Institutional Review Board Approval
This study was conducted in accordance with the Declaration of Helsinki and the Institutional Review Board. Since the survey was anonymous and did not contain any unique indentifying information or protected health information, the study qualified for category 2 exempt status according to 45 CFR 46.101(b). Information regarding Institutional Review Board approval and contact information was provided on the first page of the survey.
The first and last authors drafted an invitation letter and survey. The survey was designed using the principles outlined by Keenan  as implemented with the SurveyMethods, Inc. website software (http://www.SurveyMethods.com). Straightforward unambiguous non-open-ended questions were used when possible. The survey was designed to be comprehensive in the variety of possible responses so as to eliminate the need for free text entry. In order to reduce perceived bias of the survey for a specific type of treatment, the survey was designed to include a wide range to treatments commonly used in the treatment of ASD, both traditional treatments for seizures as well as non-traditional treatments.
The first page of the survey provided the responder with basic information regarding the purpose of the survey, a statement regarding the host institution and regulatory approval from the institution and an approximate completion time. The second page of the survey provided the responder with more specific information regarding the structure of the survey, the correct manner in which to respond to specific questions and contact information for the principal investigator if the responder wanted more information or had any questions. The survey did not record any identifying information of the responder nor did it ask for any protected health information identifiers. The survey was designed to be easy to answer with validation checks for each response and used skipping logic to make the survey easy and efficient to answer. Except where noted, answers were providing by multiple checkbox, rating scales, yes/no questions or specific fill in the blank questions. Very few open-ended questions or questions that required free text entry were used.
Some basic information was collected about each child with ASD including current age, gender, spectrum diagnosis and developmental profile, commonly associated medical conditions (a text box was provided for entering other medical conditions not listed), and the practitioners that manage the child's medical and developmental disorders.
Specific information regarding seizures was also queried. Responders indicated practitioners that managed or rule-out the seizure disorder and the type of test used to diagnose or rule out seizures. Responders indicated whether the individual with ASD had any of the following seizure types: generalized, partial complex, absence, typical or atypical Landau-Kleffner syndrome, subclinical epileptiform discharges, Lenox-Gastaut syndrome and/or infantile spasms. These choices were provided in a checkbox fashion so that multiple seizure types could be indicated. Next to each common seizure type was a detailed description of the seizure type to help the responder select the correct seizure type. There was also a checkbox for the responder to place other seizure types if the seizure types listed were not sufficient. Other information regarding seizures (including age of onset and age of resolution of seizures if the seizures resolved) was also collected.
Following collection of this basic descriptive information, the respondents were asked to indicate if their child had been treated with a wide range of treatments. Information for each treatment was collected in a sequential manner. For each treatment, a page would appear with a yes/no question asking the respondent whether a specific treatment had been used. Both generic and brand names, inclusive of all known brand names, were included in the question. If the respondent answered 'yes' they were directed to a page where they could rate the perceived effect of the treatment and list the adverse effects. If they answered 'no' the respondent was questioned about the next treatment. This skip logic in the software eliminated the need for the respondent to understand which questions needed to be answered and which needed to be skipped, thereby eliminating potential confusion that can occur with conditional questions in surveys. Additional pages asked the respondent if their child was treated for seizures with a treatment that was not mentioned. If the respondent answered 'yes', a page appeared which included a text box to enter the information regarding the treatment along with a page to rate the perceived effect of the treatment and any adverse effects.
For each treatment, the respondent was asked to rate the perceived effect of the treatment on seizures, sleep, receptive and expressive language, verbal and non-verbal communication, stereotyped/repetitive movements, rigidity, hyperactivity, attention and mood. A seven point scale was used that ranged from a substantial negative effect, a moderate negative effect, to a mild negative effect to no effect to a mild positive effect, a moderate positive effect, to a substantial positive effect. All ratings for this complex multi-dimensional construct of the perceived treatment effect were included on the same page to facilitate the respondent's use of the same psychometric scale. The rating scale was designed to be symmetric (i.e., same number of positive and negative ratings) in order to minimize response bias. Three sets of text boxes were provided to enter any adverse effect, the severity of the adverse effect and the frequency of the adverse effect. This allowed the respondent to enter up to three adverse effects.
The survey and participant invitation letter were developed with the advice of a wide variety of experts with experience in the treatment of children with ASD. A copy of the invitation letter and a link to the initial survey were sent to participants of the Elias Tembenis Seizures Think Tank (which took place at the AutismOne Meeting in Chicago in May of 2009) approximately one month before the think tank. The participants of this think talk represented a wide variety of practitioners who treat ASD (See See Additional file 1, Appendix A). Participants were asked to complete the survey as many times as necessary to get familiar with the survey. During the day-long think tank the first and last authors led a discussion querying the participants on their opinion regarding the specific information about the children with ASD and seizures, the treatments that should be surveyed and the specific effect of each treatment that should be asked as well as the wording of the survey. Over the weeks following the survey, changes were made to the initial survey and invitation letter as a result of the suggestions of the members of the think tank. Participants were again asked to complete the survey and provide additional comments by email. The survey was again modified and the participants were again asked to review the survey. After no further suggestions were made, the first and last authors recruited volunteer parents with children affected by ASD and seizures to review the survey and invitation letter. The survey and invitation letter were sent to these volunteers and their suggestions were integrated into the survey and letter and the modified survey and letter were again sent out for review to these volunteers. After no more significant comments were received, the final survey and invitation letter were prepared.
The expert group decided on including the following treatments for the survey. Traditional seizure treatments included valproic acid, phenytoin, lamotrigine, levetiracetam, caramazapine, topiramate, oxcarbazepine, pyridoxine, clonazepam, phenobarbitol, zonisamide, gabapentin, felbamate, ethosuximide, tigabine, primidone, vigabatrin, neurofeedback, ketogenic diet, Atkin's or modified Atkin's diet, steroids, vagus nerve stimulation, surgery, intravenous immunoglobulin, transcranial magnetic stimulation/direct current stimulation. Non-traditional treatments included gluten free casein free diet, specific carbohydrate diet, hyperbaric oxygen therapy, 5-Hydroxytryptophan, gamma-aminobutyric acid, dimethylglycine, taurine, chelation therapy, co-enzyme Q10, B6, gluatathione, magnesium, B12, L-carnitine/Acetyl-L-carnitine, L-carnosine, minocycline, bacopa, actos, and spironolactone.
Although children were likely provided multiple treatments at the same time, information regarding response to specific treatment was queried individually for each treatment. This assumes that each treatment is having an influence independent of the other treatments. The authors and the expert group believe this was a necessary limitation since asking about each combination of treatments would create a questionnaire that would be prohibitively long and complex. In addition, it is likely that the number of respondents with experience with specific treatment combinations would be prohibitively small for a valid analysis. From a practical point of view, most practitioners usually start and/or stop treatments independent each other so that the clinical effect (and adverse effect) can be determined for the specific treatment.
Questions Included in the Survey But Not Addressed In This Study
In addition to questions regarding seizures, the survey contained a small section in the beginning that asked about the effect of allergies and season on seizures and behavior. Additional questions within the survey also asked about over- or under-reactivity to external stimuli and the effect of treatments on such reactivity. These aspects of the survey are not addressed in this manuscript.
In order to determine a baseline for the data collected on children with ASD and seizures, a control survey was developed to gather information regarding children with ASD without seizures (henceforth described as the control survey). This control survey contained all of the questions that the seizure survey contained except for specific questions about seizures (e.g., "What type of seizures has your child been diagnosed with?"). All treatments in the seizure survey were included in the control survey but a rating for the effect of the treatment on seizures was not included. Questions were included regarding whether the child had been evaluated for seizure, what type of practitioner evaluated the child and what test, if any, had been done to rule-out seizures.
An invitation letter (See See Additional file 1, Appendix B) for the on-line survey was posted on the website and in email newsletters of the Autism Research Institute (ARI) and approximately 30 local and national ASD support groups for parents of individuals with ASD. The non-ARI support groups included Autism Speaks, several local chapters of the Autism Society of America and ARC and other local support groups for families of autism. The letter specifically asked parents of children with ASD both with and without seizures to follow one of two web links depending on whether or not their child had clinical seizures, subclinical epileptiform discharges or seizure-like activity. These web links activated different surveys located on the SurveyMethods, Inc. website (http://www.SurveyMethods.com). Identical letters were used for ARI and non-ARI websites except that the links referred to surveys that stored responses in different databases. This allowed responses to exact same seizure or control survey questions to be stored in different databases depending on whether the respondent had followed the link from the ARI or non-ARI webpage.
Survey Response Reduction
The frequencies of specific genetic conditions were very low so all responses for specific genetic disease were included as a general genetic condition response. Treatments with less than 20 total responses were excluded from all analyses. These included surgery, transcranial magnetic stimulation/direct current stimulation, tigabine, primidone, vigabatrin, neurofeedback, minocycline, bacopa, actos, and spironolactone. Treatments with less than 20 responses for the subclinical seizure group were excluded from the subclinical seizure treatment analysis. These included ethosuximide, phenytoin, clonazepam, gabapentin, zonisamide, felbamate, phenobarbitol, vagus nerve stimulator, intravenous immunoglobulin, hyperbaric oxygen therapy, dimethylglycine, gamma-aminobutyric acid, and specific carbohydrate diet. For the subclinical seizure treatment analysis, responses for the ketogenic diet and Atkin's or modified Atkin's diet were combined because of their similarity in order to prevent elimination due to too few responses.
Chi-squares were used to analyze bivariate variables. To mitigate the effect of multiple comparisons, for each set of comparisons made, the Bonferroni correction was calculated to correct the alpha cutoff. The caption of each table explains the appropriate Bonferroni correction.
Ratings were converted into an ordinal scale ranging from 1 to 7 for analysis: substantial negative effect (1), a moderate negative effect (2), a mild negative effect (3), no effect (4), a mild positive effect (5), a moderate positive effect (6), and a substantial positive effect (7). Although the response scale was ordinal, the response distribution was found to be normally distributed allowing the use of parametric analyses for treatment ratings.
Comparing every treatment to every other treatment would result in a very large number of comparisons (>500), thereby considerably increasing the probability of a type I error. To reduce the number of comparisons we investigated whether multiple treatments could be clustered together into treatments that showed the same pattern of ratings for seizures and other clinical factors (i.e., sleep, communication, behavior, attention, mood) and then compared the ratings from each treatment cluster to other treatment clusters.
For the cluster analysis, seizure types were divided into two broad categories: clinical (generalized, partial complex, absence) and subclinical (typical and atypical Landau-Kleffner syndrome, subclinical epileptiform discharges). Lenox-Gastaut syndrome and infantile spasms were not considered in this manuscript. For the cluster analysis, two summary scales were calculated to reduce the number of rating scales: receptive and expressive language and verbal and non-verbal communication ratings were averaged to create a rating called communication and stereotyped/repetitive movements, rigidity and hyperactivity ratings were averaged to create a rating called behavior. For subclinical seizures we did not include the seizure ratings in the cluster analysis as the primary manifestations of subclinical seizures are other clinical factors and seizures are not reliably detected in individuals with subclinical seizures.
Cluster analysis was conducted using Ward's technique . The Ward's technique defines the distance between treatments in terms of the between cluster variability to the within cluster variability. The Ward's technique is a hierarchical analysis that starts with n clusters, one for each treatment, and then at each step groups the most similar treatments into clusters. This procedure continues until there is one cluster containing all respondents. By examining the dendogram and several statistics (pseudo F and t), a judgment is made about the number of clusters .
The ratings of treatments within each cluster were compared to the rating from other clusters using a mixed-model analysis-of-variance (ANOVA) with two fixed-effects: cluster and seizure type (generalized, partial complex, absence for clinical seizures and Landau-Kleffner syndrome, atypical Landau-Kleffner syndrome, subclinical epileptiform discharges for subclinical seizures), and the interaction between these factors. The ANOVA was calculated using the 'glimmix' procedure of SAS 9.1 (SAS Institute Inc., Cary, NC) with respondent and seizure type as a random variable. Seizure type and the interaction were not significant in any analysis. Statistical values for the analysis of variables are presented in supplementary tables (See Additional file 2) along with the calculation for the Bonferroni correction. The statistical values for the seizure type and the interaction effects were not included in the tables since they were not significant. For selected clusters, individual treatments were compared using a similar ANOVA.
Planned contrasts were used to compare ratings. Planned contrasts were calculated using the 'estimate' command in SAS for the 'glimmix' procedure. The procedure uses both the fixed-effects and random-effects matrices to construct a matrix with an approximate t distribution. The unadjusted t-values and p-values are presented and the Bonferroni correction was used to calculate the appropriate alpha levels for each set of comparisons. Statistical values for the cluster or treatment effects of the analysis are presented in supplementary tables (See Additional file 2), along with the results of the contrasts.