ADHD in adolescents. Common pediatric concerns.
--ADHD persists into adolescence for 78% of the children diagnosed with this condition, predisposing teens to many high-risk behaviors.

Modern electroencephalography: its role in epilepsy management.
--EEG assessments are important for answering specific questions that commonly arise in the management of seizure disorders

Event-related EEG/MEG synchronization and desynchronization: basic principles.
--Quantification of ERD/ERS is demonstrated topographically and temporally on various movement experiments.

Long-term intra-individual variability of the background EEG in normals.
--Long-term intra-individual variability across 25 months for most EEG parameters, esp. total absolute power and alpha mean frequency, was less than the inter-individual variability in the normal population.

Atypical frontal brain activation in ADHD school boys and girls.
--Compared to normals, ADHD boys exhibit less & ADHD girls more right-lateralization frontally; at both 4 1/2 and 8 years of age.

Toward a psychobiology of dissociation.
--Glutamate release may be involved in dissociative states. Acute and long-lasting consequences of traumatic stress exposure are associated with hyperglutamatergic states.

Alterations in brain structure and function associated with PTSD.
--PTSD is associated with reduced volume of the hippocampus and dysfunction of medial and orbital prefrontal cortex, brain areas important in memory & emotional regulation.

Costs of alcohol and drug abuse in the United States
--Alcohol and drug abuse was $246 billion in 1992, nearly $1K for every American; alcohol abuse and alcoholism accounting for 60%.

Psychotic side effects of psychostimulants: 5-year review.
--Of 98 children who received stimulant treatment, six developed psychotic or mood-congruent psychotic symptoms during treatment.

PTSD in injured adults
--PTSD occurred in 42.3% of injured adults 6 months after trauma and was related to assault, dissociation, female gender, youth, poor mental health, and prior illness.

Varieties of impulsivity.
--A series of psychopharmacological studies which measured selectively different aspects of impulsivity suggest that several neurochemical mechanisms can influence impulsivity; impulsive behaviour has no unique neurobiological basis.

Drug abuse treatment outcome study of adolescents in three treatment modalities.
--37 juvenile drug treatment programs in 6 cities surveyed: recommendations that programs be designed to address specialized issues such as comorbid substance abuse and psychiatric problems, family dysfunction, physical and sexual abuse, gender and ethnic differences, and academic performance.

Frontal midline theta rhythms and activation of prefrontal & anterior cingulate cortex
--Using magnetoencephalogram (MEG) and EEG, the authors identified that the appearance of Fz theta during consecutive mental tasks may reflect alternative activities of the medial prefrontal cortex and anterior cingulate cortex.

PREREQUISITES FOR EITHER PRACTICUM:
• Completion of 1 EEG Spectrum International Biofeedback Training Course for Professionals
• EEG Biofeedback experience using NeuroCybernetics Instrumentation

New Neurofeedback Clinicians / New Offices

Robert McRoberts, Ph.D.
7315 Frontage Rd, Suite 110
Shawnee Mission, KS 66204
(816) 444-4887

Judy N. Chiswell, Ed.D., OTR
Brain Matters
76 Chardon Drive
Buffalo, New York 14225
(716) 632 8675
j.chiswell@att.net

Sometimes science moves slowly. In fact, most of the time it moves slowly. Science Magazine just published a "News Focus" piece on the intrusion of Bayesian analysis into statistical thinking. Thomas Bayes was a Presbyterian minister in the 18th century, whose theorem on statistical inferences was unfortunately soon eclipsed. Bayes may finally be coming into his own, after more than 200 years.

This is particularly important for us, because if Bayesian analysis becomes common currency among statisticians, proving efficacy for neurofeedback will become a lot easier. The problem with Bayesian analysis is that it is not entirely free of assumptions. Its statistical inferences therefore always remain approximations. By contrast, the prevailing method of statistical proof, term "frequentist" analysis, makes no such assumptions. By prevailing standards, one must prove efficacy by means of a chosen group of matched experimental and control subjects, and one must rule out the "null hypothesis" to a certain level of assurance, typically p=0.05. This approach obscures two important considerations, at a minimum. The first is that it is often easy to establish "statistical significance" with large group sizes, and the second is that no account is taken of the fact that one can have a small effect size with high assurance, and yet the game may not be worth the candle.

The Science article illustrates a Bayesian approach by the following example: Imagine a precocious baby seeing its first sunset. The problem is predicting the likelihood of a sunrise to follow. Absent any knowledge, the baby assigns an equal probability to the sun rising or not rising. A white marble represents the one event, a black marble the other. Both are placed in a bag. Every morning, upon sunrise, another white marble is placed in the bag. After a while, the white marbles will completely overwhelm the single black marble, and one may deduce a high probability of sunrise occurring. The fact that the initial starting condition of one black marble along with one white one represents a misjudgment of actual conditions progressively becomes negligible.

This is of course very similar to the process by which we reach assurance in the clinical world that we are actually accomplishing something. We are all closet Bayesians. In our minds, we initially assign a low probability to our ability to help in a particular condition, say bruxism, or bedwetting. So to start with, a black marble goes into our virtual bag. If we actually help the person with his or her bruxism, a white marble goes into our mental collection bag. After a while, we see the white marbles outnumbering the black ones. Throughout this process, we have never focused on bruxism itself, or solicited patients for the treatment of bruxism. We have simply been data-gathering on cases where bruxism happened to be an issue along with whatever we were addressing.

We then face the issue of whether to actually "go public" with the claim that we are able to help with bruxism. Here again, a Bayesian analysis is implicitly done. We are really making a judgment with regard to whether neurofeedback is "worth trying," rather than whether it will actually be effective. After all, we are not dealing with breast surgery. A person can always stop doing neurofeedback and will only have wasted time and money. Hence, there is not a large "hurdle" to overcome for someone to say yes to neurofeedback, and for us to recommend a trial. There is another factor at play, however, and it is the availability of alternative approaches. If there are other effective treatments for bruxism, we must ethically ask whether our expensive and time-consuming approach is in fact preferable to the alternatives. Bayesian analysis is once again appropriate here, and we implicitly make that calculation non-rigorously in our heads.

When there are no other attractive alternatives, such as in cases of fetal alcohol syndrome, autism, and attachment disorder, the mental hurdle to overcome is very small. Even if there are alternatives, they could have high "hurdle factors" on their own, e.g., brain surgery for seizures, or electroconvulsive shock therapy for depression. Here we would also end up with a strong bias toward trying neurofeedback first, even if the probability of our success is modest.

In the case of ADHD, where viable alternatives such as stimulant medication exist, the "hurdle factor" is much higher. Here it is relevant that sometimes we don't achieve training results (in TOVA terms, for example) which entirely match what can be achieved with Ritalin, particularly when it comes to reaction time and variability. This question can again be subjected to Bayesian analysis. This is one reason that TOVA data are so very important here.

This discussion becomes relevant when we consider a reviewer's response to a recently submitted paper of our latest TOVA results on over 1000 subjects. The hackneyed criticism was that the work did not follow the standard experimental design. In particular, it did not restrict itself to narrowly selected clinical categories, e.g. those actually diagnosed with ADHD. Again, this criticism can be answered with a Bayesian argument. Since we established that there was no statistically significant difference either in pre-treatment condition or in outcomes between those who had been formally diagnosed with ADHD and those who had not, the selection of only ADHD subjects was clearly not mandated. In fact, this dispenses entirely with one of the arguments always raised against us, namely whether the subjects had been well-diagnosed. The fact that the findings hold true irrespective of subtlety of diagnosis makes them much stronger. Those for whom current methods of experimental design and data analysis are sacramental aspects of their scientific religion will not be impressed by these arguments. That's why an acceptance of the Bayesian methodology for statistical inferences is so important.

In fact, the study we submitted can best be thought of as a kind of epidemiological study, with neurofeedback being the antigen that attacks the prevailing paradigm of ADHD as a biochemically mediated disorder that by its very nature can only yield to pharmacological remedies. The null hypothesis is that this antigen only purports to improve attentional variables, but does not in fact do so. This null hypothesis is cleanly ruled out by the results of the study. As mentioned above, however, we are not merely contending against a null hypothesis here. The world offers both stimulant medication and nutritional and other approaches. The Bayesian analysis of when it is appropriate to recommend neurofeedback becomes much more complicated in this larger space.

At the moment, critics of neurofeedback training still hold to the null hypothesis with respect to neurofeedback. They could readily change their minds on this without yielding the essential ground that "stimulant medication approaches are preferable." It would be a mistake, therefore, to put huge resources into ruling out the null hypothesis to their satisfaction, because that will not really get it done. (Additionally, the issue is not as much about the data as it is about mechanisms---just as in the case of continental drift. These folks have no idea how neurofeedback might actually work, given their assumptions about ADHD. So they are not inclined to be attentive to the data.)

We must instead set the higher goal of showing that we can do things that cannot be accomplished with stimulant medication, that we can meet the needs of children who are not responsive to stimulant medication, and that when we make a direct comparison with stimulant medication, our data don't suffer in comparison. That is the real battleground. And in this comparison, Bayesian analysis will allow us to extract the maximum impact from every data point, because in this approach, every data point matters.

It is in the Bayesian framework that the incidental anecdote receives full value. Going back to the case of the baby learning about the sun, the most important data point was the first morning when the sun came up. All subsequent data points were of lesser import. By contrast, the bias of existing methodologies is to discount the isolated data point until it is imbedded in a statistically sound study. This is of course nonsense. When Norman Cousins survived his predictably fatal condition, he judged that he had successfully mobilized the resources of self-recovery with humor. Well, whether or not his survival is attributable specifically to his "humor cure," his survival is still a significant data point for the existence of mechanisms of self-recovery that have yet to gain the attention they deserve from scientists.

The primary casualty of existing statistical approaches to clinical studies has been the "anecdotal case." It goes without saying that every observation of scientific anomalies in medicine starts with someone violating the mandate to "disregard the anecdotal evidence." Anecdotal data are the mulch of new scientific departures. The mandate to trash anecdotal evidence cannot possibly come out of "medicine as a scientific discipline." It can only come from "medicine as a body of accepted practice." Every instance of the vaunted "placebo effect" and every case of "spontaneous recovery" can be reframed as "the body did it," or, in the modern idiom, "the body-mind did it." That should naturally lead to the question, "How did the body-mind do it?" This question must be confronted case by disagreeable case. That's what Bayesian statistical inference will finally liberate us to do, since every event ends up being measured against prior expectations for its occurrence.

As our growing collection of case histories becomes "sufficiently large", it, combined with Bayesian analysis of the findings, will ultimately be our bulwark against critics.