Structured interactive play for autism spectrum disorder screening in toddlers: a primary care-based cohort study

Abstract

Background

Early identification of autism spectrum disorder (ASD) is crucial for improving outcomes, yet diagnosis is often delayed. Primary care faces challenges, including the limited accuracy of existing tools and scarce access to specialist assessments. This prospective cohort study evaluated the feasibility of a novel, structured examiner-child interactive play protocol as an early behavioral screening tool for ASD.

Methods

We recruited 260 toddlers aged 18–36 months. At baseline, all children participated in a structured interactive play session. Their behaviors (e.g., eye contact, turn-taking, social smiling, symbolic play) were coded from video recordings to quantify nine core indicators. After a 6-month follow-up, 217 children completed comprehensive clinical assessments. Core symptoms were evaluated using the Childhood Autism Rating Scale (CARS), and developmental level was assessed using the Bayley Scales of Infant and Toddler Development-Fourth Edition (Bayley-4). Based on clinical assessment results, participants were classified into an ASD group (n = 45) and a typically developing (TD) group (n = 165) for primary analysis. Spearman correlation analysis was used to examine the relationships between behavioral indicators, developmental levels, and symptom severity. Receiver operating characteristic (ROC) curve analysis assessed each indicator’s discriminative validity for ASD.

Results

During the structured interactive play, the ASD group performed significantly worse on all behavioral indicators than the TD group. Correlation analysis revealed that several behavioral indicators, including eye contact, turn-taking, and symbolic play, showed positive correlations with cognitive (rs = 0.30–0.51) and language ability (rs = 0.34–0.43) and negative correlations with symptom severity (rs=-0.53–-0.36). ROC analysis identified that eye-contact metrics were the strongest discriminators. Specifically, eye contact frequency yielded an AUC of 0.915 (sensitivity 96.36%, specificity 73.33%), and total eye contact duration achieved an AUC of 0.899 (sensitivity 86.67%, specificity 84.44%).

Conclusion

Behavioral observation through examiner-child structured interactive play, particularly metrics related to eye contact, effectively differentiates toddlers with iASD. This objective, low-cost method shows potential as a primary screening tool for enhancing early ASD detection in primary care settings.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12888-026-08274-9.

Introduction

Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder characterized by core features of social communication deficits and restricted, repetitive behaviors [1]. The current prevalence of ASD in the United States is approximately 1 in 31 [2], and this rate has shown a long-term increasing trend. In China, the ASD prevalence among 6-year-old children was first reported as 1.05% in 2021 [3]. Extensive research has demonstrated that early intervention can significantly improve long-term outcomes for children with ASD [4]. Although existing evidence indicates that ASD diagnosis can be stable as early as 14 months of age, the median age of diagnosis remains as high as 47 months [2, 5]. Consequently, a significant gap exists between delayed diagnosis and the critical window for intervention, highlighting the urgent need to develop and apply practical early screening tools to address this disparity.

Primary care is a core component of early childhood health management and plays a critical role in the early identification of ASD [6]. The American Academy of Pediatrics(AAP) recommends routine ASD screening at 18 and 24 months in primary care settings, accompanied by ongoing developmental monitoring [7, 8]. Within this system, widely used caregiver-reported screening tools, such as the Modified Checklist for Autism in Toddlers and its revised version (M-CHAT-R/F), are cost-effective and hold established clinical value. A recent systematic review and meta-analysis revealed that the M-CHAT and M-CHAT-R/F demonstrate a pooled sensitivity of 0.83 and a pooled specificity of 0.94 [9]. However, substantial cross-study heterogeneity in performance has been noted, which may affect their consistent utility in varied clinical and research contexts. Furthermore, because these tools rely heavily on caregiver reports, they are inherently susceptible to subjective differences in parental perception and experience. This subjectivity makes it challenging to consistently capture subtle social and communication abnormalities in ASD, underscoring the complementary need for objective, observation-based screening paradigms [10]. The gold-standard diagnostic instruments, such as the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2) [11] and the Autism Diagnostic Interview-Revised (ADI-R) [12], both require trained professionals to administer and are time-consuming, limiting their accessibility in the context of high ASD prevalence and resource-limited primary care settings. Recent advances in technologies such as magnetic resonance imaging (MRI) [13] and eye tracking [14] have provided more objective quantitative tools for early ASD identification. However, these techniques are costly, complex to operate, and require specialized analysis, making them difficult to implement in resource-limited primary care settings. Consequently, research trends are gradually shifting toward identifying early behavioral markers of ASD in real-world social contexts to develop more objective, reliable, and cost-effective identification strategies.

Signs of social abnormalities in children with ASD begin to emerge during infancy and toddlerhood [15], manifesting in areas such as social attention [16, 17], social smiling [18], response to name calling [19], action imitation [20], symbolic play [21], and emotion regulation [22, 23], with sustained impacts on their daily functioning. Studies have validated some of these behavioral characteristics through video coding in paradigms such as the still-face or caregiver-child free play [24–27].In these paradigms, the mother’s task is to play with the infant and elicit interactive behaviors in ways typical of daily life. However, the lack of experimental control in such paradigms may limit the reliability of findings and cross-study comparability. Furthermore, related studies typically require specialized behavioral analysis software for detailed coding, with relatively high technical thresholds, making direct application difficult in resource-limited, efficiency-focused primary care settings. Although evidence suggests these behavioral characteristics have some degree of cross-situational stability [28], current research based on structured examiner-child interaction paradigms remains limited. Utilizing systematically trained examiners to conduct structured interactions may offer better methodological controllability, but their clinical validity and generalizability still require further verification.

Therefore, this study developed a 5-7-minute structured examiner-child interactive play paradigm. The paradigm sequentially includes tasks such as turn-taking games, instruction following, symbolic play, action imitation, and affective response to observe core social behavior indicators of ASD. This protocol only requires simple toys and standard video recording equipment, demonstrating good standardization and clinical feasibility. This study aims to systematically examine differences in core social behavioral characteristics between ASD and Typically Developing(TD) toddlers through standardized, structured interactive play; thoroughly explore the relationships between these behavioral indicators and developmental level and ASD symptom severity; and objectively verify their discriminative efficacy for identifying ASD toddlers.

Methods

Participants

This prospective cohort study consecutively recruited 260 children aged 18–36 months (162 males, 98 females) who were receiving routine health care at the Child Health Department of Chongqing Bishan Maternal and Child Health Hospital between July 2024 and February 2025. The recruited sample comprised both children receiving routine child health care and children whose caregivers actively sought consultation due to concerns about their social communication development. At baseline, all children completed a structured examiner-child interactive play task during visits, with the entire process video-recorded. A follow-up was conducted 6 months after enrollment for the whole sample. Forty-three participants were lost to follow-up due to inability to contact or refusal to provide information, resulting in 217 participants who completed the follow-up assessment. Follow-up assessments included the Childhood Autism Rating Scale (CARS) [29], the Chinese standardized edition of the Bayley Scales of Infant and Toddler Development-Fourth Edition (Bayley-4) [30], and medical history collection. To ensure strict methodological rigor and prevent assessment bias, the evaluation processes were completely decoupled. The clinical assessors who administered the CARS and Bayley-4 were entirely blinded to the offline behavioral coding scores of the play paradigm. Based on the follow-up results, the 217 children were classified into three groups: the ASD group, the other neurodevelopmental disorders group (NDD group), and the TD group. The specific grouping criteria were as follows:

ASD group inclusion criteria: clinical confirmation of meeting DSM-5 diagnostic criteria for ASD by a senior developmental-behavioral pediatrician. Exclusion criteria: presence of opresence of a genetic or neurological disorder of known etiology (e.g., Rett syndrome, fragile X syndrome); severe sensory or motor impairment.

NDD group inclusion criteria: diagnosis of language delay (LD) or global developmental delay (GDD) confirmed by a senior developmental-behavioral pediatrician based on Bayley-4 results or clinical presentation. Exclusion criteria: presence of opresence of a genetic or neurological disorder of known etiology (e.g., Rett syndrome, fragile X syndrome); severe sensory or motor impairment.

TD group inclusion criteria: typically developing, not meeting any of the inclusion criteria for the ASD group or the NDD group. Exclusion criteria: presence of opresence of a genetic or neurological disorder of known etiology (e.g., Rett syndrome, fragile X syndrome); severe sensory or motor impairment. Figure 1 shows the flow of participant recruitment and grouping.

Fig. 1.

Participant flow diagram. ASD, autism spectrum disorder; TD, typical development; NDD, other neurodevelopmental disorders

This study was approved by the Ethics Committee of Children’s Hospital of Chongqing Medical University (Approval No. 2024 (Research) 305). Guardians of all participants provided written informed consent.

Measures

Clinical assessments

Core symptoms

The Childhood Autism Rating Scale (CARS) [29] is a standardized tool used to diagnose and assess the severity of ASD in children. The scale consists of 14 core items assessing ASD-related behaviors, with an additional 15th item for rating the overall impression of autistic features. Each item is scored on a 4-point scale (1–4), yielding a total score of 15–60. Higher scores indicate greater functional impairment.

Developmental level

The Bayley Scales of Infant and Toddler Development, Fourth Edition (Bayley-4) [30] is a core developmental instrument for assessing early cognitive, expressive language, receptive language, fine motor, and gross motor abilities in infants and toddlers. It is regarded as the gold standard for clinical assessment of early child development. Its Chinese edition has been introduced and standardized by professional institutions. This study used the Chinese standardized edition of the Bayley-4 to assess children’s developmental levels. The scale scores reflect a child’s performance on specific subtests or domains relative to a normative peer group. The scaled scores for the Cognitive, Language, and Motor domains each range from 1 to 19, with a mean of 10 and a standard deviation of 3.

Structured examiner-child interactive play paradigm

This study used a self-developed, structured interactive play paradigm to systematically assess social behavior in toddlers in primary care settings. The total paradigm duration was approximately 5–7 min. All assessments were conducted in a quiet, well-lit, standardized testing room. The child was seated at the testing table, with the caregiver positioned to the child’s right rear and the examiner facing the child directly. Before formal tasks began, the examiner conducted a 5-minute interactive warm-up with the child to establish rapport.

The paradigm consisted of five core behavioral observation tasks: (1) turn-taking game, using a ball for three rounds of push-ball interaction; (2) instruction following, assessing identification of seven body parts (eyes, nose, mouth, ears, hair, hands, feet) on a doll; (3) symbolic play, including pretend play (completing three pretend actions in sequence: covering with a blanket, feeding, and talking on a phone) and role-play (open-ended doctor-themed role-play); (4) action imitation, requiring the child to imitate the examiner’s table-wiping action; (5) affective response, observing the child’s reaction to the caregiver simulating stomach ache with a painful expression. To ensure tasks mapped onto normative developmental trajectories and captured age-specific socio-communicative deficits [31–34], the tasks were developmentally calibrated by age: the 18–24 month group completed turn-taking game, instruction following, pretend play, and action imitation; the 24–30 month group completed turn-taking game, instruction following, pretend play, action imitation, and affective response(as empathic responding becomes a stable, measurable trait during this window); and the 30–36 month group completed turn-taking game, instruction following, the more cognitively demanding role-play, and action imitation, thereby maximizing discriminative sensitivity across age layers. All assessments were performed by psychological assessors. Prior to the study, they underwent standardized training strictly focused on protocol adherence, which included mastering unified verbal prompts, standardized material presentation methods, and strict enforcement of task duration limits. Supplementary Material Table S1 provides the detailed standardized procedure. The complete interaction between each participant and the examiner was recorded from two angles (frontal and lateral) for subsequent offline behavioral coding.

Behavioral coding

At this proof-of-concept validation stage, this study coded offline videos to methodologically ensure rater blinding and explicitly operationalize all indicators. We quantified nine behavioral indicators: frequency and total duration of eye contact, turn-taking, social smiling, task completion time, instruction following, symbolic play, affective empathy, and imitation. Notably, task completion time was recorded within the social context of the turn-taking game, serving as a behavioral marker that captures the combined effects of compromised social reciprocity and cognitive inflexibility in ASD. The first five indicators were coded based on the turn-taking game task, while the remaining indicators were derived from their corresponding independent tasks. Table 1 provides the detailed definitions and scoring criteria for each indicator.

Table 1.

Description of behavioral indicators

Age range(months)	Behavioral indicator	Definition	Scoring criteria/notes
18–36	Frequency of eye contact (per min)	Ratio of the total number of valid eye contacts to total game duration.	Valid eye contact was defined as a single instance lasting > 0.5s.
	Total duration of eye contact (s/min)	Ratio of total duration of valid eye contacts to total game duration.
	Turn-taking	The examiner rolls a ball to the child, and the number of valid times the child rolls it back is recorded.	A total of 3 attempts were made. The number of valid returns was recorded. Non-return actions (e.g., throwing, slamming) were not counted.
	Social smiling	Smiling or laughing while gazing at the face.	0 points: negative emotions like crying or anger; 1 point: facial gazing without positive expression; 2 points: facial gazing co-occurring with a smile or laugh.
	Task completion time (s)	Total time required to complete the turn-taking game.	If the child refused to cooperate and no effective interaction occurred for more than 2 min, the task was terminated, and the next game began.
	Instruction following	The child was asked to point to the eyes, nose, mouth, ears, hair, hands, and feet (7 body parts) of a doll or themselves upon instruction.	The number of correctly identified target body parts was recorded.
	Imitation	After the examiner demonstrated wiping a table, the child’s behavior was observed and recorded for imitation.	0 points: no imitation; 1 point: imitation observed.
18–30	Symbolic play	The child was required to complete three pretend play tasks equentially: “covering with a blanket,” “feeding”, and “making a phone call”	Scores ranged from 0 to 3, corresponding to the number of tasks completed.
30–36	Symbolic play	In the role-play task, the child used doctor-themed props to engage in play.	0 points: no pretend play; 1 or 2 points: used some or all props for pretend play without social interaction; 3 points: complete pretend play with initiated social interaction.
24–30	Affective empathy	Upon the examiner’s instruction, the caregiver simulated a stomach ache with a pained expression, and the child’s response was recorded.	0 points: did not turn to the caregiver; 1 point: turned to the caregiver but provided no comforting behavior; 2 points: turned to the caregiver accompanied by comforting vocalizations, actions, or expressions.

Behavioral coding was completed by independent graduate students in pediatrics who received uniform training covering the operational definitions of behavioral indicators, specific scoring criteria, and standardized coding workflows. Coders were only permitted to independently code after passing a reliability threshold (ICC/Kappa > 0.80). The coding process employed a blind method, with coder A coding all videos. To evaluate inter-rater reliability, a stratified random sampling method was used, randomly selecting 30% of the samples from the ASD group and the TD group, respectively, for independent coding by coder B. Inter-rater reliability for continuous variables was assessed using the intraclass correlation coefficient (ICC), and Cohen’s Kappa was used for categorical variables. The results indicated high reliability for all indicators, as follows: frequency of eye contact (ICC = 0.82), total duration of eye contact (ICC = 0.903), turn-taking (Kappa = 0.829), social smiling (Kappa = 0.765), task completion time (ICC = 0.97), instruction following (Kappa = 0.862), symbolic play (Kappa = 0.865), imitation (Kappa = 0.814).

Statistical analysis

Statistical analysis and graphing were performed using SPSS 29.0, GraphPad Prism, and Origin software. Continuous variables were first tested for normality using the Shapiro-Wilk test. Data with normal distributions are presented as mean ± standard deviation and were compared between groups using independent samples t-tests. Non-normally distributed data are presented as median (interquartile range) and were compared using Mann-Whitney U tests. Categorical variables were presented as frequencies (percentages) and compared between groups using chi-square tests (for binary variables) and Mann-Whitney U tests (for ordinal variables). Spearman’s rank correlation was used to analyze the relationships among variables. Receiver operating characteristic (ROC) curve [35] analysis was applied to evaluate the discriminative ability of each behavioral indicator between the ASD and TD groups. The area under the curve (AUC) [36] was calculated. The optimal cut-off value was determined from the ROC curve, and the corresponding sensitivity and specificity are reported. P < 0.05 indicated that the difference was statistically significant. Due to the small sample size in the NDD group (n = 7), data from this group were subjected only to descriptive statistical analysis and were not included in subsequent group comparisons.

Results

Participant information

In this prospective cohort study, 217 out of the initially enrolled 260 children aged 18–36 months completed the follow-up. Following comprehensive clinical evaluation, participants were classified into an ASD group (n = 45), an NDD group (n = 7), and a TD (n = 165). Due to the limited sample size and insufficient statistical power in the NDD group, primary statistical analyses focused on comparisons between the ASD and TD groups, with data from children of different ages combined. Based on clinical considerations and guardians’ informed consent, 87 children underwent assessments using the CARS and the Bayley-IV (27 in the ASD group, 58 in the TD group, and 2 in the NDD group).

Table 2 shows the demographic and clinical characteristics of the ASD and TD groups. Baseline characteristics revealed no statistically significant differences between the two groups in age, age distribution, or sex composition (all P > 0.05). In terms of development and symptoms, the ASD group scored significantly lower than the TD group across all Bayley-IV subtests, including cognition, language (receptive and expressive), motor (gross and fine), social-emotional, and adaptive behavior (all P < 0.01). Additionally, the total CARS score was significantly higher in the ASD group compared to the TD group (P < 0.001).

Table 2.

The demographic and clinical characteristics of the two groups

Item	ASD group	TD group	Statistic	P
Baseline demographic characteristics( N  = 210)	n  = 45	n  = 165
Age (months)	24.38(21.68,30.32)	24.16(19.83,25.71)	Z=-1.359	0.174
Age distribution, n (%)			χ2 = 2.007	0.367
18–24 months	14(31.1)	66(40)
24–30 months	19(42.2)	69(41.8)
30–36 months	12(26.7)	30(18.2)
Sex, n (%)			χ2 = 3.168	0.075
Male	34(75.6)	101(61.2)
Female	11(24.4)	64(38.8)
Clinical characteristics (N = 85)	n = 27	n = 58
Bayley-IV subtest scores
Cognitive	8.63 ± 2.24	11.74 ± 2.10	t=-6.231	< 0.001
Receptive language	6.89 ± 3.56	11.79 ± 2.08	t=-6.655	< 0.001
Expressive language	6.41 ± 2.50	11.03 ± 2.18	t=-8.701	< 0.001
Gross motor	10.11 ± 2.03	11.50 ± 1.82	t=-3.161	0.002
Fine motor	9.67 ± 2.66	11.95 ± 2.45	t=-3.894	< 0.001
Social-emotional	7(5,9)	9(7,11)	Z=-3.182	0.001
Adaptive behavior	40.59 ± 9.34	49.69 ± 10.04	t=-3.976	< 0.001
CARS total score	29(27,32)	18(16,19)	Z=-7.419	< 0.001

Continuous variables are presented as mean ± standard deviation or median (interquartile range), depending on the data distribution; categorical variables are expressed as number (percentage). CARS, The Childhood Autism Rating Scale; Bayley-IV, The Bayley Scales of Infant and Toddler Development, Fourth Edition; ASD, autism spectrum disorder; TD, typical development

Comparison of behavioral indicators between ASD and TD groups during examiner-child interactive play

Table 3 shows that, aside from the action imitation task, which excluded 14 cases due to equipment synchronization issues (final n = 196), the analyses for the turn-taking, instruction following, and symbolic play tasks were based on a sample of 210 cases. The results showed that, compared to the TD group, the ASD group exhibited significantly lower eye contact frequency, shorter total duration of eye contact, fewer turn-taking instances, and less social smiling during the turn-taking game, and required longer task completion time (all P < 0.05). Furthermore, the ASD group performed significantly worse than the TD group on the instruction following, symbolic play, and action imitation tasks (all P < 0.01). Additionally, in the 24–30-month-old subsample, after excluding 16 cases from the affective response task due to equipment synchronization issues, analysis of the 72 valid cases revealed that the ASD group scored significantly lower than the TD group on affective empathy (P < 0.001).

Table 3.

Comparison of behavioral indicators between the ASD and TD groups

Behavioral indicator	ASD group	TD group	Z/χ2	P
Turn-taking, instruction following and symbolic play tasks ( N  = 210)	n  = 45	n  = 165
Frequency of eye contact(per min)	3.46(2.16,5.69)	11.43(7.71,14.17)	Z = 8.530	< 0.001
Total duration of eye contact(s/min)	4.39(2.16,6.90)	14.29(9.53,20)	Z=-8.194	< 0.001
Turn-taking			Z=-7.071	< 0.001
0	13(28.9)	2(1.2)
1	4(8.9)	2(1.2)
2	7(15.6)	11(6.7)
3	21(46.7)	150(90.9)
Social smiling			Z=-2.242	0.025
Score 0	7(15.6)	1(0.6)
Score 1	6(13.3)	26(15.8)
Score 2	32(71.1)	138(83.6)
Task completion time (s)	53(39,79.5)	35(26,55)	Z=-3.876	< 0.001
Instruction following(No. correct)	5(1,7)	7(6,7)	Z=-3.576	< 0.001
Symbolic play			Z=-4.163	< 0.001
Score 0	11(24.4)	15(9.1)
Score 1	12(26.7)	19(11.5)
Score 2	15(33.3)	64(38.8)
Score 3	7(15.6)	67(40.6)
Action imitation task(N = 196)	n = 41	n = 155
Imitation			χ2 = 8.837	0.003
Score 0	16(39)	27(17.4)
Score 1	25(61)	128(82.6)
Affective response task(N = 72)	n = 13	n = 59
Affective empathy			Z=-3.515	< 0.001
Score 0	8(61.5)	6(10.2)
Score 1	4(30.8)	34(57.6)
Score 2	1(7.7)	19(32.2)

Continuous variables are presented as median (interquartile range); categorical variables are expressed as frequency (percentage) [n (%)]. ASD, autism spectrum disorder; TD, typical development

Relationship between behavioral indicators, developmental levels, and social symptom severity

Spearman correlation analysis revealed that during structured interactive play, toddlers’ core social communication behaviors (eye contact, turn-taking, instruction following, and symbolic play) were positively correlated with Bayley-IV cognitive, language, and adaptive behavior scaled scores and negatively correlated with the total CARS score. Among these, eye contact frequency and total eye contact duration showed the strongest negative correlations with the total CARS score (rs = -0.53, P < 0.0001). In contrast, task completion time exhibited the opposite association patterns. No significant correlations were found between social smiling and any of the assessment indicators (Fig. 2). Table 4 presents the complete correlation coefficient matrix.

Fig. 2.

Correlation heatmap. The x-axis represents the CARS total score and Bayley-4 subscale scores, and the y-axis represents behavioral indicators. Color intensity reflects the magnitude of the correlation coefficients, with red and blue indicating positive and negative correlations, respectively. Significance levels are marked with asterisks in each cell: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

Table 4.

Spearman correlations of behavioral indicators with symptom severity and developmental scale scores(rs-value)

Behavioral indicator	CARS score	Bayley-IV subtest scores
		Cognitive	Receptive language	Expressive language	Gross motor	Fine motor	Social-emotional	Adaptive behavior
Frequency of eye contact	-0.53****	0.36***	0.37***	0.40***	0.19	0.26*	0.22*	0.32**
Total duration of eye contact	-0.53****	0.30**	0.34**	0.35***	0.13	0.22*	0.20	0.27*
Turn-taking	-0.48****	0.37***	0.37***	0.38***	0.14	0.19	0.18	0.36***
Social smiling	-0.10	0.05	0.14	0.08	-0.09	-0.07	-0.05	-0.00
Instruction Following	-0.36***	0.46****	0.43****	0.37***	0.11	0.29**	0.17	0.25*
Task completion time	0.42****	-0.39***	-0.28**	-0.28**	-0.08	-0.21*	-0.15	-0.29**
Symbolic play	-0.39***	0.51****	0.41****	0.39***	0.15	0.24*	0.20	0.27*

CARS, Childhood Autism Rating Scale; Bayley-IV, Bayley Scales of Infant and Toddler Development, Fourth Edition; Significance levels: *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

Analysis of behavioral indicators’ screening efficacy

ROC curve analysis (Fig. 3; Supplementary Material Table S2) showed that, among the examined indicators, eye contact frequency and total eye contact duration demonstrated good discriminative efficacy in distinguishing between ASD and TD toddlers. The area under the curve (AUC) for eye contact frequency was 0.915 (95% CI: 0.864–0.967), with a sensitivity of 96.36% and a specificity of 73.33%. The AUC for total eye contact duration was 0.899 (95% CI: 0.839–0.959), with a sensitivity of 86.67% and a specificity of 84.44%.

Fig. 3.

ROC curve analysis. ROC, Receiver Operating Characteristic; FEC, frequency of eye contact; TEC, total duration of eye contact; TT, turn-taking; IF, instruction following; SP, symbolic play; TCT, task completion time

Exploratory observation of other neurodevelopmental disorders

Beyond the primary analysis, this study also identified an NDD group consisting of 7 children. Descriptive data on this group’s participant characteristics and behavioral indicators are presented in Supplementary Material Table S3. Observations showed that this group’s distribution patterns for some behavioral indicators were intermediate between those of the ASD and TD groups. Due to the group’s limited sample size (n = 7), it was not included in formal between-group statistical comparisons. These data provide exploratory findings that can serve as a reference for subsequent larger-sample studies.

Discussion

This primary care-based prospective cohort study systematically investigated the value of brief behavioral observation through structured examiner-child interaction for early identification of toddlers with ASD. Results demonstrated that, compared to TD toddlers, the ASD group showed significant deficits across a range of core social communication behaviors, including eye contact, turn-taking, instruction following, and symbolic play. Crucially, eye contact (both frequency and total duration) demonstrated excellent discriminative ability (AUC > 0.899). These findings strongly support the potential of this brief, objective, and low-cost behavioral observation method as a primary care-level screening tool, offering a feasible solution to address current gaps in early ASD identification resources.

First, this study found that toddlers with ASD exhibited widespread deficits in social communication behaviors during structured interactions. Most critically, the reduced eye contact finding is consistent with results from detailed behavioral coding [25, 37] and eye-tracking technology [38, 39] in laboratory settings. This consistency across different research contexts and methodologies strongly reinforces the status of social attention deficits as the stable and core early behavioral marker of ASD. Theoretically, this result provides strong support for the social motivation theory [40, 41], which posits that individuals with ASD may have diminished social motivation due to reduced perceived reward value of social stimuli, thereby making it difficult for them to initiate and maintain spontaneous eye contact. Furthermore, the ASD group showed widespread difficulties in turn-taking, instruction following, and action imitation, collectively revealing impairments in their higher-level “joint action” abilities [42–44]. Successful execution of joint action requires integrating multiple socio-cognitive components and motor skills, including joint attention, turn-taking, intention understanding, motor planning, and prediction. These abilities depend on early developmental sensorimotor processes [45–47], which are commonly identified as vulnerable areas in the neurodevelopment of ASD. Similarly, ASD toddlers demonstrated less symbolic play, with play content that lacked diversity and complexity. This finding aligns with previous research [27, 34, 48] and supports both the Theory of Mind [49] and symbolic deficit hypotheses [50] in ASD. The results also suggest that interventions focused on enhancing symbolic play ability may have a positive impact on core symptoms.

Furthermore, another significant contribution of this study is the quantitative evidence for the clinical applicability of these behavioral indicators, provided through ROC analysis. The extremely high sensitivity (96.36%) demonstrated by the frequency of eye contact indicates its critical advantage in minimizing missed diagnoses, which is crucial for early screening. Concurrently, the high specificity (84.44%) of the total eye contact duration helps control the false-positive rate, thereby avoiding unnecessary referrals and family anxiety. Previous studies in high-risk ASD populations (e.g., siblings of individuals with ASD) have also highlighted that the frequency and total eye contact duration during infant-mother interaction can help predict autism outcomes [51, 52]. This excellent performance, combined with the feasibility of implementing this paradigm in primary care settings, strongly supports its use as a core component of a primary ASD screening tool.

Spearman correlation analysis provided further evidence for the clinical significance of these behavioral indicators at a deeper level. The study revealed that core indicators, including eye contact, turn-taking, instruction following, and symbolic play, were significantly positively correlated with cognitive, language, and adaptive behavior scaled scores on the Bayley-IV, while showing significant negative correlations with symptom severity assessed by the CARS. This clear pattern of associations indicates that these micro-level social behaviors are not isolated but rather reflect the overall neurodevelopmental level of the toddlers, with their deficits closely linked to the severity of core ASD symptoms. Notably, children in the ASD group required more time to complete the tasks, and the task completion time showed an inverse correlation pattern with other behavioral indicators. This indicator captures the synergistic effect of impaired social reciprocity and cognitive rigidity in real-time interactions, offering a more nuanced perspective on the heterogeneity of ASD.

Furthermore, this study also revealed several findings worthy of further investigation. For instance, despite a reduced sample size due to technical issues, the trend of poorer performance on the affective response task within the ASD group suggests that difficulties in emotional empathy [31, 53] are also noteworthy early behavioral features, which should be validated in larger future samples. Notably, although social smiling showed significant between-group differences, it was not significantly associated with developmental levels in any domain or with symptom assessment scores. This finding is somewhat inconsistent with the general expectation that positive emotional expression is associated with favorable socio-emotional development [54, 55]. We speculate that social smiling, as an explicit emotional expression, may be more susceptible to immediate contextual factors such as the assessment setting or the presence of an unfamiliar assessor. Its state stability and trait-like properties might be lower than those of more biologically rooted social attention indicators, such as eye contact. Consequently, although social smiling is a characteristic of social presentation in toddlers with ASD, its reliability as an independent and stable screening marker may be limited. This finding suggests that future efforts to refine behavioral marker combinations should prioritize indicators with higher cross-situational stability.

A key strength of this study lies in its successful validation of the feasibility of implementing a structured assessor-child interaction paradigm in primary care settings. Traditional laboratory-based, real-time parent-child free-play paradigms, while ecologically valid, often lack standardized procedures and require high coding expertise [24, 25]. The standardized interaction tasks used in this study, administered by trained assessors, effectively controlled the basic interaction framework, significantly enhancing the assessment’s reproducibility and comparability. More importantly, the behavioral features extracted under these conditions showed high consistency with findings from laboratory-based studies. It strongly indicates that core social behavior markers exhibit strong cross-situational stability, significantly boosting confidence in the broader applicability of this screening method.

It should be emphasized that the video recording and offline manual coding utilized in this study were strictly employed for methodological validation—specifically, to ensure rater blinding, establish reliable operational definitions, and calculate inter-rater reliability. To eliminate the operational burden in routine clinical settings, our future translational efforts will pursue two parallel pathways. First, for immediate clinical implementation, we plan to develop a highly standardized “live scoring checklist” that psychological assessors can complete during or immediately after the 5–7 min interaction, mirroring established tools like the Screening Tool for Autism in Two-Year Olds (STAT) [56]. Second, as a long-term goal, we aim to integrate computer vision and artificial intelligence (AI) to develop an automated behavioral scoring system. This approach would seamlessly extract and score core indicators (e.g., eye contact and motor features) from videos, preserving the objectivity of the assessment without increasing the workload of clinical staff.

However, this study has several limitations. First, limited by the sample size, we could not analyze the NDD group, restricting in-depth exploration of behavioral phenotypes across different developmental trajectories. Second, regarding clinical data completeness and comorbidity analysis, while all 45 ASD cases were formally diagnosed by experienced developmental-behavioral pediatricians according to DSM-5 criteria, only 27 completed the Bayley-4 assessment. In routine, real-world clinical pathways, comprehensive cognitive testing is not a mandatory prerequisite for ASD diagnosis and is often selectively prescribed based on the child’s specific clinical presentation. Among those assessed, only 3 children (11.1%) presented with comorbid GDD. Due to this extremely small subgroup, we lacked sufficient statistical power to conduct a meaningful comparative analysis between “ASD with GDD” and “pure ASD.” Notably, this low rate of comorbid GDD highly aligns with the core purpose of our tool: primary care-level early screening. Recruited primarily from routine well-child visits, our target population naturally consists predominantly of “pure” or milder ASD cases without severe developmental delay phenotypes, distinct from the complex cases typically seen in specialized tertiary clinics. Nevertheless, the inability to completely rule out the confounding effects of GDD on behavioral performance remains an objective limitation. Third, the recruitment strategy inherently included a mix of general pediatric populations and parents actively seeking consultation, which may have introduced selection bias. Future research needs to optimize follow-up procedures in larger cohorts with well-stratified GDD subgroups to improve assessment completion rates and enhance the reliability of conclusions. Finally, the single-center design limits the generalizability of the results, requiring multi-center research to further validate their external applicability. Given these limitations, future research should also explore the integration of multimodal physiological indicators, such as eye-tracking and EEG, to construct more precise early screening models.

Conclusion

Behavioral observation through structured examiner-child interactive play, particularly indicators related to eye contact, effectively differentiates toddlers with ASD. This objective, low-cost method shows potential as a primary screening tool for enhancing early ASD detection in primary care settings. Future research should validate its efficacy in multi-center studies, conduct subgroup analyses within expanded cohorts (including clearly defined NDD), and accelerate its clinical translation through live scoring checklists and AI-assisted automated assessment systems.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AAP

American Academy of Pediatrics

ADI-R

Autism Diagnostic Interview-Revised

ADOS-2

Autism Diagnostic Observation Schedule, Second Edition

AI

Artificial Intelligence

ASD

Autism Spectrum Disorder

AUC

Area Under the Curve

Bayley-4

Bayley Scales of Infant and Toddler Development, Fourth Edition

CARS

Childhood Autism Rating Scale

DSM-5

Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition

GDD

Global Developmental Delay

LD

Language Delay

M-CHAT-R/F

Modified Checklist for Autism in Toddlers, Revised with Follow-Up

NDD

Other Neurodevelopmental Disorders

ROC

Receiver Operating Characteristic

STAT

Screening Tool for Autism in Two-Year Olds

TD

Typically Developing

Author contributions

AD, HBY, WQH, & ZDY were responsible for data collection and analysis, as well as drafting and revising the manuscript. LTY, YT & CJ conceptualized and designed the study, critically revised the manuscript, and provided overall supervision. ZY, ZH, XY, NXY & CY contributed to data collection. All authors agree to be accountable for all aspects of the work. All authors read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (Grants #82304119 to T.Y., and #82372559 to J.C.), the China Postdoctoral Science Foundation (Grant #2024MD764064 to Q.-H.W.), the Special Funding for Postdoctoral Research Project of Chongqing (Grant #2024CQBSHTB3014 to Q.-H.W.), and the Chief Medical Expert Studio of Chongqing (Grant #YWBF[2018]263 to T.-Y.L.).

Data availability

The datasets generated and analysed during the current study are not publicly available due to privacy and ethical restrictions but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The studies involving humans were approved by the Children’s Hospital of Chongqing Medical University Institutional Review Board, with approval number (2024) IRB (study) 305. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation in this study was provided by the participants’ legal guardians/next of kin.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.