An Exploratory Analysis of Eye-Tracking During a Virtual Reality Social Cognition Intervention for Children with Autism Spectrum Disorders

Abstract

The diagnosis of autism spectrum disorder is based on clinical judgment, as there are no clearly identified markers to determine the presence of this condition. Gaze patterns have been proposed as a potential biomarker for autism. This study aims to conduct an exploratory analysis of the eye-tracking data collected during a virtual reality-based intervention for social cognition in autistic children. Specifically, we evaluated the variations in social orientation toward social stimuli, the association of gaze patterns with autistic traits, theory of mind (ToM) and task performance, and the mediating effect of attentional mechanisms on the relationship between social cognition performance and autism symptomatology. Our findings identified an increase in social orientation time toward social stimuli, but eye-tracking measures did not significantly predict autism symptom severity or ToM ability. The mediation analysis also failed to find a significant mediating effect of gaze patterns on the relationship between task performance and autism severity. This study points to VR as a promising tool for improving social orienting in autistic children, although there is a need to further investigate the potential of eye-tracking measures as a behavioral marker for predicting social cognition performance.

Keywords

autism spectrum disorders eye-tracking data gaze patterns social cognition theory of mind virtual reality

Introduction

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by differences in social communication and restricted and repetitive behaviors. Although its precise causes remain unclear, converging evidence suggests interactions between genetic and environmental factors, resulting in a highly heterogeneous phenotype. Diagnosis relies on clinical judgment, as there are no clearly identified markers to determine the presence of this condition,¹ prompting growing interest in identifying genetic, cognitive, physiological, and behavioral markers.^2–4

Eye-tracking has emerged as a promising behavioral tool for studying attentional processes in autism due to its feasibility, non-invasive nature, and sensitivity to atypical gaze patterns. It provides objective indices of social attention and attentional flexibility, both widely studied in ASD.^5–8 Social attention—the tendency to orient toward socially relevant cues such as faces—emerges early in development and supports subsequent cortical specialization.⁹ Autistic individuals, however, show reduced preference for social stimuli.^10–13 Attentional flexibility, an executive function domain, refers to shifting attention according to contextual demands. Research consistently reports differences in gaze allocation in autism,^14–23 including reduced eye fixation and increased focus on the mouth area or background during face viewing,^24–27 suggesting altered prioritization of socially relevant information.^28–30

Theory of Mind (ToM)—the ability to attribute mental states to others—is another core component of social cognition frequently affected in autism.³¹ Reduced ToM abilities may influence how individuals interpret others’ intentions, emotions, and actions.

Virtual reality (VR) offers controlled yet ecologically valid social environments for studying and training social cognition.³² VR-based interventions have demonstrated improvements in social skills, cognition, imitation, and interaction in autistic individuals, with some longitudinal benefits.^33–36 However, studies integrating VR with eye-tracking to examine dynamic social attention are scarce. It remains unclear whether gaze patterns could change after a VR intervention, how they relate to autistic traits, ToM, and task performance, or whether attentional mechanisms mediate links between social cognition and symptom severity.

This exploratory study addressed these gaps by examining gaze behavior in autistic children completing a semi-immersive VR social cognition intervention. We investigated (a) pre–post changes in social orienting, (b) associations between gaze, autistic traits, ToM, and task performance, and (c) whether attentional mechanisms mediate relations between social cognition and symptom severity.

Methods

Participants

Thirty-five participants were recruited from a cohort of children aged 6–8 years undergoing clinical follow-up at Hospital Universitari Mútua Terrassa (Barcelona, Spain). Two withdrew prior to the intervention due to anxiety and nine did not complete the program, resulting in 24 intervention completers. Inclusion criteria were (a) a diagnosis of ASD according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, Text Revision; (b) classification as autistic based on the Autism Diagnostic Observation Schedule, Second Edition (ADOS-2); (c) age 6–8 years; (d) an intelligence quotient (IQ) >85 assessed with the Wechsler Intelligence Scale for Children, Fifth Edition (WISC-V) or Wechsler Preschool and Primary Scale of Intelligence, Fourth Edition, with adequate verbal comprehension; and (e) normal or corrected vision. Ethical approval was obtained from the Comité de Ética de Investigación con Medicamentos de la Fundació Assistencial Mútua Terrassa (no. 10/2023). Written informed consent and child assent were obtained in accordance with the Declaration of Helsinki.

Measures

Participants were assessed with the ADOS-2³⁷ and the WISC-V,³⁸ and the ADOS Calibrated Severity Score (ADOS-CSS) was used to measure autism severity. All instruments demonstrated strong psychometric properties, with interrater and test–retest reliability ≥0.80 and high internal consistency (α ≈ 0.88–0.96).^37,38

The Social Responsiveness Scale-2 (SRS-2)^39,40 and the Theory of Mind Task Battery (ToMTB),⁴¹ were administered pre- and postintervention. SRS-2 shows excellent internal consistency (α ≈ 0.90–.97) and test–retest reliability (r ≈ 0.80–0.90). The ToMTB also demonstrates acceptable internal consistency (α ≈ 0.60–0.80).^40,41 Baseline clinical and demographic data were available for 33 participants (see the Supplementary Data).

Apparatus

The semi-immersive VR intervention took place at Hospital Universitari Mútua de Terrassa using an 86-inch display, central computer, laptop, and an Azure Kinect DK device.⁴² Display geometry was defined using AprilTag 2 markers.⁴³ Gaze patterns were collected using a Pupil Labs Invisible device.⁴⁴

Procedure

The analyzed task was part of a broader VR-based adaptive intervention targeting social cognition within the ADAPTEA protocol (see Supplementary Data). For the present study, participants completed a fixed, non-adaptive version. Analyses focused on an emotion-recognition task.

Eye-tracking measures

Gaze data were categorized into social (human figures) and non-social (background) areas of interest (AOIs). Social orienting was indexed as percentage of fixation time on social AOIs.^5,14,19,27 Attentional mechanisms were quantified using time to first fixation on social AOIs, peak fixation duration for social and background AOIs,^6,14,19,22 and fixation and saccade counts.^8,28–30 These metrics align with established findings of reduced social orienting and altered attentional patterns in autism.^{14,19,23,27,29} Peak fixation duration was prioritized over mean fixation duration to mitigate the influence of artifacts.⁴⁵

Data processing and statistical analyses

Eye-tracking data were extracted and preprocessed in Visual Studio Code and analyzed in IBM SPSS Statistics. Raw gaze streams were visually inspected for artifacts. Fixations and saccades were identified using Pupil Labs’ dispersion-based algorithm with manual verification. AOIs were manually defined and validated (10 percent frame review per participant).

Statistical analyses included paired-samples t-tests, Pearson correlations, hierarchical multiple regression, and mediation analysis, using 5,000 bootstrap resamples. For hierarchical regressions, predictors were entered in three steps: demographic covariates (age, IQ, presence of attention-deficit/hyperactivity disorder [ADHD]) in Step 1; social-cognitive variables in Step 2; and eye-tracking variables in Step 3.

Sample size and analytic samples

Thirty-five participants were recruited to account for anticipated attrition and data loss. Twenty-four completed the intervention; 18 provided valid paired eye-tracking data for primary pre–post analyses. With n = 18, paired-sample t-tests (α = 0.05) provide ∼80 percent power to detect within-subject effects of dz ≈ 0.70. Regression and mediation analyses were considered exploratory due to sample size constraints.

Eye-tracking analyses required successful calibration, sufficient tracking ratio, and valid trial completion. Sample sizes varied by analysis: pre–post comparisons included participants with valid paired data, correlations included all available cases (maximum n = 22), and regression/mediation models used complete cases.

Results

Paired-sample t-test

Results show a significant increase in the percentage of social orienting from baseline to endline, t(17) = −2.123, p = 0.049, d = −0.50, and a longer time to first fixation on social stimuli from baseline to the last session, t(17) = −3.915, p = 0.001, d = −0.923.

Correlation analyses

ADOS-CSS was significantly associated with longer peak fixation duration on background AOIs (p < 0.05). SRS Social Cognition correlated with autism severity (p < 0.05). ToMTB negatively correlated with the number of fixations and saccades. Social orienting was strongly associated with peak fixation duration to social AOIs (r = 0.844) and negatively associated with peak fixation duration on background AOIs, and number of fixations and saccades (all p < 0.05). Longer first-fixation latency to social stimuli was associated with diminished social orienting and more fragmented visual exploration (see Tables 1, 2, and 3 for more details).

Table 1.

Correlation Coefficients Among Clinical Measures (n = 22 Participants)

Variable	1	2	3	4	5
1. ADOS-CSS	—	0.984^a	0.253	0.378	0.248
2. ADOS Total Score	0.984^a	—	0.339	0.435^b	0.186
3. SRS Total Score	0.253	0.339	—	0.921^a	–0.152
4. SRS Social Cognition	0.378	0.435^b	0.921^a	—	–0.126
5. ToMTB	0.248	0.186	–0.152	–0.126	—

Correlation is significant at the 0.01 level (two-tailed).

Correlation is significant at the 0.05 level (two-tailed).

ADOS, Autism Diagnostic Observation Schedule; ADOS-CSS, Autism Diagnostic Observation Schedule—Calibrated Severity Score; SRS, Social Responsiveness Scale; ToMTB, Theory of Mind Task Battery.

Table 2.

Correlation Coefficients Among Eye-Tracking Measures (n = 22 Participants)

Variable	1	2	3	4	5	6	7	8	9	10
1. Number of fixations	—	1.000^a	0.768^a	–0.521^b	–0.449^b	0.099	0.097	–0.106	0.835^a	–0.108
2. Number of saccades	1.000^a	—	0.769^a	–0.523^b	–0.451^b	0.101	0.096	–0.108	0.836^a	–0.108
3. Time to first fixation on social AOIs	0.768^a	0.769^b	—	–0.687^a	–0.522^b	0.457^b	–0.201	–0.236	0.947^a	–0.236
4. Percent of social stimuli	–0.521^b	–0.523^a	–0.687^a	—	0.844^a	–0.692^a	0.272	0.727^a	–0.751^a	0.727^a
5. Social peak duration time	–0.449^b	–0.451^a	–0.522^a	0.844^a	—	–0.542^a	0.057	0.540^a	–0.606^a	0.540^a
6. Background peak duration time	0.099	0.101	0.457^a	–0.692^a	–0.542^a	—	–0.555^a	–00.573^a	0.414	–0.573^a
7. Percent right answers	0.097	0.096	–0.201	0.272	0.057	–0.555^a	—	0.065	–0.153	0.065
8. Total fixations (social AOIs)	–0.106	–0.108	–0.236	0.727^a	0.540^a	–0.573^a	0.065	—	–0.307	—
9. Total fixations (background AOIs)	0.835^a	0.836^b	0.947^b	–0.751^a	–0.606^a	0.414	–0.153	–0.307	—	–0.307

Correlation is significant at the 0.01 level (two-tailed).

Correlation is significant at the 0.05 level (two-tailed).

% of social stimuli, percentage of time looking at social stimuli; % right answers, percentage of right answers; AOIs, areas of interest; background peak duration time, peak duration time toward nonsocial stimuli; social peak duration time, peak duration time toward social stimuli; total fixations (background AOIs), number of fixations toward nonsocial stimuli; total fixations (social AOIs), number of fixations toward social stimuli.

Table 3.

Correlation Coefficients Between Clinical Measures and Eye-Tracking Variables (n = 22 Participants)

Predictor	Number of fixations	Number of saccades	Time to first fixation on social AOIs	% of social stimuli	Social peak duration time	Background peak duration time	% right answers	Total fixations (social AOIs)	Total fixations (background AOIs)
ADOS-CSS	–0.088	–0.086	0.117	–0.251	–0.283	0.502^a	–0.276	–0.032	0.110
ADOS total	–0.110	–0.109	.078	–0.203	–0.254	0.490^a	–0.273	–0.024	0.067
SRS total	–0.113	–0.113	–0.076	.084	–0.034	0.048	–0.033	–0.035	–0.129
SRS Social Cognition	–0.170	–0.170	–0.131	.208	0.019	–0.021	0.019	0.136	–0.205
ToMTB	–0.221	–0.221	.000	–0.048	–0.036	0.206	0.174	–0.190	0.023

Correlation is significant at the 0.05 level (two-tailed).

Predictors of autism symptom severity

In Step 1, the control variables did not significantly predict ADOS-CSS, F(3, 20) = 0.176, p = 0.911. Step 2 did not significantly improve the model, ΔR² = 0.228, F(2, 18) = 1.227, p = 0.337. Adding eye-tracking variables in Step 3 did not significantly improve prediction, F(10, 13) = 0.996, p = 0.492 (Tables 4 and 5).

Table 4.Hierarchical Regression Models Predicting ADOS-CSS

Model	R	R²	Adjusted R²	SE of estimate	ΔR²	F change	df1	df2	p (F change)	Model F	p (model)
1	0.160	0.026	–0.120	1.18	—	0.176	3	20	0.911	0.176	0.911
2	0.504	0.254	0.047	1.09	0.229	2.758	2	18	0.090	1.227	0.337
3	0.659	0.434	–0.002	1.12	0.180	0.825	5	13	0.554	0.996	0.492

Durbin–Watson (final model) = 1.90.

Table 5.

Hierarchical Regression Predicting ADOS-CSS with Bootstrapped Estimates (95 % CIs)

Predictor	B	SE (boot)	t	p	95% CI lower	95% CI upper
Step 1
Intercept	4.841	4.044	1.197	0.235	–1.555	14.119
ADHD (yes/no)	–0.072	0.581	–0.120	0.898	–1.034	1.239
Age (months)	–0.010	0.028	–0.346	0.723	–0.065	0.042
IQ	0.013	0.041	0.397	0.715	–0.087	0.066
Step 2
Intercept	3.247	3.949	0.822	0.337	–2.710	12.580
ADHD (yes/no)	–0.310	0.528	–0.548	0.508	–1.138	0.960
Age (months)	–0.014	0.027	–0.480	0.586	–0.068	0.040
IQ	–0.010	0.038	–0.301	0.784	–0.093	0.049
SRS Social Cognition	0.039	0.024	2.076	0.108	–0.003	0.089
ToMTB	0.134	0.119	1.126	0.273	–0.085	0.379
Step 3
Intercept	1.255	12.594^b	0.263	0.866^b	–16.239^b	19.765^a
ADHD (yes/no)	–0.238	1.326^b	–0.406	0.780^b	–1.969^b	2.329^a
Age (months)	–0.004	0.066^b	–0.127	0.912^b	–0.105^b	0.099^a
IQ	–0.005	0.094^b	–0.133	0.896^b	–0.127^b	0.130^a
SRS Social Cognition	0.036	0.043^b	1.649	0.350^b	–0.033^b	0.118^a
ToMTB	0.079	0.268^b	0.658	0.649^b	–0.375^b	0.457^a
Time to first fixation (social AOIs)	0.000	0.000^b	0.182	0.890^b	–0.000^b	0.000^a
Peak duration time (social AOIs)	–0.000	0.000^b	–0.321	0.825^b	–0.001^b	0.001^a
Peak duration time (background AOIs)	0.000	0.001^b	1.288	0.532^b	–0.001^b	0.001^a
Number of saccades	–0.000	0.007^b	–0.009	0.994^b	–0.015^b	0.013^a
Social orienting	0.009	0.064^b	0.278	0.849^b	–0.115^b	0.112^a

^aUnless otherwise noted, bootstrap estimates are based on 5,000 samples.

^bBased on 4,992 bootstrap samples.

ADHD, attention-deficit/hyperactivity disorder; AOIs, areas of interest; IQ, intelligence quotient; social orienting, percentage of total fixation time toward social stimuli; SRS, Social Responsiveness Scale; ToMTB, Theory of Mind Task Battery.

Predictors of ToM ability

In Step 1, the control variables did not significantly predict ToM ability, F(3, 20) = 2.335, p = 0.105. Adding ADOS-CSS in Step 2 did not significantly improve the model. In Step 3, the inclusion of eye-tracking variables increased the explained variance to 44.3 percent, but this change was not significant, ΔR² = 0.145, F(5, 14) = 0.731, p = 0.612 (Tables 6 and 7).

Table 6.

Hierarchical Regression Models Predicting Theory of Mind Performance

Model	R	R²	Adj. R²	SEE	ΔR²	Fchg	p(fchg)	df1	df2
1	0.509	0.259	0.148	2.38	—	2.34	0.105	3	20
2	0.545	0.297	0.149	2.38	0.038	1.03	0.323	1	19
3	0.665	0.443	0.085	2.47	0.145	0.73	0.612	5	14

Durbin–Watson (final model) = 2.48.

Table 7.

Hierarchical Regression Predicting Theory of Mind Performance (Bootstrapped Estimates, 95% CIs)

Predictor	B	SE (boot)	t	p	95% CI lower	95% CI upper
Step 1
Intercept	–7.749	8.102	–0.957	0.321	–21.858	10.043
ADHD (yes/no)	–0.271	1.351	–0.201	0.846	–2.767	2.558
Age (months)	0.115	0.060	1.917	0.081	–0.008	0.234
IQ	0.098	0.076	1.289	0.158	–0.094	0.211
Step 2
Intercept	–9.962	8.784	–1.135	0.229	–25.752	8.462
ADHD (yes/no)	–0.239	1.360	–0.176	0.861	–2.786	2.628
Age (months)	0.119	0.060	1.983	0.064	–0.008	0.235
IQ	0.092	0.077	1.195	0.179	–0.086	0.221
ADOS-CSS	0.457	0.418	1.093	0.219	–0.519	1.157
Step 3
Intercept	–5.764	20.695^b	–0.278	0.655^b	–32.926^b	47.204ª
ADHD (yes/no)	–0.094	2.344^b	–0.040	0.956^b	–4.599^b	4.196ª
Age (months)	0.134	0.100^b	1.340	0.161^b	–0.079^b	0.290ª
IQ	0.096	0.139^b	0.691	0.270^b	–0.215^b	0.316ª
ADOS-CSS	0.298	1.082^b	0.275	0.652^b	–1.288^b	3.049ª
Time to first fixation on social AOIs	0.000	0.000^b	0.141	0.867^b	0.000^b	0.000^a
Peak duration time (social AOIs)	0.000	0.001^b	0.455	0.651^b	–0.001^b	0.002ª
Peak duration time (background AOIs)	–0.000	0.001^b	–0.112	0.911^b	–0.002^b	0.002ª
Number of saccades	–0.009	0.014^b	–0.643	0.424^b	–0.039^b	0.018ª
Social orienting	–0.068	0.111^b	–0.613	0.451^b	–0.275^b	0.154^a

^aUnless otherwise noted, bootstrap estimates are based on 5,000 samples.

^bBased on 4,999 bootstrap samples.

Predictors of social orienting

Model 1 was not significant, F(3, 20) = 0.587, p = 0.630. Adding social cognition variables in Step 2 did not significantly improve prediction, F(5, 18) = 0.719, p = 0.618, R² = 0.166. The final model, including ADOS-CSS, also was not statistically significant, F(6, 17) = 1.037, p = 0.436 (Tables 8 and 9).

Table 8.

Hierarchical Regression Models Predicting Social Orienting

Model	R	R²	Adjusted R²	SE of estimate	ΔR²	F change	df1	df2	p (F change)	Model F	p (model)
1	.285	.081	–0.057	22.39	—	0.587	3	20	0.630	0.587	0.630
2	.408	.166	–0.065	22.48	0.085	0.922	2	18	0.416	0.719	0.618
3	.518	.268	.010	21.67	0.102	2.360	1	17	0.143	1.037	0.436

Durbin–Watson (final model) = 1.90.

Table 9.

Hierarchical Regression Predicting Social Orienting (Bootstrapped Estimates, 95% CIs)

Predictor	B	SE (boot)	t	p	95% CI lower	95% CI upper
Step 1
Intercept	–50.942	76.512	–0.666	0.474	–186.001	114.406
ADHD (yes/no)	2.594	12.338	0.210	0.832	–24.068	23.188
Age (months)	0.524	0.587	0.893	0.362	–0.588	1.688
IQ	0.477	0.616	0.774	0.362	–0.650	1.851
Step 2
Intercept	–90.813	77.159	–1.177	0.198	–237.389	74.266
ADHD (yes/no)	–0.739	13.530	–0.055	0.956	–31.927	20.428
Age (months)	0.820	0.574	1.428	0.14	–0.322	1.938
IQ	0.529	0.702	0.754	0.408	–0.808	1.961
SRS Social Cognition	0.408	0.434	0.940	0.305	–0.333	1.407
ToMTB	–1.547	2.140	–0.723	0.432	–5.603	2.705
Step 3
Intercept	–67.417	81.359	–0.828	0.328	–206.518	94.228
ADHD (yes/no)	–2.974	13.351	–0.223	0.812	–31.562	19.269
Age (months)	0.721	0.568	1.270	0.192	–0.416	1.832
IQ	0.460	0.672	0.684	0.407	–0.883	1.691
SRS Social Cognition	0.686	0.475	1.444	0.114	–0.317	1.697
ToMTB	–0.584	2.086	–0.280	0.745	–5.172	3.173
ADOS-CSS	–7.207	5.729	–1.259	0.149	–17.609	4.575

Unless otherwise noted, bootstrap estimates are based on 5,000 samples.

IQ, Intellectual Quotient.

Predictors of task performance

In Step 1, the control variables did not significantly predict performance, F(3, 18) = 0.350, p = 0.790. Step 2 did not significantly improve the model, ΔR² = 0.008, F change (2, 16) = 0.214, p = 0.951. Model 3 also resulted in not being statistically significant, ΔR² = 0.130, F change(1, 15) = 2.425, p = 0.140 (Tables 10 and 11).

Table 10.

Hierarchical Regression Predicting Task Performance

Model	R	R²	Adjusted R²	SE of estimate	ΔR²	F change	df1	df2	p (F change)	Model F	p (model)
1	.235	0.055	–0.102	0.414	—	0.350	3	18	0.79	0.350	0.79
2	.251	0.063	–0.230	0.438	.008	0.065	2	16	0.937	0.214	0.951
3	.440	0.193	–0.130	0.420	.130	2.425	1	15	0.14	0.599	0.727

Durbin–Watson (final model) = 1.718.

Table 11.

Hierarchical Regression Predicting Task Performance (Bootstrapped Estimates, 95% CIs)

Predictor	B	SE(boot)	t	p	95% CI lower	95% CI upper
Step 1
Intercept	0.466	1.351	0.345	0.718	–2.384	2.682
ADHD (yes/no)	0.091	0.204	0.446	0.633	–0.285	0.535
Age (months)	0.004	0.009	0.444	0.608	–0.012	0.025
IQ	0.010	0.012	0.833	0.309	–0.010	0.037
Step 2
Intercept	0.546	1.707	0.320	0.678	–3.293	2.977
ADHD (yes/no)	0.103	0.261	0.395	0.707	–0.411	0.594
Age (months)	0.003	0.012	0.265	0.791	–0.017	0.030
IQ	0.008	0.015	0.533	0.501	–0.019	0.041
ToMTB	0.017	0.056	0.304	0.727	–0.096	0.128
SRS Social Cognition	0.000	0.010	0.043	0.957	–0.019	0.021
Step 3
Intercept	1.083	1.819	0.596	0.444	–2.765	4.543
ADHD (yes/no)	0.080	0.271	0.295	0.771	–0.429	0.627
Age (months)	0.000	0.012	–0.024	0.981	–0.020	0.027
IQ	0.006	0.017	0.353	0.652	–0.033	0.035
ToMTB	0.044	0.061	0.721	0.423	–0.046	0.190
SRS Social Cognition	0.006	0.012	0.509	0.612	–0.013	0.035
ADOS-CSS	–0.145	0.121	–1.198	0.214	–0.396	0.079

Bootstrapped estimates are based on 5,000 samples, except where noted.

Step 3 coefficients based on 4,944 samples.

Mediation analysis

A mediation analysis (PROCESS Model 4) tested whether background peak duration time mediated the association between percentage of correct answers during the VR intervention and ADOS-CSS. Percentage of correct answers significantly predicted background peak duration time (b = −5,961.01, p = 0.0047), but background peak duration time did not predict ADOS-CSS (b = 0.0001, p = 0.2235). The direct effect was nonsignificant (b = −0.2620, p = 0.7099). The indirect effect was b = −0.4854 (BootSE = 0.7006), 95 percent CI [−1.7546, 0.2649], indicating no mediation (Figure 1).

FIG. 1.

Mediation analysis.

Discussion

This exploratory study examined changes in social orientation following a VR social cognition intervention and explored associations between gaze patterns, autistic traits, ToM, and task performance. Social orientation increased significantly across sessions, supporting the hypothesis that the intervention may enhance social attention. The expected association between ADOS-2 scores and the SRS Social Cognition subscale was replicated.^46,47 ADOS-2 and ADOS-CSS were associated with greater baseline attention to background stimuli, suggesting reduced social motivation.¹⁴ Longer time to first social fixation was linked to lower engagement and more fragmented attention, consistent with prior findings.⁴⁸

Although autism severity was correlated with longer peak fixation durations on background AOIs, regression analyses revealed no significant predictive relationships between gaze, social cognition, and autistic traits. This dissociation supports models proposing that implicit attentional processes (eye-tracking) and explicit reasoning (ToM tasks, clinical ratings) reflect distinct but interacting systems.^49,50 The limited sensitivity of ADOS-CSS⁵¹ and the absence of direct social cognition assessment in ADOS Module 3 may partly explain these null findings. ToM performance depends on language and higher-order cognition,^52–54 which may not be captured by gaze metrics. In addition, eye movements are influenced by scene salience and task demands,^55,56 potentially obscuring individual differences.

The mediation model aligned with this interpretation: background fixation duration did not explain links between performance and symptom severity, although limited sample size likely reduced power. Overall, the dissociation between clinical severity, social cognition, and gaze behavior supports partially independent but interacting systems underlying social functioning. Eye-tracking captures rapid, context-sensitive processes, whereas ADOS-CSS and ToM index more stable traits.

Despite limited predictive associations, the increase in social orienting highlights eye-tracking’s sensitivity to short-term attentional shifts. The association between background attention and autism severity suggests that non-social gaze allocation may represent a feasible biomarker. VR combined with eye-tracking shows promise as a complementary tool for profiling social attention. Future research should examine whether gaze changes predict treatment response, whether adaptive VR enhances engagement, and whether integrating eye-tracking with physiological or neural measures improves characterization of social-cognitive profiles.

Limitations

The small sample reduced statistical power, and ceiling ToMTB scores limited the detection of change. Underrepresentation of autistic females, exclusion of individuals with intellectual disabilities or behavioral disorders, and absence of a control group restrict generalizability. Social orientation metrics also limited precision, as gaze to specific facial features could not be analyzed.

Conclusions

VR appears promising for capturing social orienting in autistic children, although eye-tracking measures did not yield strong predictive associations. These findings highlight the need to distinguish between task-specific attentional dynamics and trait-level clinical features when interpreting gaze behavior in autism. Larger, more diverse samples are needed to further evaluate eye-tracking as a behavioral marker of autism-related social differences.

Footnotes

Acknowledgments

The authors acknowledge the valuable contribution of Luna Maddalon and Maria Eleonora Minissi for their assistance in facilitating access to the data used in this study and their support throughout all the process.

Authors’ Contributions

S.G.M.: Conceptualization, data curation, formal analysis, methodology, investigation, visualization, writing—original draft. G.L.R.: Resources, writing—review and editing. M.A.R.: Conceptualization, methodology, funding acquisition, supervision, validation, and writing—review and editing. A.H.Z.: Conceptualization, methodology, funding acquisition, project administration, supervision, validation, and writing—review and editing.

Author Disclosure Statement

The authors declare no potential conflicts of interest.

Funding Information

This work was supported by the Ministerio de Ciencia, Innovación y Universidades (Proyectos de I + D + i Retos Investigación, grant number PID2020-116422RB-C22).

Supplemental Material

References

1. Klin

. Biomarkers in autism spectrum disorder: Challenges, advances, and the need for biomarkers of relevance to public health. Focus (Am Psychiatr Publ) 2018;16(2):135–142.

2. Charman

, Loth

, Tillmann

, et al. The EU-AIMS Longitudinal European Autism Project (LEAP): Clinical characterisation. Mol Autism 2017;8:27.

3. Isaksson

, Tammimies

, Neufeld

, et al.; EU-AIMS LEAP group. EU-AIMS Longitudinal European Autism Project (LEAP): The autism twin cohort. Mol Autism 2018;9(1):26.

4. Loth

, Charman

, Mason

, et al. The EU-AIMS Longitudinal European Autism Project (LEAP): Design and methodologies to identify and validate stratification biomarkers for autism spectrum disorders. Mol Autism 2017;8:24.

5. Chevallier

, Parish-Morris

, McVey

, et al. Measuring social attention and motivation in autism spectrum disorder using eye-tracking: Stimulus type matters. Autism Res 2015;8(5):620–628.

6. Goold

, Murphy

, Goodale

, et al. Faster social attention disengagement in individuals with higher autism traits. J Clin Exp Neuropsychol 2022;44(10):755–767.

7. Gliga

, Jones

EJH

, Bedford

, et al. From early markers to neuro-developmental mechanisms of autism. Dev Rev 2014;34(3):189–207.

8. Craig

, Margari

, Legrottaglie

, et al. A review of executive function deficits in autism spectrum disorder and attention-deficit/hyperactivity disorder. Neuropsychiatr Dis Treat 2016;12:1191–1202.

9. Jones

EJH

, Venema

, Lowy

, et al. Developmental changes in infant brain activity during naturalistic social experiences. Dev Psychobiol 2015;57(7):842–853.

10.

10. Johnson

, Dziurawiec

, Ellis

, et al. Newborns preferential tracking of face-like stimuli and its subsequent decline. Cognition 1991;40(1–2):1–19.

11.

11. Di Giorgio

, Frasnelli

, Rosa Salva

, et al.; NIDA-Network. Difference in visual social predispositions between newborns at low- and high-risk for autism. Sci Rep 2016;6(1):26395.

12.

12. Dawson

, Meltzoff

, Osterling

, et al. Children with autism fail to orient to naturally occurring social stimuli. J Autism Dev Disord 1998;28(6):479–485.

13.

13. Dawson

, Webb

, McPartland

. Understanding the nature of face processing impairment in autism: Insights from behavioral and electrophysiological studies. Dev Neuropsychol 2005;27(3):403–424.

14.

14. Chita-Tegmark

. Social attention in ASD: A review and meta-analysis of eye-tracking studies. Res Dev Disabil 2016;48:79–93.

15.

15. Falck-Ytter

, von Hofsten

. How special is social looking in ASD: A review. Prog Brain Res 2011;189:209–222.

16.

16. Chawarska

, Volkmar

, Klin

. Limited attentional bias for faces in toddlers with autism spectrum disorders. Arch Gen Psychiatry 2010;67(2):178–185.

17.

17. Fedor

, Lynn

, Foran

, et al. Patterns of fixation during face recognition: Differences in autism across age. Autism 2018;22(7):866–880.

18.

18. Forby

, Anderson

, Cheng

, et al. Reading the room: Autistic traits, gaze behaviour, and the ability to infer social relationships. PLoS One 2023;18(3):e0282310.

19.

19. Frazier

, Strauss

, Klingemier

, et al. A meta-analysis of gaze differences to social and nonsocial information between individuals with and without autism. J Am Acad Child Adolesc Psychiatry 2017;56(7):546–555.

20.

20. Griffin

, Scherf

. Does decreased visual attention to faces underlie difficulties interpreting eye gaze cues in autism? Mol Autism 2020;11(1):60.

21.

21. Harrop

, Jones

, Zheng

, et al. Visual attention to faces in children with autism spectrum disorder: Are there sex differences? Mol Autism 2019;10:28.

22.

22. Haskins

, Mentch

, Botch

, et al. Reduced social attention in autism is magnified by perceptual load in naturalistic environments. Autism Res 2022;15(12):2310–2323.

23.

23. Reisinger

, Shaffer

, Horn

, et al. Atypical social attention and emotional face processing in autism spectrum disorder: Insights from face scanning and pupillometry. Front Integr Neurosci 2019;13:76.

24.

24. Falck-Ytter

. Gaze performance during face-to-face communication: A live eye tracking study of typical children and children with autism. Res Autism Spectr Disord 2015;17:78–85.

25.

25. Falck-Ytter

, Von Hofsten

, Gillberg

, et al. Visualization and analysis of eye movement data from children with typical and atypical development. J Autism Dev Disord 2013;43(10):2249–2258.

26.

26. Klin

, Jones

, Schultz

, et al. Visual fixation patterns during viewing of naturalistic social situations as predictors of social competence in individuals with autism. Arch Gen Psychiatry 2002;59(9):809–816.

27.

27. Papagiannopoulou

, Chitty

, Hermens

, et al. A systematic review and meta-analysis of eye-tracking studies in children with autism spectrum disorders. Soc Neurosci 2014;9(6):610–632.

28.

28. Dehnavi

, Khan

. Executive function among adults with autism spectrum disorder: An eye-tracking study. Appl Neuropsychol Adult 2024:1–13.

29.

29. Nayar

, Shic

, Winston

, et al. A constellation of eye-tracking measures reveals social attention differences in ASD and the broad autism phenotype. Mol Autism 2022;13(1):18.

30.

30. Sasson

, Turner‐Brown

, Holtzclaw

, et al. Children with autism demonstrate circumscribed attention during passive viewing of complex social and nonsocial picture arrays. Autism Res 2008;1(1):31–42.

31.

31. Frith

, Frith

. Theory of mind. Curr Biol 2005;15(17):R644–R646.

32.

32. Abbas

, O’Connor

, Ganapathy

, et al. What is virtual reality? A healthcare-focused systematic review of definitions. Health Policy and Technology: Elsevier B.V 2023;12(2):100741.

33.

33. Calderone

, Militi

, Latella

, et al. Harnessing virtual reality: Improving social skills in adults with autism spectrum disorder. J Clin Med 2024;13(21):6435.

34.

34. Jacob

, Pillay

. A systematic and meta-analysis review of augmented reality interventions for individuals with autism spectrum disorder. Ubiquitous Learning 2024;17(2):245–264.

35.

35. Yang

, Wu

, Ma

, et al. Effectiveness of virtual reality technology interventions in improving the social skills of children and adolescents with autism: Systematic review. J Med Internet Res 2025;27:e60845.

36.

36. Zhao

, Zhang

, Lu

, et al. Virtual reality technology enhances the cognitive and social communication of children with autism spectrum disorder. Front Public Health 2022;10:1029392.

37.

37. Lord

, Rutter

, DiLavore

, et al. Autism Diagnostic Observation Schedule, Second Edition. Western Psychological Services: Torrance, CA; 2012.

38.

38. Wechsler

. Wechsler Intelligence Scale for Children – Fifth Edition (WISC-V). NCS Pearson: San Antonio, TX; 2014.

39.

39. Constantino

, Davis

, Todd

, et al. Validation of a brief quantitative measure of autistic traits: Comparison of the social responsiveness scale with the autism diagnostic interview-revised. J Autism Dev Disord 2003;33(4):427–433.

40.

40. Constantino

, Gruber

. Social Responsiveness Scale, Second Edition (SRS-2). Western Psychological Services: Torrance, CA; 2012.

41.

41. Hutchins

, Prelock

, Chace

. Test-Retest reliability of a theory of mind task battery for children with autism spectrum disorders. Focus Autism Other Dev Disabl 2008;23(4):195–206.

42.

42.Microsoft. Azure Kinect DK. Microsoft Corporation; 2020.

43.

43. Wang

, Olson

. AprilTag 2: Efficient and robust fiducial detection. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS); Daejeon, Korea; 2016.

44.

44. Tonsen

, Baumann

, Dierkes

. A High-Level Description and Performance Evaluation of Pupil Invisible. ArXiv 2009.

45.

45. Mason

, Shic

, Falck-Ytter

, et al.; LEAP Team*. Preference for biological motion is reduced in ASD: Implications for clinical trials and the search for biomarkers. Mol Autism 2021;12(1):74.

46.

46. Isaksson

, Van’t Westeinde

, Cauvet

, et al. Social cognition in autism and other neurodevelopmental disorders: A co-twin control study. J Autism Dev Disord 2019;49(7):2838–2848.

47.

47. Bölte

, Westerwald

, Holtmann

, et al. Autistic traits and autism spectrum disorders: The clinical validity of two measures presuming a continuum of social communication skills. J Autism Dev Disord 2011;41(1):66–72.

48.

48. Kaliukhovich

, Manyakov

, Bangerter

, et al. Visual preference for biological motion in children and adults with autism spectrum disorder: An eye-tracking study. J Autism Dev Disord 2021;51(7):2369–2380.

49.

49. Jenner

, Farran

, Welham

, et al. The use of eye-tracking technology as a tool to evaluate social cognition in people with an intellectual disability: A systematic review and meta-analysis. J Neurodev Disord 2023;15(1):42.

50.

50. Callenmark

, Kjellin

, Rönnqvist

, et al. Explicit versus implicit social cognition testing in autism spectrum disorder. Autism 2014;18(6):684–693.

51.

51. Anderson

, Liang

, Lord

. Predicting young adult outcome among more and less cognitively able individuals with autism spectrum disorders. J Child Psychol Psychiatry 2014;55(5):485–494.

52.

52. Frith

, Frith

. Implicit and explicit processes in social cognition. Neuron 2008;60(3):503–510.

53.

53. André

, Maintenant

. About the relationship between executive function, theory of mind, and language abilities in children with autism: A systematic review. Appl Neuropsychol Child 2024:1–18.

54.

54. Yu

Y-T

, Li

H-J

, Tsai

C-H

, et al. Cool Executive Function and Verbal Comprehension Mediate the Relation of Hot Executive Function and Theory of Mind in Children with Autism Spectrum Disorder. Autism Res 2021;14(5):921–931.

55.

55. Freeth

, Chapman

, Ropar

, et al. Do gaze cues in complex scenes capture and direct the attention of high functioning adolescents with ASD? Evidence from eye-tracking. J Autism Dev Disord 2010;40(5):534–547.

56.

56. Guillon

, Hadjikhani

, Baduel

, et al. Visual social attention in autism spectrum disorder: Insights from eye tracking studies. Neurosci Biobehav Rev 2014;42:279–297.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.02 MB

0.00 MB