The recently published study by Rossignol and colleagues about hyperbaric oxygen therapy (HBOT) for Autism has generated lots of commentary and is sure to lead to more. Because it is a treatment study and employs more careful methods than are common in many of the therapies promoted these days, I sat up and said, “Hmm. I ought to read this one.”
So I did. And I found it to be, indeed, a cut above much of the ersatz research that’s passed off as evidence in the Autism arena. But, I found some concerns, too.
Those concerns led me to poke about a bit on the Internet to see whether there were any others who were raising questions. There are. And I still have some more poking to do. But, I thought I ought to record my concerns. Thus this post.
Steven Novella of the New England Skeptical Society (and Evidence-based Medicine) discussed several issues about the study. He questions whether
- The sample is large enough to merit robust statements about effects. One reason for employing large samples is to ensure that researchers do not overlook small but significant effects; in this case, there were substantial effects on some measures despite the small number of participating children. Although this point mitigates the importance of small size of the sample in this study, there is a pesky issue about the variables on which the researchers found effects.
- The participating parents may actually have been aware of whether their children were receiving the active treatment; that is, they may not have been “blind” to about the treatment. Although this is a problem in itself, it is compounded by the fact that some of the measures of improvement in children (rating scales) were provided by the parents.
- The brevity of the time period in the study raises questions about whether, if there is a true effect, the benefits would be sustained. That is, the study could have discovered true-but-temporary improvements.
Over at Leftbrain Rightbrain, Do’C noted a concern with the nature of the independent variable (i.e., “treatment”). Do’C examined the prevailing barometric pressure at the locations of some of the sites where the therapy was administered and found that at higher-altitude sites it was likely that the children and their parents were subjected to lower levels of pressure than is reported in the study. To some extent, this may simply be a matter of rounding error; more importantly, it may indicate that even lower dose levels may result in theraputic effects. (Note that I used the word “may” in a neutral, plain way there.)
These are important and reasoned considerations when reviewing the Rossignol et al. (2009) study. There are other issues on which other commentators may have touched, but that I’ve missed. Among those others concerns are several that I develop in the following paragraphs.
The participants included 62 children (and their families) drawn from 66 who were evaluated. They were drawn from 6 clinics apparently affiliated with the authors. I’d like to know more about this sample. Because many of the authors of the study are affiliated with clinics offering HBOT, it is likely that the sample is composed mostly of families that sought treatment at these clinics. The extent to which these families are representative of the families of children with autism is unknown. To the extent that they represent a specially selected sample of the population, the participants in this study make it impossible to generalize the results of the study to the full population. It would be helpful if the authors would explain how the participants were selected and provide additional data about the characteristics of the applicants for their services. In addition, it is important that the research team disclose fully their relationships with the participants’ families.
The measures used in the HBOT study are reasonable choices. The Clinical Global Impression (CGI) Scale is a widely used rating system. The Aberrant Behavior Checklist—Community (ABC) is an extension of an instrument originally developed for use as a means for capturing diagnostic indicators in in-patient settings where professional caregivers made ratings; in this instance, it was rated (with precedence) by parents or primary caregivers as on outcome measure. In contrast, the Autism Treatment Evaluation Checklist was designed to provide an estimate of treatments’ effectiveness; the developers’ Web site (Autism Research Institute) reports overall reliability greater than .90 (which is good) and variable but pretty high reliability for the subscales (.82-.94). I have not had the time to complete a careful review of the psychometric characteristics of these instruments (reliability, validity), but my general impression is that they are basically trustworthy. (If someone would drop a comment based on reviews from strong sources, that would help.) Perhaps the most substantial limitation to all of these instruments is that they are ratings made by people who are familiar with the participating children; they are essentially subjective rather than objective. To be sure, both children in the treatment and the control groups are rated subjectively, reducing the potential problems with the subjectivity (they are presumably equally subjective), but they do not provide the kind of objective data that produces powerful results.
Rossignol and colleagues employed procedures to mask the presence or absence of the active treatment (e.g., simulated HBOT for the controls; covering control switches). They also tested the conditions with a small number of adults and reported that these people did not report they could distinguish the conditions. These steps increase the trustworthiness of the manipulations.
|Measure||Experimental Group Mean||Comparison Group Mean|
|ABC Post-test||46.4 ± 24.7||45.5 ± 17.3|
|ATEC Post-test||65.9 ± 16.4||70.1 ± 21.9|
Table of Means (with SEM) for total or overall score on two of the measures.
I am not a biostatistician, so my reflections on the procedures for analyzing the data are only observations, not a formal critique of the analysis. (Where are the epidemiologists on this? Epi Wonk? Photon?) Because the study was a randomized trial, it seems to me that the foremost analyses should be those comparing the levels of the dependent variables between the two groups. However, for each of the dependent variables, the authors lead their results with changes from pre- to post-test. The differences between the groups on the total scores (i.e., the most reliable scores) on the ABC and the ATEC are not significant. Table 1 shows the means and Figure 1 represents them graphically (with error bars); the data are from a supplemental table in Rossignol et al.
It appears that the authors depended on the gain scores to make their case. In addition, they conducted many statisticl tests, thereby increasing the chances that they would find at least some significant findings. Do they have a study-wise control for the number of significance tests they ran?
Here are my tentative conclusions:
- Does this study demonstrate that the therapy is effective? No.
- Is this study so fatally flawed that it should be ignored? No.
- Should parents seek and professionals recommend HBOT based on this study? No.
- Does the therapy merit additional studies using rigorous methods? Yes.
Some research suggestions
To establish that HBOT provides benefits to children with autism, it will be important to have studies of it that
- Examine representative samples of children with enough participants to permit testing of both alternative hypotheses (i.e., multiple control groups) and potential participant characteristics that might interact with the HBOT treatment (e.g., degree of involvement).
- Collect objective measures of behavior of importance. (This is not to say that those in the Rossignol et al. study are not important, just that additional measures would help understand the effects of the therapy.) Among those that should probably be considered would be indicators of learning performance (e.g., trials to criterion on some sensible tasks), objective measures of the qualities of social interaction, forth, frequencies of stereotypies, and so forth.
- Incorporate controls that examine critical alternative explanations (e.g., parents are not blind to treatment conditions) Although it was valuable to include what appears to be a placebo therapy in the the original study, to assure scientists that the active ingredient is, indeed, the HBOT it is necessary to have conditions that discount other features of the therapy.
- Conducted by independent researchers, those who will receive no financial gain if the therapy proves effective.
- Dr. Novella’s entry, “Hyperbaric Oxygen for Autism“;
- Do’C’s post, “Autism, HBOT, and the new study by Rossignol et al.“;
- ARI’s page on the ATEC.
References about the utility of the ABC
Updated 17 November 2009: Corrected typographical and style errors.