treat the different cohorts as homogeneous, a wildly inappropriate assumption. As soon as individualization of protocols is allowed, then it does not take long to discover that responses to training are sometimes highly specific in terms of reinforcement frequency as well. One of the key differences between the clinical world and the research world is that the astute clinician does not continue to beat a dead horse, whereas the researcher remains wired to his original research design. If clinical success means abandoning the assumption of static protocols, then the clinician loses no sleep over the matter. But back to the story.
The surprising finding was that changes in ERPs started showing up even in the first session, and were confirmed in the fifth session. Changes in ERPs and ERSPs distinguished training groups from the sham group. Effects on ERSPs were lateralized, being stronger on the training side. Resting eyes-closed band amplitudes in theta, alpha, SMR, and beta all distinguished between the active and the sham groups.
The study implies that physiological measures can readily distinguish active training from sham. This, it seems to me, disposes of the placebo argument wholesale. Lasting effects are seen which differentiate protocols from each other, and collectively differentiate them from sham. Why then do we still offer the greatest deference to those who would still, at this late stage, breathe life into the placebo hypothesis of neurofeedback?
So we come next to the proposal out of Ohio State to perform the definitive sham-controlled study of neurofeedback for ADHD, and many in our field are still in fibrillation about the prospect of having our collective fate tied to the outcome of a single such study. Now Nick Lofthouse certainly appears to be a competent, well-meaning researcher. But what is the best possible outcome of such an enterprise? It is the finding that EEG feedback, done in the traditional manner, is indeed better than doing nothing. This will hardly be enough to move the NIH off its pharmaco-centric posture. A lot of effort is being mounted here for small beer.
What if we took the approach of arguing the case for EEG neurofeedback in a physiological frame? The basic argument that physiological function can be enduringly altered with feedback on physiological variables has by now been very well established by the biofeedback community. Is the EEG among the physiological variables that are suitable for such training? All of the EEG feedback studies collectively make the case that it is. Now most biofeedback modes are aiming at system functioning rather than specific diagnoses, most particularly the auto-regulation of the autonomic nervous system. Are matters different in EEG feedback? They can be, of course, as in the targeting of something so specific as dyslexia or auditory processing disorder. But that is not where we have distinguished ourselves to date, by and large. In most current applications of EEG feedback, the benefit derived with respect to the discrete disorders is best explained in terms of better functioning of core regulatory systems, i.e. at the systems level. Here we are referring to improved regulation of tonic arousal level, to more appropriate regulation of affect, to better autonomic balance, to enhanced cerebral stability, to appropriate interoception, to improved executive function, etc.
And how is the cause of improved functioning at the systems level served by EEG feedback? A good working hypothesis is that we are directly affecting the functional connectivity of our resting state networks. This is readily testable. In support of this hypothesis one could already offer all of the evidence that neurofeedback—by various approaches—dramatically alters, and generally improves, coherence relationships within cortex. This occurs irrespective of whether or not coherence relationships are explicitly targeted in the training. This finding is neither tied to diagnosis nor to disorder. It is not tied to level of initial deficit. Most notably, it is not tied to protocol. Some protocols indeed work better than others, but those that are effective clinically can all be shown to affect coherence relationships favorably.
This means that group studies are not actually needed to prove the essential claims of EEG feedback. It means that all of the case data accumulated by clinicians to date can collectively make the case for the core hypothesis. This is in effect what has happened. Our evolving convictions about what is possible with neurofeedback have been consolidated largely without the benefit of group studies with standard controlled designs. Likewise protocol development has progressed—where it has done so—entirely on the basis of individual case observations.
This is not a flaw. Protocol refinement is really better served by working with individuals over time, with intra-individual variability as the most significant confound. The question of optimizing reward frequencies, for example, has to be examined within individuals rather than via group studies. And protocol refinement is in the best of hands with real clinicians, as opposed to naïve researchers recruited from academia. The research world has not come to terms with this, so in that respect we have been fortunate in not having the NIH regiment the development of this field over the past few decades.
The core claim, then, is that neurofeedback can positively influence any brain function that is governed in any fashion by our resting state networks, provided of course that structural connectivity networks are sufficiently intact to support the change. Again, all of the clinical work to date can be brought to bear in support of this proposition.
With this background we turn to the presentation by Martijn Arns on the value of using physiological variables such as ERPs to document change with neurofeedback for ADHD and depression. No quarrel there from this quarter. Now with regard to the work with depression, Arns referred to results achieved with repetitive Transcranial Magnetic Stimulation (rTMS) rather than neurofeedback. He motivated this by declaring that neurofeedback was ineffective in application to depression. I was startled by this declaration and took the matter up with him after the talk. He pointed to the lack of published group studies in his defense. I counter-argued that he was making an affirmative statement, which required its own affirmative evidence. As we are constantly reminded, absence of evidence is not evidence of absence. The proper statement would have been to declare the matter as still unsettled.
Now on the basis of the ‘grand hypothesis’ above, one would certainly expect neurofeedback to be effective for depression. If it is not, then that would indeed be fatal to the maximal position. So what is the evidence in this regard? It turns out that the first reports of benefit of standard SMR/beta neurofeedback for depression came out of Sterman’s laboratory, where UCLA graduate students recruited for the training routinely reported an easing of depressive symptoms. Sterman never made much of this because that had not been part of the formal hypothesis, but one might also argue that the evidence is more robust for not having been looked for at the outset. Margaret Ayers certainly saw the recovery from depression in connection with her work with TBI. We saw benefit for depression in our early work as well, with a slight modification of the Ayers protocol, and we started saying so openly in the 1990-1 timeframe.
This declaration was met with considerable skepticism and even hostility at the time. This was when Lubar was still making the case that neurofeedback was for ADHD and not for anxiety and depression. Not long thereafter, Peter Rosenfeld gave the idea of neurofeedback for depression scientific respectability with his hemispheric asymmetry protocol, modeled after Davidson’s work on the hemispheric bias with respect to approach and withdrawal. The training didn’t really work very well in the general case, so the technique has gradually faded from view. That development might in turn have dampened the expectation that neurofeedback would be helpful for depression. If the ‘scientific’ protocol didn’t work very well, then what hope was there?
Meanwhile, however, the early approaches just kept on working. More recently, Cory Hammond has come out with a depression protocol that looks remarkably like what we were doing twenty years ago—namely beta1 training with a frontal or pre-frontal, left-hemisphere bias. That stands as confirmation of a sort for our early protocols. For our part, we realized that the most intractable cases of depression, namely agitated depression and suicidal depression, required mainly right-hemisphere training, which took us some years to develop. More recently, data on veterans with PTSD shows depression scores being cut in half within ~six sessions in close to 80% of trainees. Suicidality has remitted in nearly all cases to date. Within the clinical community, depression is not being discussed as a thing apart. It is not either more or less intractable than other conditions encountered in EEG training. So depression is not a counter-example to the more inclusive vision of neurofeedback as potentially aiding all conditions grounded in the disregulation of our resting state networks.
In reflection on the ISNR Conference, I was most riveted on the invited talks, but when it comes to progress within our discipline, the critical contributions are being made by the people who have committed themselves to this field. This field has been built, and continues to be built, largely on the contributions of scientist/practitioners and their technical support teams, the developers and manufacturers. The discipline has been built from the bottom up, and it is moving into a ‘crowd-sourcing’ phase, with the enlargement of the central core of contributors. This is a very healthy development, and there is no need for the NIH to come in for a do-over and a re-write.
That is not to say that there isn’t a useful role that the NIH might play. In his review of imaging technologies in Science in 2009, Karl Friston lamented that by their very nature as resting state networks, these could not be readily subjected to experimental manipulation without disruption. One is reminded of the dilemma that a quantum mechanical system cannot be probed without affecting it. Those who have followed that field know that that is no longer entirely true. And neither is it true of our resting state networks. They can be probed with neurofeedback. The technique would be used to subtly induce state shifts, and the resulting changes in resting state functional conformation tracked. This would establish the sound science that would benefit this field. As for the clinical frontier, the research perspective should be broadened, not narrowed. Perspective is needed at top levels on the full dimensions of current diverse practice within the field. An anthropological approach should be taken in which researchers formally study the client-clinician dyad in its state of nature with the full variety of neurofeedback approaches currently flourishing in clinical application. It will be quickly appreciated that neurofeedback will alter the face of mental health profoundly. Given the current state of our society, that cannot happen soon enough.