Three authors meticulously adhered to the guidelines for systematic reviews and meta-analyses (36, 37) and conducted an exhaustive search in two databases: Scopus and PubMed, with the search culminating on August 10, 2023. The search terms were designed based on strategies developed by Hedges’ team (38), encompassing types and means of intervention—such as AAI, including commonly practiced Equine-Assisted Therapy (EAT), Dog-Assisted Therapy (DAT), etc.—target population: ASD, and article types: articles published in English-language peer-reviewed journals (for detailed search strategy, see Appendix 1). In addition, supplementary searches were conducted on Google Scholar,¹ and related systematic reviews and meta-analyses were reviewed to ensure comprehensive inclusion of qualified publications. Following the search, two authors independently reviewed the titles and abstracts of the included articles before proceeding to full-text evaluation. In cases of discrepancy, a third author would make the final decision on inclusion. All screening processes were conducted using Excel (Microsoft, Redmond, Washington State, United States of America) and Endnote X9 (Clarivate Analytics, California State, United States of America) (for included and excluded literature, see Appendices 2–4). The detailed protocol was registered on PROSPERO with registration number: CRD42023431425 (for adjustments, see Appendix 5).

The study rigorously followed the recommended PICO (Participant-Intervention-Comparison-Outcome) criteria for defining inclusion and exclusion standards. (1) Participants: individuals with a clinical or professionally diagnosed ASD, in accordance with the Diagnostic and Statistical Manual of Mental Disorders (DSM), International Classification of Diseases (ICD), Autism Diagnostic Observation Schedule (ADOS), or Social Communication Questionnaire (SCQ). No age restrictions were applied in our study. (2) Interventions: AAI, including commonly used therapy animals like horses, dogs, and dolphins. (3) Comparisons: the literature should include comparisons between AAAT and other treatments or a blank control. Pre- and post-comparisons are also eligible for inclusion. (4) Outcomes: the selected studies must indicate the improvement in at least one core or common symptom of ASD, such as overall symptom severity, social function deficits, or restrictive/repetitive behaviors. Additionally, we will also analyze other traits significantly correlated with ASD, like global cognitive ability (IQ), executive function (adaptive behavior), language expressive ability, and other disruptive or abnormal behaviors associated with ASD. (5) Types of literature: the literature should comprise clinical trial articles published in peer-reviewed journals in English, including randomized controlled trials (RCTs) and observational studies. Given the high probability of non-randomized controlled trials (non-RCTs) in this field, as indicated by our preliminary analyses, our study also included non-RCTs to provide a comprehensive evaluation of the overall intervention effectiveness.

All included studies will be imported into Excel (Microsoft, Redmond, Washington State, United States of America) for data extraction, segregated into four primary categories: (1) Bibliographic information: this includes the first author, publication year, and journal of the study. (2) Participant information: comprises the diagnostic outcome of the population involved (e.g., ASD, ADHD), the diagnostic criteria used (e.g., DSM, ICD), the total number of subjects in the experimental and control groups, as well as the sample’s mean age, age range, and sex. It further includes the sample’s mean IQ and IQ range. (3) Experimental Information: This involves the type of intervention and control (e.g., EAT, DAT), the kind of study (RCT or Non-RCT), the duration of the intervention, the outcomes post-intervention, and the effect sizes, standard errors (SE), and confidence intervals (CI). It also contains information about the review or approval authorities for the intervention. (4) Quality of Included Studies: Incorporates an analysis of publication bias, quality assessment measures, registration protocols, and data availability.

Upon data extraction, we will utilize Stata 17 (Stata Corp, Texas State, United States of America) for data analysis. It is noteworthy that in this study, the mean difference (MD) will be uniformly used as the outcome measure. Thus, we will primarily extract the mean and standard deviation (SD) from the original articles to compute the MD. If the original articles report only the standardized mean difference (SMD), these will be converted into MD. When the meta-analysis includes multiple outcomes, standard summarization methods will be employed. Additionally, the study adheres to the common I² assessment for data heterogeneity. If I² > 50%, we consider the heterogeneity to be high and will conduct heterogeneity tests or subgroup analyses; if I² < 50%, the heterogeneity is considered within a reasonable range.

Given that both RCTs and Non-RCTs are included in this study, we will employ the Physiotherapy Evidence Database scale (PEDro) and the Cochrane Risk of Bias Tool (RoB) to evaluate the quality of RCTs, and the ROBINS-I (Risk of Bias In Non-randomized Studies – of Interventions) scale for the quality of the included non-RCTs. The PEDro scale is specifically designed to evaluate the methodological quality of RCTs in the field of physical therapy. Comprising 11 items, including random assignment, concealment of allocation, baseline similarity, and blinding of patients, therapists, and assessors, among others, this method provides an accurate reflection of the internal validity of trial outcomes. The RoB covers seven specific areas of bias, including random sequence generation, allocation concealment, and blinding of participants and personnel, offering a nuanced examination of potential biases in RCTs, thus complementing the PEDro scale and enhancing the robustness of the review. ROBINS-I assesses bias in seven domains: confounding, participant selection, classification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selection of reported results, thereby providing a comprehensive overview of the reliability and validity of non-RCTs, and ensuring their results are appropriately weighted in the meta-analysis. Combining these tools allows for a thorough, detailed evaluation of both randomized and non-randomized controlled trials, ensuring that potential sources of bias are considered in our analysis, thereby providing a foundation for the quality and strength of the evidence.

A comprehensive search of the Scopus and PubMed databases yielded 341 publications that met the initial inclusion criteria. Subsequently, a stringent selection process narrowed this to 38 articles that satisfied the final inclusion standards (for details on inclusion and exclusion, see Appendices 4, 5). Additionally, during the review of 57 previously published systematic reviews and meta-analyses, seven more publications were included (see Figure 1). Ultimately, 45 studies met the selection criteria. These 45 studies encompassed 1,212 ASD patients, averaging approximately 27 individuals per study; specifically, 859 were male, 328 were female, and three studies included 25 individuals without reporting sex. Among the participants, 51 were adults, 1,077 were minors, and 89 participants were unclassifiable as either. Regarding the duration of the interventions, 7 studies (15.22%) did not report this data. The remaining 38 studies had intervention durations ranging from 120 to 1,920 min, totaling 25,705 min, with an average intervention time per study of 676 min (for detailed information, see Table 1 and Supplementary Table S1). The animals used for assisted therapy included dogs, horses, guinea pigs, and dolphins. Horses were the most used species, featured in 32 out of the 45 studies, constituting 71.11% of all studies, followed by dogs at 20%. Regarding geographical distribution, 18 studies (40%) were conducted in the USA, 5 in Spain (11.11%), and 3 each in China and Australia. Developmental countries accounted for only 17.78% of all the research, suggesting a significant imbalance in the distribution of such studies.

Moreover, a detailed examination using Web of Science revealed exponential growth in the field of AAAT from 1980 to 2022. This growth can be segmented into three phases: the first phase from 1980–1997 saw the nascent stages of AAAT, with fewer than ten publications per year and limited citation impact. The second phase spanned from 1998–2010, during which the annual publication rate increased from 2–3 to over 30, and citations swelled to 11,902 by 2010. AAAT for ASD patients started to appear in the literature around 2002. The third phase started in 2011, with even more rapid growth in research and citations, including more than 20 annual publications focused on AAAT for ASD. The analysis of the burst intensity of the keywords revealed that the outbreak of the field started from 2013, and there were 23 keywords with a burst intensity of more than 2.50, and the highest attention was paid to the means of intervention constituted by animals such as horse, dog, etc., the population of intervention constituted by spectrum disorder, disease, elderly patients, etc., and the assisted therapy, physical therapy, and other keywords (see Figure 2).

Because one study was missing quality data, we evaluated 19 RCTs and 25 Non-RCTs. For RCTs, the PEDro and Cochrane Risk of Bias Tools were used for quality assessment, and for Non-RCTs, the ROBINS-I scale was employed. Among the 19 included RCTs, only three studies were free from high risk in all six subcategories: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective reporting. The remaining 17 studies had some level of high risk, especially in the subcategories of blinding of participants and personnel and blinding of outcome assessment. This is mainly due to the inherent challenges of conducting double-blind experiments in this type of research (see Figure 3). According to the PEDro scale, eight studies scored 7 points and were considered high quality, eleven studies scored between 4–6 points and were deemed fair quality, and one study scored 3 points, indicating poor quality. For the 25 assessed non-RCTs, four were rated as moderate risk, thirteen as serious risk, and eight as critical risk. None of the included non-RCTs had a low risk level (see Figure 3, Table 2, Appendices 6, 7). Overall, the studies included in this research generally demonstrated a high risk of bias and moderate quality levels.

In assessing the effectiveness of AAAT, standardized scales serve as the most prevalent methodologies. The 45 incorporated studies collectively employed a total of 24 distinct scales to assess core and common symptom domains in ASD patients, including social functioning and responsiveness, behavioral competencies, and motor skills, as well as stress levels.

For evaluating social functioning and responsiveness, instruments such as the Social Responsiveness Scale (SRS) (30, 34, 39–41), Pediatric Evaluation of Disability Inventory Computer Adaptive Test, Autism Spectrum Disorder Module (PEDI-CAT ASD) (41), Psychoeducational Profile Revised (PER-R) (42), Sensory Profile (SP) (30), Stone’s Social Skills Scale (SSSS) (43), and the Receptive and Expressive Vocabulary Test (REVT) (44) were commonly utilized. These instruments gauge various dimensions, such as social cognition, linguistic capabilities, and the capacity for daily social activities in ASD patients.

To assess motor functioning and behavioral competencies, scales like Aberrant Behavior Checklist–Community (ABC-C) (32), VABS (32), BOT (32), SIPT (32), Test of Gross Motor Development-Third Edition (TGMD-3) (45), Vineland (46), Developmental Coordination Disorder Questionnaire, as revised in 2007 (DCDQ’07) (46), and Interaction Emotions Motor Skills (IEMS) (46) were employed as primary tools.

Furthermore, the evaluation of stress states in ASD patients included scales that measure perceived stress, such as the Perceived Stress Scale (PSS) (34), the Symptom Checklist-90-Revised (SCL-90-R) (34) for psychological and physical symptoms, the ABC-C (39, 41) for irritability and hyperactivity, and the Rosenberg Self Esteem Scale (RSES) (34) for self-esteem levels. Physiological indicators, specifically cortisol levels, were also assessed (39, 41, 47–49).

In studies adopting descriptive statistics, measures such as the Canadian Occupational Performance Measure (COPM) (40), on-task behavior, and Goal Attainment Scaling (GAS) were primarily employed to ascertain whether AAI would impact the rate of goal achievement and task completion time among ASD participants. This served to validate the potential improvements AAI could confer on social behaviors and task performance for individuals with ASD.

Social responsiveness and functioning are pivotal symptom domains in individuals with ASD. Within the scope of this study, four research articles employed the SRS to comprehensively analyze the social responsiveness and functioning of ASD patients. Of these, three studies utilized Barn Activities (BA) as the control group (25, 30, 39), and one employed Occupational Therapy (OT) (41). The SRS dissects social functioning into five sub-categories: social awareness, social cognition, social communication, social mannerism, and social motivation. According to the meta-analysis, in comparison with control groups, AAAT exhibited a significant improvement in the social communication domain for ASD patients (MD = −4.96, 95% CI [−7.49, −2.44]). However, no improvements were observed in social awareness (MD = −1.63, 95% CI [−4.07, 0.81]), social cognition (MD = −3.60, 95% CI [−9.36, 2.17]), social mannerism (MD = −0.73, 95% CI [−2.55, 1.09]), and social motivation (MD = −1.21, 95% CI [−2.56, 0.13]) (see Figure 4 and Supplementary Table S2). Even after eliminating studies with high sources of heterogeneity, adjusted results continued to show no improvements in social awareness (MD = −0.47, 95% CI [−1.44, 0.50], I² = 0.00%) or social cognition (MD = −1.55, 95% CI [−3.26, 0.15], I² = 0.00%) (see Appendix 8).

Another critical area of social functioning is linguistic capability, a significant impediment for ASD patients. Among the 45 studies included, only two provided comprehensive data using SALT for analyzing language abilities in ASD patients (25, 39). The analysis indicated a significant improvement in the utilization of different words (MD = 20.48, 95% CI [7.41, 33.55]) and the existing vocabulary (MD = 59.91, 95% CI [22.60, 97.21]) in ASD patients undergoing AAI (see Figure 4B).

These findings align generally with previous research, indicating improvements in social motivation and performance (30, 41, 43, 50), enhanced language expression and communication skills (32, 39, 44, 51–54), and elevated emotional understanding, empathy, and perspective-taking abilities (43, 50, 55). It also confirmed the efficacy of AAAT in enhancing social functioning (39, 56). In summary, AAI interventions appear to facilitate some improvement in the social and linguistic abilities of ASD patients. However, current meta-analytical evidence strongly supports the impact of AAI only on social communication, leaving improvements in other domains a subject of ongoing debate.

Behavioral impairments are among the core symptoms manifest in individuals with ASD. Research reveals that approximately 83% of children with ASD demonstrate varying degrees of developmental delays in behavioral ability (45). Such deficits curtail the likelihood of these individuals participating in physical and social activities, exacerbating social and familial isolation and consequent anxiety (57). This further escalates the manifestation of repetitive and restrictive behaviors. The current study predominantly employed the ABC-C and VABS scales to gauge behavioral functions in ASD patients. Five studies that utilized the ABC-C and provided comprehensive data (25, 39, 41, 56, 58) were incorporated. According to meta-analysis, AAAT significantly decreased irritability (MD = −2.38, 95%CI [−4.06, −0.71]) and hyperactivity (MD = −4.03, 95%CI [−6.17, −1.89]) in ASD patients. No significant improvements were noted in subdomains, such as lethargy (MD = −1.12, 95%CI [−3.92, 1.68]), stereotypical (MD = −0.23, 95%CI [−1.27, 0.80]), or inappropriate speech (MD = −0.28, 95%CI [−0.99, 0.42]) (see Figure 5).

Additionally, four studies employed a pre-post design using VABS to measure behavioral function in ASD patients (25, 32, 46, 55). The evidence indicates significant improvements in the socialization subdomain (MD = 7.47, 95%CI [0.50, 14.44]) following AAAT, whereas no significant changes were observed in the communication (MD = 5.54, 95%CI [−1.59, 12.68]), daily living skills (MD = 7.56, 95%CI [−0.66, 15.78]), or adaptive total score (MD = 3.05, 95%CI [−0.43, 6.53]) (see Figure 5B).

These findings are consistent with earlier research that attests to the significant improvement of behavioral capacities in ASD patients through AAAT programs (25, 50, 59). Such interventions have also been found to reduce problem behaviors and aggression (60, 61), as well as enhance self-regulation, adaptive behavior, and coordination (32, 46, 54, 55, 62). Importantly, AAAT has been noted to significantly increase the minutes of mild physical activity and substantially reduce sedentary minutes in ASD patients (30, 63). However, some studies suggest that improvements in task completion rates and times in AAAT-involved ASD patients do not differ significantly from those in non-animal-involved occupational therapy control groups (28). Further evidence is needed to ascertain the comparative efficacy of AAI over other therapies.

Prior research has substantiated those individuals with ASD experience higher levels of stress and lower self-esteem compared to the general population (2, 34, 49). Stress positively correlates with depression, anxiety, and cardiovascular diseases (64), while self-esteem strongly inversely correlates with stress (65). Long-term exposure to high stress levels can exacerbate cognitive flexibility challenges in ASD patients, thereby hindering the acquisition of basic social skills (66). Therefore, mitigating stress can effectively improve the depressive and anxious states of ASD patients and augment their self-esteem and social interest. On stress-related outcome measures, AAAT has been shown to alleviate both physiological and subjective stress states in ASD patients as reported by O’Haire, Wijker et al., Peters et al., among others (41, 47–49). However, current evidence falls short in confirming the long-term efficacy of AAAT in stress reduction for ASD patients.

The dynamics of parent–child relationships and parental stress serve as significant mediators in the treatment of individuals with ASD. Prior research has highlighted elevated levels of stress and frustration experienced by caregivers of ASD patients (67). Such stress levels are instrumental in shaping the management and caregiving strategies for ASD. Among the studies incorporated in this research, two specifically addressed this criterion. Zoccante et al. (46) found that EAT yielded no significant effect in ameliorating parent–child relationships or alleviating parental stress. Conversely, Ozyurt et al. (68) reported that AAT contributed positively to the mental health and familial functionality of mothers of children with ASD. At present, the scarcity of studies on this outcome measure makes it challenging to arrive at conclusive judgments.

Another crucial metric for assessing interventions for ASD encompasses follow-up outcomes. However, only seven studies reported on these. Llambias et al. (8) documented those children with ASD who participated in AAT programs maintained significantly improved social engagement skills. Wijker and associates (34, 58) emphasized that post-AAT, the reduction in perceived stress and agoraphobia among ASD patients continued to be effective, and improvements in social skills persisted. However, both Viau et al. and Wijker et al. noted that despite the decline in cortisol levels among ASD patients following AAT, no long-term effects were observed (48, 49).

Furthermore, Ward et al. (33), O’Haire et al. (16), and Holm et al. (69) conducted intermittent experiments, where initial AAAT sessions resulted in notable improvements in social functioning and substantial reductions in social withdrawal behaviors. However, the outcomes varied during the interlude between AAAT sessions. When AAAT was resumed, the improvement resurged. Holm et al. (69) also emphasized the significance of dosage in AAT, asserting that increasing the dosage of AAAT is instrumental for enhancing target behaviors in ASD patients. Nevertheless, no conclusive evidence is currently available to determine the optimal dosage of AAAT for maximal efficacy.