Mental Health Outcomes Associated with Applied Behavior Analysis in a U.S. National Sample of Privately Insured Autistic Youth

Aguirre Mtanous, N. G., Koenig, J., Nikahd, M., Effertz, S. E., Silinonte, S., Hyer, J. M., Hand, B. N., & Bishop, L. (2026). Mental health outcomes associated with applied behavior analysis in a US national sample of privately insured autistic youth. Autism, 30(2), 484-494. https://doi.org/10.1177/13623613251390604

Reviewed by:
Svein Eikeseth, PhD, Oslo Metropolitan University
Gina Green, PhD, BCBA-D, San Diego, California
Eric Larsson, PhD, LP, BCBA-D, University of Minnesota and Lovaas Institute

Why Research this Topic?

Research Synopses - Psuchological, Educational and Therapeutic Interventions Applied behavior analysis (ABA) is one of the most widely used intervention approaches for autistic individuals. Many studies have examined whether ABA interventions can improve communication, learning, adaptive behavior, and other important developmental outcomes. Far fewer studies have examined whether ABA interventions are associated with later mental health outcomes.

That question has become increasingly important because some autistic adults have reported that ABA-based services they received as children were distressing or harmful (McGill & Robinson, 2021). Those concerns deserve careful scientific attention. It is important to recognize, however, that only certain types of studies can produce credible evidence that an intervention was responsible for apparent outcomes. Those studies have several characteristics: The research question(s) and outcome measures are specified in advance. The study is prospectve, meaning that it is conducted from a point in time forward. It is controlled in that it compares the effects of the intervention of interest against the effects of no intervention or an alternative intervention. In studies using group research designs, one group of participants (the treatment group) receives the intervention of interest for a period of time while another similar group (the control or comparison group) receives no specific intervention or an alternative intervention. The outcome measures are administeredd to all participants when the study begins and again when it ends. Because the study is conducted in real time, the researcher can monitor how the treatment and control or comparison conditions are implemented and can try to prevent or minimize factors that could affect outcomes for either group, such as other interventions or life events. If at the end of the study the measures show that the treatment group had better outcomes than the control or comparison group and the difference meets certain statistical criteria, the logical conclusion is that the difference was likely due to the intervention.

In contrast, retrospective studies involve looking back at information that was recorded in the past, often for purposes other than research (e.g., health care or education records). Even if the researcher compares the records of individuals who were said to have received a particular intervention in the past and other individuals who presumably did not receive that intervention, such a study cannot be described as controlled in the scientific sense. That is because the researcher can only work with the information that is available in the records. They cannot go back in time to see how the intervention was delivered and often have little or no information about other interventions or events the participants experienced. For those and other reasons, retrospective studies may identify variables that are associated (correlated), such as an intervention and certain outcomes, but they do not provide evidence that one variable caused another.

The retrospective study by Aguirre Mtanous and colleagues (2026) attempted to contribute to the discussion about the long-term effects of ABA interventions by examining a large U.S. health insurance claims database for information as to whether autistic youths with recorded insurance claims for ABA services had different rates of selected mental health outcomes than matched autistic youths without recorded insurance claims for ABA services. The researchers focused on four outcome measures: post-traumatic stress disorder diagnosis, suicidality, mental health hospitalization, and length of stay during mental health hospitalization. They also examined whether different amounts of ABA intervention were associated with different mental health outcomes.

What Did the Researchers Do?

Aguirre Mtanous and colleagues conducted a retrospective cohort study using the IBM MarketScan Commercial Claims and Encounters Database. That database includes health care claims for individuals with employer-sponsored private health insurance in the United States. The study used records from 2012 to 2020.

The sample was drawn from the private health insurance records of children and youth younger than 18 years at their first observed encounter. Inclusion required an autism diagnosis recorded in the insurance claims and at least 12 months of continuous private health insurance enrollment. For individuals with at least one recorded claim for outpatient ABA services, eligibility additionally required at least six months of continuous enrollment before the first such claim.

The researchers identified 17,120 autistic youths with recorded insurance claims for ABA services and matched them to 17,120 autistic youths without recorded insurance claims for ABA services. The groups were matched on sex, age, U.S. region, rurality, follow-up time, and recorded secondary diagnoses of intellectual disability, mood disorder, anxiety disorder, and psychotic disorder. After matching, each matched youth without recorded insurance claims for ABA services was assigned a comparison time point based on the date of the corresponding youth’s first recorded ABA claim. The researchers excluded matched pairs if either youth had one of the target outcomes during the pre-ABA period.

The researchers counted outcomes occurring after the first recorded ABA claim or after the matched comparison time point for youths without recorded insurance claims for ABA services. That is an important detail: the study examined outcomes after the first recorded ABA claim, not after the recorded ABA services had ended. The researchers also grouped youths with recorded insurance claims for ABA services into four “dose” categories based on the number of ABA “visits” billed to private health insurance. The median numbers of ABA visits in those dose categories were 19, 198, 450, and 895.

What did the Researchers Find?

The researchers reported that, on average, youths with recorded insurance claims for ABA services were more likely to have at least one recorded mental health hospitalization than matched youths without recorded insurance claims for ABA services. After the index date, 408 of 17,120 youths with recorded insurance claims for ABA services had at least one mental health hospitalization compared with 341 of 17,120 matched youths without recorded insurance claims for ABA services – approximately 2.4% versus 2.0%. After statistical adjustment, the odds ratio of 1.30 indicated that youths with recorded insurance claims for ABA services had 30% higher odds of having at least one hospitalization than matched youths without recorded insurance claims for ABA services. That does not mean that 30% more youths were hospitalized: the observed difference between the groups was approximately 0.4 percentage points.

The researchers also reported an incidence rate ratio of 1.32. Unlike the odds ratio, which concerned whether a youth had at least one hospitalization, the incidence rate ratio compared the number of hospitalization events per unit of followup time. Each hospitalization was counted separately, so a youth hospitalized three times contributed three events. The result indicated that, for the same amount of follow-up time, the estimated rate of hospitalization events was 32% higher among youths with recorded ABA claims than among matched youths without such claims. That relative comparison between claims-defined groups, however, does not establish that ABA caused the difference.

The study found no statistically significant differences between youths with recorded insurance claims for ABA services and matched youths without such claims in the odds of a recorded post-traumatic stress disorder (PTSD) diagnosis, suicidality, or length of hospital stay. The researchers also did not find a statistically significant relationship between ABA dose and any of the mental health outcomes.

The findings are therefore narrower than what is implied by a simple statement that ABA intervention was associated with poor mental health. The main statistically significant finding was limited to mental health hospitalization occurrence and frequency.

What are the Strengths and Limitations of the Study?

This study has several strengths. It used a very large national sample of health care records from privately insured autistic youths. The researchers used propensity-score matching to create a comparison group: each youth with recorded insurance claims for ABA services was matched with a youth without recorded insurance claims for ABA services who was similar on selected characteristics available in the database. That procedure can make groups comparable on the measured characteristics, although it cannot account for important characteristics that were not measured or recorded. The researchers also established a defined comparison point: the first ABA “visit” recorded in the billing claims and a corresponding matched pseudo-start date for youths without recorded insurance claims for ABA services. The researchers excluded a matched pair if either youth had one of the study outcomes recorded during the pre-index period. Those methodological choices strengthened the study relative to a simple comparison of the health care records of all youths with recorded insurance claims for ABA services and the records of all youths without such claims.

The study also addressed an important and under-researched question. Concerns about possible harms of ABA interventions are often discussed, but scientific studies of mental health outcomes following participation in ABA services are limited and adverse events have not been monitored or reported adequately in autism intervention research in general (Bottema-Beutel et al., 2021). The study helps identify questions that should now be examined in controlled prospective studies with valid and reliable outcome measures and better matching of treatment and control or comparison groups.

The study has substantial limitations, however. First, it was a retrospective observational study based on existing insurance claims, not a prospective controlled study with researcher-assigned conditions. The researchers did not assign youths to receive ABA services or a comparison condition and then follow them forward. Instead they analyzed billing records after the services and outcomes had already occurred.

Youths with recorded claims for ABA services may have differed in important ways from matched youths without such claims. For example, youths referred for ABA services or whose parents sought insurance coverage for those services may have had greater autism-related support needs, lower adaptive functioning, greater communication difficulties, more severe challenging behavior, greater emotional dysregulation, or more acute family needs than youths who were not referred for ABA services. Those characteristics may have increased their risk of hospitalization independently of any ABA services they received. Although propensity-score matching reduced differences in measured characteristics, it could not account for factors that were unmeasured or captured inadequately in the insurance claims.

That problem is known as confounding by indication (Kyriacou & Lewis, 2016). In plain language, the factors that lead someone to receive a service may themselves be related to later outcomes. If greater baseline needs increased both the likelihood of having recorded ABA claims and the likelihood of later hospitalization, the data could show an association between ABA claims and hospitalization even if the ABA services did not increase hospitalization risk. As we noted previously, retrospective record reviews can identify associations but they cannot establish that one variable caused another.

Relatedly, Aguirre Mtanousand colleagues did not provide strong evidence that the two groups were clinically equivalent before the first recorded ABA claim. The groups were matched on sex, region, rurality, age at first observation, followup time, intellectual disability, mood disorder, anxiety disorder, and psychotic disorder. However, matching on the clinical characteristics was based on diagnoses recorded at least once at any time during the study period. Thus a diagnosis first recorded after the first recorded ABA claim could have been used for matching or adjustment. For example, a child’s first ABA claim might appear in January 2020, while an anxiety diagnosis first appears in the claims data in September 2020. If that anxiety diagnosis was used to match the child with a comparison participant, the analysis would treat information recorded after ABA began as though it were available beforehand. That is a serious limitation. To avoid incorporating post-exposure information, matching and adjustment should have used only characteristics that were documented before the index date.

Third, health insurance claims data provide a very limited picture of an individual’s functioning. The study did not include standardized measures of intellectual skills, language skills, adaptive behavior, autism severity, challenging behavior, emotional regulation, family stress, socioeconomic factors, race or ethnicity, or school placement. Those are highly relevant variables that should be considered when evaluating referrals for and effects of ABA services as well as mental health and other outcomes.

Fourth, both autism status and the nature and amount of ABA services were inferred from administrative documents—health insurance billing claims and associated records—rather than being verified independently (Grosse et al., 2022). The researchers did not confirm autism diagnoses through clinical assessments or determine whether services billed with ABA procedure codes had the defining features of ABA, were delivered as intended, or were overseen by appropriately qualified professionals (Dubuque et al., 2021). The health care records did not show what ABA services were provided, their quality, the behaviors targeted for each youth, or whether procedures were implemented with fidelity. Nor did they show whether practitioners monitored the youths’ assent (their ongoing willingness to participate, expressed verbally or through behavior) and distress, or whether restrictive procedures were used. The authors acknowledged that the database did not capture the quality or characteristics of ABA services. Consequently, some participants may have been misclassified with respect to autism diagnosis, the presence or absence of recorded ABA claims, or both.

Fifth, matched youths without recorded insurance claims for ABA services may nevertheless have received ABA services. Again, the database captured only services that had been billed to private health insurance. Youths classified as having no recorded insurance claims for ABA services may have received ABA services through schools, other payers, or out-of-network providers. They may have received services that had the characteristics of ABA but were coded as other kinds of services. The authors acknowledged that services that were not billed to private health insurance, such as school-based ABA services, may have been missed. Additionally, some individuals in both groups may have received interventions or had experiences that contributed to the outcomes these researchers considered. In a sound prospective study, researchers typically try to control for those kinds of factors (“extraneous variables”) or at least attempt to document their occurrences in real time over the course of the study. Aguirre Mtanous and colleagues could not go back in time to do that, and it does not appear that the health care records they used in their study provided information about such factors.

Sixth, the timing of the recorded ABA services and the documentation of outcomes was unclear. Outcomes recorded after the first ABA claim may have occurred while ABA services were still being recorded, not after the services had ended. In addition, the total ABA “dose” may have included visits billed after a mental health hospitalization, making any association between dose and outcomes difficult to interpret.

Finally, the absolute difference in mental health hospitalizations between the two groups was small. The proportion with at least one hospitalization was about 2.4% among youths with recorded insurance claims for ABA services and 2.0% among matched youths without recorded insurance claims for ABA services. A relative difference of 30% sounds large, but the absolute difference was about 0.4 percentage points. Both relative and absolute differences are important for families and policymakers to understand. The reported odds ratio describes the relative difference in the odds of having a mental health hospitalization, while the incidence rate ratio describes the relative difference in how often hospitalizations occurred. Absolute figures, such as the number of children hospitalized and the number of hospitalizations per 1,000 children in each group, would show the size of those differences in practical terms.

What do the Results Mean?

The study provides some evidence of an association between recorded insurance claims for ABA services and later mental health hospitalization in a large sample of privately insured autistic youths. It does not show that ABA caused mental health hospitalizations. The most appropriate interpretation is that, on average, youths with recorded insurance claims for ABA services had slightly higher odds and rates of later mental health hospitalization than matched youths without recorded insurance claims for ABA services. Because the study was retrospective and observational, it does not indicate whether the difference was caused by ABA services, pre-existing differences between the groups, differences in health care access or referrals, other life events, or other factors.

The absence of statistically significant differences for PTSD and suicidality is also important. The researchers did not find higher odds of recorded PTSD diagnoses or suicidality among youths with recorded insurance claims for ABA services than among matched youths without such claims. Accordingly, the study did not provide statistically significant evidence of an association between recorded ABA claims and those outcomes.

The lack of a dose-response relationship is also relevant. If ABA interventions were strongly associated with mental health harms and if billed visits accurately measured exposure, one might expect poorer outcomes among participants with more recorded ABA visits. Aguirre Mtanous and colleagues did not find that pattern. Because the dose measure was crude and the groups may have differed in unmeasured ways, that result does not rule out the possibility that some forms of intervention may be harmful. It does mean that the study provides no evidence that more recorded ABA visits produced worse mental health outcomes for the selected participants, much less for all youths diagnosed with autism.

Families and service providers should be cautioned not to read this study as showing that ABA interventions are generally harmful for people with autism or that they were harmful for the youths whose records were examined by Aguirre Mtanous and colleagues. Given its limitations, this retrospective record review identifies an association that should be investigated in prospective studies with stronger designs and more complete measurement. Simultaneously, this study should not be dismissed. It highlights the need for more research and better clinical monitoring of any adverse events and side effects of ABA and other interventions for people with autism (Bottema-Beutel et al., 2021; Schuck et al., 2024). Future controlled studies using group research designs should measure all participants’ functioning before intervention begins, including autism severity, communication skills, social skills, adaptive behavior, challenging behavior, psychiatric symptoms, and family stress. Researchers should also clearly describe the ABA and other interventions received by participants in both groups, whether interventions were individualized, whether assent and distress were monitored, and whether treatment goals were meaningful to each individual and their family.

The report by Aguirre Mtanous and colleagues also raises a broader clinical issue: intervention outcomes should include more than skill acquisition and reduction of challenging behaviors. Overall health, safety, emotional well-being, distress, assent, quality of life, family priorities, and the individual’s experience should be measured (Wolf, 1978). High-quality ABA services should be compassionate, individualized, developmentally appropriate, and focused on producing meaningful outcomes—not simply reducing behaviors that others find inconvenient (Taylor et al., 2019).

In summary, the study by Aguirre Mtanous and colleagues raises important questions about interventions and mental health outcomes. However, it cannot tell us whether ABA caused the mental health outcomes recorded for the youths whose records were used in the study. Its findings should be evaluated in well-controlled prospective studies. The study also suggests the need for safeguards to identify and minimize potential harms in practice. The appropriate conclusion is that a slightly higher proportion of youths with recorded insurance claims for ABA services experienced a mental health hospitalization and their adjusted hospitalization rate was higher than the rate for matched youths without recorded ABA claims. The researchers could not determine whether those differences were due to ABA, other interventions, pre-existing differences, or other experiences.

References

Bottema-Beutel, K., Crowley, S., Sandbank, M., & Woynaroski, T. G. (2021). Adverse event reporting in intervention research for young autistic children. Autism, 25(2), 322-335. https://doi.org/10.1177/1362361320965331

Dubuque, E. M., Yingling, M. E., & Allday, R. A. (2021). The misclassification of behavior analysts: How national provider identifiers (NPIs) fail to adequately capture the scope of the field. Behavior Analysis in Practice, 14(1), 214-229. https://doi.org/10.1007/s40617-020-00451-w

Grosse, S. D., Nichols, P., Nyarko, K., Maenner, M., Danielson, M. L., & Shea, L. (2022). Heterogeneity in autism spectrum disorder case-finding algorithms in United States health administrative database analyses. Journal of Autism and Developmental Disorders, 52(9), 4150-4163. https://doi.org/10.1007/s10803-021-05269-1

Kyriacou, D. N., & Lewis, R. J. (2016). Confounding by indication in clinical research. JAMA, 316(17), 1818-1819. https://doi.org/10.1001/jama.2016.16435

McGill, O., & Robinson, A. (2021). “Recalling hidden harms”: Autistic experiences of childhood applied behavioural analysis (ABA). Advances in Autism, 7(4), 269-282. https://doi.org/10.1108/AIA-04-2020-0025

Schuck, R. K., Baiden, K. M. P., Wang, M., Olis, S., Ingram, C., & Fisher, G. (2024). Assessment of adverse events, side effects, and social validity in evidence-based behavioral interventions for autistic students. Review of Research in Education, 48(1), 154-190. https://doi.org/10.3102/0091732X241281268

Taylor, B. A., LeBlanc, L. A., & Nosik, M. R. (2019). Compassionate care in behavior analytic treatment: Can outcomes be enhanced by attending to relationships with caregivers? Behavior Analysis in Practice, 12(3), 654-666. https://doi.org/10.1007/s40617-018-00289-3

Wolf, M. M. (1978). Social validity: The case for subjective measurement or how applied behavior analysis is finding its heart. Journal of Applied Behavior Analysis, 11(2), 203-214. https://doi.org/10.1901/jaba.1978.11-203

Reference for this article:

Eikeseth, S., Green, G., & Larsson, E. L. (2026). Research Synopsis: Mental health outcomes associated with applied behavior analysis in a U.S. national sample of privately insured autistic youth. Science in Autism Treatment, 23(10)

Related Research Synopses:

Other Related ASAT Reviews and Articles:

#BehaviorAnalysts #MedicalProviders #Researchers #SavvyConsumer