6

Characteristics of ACD Participants

Since the initiation of the Comprehensive Autism Care Demonstration (ACD), the Defense Health Agency (DHA) of the Department of Defense (DoD) has tracked the number of ACD participants from fiscal year to fiscal year. See Figure 6-1 for trends in ACD participants and costs per participant from FY 2015 to FY 2022. Annual reports to Congress summarize total program costs, average cost per participant, and the number of ACD participants, distinguishing between active-duty family members (ADFMs) and non-active-duty family members (NADFMs; DoD, 2019a, 2024b). Total costs for applied behavior analysis (ABA) services through the ACD have ranged from $161.5 million (FY 2015) to $454.4 million (FY 2021). The reports to Congress also capture total expenditures among ACD participants for other TRICARE medical services common for autistic individuals including physical, speech, and occupational therapies as well as prescription medications. In the most recent report to Congress (DoD, 2024b), services provided under the TRICARE Extended Care Health Option program are also reflected.

The reports to Congress consider what percentage of beneficiaries diagnosed with autism utilize the ACD (this has ranged from 52% to 57% for ADFMs and from 23% to 29% for NADFMs from FY 2015 to FY 2021). The age distribution of ACD participants has been generally consistent from year to year, with nearly 99% younger than age 21. For the most recent report, the median participant age is reported as 7 years, the average age is 8.0 years, and the most common age of participants is 5 years. Roughly four out of five participating beneficiaries are male (DoD, 2024b).

The reports have tracked the increase in certified ABA providers as submitted by the Managed Care Support Contractors. In 2015, there were 6,887 TRICARE-authorized ABA supervisors across all TRICARE regions, and there were 523 assistants and 19,421 behavior technicians supporting authorized ABA supervisors (DoD, 2016). In 2021, there were 20,064 TRICARE-authorized ABA supervisors across East and West TRICARE regions, and there were 1,433 assistants and 61,554 behavior technicians supporting authorized ABA supervisors (DoD, 2023c).

The remainder of this chapter presents findings on ACD participants from the committee’s own analyses of data requested through DHA on ACD participants and relevant claims information processed through TRICARE. The committee was tasked with conducting its own data analyses, using data similar to those used by DHA in its reports to Congress, to address questions in its charge (see Chapter 1, Box 1-3). The committee engaged a subcontractor, Westat, to carry out the data analyses. Per contractual requirements and approval by the National Academies’ Institutional Review Board (IRB), the committee did not have access to the protected data provided by DoD and relied on Westat’s analytic summaries for its deliberations and report. See Appendix D for additional information on analytical methods, assumptions, and data limitations.

DESCRIPTION OF DATA REQUEST

The data requested through DHA from the Military Health System Data Repository included patient-level encounter data, claims data, and ACD records data from calendar years 2018 through 2023 for individuals aged 1 to 18 years old with autism. This time period was deemed sufficient to make comparisons of ACD participants and non-participants and reflected periods before and after the 2021 policy changes. For the data request, autism was defined by an F84.0 International Classification of Diseases (ICD)-10 code between 2018 and 2023 in the patient record. Administrative data from claims and electronic health records (EHRs) included approximately 300 fields selected from seven DHA and DoD datasets. The seven datasets requested included the following:

1. Comprehensive Ambulatory/Provider Encounter Record (CAPER);
2. TRICARE Encounter Data–Institutional (TED-I);
3. TRICARE Encounter Data–Non-Institutional (TED-NI);
4. Standard Inpatient Data Record (SIDR);
5. GENESIS, an EHR system with admission and demographic records;
6. Pharmacy Data Transaction Service (PDTS); and
7. Defense Enrollment Eligibility Reporting System (DEERS).

Combined, these datasets include all ambulatory and inpatient encounters, health services from the private sector, and medication and demographic information for TRICARE beneficiaries aged 1 to 18 over the study period. Important data elements in these tables include ICD-10 codes identifying autism and other relevant co-occurring health conditions and procedure codes identifying ABA and other relevant healthcare services. DEERS data include records for demographic variables, such as birth date, race/ethnicity, sex, sponsor rank, and residence ZIP code.

ACD data include assessment scores for participants. Composite and available subdomain scores from the Pervasive Developmental Disorder Behavior Inventory (PDDBI); the Vineland Adaptive Behavior Scales, Third Edition (Vineland-3); Social Response Score (SRS); and Parenting Stress Index–Short Form (PSI-SF) were provided to establish multiple assessment points (see Chapter 3, Box 3-2). Required assessment and reporting policies for the ACD as well as engagement of participating families changed during the period of interest. This affected the availability of assessment and subdomain scales over the study period. According to the TRICARE operations manual, these assessments are to be completed and submitted periodically by ABA providers, and the results are to be incorporated into participants’ treatment plans by ABA providers and the comprehensive care

period from calendar years 2018 to 2023. There were 109,073 unique individuals in the sample. Figure 6-2 shows a breakdown of this sample into continuous and intermittent users of ABA and non-participants in the ACD. The larger non-participant group contained more individuals across the age range, whereas the ABA user groups were predominantly younger children. See Figure 6-3 for a breakdown of ages within these subgroups.

• Continuous participants were defined as beneficiaries with an F84.0 ICD-10 code in their medical records and at least one ABA claim per month for at least six consecutive months between January 1, 2018, and December 31, 2023.

• Intermittent participants were defined as beneficiaries with an F84.0 ICD-10 code in their medical records and at least one ABA claim between January 1, 2018, and December 31, 2023. The intermittent participants infrequently reached six consecutive months of ABA therapy.
• Non-participants were defined as beneficiaries with an F84.0 ICD-10 code but with no ABA-related claims in their medical history during this this time frame.

Within each of the continuous and intermittent ACD groups, ABA intensity for each participant was calculated as the sum of ABA hours for six months divided by the weeks in the ACD during the six months. All participants were anchored from their first ABA claim. Hours of ABA were calculated using the number of Current Procedural Terminology (CPT) codes billed and the time associated with them (in minutes). All available CPT codes for treatment (including family guidance) were summed. Category III codes in 2018 were recorded in 30-minute increments. Category I codes, used from 2019 to 2023, were recorded in 15-minute increments (see Chapter 4, Box 4-2 for description of Category I CPT codes for treatment).¹

The committee examined several demographic characteristics among participant groups, including age, sex, ethnicity, race, sponsor rank, branch and component, residence region and urbanicity, and whether individuals appeared to move during the study period. We examined these characteristics by age groups as well as by participation or non-participation in the ACD (as well as by amount of ABA for participants). Since there were a lot of missing or indeterminate data (see Appendix D), only noteworthy findings are shared here.

Across age groups and participant status, the sample of autistic children consisted of approximately one-quarter females and three-quarters males (26% vs. 74% for total sample), and approximately three-fifths lived in the eastern region of the United States and one-third in the western region (59% vs. 31% for the total sample of autistic individuals). The remainder were overseas or unknown.

In our sample, there were 62,368 (58%) individuals from ADFMs and 43,602 (41%) individuals from NADFMs. ACD participants were primarily from ADFMs; 41% of ADFM beneficiaries used ABA services and 20% of NADFM beneficiaries used ABA services in the 2018–2023 time period. There were noticeable shifts in the ratio of individuals from ADFMs to those from NADFMs compared across age groups and participant status (see Figure 6-4).

Among military-connected children ages 1–18 years old with an autism diagnosis in the period 2018–2023, the committee found that 35,034 (32%) of these children had a claim for ABA services, but only 26,131 (24%) received ABA continuously for at least six months. More than half (58%) were under age six and a majority (89%) of ABA recipients were under age 12. The committee identified subsets of regular or continuous ABA recipients (75%) and intermittent ABA recipients (25%). The committee observed that continuous recipients tended to be younger children, to have more co-occurring conditions like delayed milestones, and to have greater utilization of other services in addition to ABA. Notably, however, mental health issues were more common among those not receiving ABA, particularly for older children. This trend was seen among school age children as well. Comparisons for the total sample and for the youngest (0 to 2 years old) and oldest (12 to 18 years old) are shown here (see Table 6-1,

Table 6-2, and Table 6-3). Additional comparisons of clinical characteristics for different age groups as well as a breakdown of ICD-10 and CPT codes for clinical characteristics can be found in Appendix D.

Among ABA recipients, the number of hours per week of ABA treatment trends low (60% of ABA recipients receive less than five hours/week) and 20+ hours/week is rare (4% of ABA recipients). Even among the youngest children, the majority receive less than 10 hours/week. See Figure 6-5 for a distribution on the average amount of ABA by age and Figure 6-6 for a distribution of ABA intensity among 26,131 continuous users. Table 6-4 compares the presence of other services and other conditions between groups defined by amount of ABA. Based on these data, increasing amounts of ABA are associated with increasing use of other services. Notable differences in co-occurring conditions include the presence of delayed milestones higher among higher-intensity ABA users contrasted with the presence of attention-deficit/hyperactivity disorder (ADHD) and anxiety (and other mental health conditions) higher among lower-intensity ABA users.

TABLE 6-3 Clinical Characteristics for Ages 12 to 18 Years Old

NOTES: ADHD = attention-deficit/hyperactivity disorder; ENT = ear, nose, and throat; OCD = obsessive-compulsive disorder; OT = occupational therapy; PT = physical therapy.

Planning Analytic Approach

While waiting for the data to be delivered to Westat, the committee engaged in developing and refining analytic approaches that would address questions in its statement of task (see Chapter 1, Box 1-3). This section highlights some of the considerations and questions raised. Notably, potential approaches focused on analysis of the effects of ABA and review of health outcomes and whether it would be possible to determine the prevalence of autism within this military sample. The following research questions were considered:

• What drives participation in the ACD? What are differences between individuals with an autism diagnosis participating in the ACD and those not participating?
• What drives the amount of ABA received and length of participation in ACD? What is the frequency of ABA utilization in the population participating in the ACD?
• What factors contribute to the adherence of ABA for TRICARE beneficiaries under 18?

• Do the scores on required assessment tools change over time? What factors contribute to changes in scores? Can the change in scores be linked to progress on health outcomes from ABA?
• Do scores on different assessment tools reasonably compare? To what extent are they measuring progress on similar or different autism domains?
• What is an adequate calculation of amount of ABA from filed claims (procedures codes)?
• What are adequate proxies for important covariates missing from the databases?
• What is the prevalence of autism in the military population?

The following covariates (some unchanging characteristics, others varying in time) were considered:

• Demographics (age, sex, race/ethnicity, location)
• Level of autism severity

• Time from diagnosis to initiate ABA
• State of participation in ACD—continuous or categorized in other ways? (How to quantify start and end dates?)
• Comorbidities—other chronic medical conditions (Which are most relevant to autism and delivery of ABA?)
• Co-occurring (or prior) interventions or healthcare services (Which are most relevant to autism and delivery of ABA?)
• Co-occurring (or prior) medications (Which are most relevant to autism and delivery of ABA?)
• Who completed the assessment tools
• Caregiver training and participation
• Family satisfaction

Descriptive Analysis of Assessments

As discussed in Chapter 3, DHA began collecting data on ACD participants from a set of standardized assessment tools in January 2017. The specific tools and frequency of administration changed over time, but the collection of assessment data has continued through 2025. DHA published findings from its analyses of these data in its reports to Congress, beginning with FY 2018 Quarter 4. This set of tools was introduced in the ACD in response to Congress’s request for reports on health-related outcomes (see Appendix E, Senate Report 114-255, S. 2943, p. 205) and is referred to as “outcome measures” in the TRICARE operations manual. The managed care service contractors have been directed to help ABA providers use assessment scores from the Vineland-3, SRS, and PDDBI in treatment planning (DHA, 2023a). Assessments of parent/caregiver stress are also required through the PSI-SF or the Stress Index for Parents of Adolescents (SIPA); however, these are not to be used for a beneficiary’s treatment planning.

The committee heard from researchers, ABA providers, and families who all raised questions regarding the required set of assessment tools and the data analyses conducted by DHA (National Academies of Sciences, Engineering, and Medicine [NASEM], 2024a, 2025). As such, the committee sought to understand these concerns and the utility of these tools.

The committee’s data request for ACD records data included scores and subscores from the PDDBI, Vineland-3, SRS, and PSI-SF. Table 6-5 provides an example of the number of assessment scores recorded 2018 to 2023 for the 36,420 ACD participants. Subscores from these assessment tools (such as the Vineland-3 subscores on Socialization, Communication, Daily Living Skills, and External Maladaptive Behaviors) were only recorded 2021 to 2023.

With the availability of assessment scores and subscores, the committee considered testing the association between the amount of ABA received and changes in scores and subscores on the required assessment tools using

TABLE 6-5 Summary of Availability of Assessment Scores for ACD Participants (N = 36,420)

NOTE: For the PDDBI assessments, there were 3,577 unique individuals missing assessment dates for all assessments. For the Vineland assessments, there were 7,649 unique individuals missing assessment dates for all their assessments. The vast majority of these were all prior to 2021.

multivariate and mixed effects linear regression and defining the target outcome as the difference between the last and first assessment. The amount of ABA would be defined as the summed hours of ABA obtained from procedure codes that lie between assessment periods. Important covariates would have to be included. The committee considered whether the PDDBI baseline score could be used as a categorical variable to reflect autism severity (see Figure 6-7 for distribution of PDDBI teacher composite scores of the initial assessment [baseline] for ACD participants). However, due to time constraints, data limitations, and concerns about the quality of the data collected from these assessment tools, the committee did not pursue thorough testing of the association between the amount of ABA received and outcomes as measured by the assessment tools. It did some preliminary modeling of the change in the PDDBI REXSCA (receptive/expressive social communication abilities dimension) subscore and for the Vineland-3 ABC composite score. It also compared mean scores and subscores against demographic characteristics and did not find any important variations. See Appendix D for details on this approach and list of assessment subscores that the committee viewed as most relevant to delivery of ABA.

The committee argues against the continuation of data collection from these assessment tools. We generally think the dataset is not useful enough to address questions around ABA, autism, or the effectiveness of the ACD program to be worth the burden of administering the assessments periodically to participating families.

Furthermore, for clinical purposes, DHA’s selection of the required assessment tools is out of step with the current assessment practices to evaluate

individual progress with ABA (see Chapter 4). A significant concern is that the same battery of assessments is required for all beneficiaries regardless of their characteristics (e.g., age, functioning, autism severity level, treatment goals), and this places additional burden on ABA providers to carry out these assessments in addition to those chosen to be developmentally appropriate for individualized treatment planning. Another concern is that they are all indirect measures and rely heavily on parent reporting. Entire sections of these tools may be difficult to complete for parents who have children with moderate or profound cognitive impairment. As a set, there is construct overlap (e.g., basic social communication is over-sampled), and as such, it is not adequately capturing all of the autism domains that would exist among the autistic population. The set also does not adequately cover all of the co-occurring conditions relevant to one’s functioning and is lacking broader coverage of quality-of-life indicators.

Of note, the use of scores from standardized tools to measure change is often misunderstood. Maintaining the same score over time, for example on the Vineland-3, means that the child gained skills at the same rate as neurotypical children during the same period. Small increases in raw scores may represent meaningful change, while the corresponding standard score may not increase or could even decrease while progress is being made. Any reassessments using such standardized tools are typically recommended in