Systematic review: distribution of age and intervention modalities in therapeutic clinical trials for autism spectrum disorder

Abstract

The prevalence of ASD remains relatively stable across the lifespan, necessitating a quantitative understanding of how intervention clinical research is applied across age groups. Here we report a systematic review of treatment studies between 2013 and 2017, enrolling 11,213 subjects with ASD in 218 studies. Individuals under 18 years old constituted the majority of studies (84%) and subjects (92%). Subjects under 18 years old were more likely to be enrolled in behavioral studies (OR (CI) = 1.34 (1.17-1.54)), and less likely to be enrolled in pharmacological (OR = 0.60 (0.52-0.69)) studies than subjects ≥ 18 years old. Identified disparities in both intervention modalities and outcome measures should serve to guide future research priorities.

Introduction

A substantial proportion of individuals living with ASD are adults, and the importance of a lifespan approach to this condition is increasingly recognized (Buescher et al. 2014; Howlin and Taylor 2015). This understanding is critical given recent suggestion that the prevalence of ASD is rising in childhood cohorts (Baio et al. 2018), translating into greater service need for adults with ASD (Brugha et al. 2011; Dudley et al. 2018). It is likely that age interacts with the complex biological, psychological, and social variables associated with ASD, necessitating empirical evaluation of interventional strategies at distinct time points during development. For school-aged individuals with ASD, a range of empirically validated interventions exists across the biopsychosocial spectrum. This is likely reflective of a recognition that ASD is present from the earliest periods of development, and that optimal outcomes may be contingent on the early delivery of best available interventions (Smith 1999). Biological interventions include the use of medications, which are frequently used to target associated symptom dimensions such as aggression, anxiety, or attention (Mandell et al. 2008). Psychosocial interventions include empirically validated educational strategies, and ASD-specific therapies with a strong basis in behavioral theory (Reichow et al. 2018).

The literature becomes sparse as clinicians seek guidance on supporting individuals with ASD of transitional age and beyond (Warren et al. 2012). From the time they leave school, individuals with ASD begin to encounter a diverse set of evolving environmental stressors, including a loss of structure, the pursuit of a variable amount of independence, and a potential increased awareness as to how their cognitive and social style may impact their personal and vocational opportunities (Lounds Taylor et al. 2017). In parallel, well-developed sources of support are replaced with highly idiosyncratic or potentially non-existent systems of care (van Schalkwyk and Volkmar 2017). Issues regarding relationships, sexuality, family, and aging may become salient over time, and intersect with the unique cognitive and social style of adults with ASD to produce additional needs for clinical support. Clinical trials exist which describe the effects of antipsychotic medications for the management of irritability and aggressive behavior in adults with ASD, but psychopharmacologic research appears to be otherwise limited in this age group.

The unique clinical needs of adults with ASD have been emphasized in recent dedicated volumes (Volkmar et al. 2014), and at least two attempts have been made at describing critical research agendas (Howlin and Taylor 2015; Warren et al. 2012). Depending on level of functioning and individual goals and preferences, the needs of adults with ASD may be diverse (van Schalkwyk and Volkmar 2017). Adults with ASD may need support to achieve optimal vocational outcomes or succeed in tertiary education, but there is little systematic study of how this is best achieved (Taylor et al. 2012; Vanbergeijk et al. 2008). Although adults appear to have fewer behavioral symptoms than during adolescence, the rate of improvement appears to slow in adulthood, questioning the efficacy of currently utilized strategies compared to school-based interventions (Taylor and Seltzer 2010). Despite this recognition, the parity of enrollment between children and adults in treatment studies remains unknown. No research to date has used a systematic approach to quantify the presumed difference in volume of interventional research between these two groups. In this analysis, we sought to identify disparities in the study of interventional modalities for ASD subjects across the lifespan. By using a systematic and quantitative analytic approach, we sought to generate data which could expand on the previously recognized general impression of a disparity, to a focused, evidence-based description as to both the presence and extent of such a disparity, with the goal of informing research policy and initiatives.

Methods

This systematic review is reported in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al. 2009).

Literature search

Literature searches were conducted to broadly identify all prospectively conducted studies between 2013 and 2017 that tested any intervention modality hypothesized to be potentially therapeutic for individuals with ASD. The literature search was limited to this five-year period to allow sufficient sample size to make valid conclusions, while at the same time to be recent enough to capture current trends in the field. Secondarily, the release of DSM-5 in 2013 likely resulted in the term “autism spectrum disorder” capturing more individuals than prior to 2013. On March 20, 2018 the PubMed database was searched using the search term (((("Autistic Disorder" [MeSH]) OR "Autism Spectrum Disorder" [MeSH]) AND "Clinical Trial" [Publication Type]) AND "Humans"[MeSH]) NOT "Review" [Publication Type]. This search strategy enabled identification of closely related terms. For instance, an entry term included under “Clinical Trial” [Publication Type] is also “Intervention Trial”. Similarly, the search term “Autism Spectrum Disorder” lies above the term “Asperger Syndrome” in the MeSH hierarchy, and therefore would capture all publications with Asperger Syndrome or any of its entry terms, such as “Asperger’s Disease” or “Asperger’s Disorder”, etc. To increase the likelihood of capturing single case design studies, we also searched the PubMed database for studies published using the term ("Autistic Disorder" [MeSH] OR "Autism Spectrum Disorder"[MeSH]) AND "single case" AND "Humans"[MeSH] NOT "Review" [Publication Type]. The Cochrane Controlled Register of Trials was also searched using the search term (“autistic disorder” OR “autism spectrum disorder”) AND “clinical trial”. Results of both searches were limited to studies published during the five-year period between 2013 and 2017. Duplicates were identified and excluded.

Inclusion criteria

Studies that satisfied all of the following criteria were included. (1) The study must be prospective, meaning a study in which the investigators defined a subject group and intervention prior to recruitment and data collection. Studies did not have to be pre-registered. (2) The study must specifically enroll subjects with a diagnosis of ASD or subjects determined to be at high genetic or environmental risk for ASD. (3) The study must test an intervention to specifically address either:

1. Core or commonly-associated behavioral, emotional, or neurocognitive symptoms of ASD;

2. Physiological measures hypothesized to play a pathophysiological role in the development or maintenance of ASD;

3. Function of individuals with ASD in any realm; and/or

4. Prevention of ASD in individuals at environmentally or genetically high risk of developing ASD.

(4) The study must report an outcome measure specifically from subjects diagnosed with ASD or at high risk for ASD. (5) The study must report information on the mean or median age or age range of enrolled subjects. (6) The study must be published in English.

Exclusion criteria

Publications or studies meeting any of the following criteria were excluded. (1) The publication is a commentary, review, conference abstract, meta-analysis, or secondary analysis of previously published study. (2) The study does not enroll subjects diagnosed with ASD or at high risk for ASD. (3) The study does not report an outcome measure from a subject with ASD (a frequent example is a parenting intervention study without any assessment or measure from the child diagnosed with ASD). (4) Full text of the study is unable to be acquired or does not provide sufficient information to make an inclusion determination. (5) The study is ultimately retracted. (6) The study is not published in English. There was no minimum number of subjects for study inclusion.

Data extraction, classification, and quality assessment

Because the overall purpose of this study is to quantify the distribution of planned interventional modalities for individuals with ASD, the following data were extracted from each included study by the first author (ASL). (1) The number of enrolled subjects with ASD. (2) A measure of central tendency (MCT) of the age of subjects enrolled in the study. In almost all cases, this was the mean age, which was then rounded to the nearest year. (3) In studies that did not provide a MCT for the enrolled study population but provided an age range (occurring in 5 of 218 studies (2.3%) and 148 of 11,213 enrolled subjects (1.3%)), a mean age was estimated for the subjects by calculating the midpoint in the age range. (4) The type of intervention was categorized as follows:

1. Behavioral, whereby the subject is exposed to an intervention designed specifically to increase positive behavior, which commonly involved social skills training or training to maintain or maximize skills and functioning in everyday life. This category frequently involved parent behavior training, and studies of this nature were included if an outcome measure of the child with ASD was reported.

2. Pharmacological, whereby the subject is orally, transdermally, or parenterally administered a defined and purified pharmacological agent.

3. Somatic, whereby a non-pharmacological or dietary intervention that exerts a hypothesized physical influence on the body of the subject with ASD is performed. Examples include brain stimulation (i.e. transcranial magnetic stimulation or transcranial direct current stimulation), therapeutic devices (i.e. weighted blankets, mattress technologies), and stem cell infusions.

4. Dietary, whereby the subject’s diet is modified in a controlled manner to either include or exclude specific components. This differs from pharmacological studies in that specifically defined agents are delivered by dietary intake rather than in a purified form;

(5) The primary outcome or major outcome of interest was categorized. In many cases, no primary outcome was explicitly stated. In these cases, the first ASD subject-relevant outcome reported within the results section was typically classified. Primary outcomes were grouped into the following classifications:

1. Subjective ratings of core ASD symptoms (social deficits, language impairment, repetitive behavior) and symptoms or behaviors commonly associated with ASD (anxiety, hyperactivity, irritability and challenging behaviors, etc.). Clinical global impression was included in this category.

2. Functional, whereby a measure of the individual with ASD’s ability to participate in daily life is reported. Examples included ability to participate in a dental examination, interventions to support communication, and programs designed to improve problem solving in the workplace.

3. Physiological and neurocognitive, whereby the effects of an intervention on a diverse set of measured physiological outcomes (i.e. eye tracking, neural connectivity measured by neuroimaging, skin conductance, gait parameters, plasma cytokines, etc.) or laboratory-measured neurocognitive tasks (i.e. verbal problem solving, attention scale on the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS), facial emotion labeling, etc.) hypothesized to be relevant to ASD are reported.

4. Intervention development, whereby the primary focus of the study is to determine whether an intervention is feasible for the subject with ASD. This category was rare (5 of 218 studies (2.3%) and 116 of 11,213 enrolled subjects (1.0%)).

Inter-rater agreement for the classification of studies into intervention and outcome categories between the study’s authors was excellent, as denoted by a Cohen κ = 1. The quality of each included study and risk of bias was not specifically assessed as this study’s aim is to identify overall patterns in the type of intervention across the lifespan, and no meta-analytic procedure was performed to combine the quantitative outcomes of studies.

Statistical analysis

Fisher exact test, Mann-Whitney test, and Kruskal-Wallis test followed by Dunn’s post-test were used to compare categorical variables, two groups of unmatched continuous variables, and three or more groups of unmatched continuous variables, respectively. The binomial test was used to compare the study distribution by age observed from our systematic literature review with the expected ASD prevalence in the United States in individuals <18 years old and ≥18 years old (Buescher et al. 2014). Odds ratios and 95% confidence intervals were generated to compare the odds of a specific intervention in individuals < 18 years old to a reference age range of ≥ 18 years old. Microsoft Excel version 14 and GraphPad Prism version 8 were used for statistical analyses and plotting.

Results

Literature search

The details of our literature selection process are shown in Figure 1. We initially identified 331 citations in PubMed. A search of the Cochrane Controlled Register of Trials returned 142 citations, of which 21 were unique to the results from the PubMed search, for a total of 352 unique citations. After title and abstract review, 306 full-text articles were reviewed for inclusion and exclusion criteria. Ultimately, 218 full-text articles were included in this systematic review. Data extraction included number of enrolled subjects, measure of central tendency (typically mean) of the age of enrolled subjects, the classification of the intervention, and the classification of the primary or major study outcome (detailed in methods). A spreadsheet detailing all 352 unique citations with abstracted study details for included studies as well as reasoning for exclusion for excluded studies is available as supplementary material (Supplemental Table 1).

Figure 1.

PRISMA diagram outlining disposition of studies identified by literature search. Diagram was generated using the PRISMA Flow Diagram Generator (http://prisma.thetacollaborative.ca).

Characteristics of identified subjects

Table 1 reports the characteristics of subjects from systematically identified studies. We identified 218 unique studies, enrolling a total of 11,213 subjects, with mean ages ranging from 1 to 42 years old (Fig. 2a, b). The vast majority (184 of 218 (84%)) of studies enrolled subjects with a mean age under 18. Similarly, 92% of all subjects were enrolled in studies with a mean age under 18, in stark contrast to a recent prevalence estimation that only about one-third of the total population of individuals with ASD in the United States are under 18 years of age (Buescher et al. 2014). Studies enrolling subjects under 18 years old were also significantly larger than those enrolling subjects 18 years old or older (median number of subjects: 40 (Interquartile range (IQR): 19-65) vs. 21 (IQR: 15-40), respectively).

Table 1.

Age distribution of subjects with autism spectrum disorder enrolled in therapeutic studies, 2013-2017.

Characteristic	Total sample	< 18 years mean subject age	≥ 18 years old mean subject age	p valuea < 18 years vs. ≥ 18 years
Number of studies (% total)	218	184 (84)	34 (16)	< 10−4
Number of subjects (% total)	11,213	10,285 (92)	928 (8)	< 10−4
Median number of enrolled subjects per trial (IQR)	39 (18-60)	40 (19-65)	21 (15-40)	0.0035

IQR: interquartile range

^ap value for number of studies and number of subjects calculated by comparing expected numbers of studies/subjects from ASD prevalence data (Buescher et al. 2014) vs. observed from systematic literature search.

Figure 2.

Histogram demonstrating the number of studies (a) and subjects (b) plotted against the mean enrolled subject age in all identified therapeutic intervention studies for individuals with ASD. Vertical dotted line is at 18 years of age. Plots demonstrating the cumulative distribution of studies by mean enrolled subject age for intervention modality subtypes (c) and primary study outcomes (d). P values are from Mann-Whitney test (c) or Kruskal-Wallis test followed by Dunn’s post-test (d).

Intervention modalities and outcome measures

Tables 2 and 3 report the distribution of intervention types and outcome measures across the total identified study population, and compares distributions between individuals less than 18 years of age or 18 years and older. Both number of studies (Table 2) and number of subjects (Table 3) were analyzed. Most interventions were behavioral or pharmacological (> 90% of studies and subjects). As compared to individuals 18 years old or older, individuals under 18 years old were more likely to be enrolled in behavioral studies and less likely to be enrolled in pharmacological studies. Furthermore, the age distribution between behavioral and pharmacological studies differed significantly (p = 0.0005, Mann-Whitney test, Fig. 2c). Somatic and dietary interventions involved less than 10% of studies and subjects. The most common outcome measure was core and related ASD symptoms and behaviors, reported as a major outcome of interest in 66% of studies. However, this outcome was significantly more likely to be reported in studies enrolling younger individuals than older individuals with ASD. Conversely, functional and physiological/neurocognitive outcomes were significantly more likely to be reported in studies enrolling older individuals with ASD. The age distribution between outcome measures also differed significantly (Core/related vs. functional, p = 0.0016; Core/related vs. physiological/neurocognitive, p < 10⁻⁴; Functional vs. physiological/neurocognitive, p > 0.99, Kruskal-Wallis test followed by Dunn’s post-test, Fig. 2d).

Table 2.

Distribution of ASD therapeutic studies by intervention modality and outcome of interest between 2013-2017.

Characteristic	Total number of studies (%)	Number of studies < 18 years mean subject age (%)	Number of studies ≥ 18 years old mean subject age (%)	Odds ratio (95% CI) for < 18 year old relative to ≥ 18 year old	p value < 18 years vs. ≥ 18 years
Intervention classification
Behavioral	128 (59)	112 (61)	16 (47)	1.75 (0.84-3.65)	0.18
Pharmacological	70 (32)	53 (29)	17 (50)	0.41 (0.19-0.85)	0.026
Somatic	13 (6)	12 (7)	1 (3)	2.30 (0.28-18.31)	0.70
Dietary	7 (3)	7 (4)	0 (0)	2.92 (0.16-52.24)	0.60
Outcome of interest
Core and related ASD symptoms and behaviors	144 (66)	132 (72)	12 (35)	4.65 (2.15-10.08)	0.0001
Function	27 (12)	17 (9)	10 (29)	0.24 (0.10-0.59)	0.0029
Physiological and neurocognitive	42 (19)	30 (16)	12 (35)	0.36 (0.16-0.80)	0.016
Intervention development	5 (2)	5 (3)	0 (0)	2.11 (0.11-39.12)	> 0.99

CI, confidence interval. In some cases total percentages do not add to 100% due to rounding.

Table 3.

Distribution of subjects in ASD therapeutic studies by intervention modality and outcome of interest between 2013-2017.

Characteristic	Total number of subjects (%)	Number of subjects < 18 years mean subject age (%)	Number of subjects ≥ 18 years old mean subject age (%)	Odds ratio (95% CI) for < 18 year old relative to ≥ 18 year old	p value < 18 years vs. ≥ 18 years
Intervention classification
Behavioral	6858 (61)	6351 (62)	507 (55)	1.34 (1.17-1.54)	< 10−4
Pharmacological	3544 (32)	3151 (31)	393 (42)	0.60 (0.52-0.69)	< 10−4
Somatic	386 (3)	358 (3)	28 (3)	1.16 (0.78-1.71)	0.51
Dietary	425 (4)	425 (4)	0 (0)	80.13 (5.00-1284)	< 10−4
Outcome of interest
Core and related ASD symptoms and behaviors	8828 (79)	8457 (82)	371 (40)	6.95 (6.03-8.00)	< 10−4
Function	762 (7)	467 (5)	295 (32)	0.10 (0.086-0.12)	< 10−4
Physiological and neurocognitive	1507 (13)	1245 (12)	262 (28)	0.35 (0.30-0.41)	< 10−4
Intervention development	116 (1)	116 (1)	0 (0)	21.27 (1.32-342.40)	< 10−4

CI, confidence interval.

Discussion

Our results confirm and extend previous research (Howlin and Taylor 2015; Warren et al. 2012; Wong et al. 2015) by quantitatively demonstrating a dramatic age disparity in the enrollment of individuals with ASD in therapeutic clinical trials, with more than 11 individuals under 18 years old enrolled for every enrolled individual 18 years old or older. In addition to the overall disparity, certain key findings are of interest. Firstly, it is notable that adults are more likely to be enrolled in pharmacologic rather than behavioral studies. It is likely that this finding results in part from ethical and regulatory processes, by which truly experimental drugs are more likely to be first trialed in consenting adults than youth. Given the neurodevelopmental nature of ASD, current pharmacologic paradigms are not plausibly expected to alter the underlying brain-based features of the disorder in older subjects, and medications are typically understood as being useful for management of comorbid symptom dimensions. Although this is a critical focus of study, it should not be pursued to the exclusion of other types of intervention. Increasingly, efforts are being made to support adults with ASD using treatment paradigms that rely on evidence collected from youth, including social skills groups (Howlin and Yates 1999), educational interventions in tertiary education settings, vocational supports (Nicholas et al. 2015), and psychotherapy (Van Schalkwyk and Volkmar 2015). The presence of such interventions in practice is reflective of the urgent clinical need for diverse supports for these individuals, and underlies the limitations of an evidence-base that is both narrow in extent and focus.

A second notable finding is the predominance of functional and neurocognitive outcomes relative to measures of core ASD symptomology in young adults and adults with ASD. This approach may be justified from a point of departure that emphasizes the importance of early intervention as critical to altering the ultimate trajectory for youth with ASD, with less expectation that core symptoms can be altered later in life. In reality, a neurobiological mechanistic understanding of ASD is not yet at the point where much can definitively be said regarding the window period for effecting significant and enduring change. At least one prominent neural model of ASD considers key differences in connectivity, with a pattern of increased connectivity between local circuits, and poorer connectivity between more spatially separated brain regions (Scott-Van Zeeland et al. 2010). These findings are not consistent across the lifespan (Ecker et al. 2015), and it is possible that as individuals age, different brain-based mechanisms may come into play by which they may compensate for their core differences in cognitive and relational style. This is especially important when one considers the off-label use of medication, primarily for behavioral symptoms, to treat patient groups outside of the age group in which the medication was originally trialed, such as very young children (Mandell et al. 2008; Persico et al. 2015). Further, it is likely that the logistics of studying physiological and neurocognitive measures using imaging or other instrumentation is far easier in older subjects.

A potential limitation of this study is selective inclusion bias, given that we did not perform a meta-analytic calculation of intervention effect sizes that would enable a quantitative estimate of bias, nor did we include unpublished studies. Along these lines, if intervention studies enrolling adults are less likely to demonstrate positive outcomes than those studies enrolling children, and therefore less likely to be published, our findings would inflate the disparity between studies across age groups. Even if this were the case, it is unlikely to account for the marked difference in number of studies between age groups. Similarly, we did not conduct an evaluation of the quality of study methodology, as our goal was to quantify which subclass of interventions were being applied to which age groups. Inclusion of solely peer-reviewed studies may provide some buffer against poorly executed research, however it should be noted that there have been ethical and methodological concerns regarding some of the research avenues classified as “somatic” in the present study. Furthermore, it should be acknowledged that functional outcomes might blur with core ASD symptom outcomes, especially when language and communication are measured, and should therefore be considered approximations. The large number of studies identified was a strength that likely mitigates any meaningful unidirectional bias in categorization.

Beyond strong theoretical arguments as to the need for a large and specific evidence base for supporting adults with ASD, outcomes research makes the case altogether clear. Adults with ASD are at ongoing risk for poor social and educational outcome (Shattuck et al. 2012), and their care poses a significant challenge to available healthcare delivery systems. Our findings however suggest growing awareness has not yet led to an increase in clinical research. Clinical research in adults with ASD is subject to multiple barriers (Burke et al. 2018). ASD is traditionally a focus of pediatricians and child psychiatrists/psychologists; as such, existing research programs may prioritize youth recruitment with emphasis on early intervention. Parental motivation to facilitate study engagement may wane with age, and the fragmented nature of adult ASD clinical services complicates effective outreach.

Given the magnitude of the disparity, future research initiatives and policy should focus on surmounting the barriers to the inclusion of these individuals to better understand how they are optimally supported. Future work should include a focus both on novel interventions, and for more systematic study of the diverse psychosocial interventions that are already being employed in order to attempt to support these individuals.