Table 1.

Characteristics and discriminating ability of scalable digital tools to assess ASD risk during early childhood.

Citation, specific measure	Participant details a (Mean age, sample size, gender distribution)	Device specifications	Experimental setup	Primary metric (s) and summary results
DSM-5 criteria: Social communication/Social interaction
Anzulewicz et al. (2016) Specific measure: ‘Sharing’ and fine-motor drawing movements Country: United Kingdom	1. ASD group a Mage (SD) = 53 (11) months, N = 35, %male = 67.57 2. TD group Mage (SD) = 55 (11) months, N = 45, %male = 71.11	Tablet computer (iPad mini) running standard iOS version 7.0	Task description: Two commercially available computerised games. Child response: Game 1 (Sharing): Slice a piece of food by tapping on it and distributing it evenly among four characters. Characters expressed positive/negative emotions based on equal/unequal distribution of food. Game 2 (Creativity): Trace a picture followed by colouring using finger motions. #trials: 1 Setting: Not specified Duration: 5 min	In total, 262 features from touch data (gestures on screen) and tablet’s inertial movement sensors acquired. Machine learning (ML) using touch and sensor data to predict child’s diagnostic group (TD vs ASD). Significant differences in the following features: 1. Impact Force and Gesture Pressure: ASD > TD 2. Distribution of forces into the device: (gyroscope data) patterns of force distribution different 3. Mean gesture velocity: ASD > TD 4. Mean area occupied by a gesture: ASD > TD 5. Gestures in distal parts of the screen: ASD > TD 6. Minimum duration of a screen tap: ASD < TD 7. ML prediction accuracy: Max AUC = 0.93, sensitivity = 0.83, specificity = 0.85
Ruta et al. (2017) Specific measure: Social preference Country: Italy	1. ASD group Mage (SD) = 39.9 (11.5) months, N = 21, %male = 85.7 2. TD group Mage (SD) = 45.5 (10.7) months, N = 37, %male = 48.6	Tablet computer (iPad)	Task description: Deliberate choice task with two pictures – non-social (toy-train) and social (smiling face) hidden under buttons. Button position was randomised in each trial. Control trials comprised scrambled images of the same pictures. Child response: Tap on button to show picture of choice. #trials: 8 per condition Setting: Laboratory/Home Duration: 5 min	1. Proportion of button taps to access social image: Significant difference (ASD < TD) No difference in control trials (scrambled images)
Chetcuti et al. (2019) Specific measure: Social imitation of simple versus complex motor tasks Country: Australia	1. ASD group Mage (SD) = 41.77 (8.5) months, N = 35, %male = 80 2. TD group Mage (SD) = 44.7 (13.64) months, N = 20, %male = 65	Tablet computer (iPad) with iPad application Slide & Spin	Task description: Four on-screen targets presented on the screen, manipulable using tap, drag, swipe and rotate actions. Child response: Imitate five-action sequence under four conditions (2 social × 2 motor complexity). Social comprised a socially responsive versus aloof instructor. Motor complexity involved low (five consecutive taps) versus high (multiple motor actions) complexity sequences. #trials: 1 per condition Setting: Laboratory/Home Duration: Not specified	1. Number of correct imitations by condition: Significant difference in high-motor demand task (ASD < TD). No difference in low-motor demand task. No difference based on social condition.
Carlsson et al. (2018) Specific measure: False-belief understanding Country: Sweden	1. ASD group a Mage (SD) = 91.19 (10.8) months, N = 52, %male = 79.41 2. TD group Mage (SD) = 89.99 (10.8) months, N = 98, %male = 51.02	Tablet computer (specifications not provided)	Task description: Watch a short film adapted from Sally–Anne task. Facilitating effect of language support assessed using three auditory conditions: (1) narrative; (2) silent and (3) interference. Child response: Respond to questions by tapping on one of two yellow circles (correct and incorrect ROI). Data saved in the device. #trials: 2 per condition Setting: Clinic (ASD); Quiet school room (TD) Duration: 3–4 min	1. Task completion: 100% in TD group, 75% in ASD group 2. Accuracy (two successful trials per condition): Significant difference (ASD < TD) in narrative and silent conditions; same trend (but not significant) in interference condition
Jones et al. (2018) Specific measure: Statistical learning Country: USA	1. ASD group a Mage (SD) = 64.67 (16.07) months, N = 56, %male = 80.36 2. TD group Mage (SD) = 60.11 (16.19) months, N = 68, %male = 55.88	Tablet computer (iPad)	Task description: Statistical learning task: sequence of two images (cue followed by target/distractor) presented for 2 s each. Cues were either high frequency (HF) or low frequency (LF), indicating likeliness of the next image to be a target or distractor. Child response: Tap on the target image, avoid distractor image. #trials: 7 runs of 24 trials each; 84 trials preceded by LF cue and HF cue, respectively. Setting: Laboratory Duration: Not specified	1. Accuracy: No difference. 2. Reaction time: Significant difference (ASD > TD). Unique RT patterns in TD (quadratic pattern in LF but not HF) versus ASD (linear pattern in both conditions). 3. Bayes classification to determine degree to which ASD child’s RT pattern was similar to TD group: ASD children with less severe autism symptoms (Social Responsiveness Scale-2) had similar learning profile to TD group.
Campbell et al. (2019) Specific measure: Social attention and response to name Country: USA	1. ASD group Mage (SD) = 26.19 (4.07) months, N = 22, %male = 77.27 2. TD group Mage (SD) = 21.91 (3.78) months, N = 82, %male = 58.54. Includes eight children with a diagnosis of language or developmental delay sufficient to qualify for speech or developmental therapy	Tablet computer (iPad)	Task description: Set of developmentally appropriate videos presented on the screen (cascading bubbles, mechanical bunny, animal puppets interacting with each other). During three of the videos, ‘Call Name’ appeared on the screen. The examiner, standing behind the child, called the child’s name loudly. Child response: The front camera on the tablet recorded the child’s video in response to name being called. ML algorithm automatically tracked head position. #trials: 3 prompts to call child’s name Setting: Laboratory Duration: 5 min	1. Task engagement (time looking at videos or people in the room): Significant difference (ASD < TD) 2. Orienting to name: a. At least once across three trials: No difference b. Multiple times across three trials: Significant difference (ASD < TD) c. Latency to orient to name: Significant difference (ASD > TD)
Gale et al. (2019) Specific measure: Social preference Country: Norway	1. ASD group Mage (SD) = 59.3 (18.3) months, N = 27, %male = 77.8 2. TD group Mage (SD) = 34.5 (12.3) months, N = 40, %male = 55	Tablet computer (Samsung Galaxy)	Task description: Study 1 and 2: Blurred videos of social (faces of people and dogs) and non-social (abstract moving geometric patterns) stimuli presented side-by-side. Study 3: Any one stimulus presented to assess reinforcement strength. Child response: Study 1 and 2: Tap on (blurred) video of choice. Chosen video grew larger and became clearly visible for 2 s. Study 3: Tap on stimuli (social or non-social) multiple times (progressively increasing – 2, 4, 6 times across trials – to view video clearly). #trials: 8 sessions Setting: child’s home, nursery or clinic Duration: 90 s for each session	1. Study 1 and 2: Proportion of taps on social videos: Significant difference (ASD < TD) 2. Study 3 (reinforcement strength): a. Proportion of taps to access non-social stimuli: Significant difference (ASD > TD). b. Proportion of time spent on session showing non-social stimuli: Significant difference (ASD > TD) c. Breakpoint of non-social video session (number of times the child clicked the non-social video before giving up): Significant difference (ASD > TD).
Carpenter et al. (2021) Specific measure: social-emotional reciprocity/facial expressions Country: USA	1. ASD group Mage (SD) = 26.2 (4.1) months, N = 22, %male = 77.27 2. TD group Mage (SD) = 21.7 (3.8) months, N = 74, %male = 58 3. DD (Non-ASD delay) Mage (SD) = 23.9 (3.7) months, N = 8, %male = 62	Tablet computer (iPad)	Task description: Same as Campbell et al. (2019). Set of developmentally appropriate videos presented on the screen. Social (woman singing nursery rhymes) and non-social (noise-making toy) videos were also presented on the left and right sides of the screen, respectively. Child response: Front camera captured child’s video. ML algorithm predicted the probability of facial expressions (positive, neutral or other) for each 3-s window. #trials: 1 Setting: Clinic Duration: 5 min	ASD group displayed increased frequency of neutral expression compared to the non-ASD group. 1. Area under the ROC curve (predict ASD diagnosis based on two metrics – mean probability of facial expressions for 3-s windows and frames not attending for each movie): (1) Bubbles_1: 0.75, Bunny: 0.81, Puppets: 0.78, Rhymes: 0.83 Bubbles_2: 0.79. Model for ‘Rhymes’ movie yielded the strongest predictive ability
Zhao and Lu (2020) Specific measure: Imitation of facial expressions Country: China	1. ASD group Mage (SD) = NA (NA) months; Range: 36–72 months, N = 10, %male = 50 2. TD group Mage (SD) = NA (NA) months; Range: 60–96 months, N = 10, %male = 50	Varying: mobile phones (both Android and iOS), personal computer (Windows 10) with Windows server	Task description: A software prompted children to imitate seven different facial expressions through pictures and sounds (happy, sad, angry, disgust, surprise, fear, neutral). Child response: Camera recorded children’s facial expressions. ML algorithm estimated probabilities of the child making each of the seven expressions. #trials: 3 per expression Setting: Quiet school room Duration: Not specified	1. Accuracy to correctly imitate on-screen expressions: Significant difference (ASD < TD). ASD group performance most compromised for disgust, neutral, surprise and fear.
Bovery et al. (2021) Specific measure: Social preference and attention Country: USA	1. ASD group Mage (SD) = 26.19 (4.07) months, N = 22, %male = 77.27 2. TD group Mage (SD) = 21.91 (3.78) months, N = 82, %male = 58.54 Includes eight children with a diagnosis of language delay or developmental delay sufficient to qualify for speech or developmental therapy	Tablet computer (iPad fourth generation). Front camera recorded video at 1280 × 720 resolution and 30 frames/s	Task description: 1-min video containing social (singing women) and non-social (dynamic toys with sound) stimuli. Both types of stimuli changed at pre-specified times within the movie (a different woman or toy appeared at different times). Child response: Webcam captured children’s videos while they watched the movie. ML algorithm 1 – automatically detected head position. ML algorithm 2 – predicted where on the screen the child looked (left, right or indeterminate) #trials: 1 Setting: Clinic Duration: 1 min	1. Attention on screen (number of frames in which child looked at the screen): Significant difference (ASD < TD). 2. Attention to social versus non-social: No difference 3. Attention shift to novel stimuli when stimuli changes: Significant difference (ASD < TD).
Lu et al. (2019) Specific measure: Abstract rule learning in social and non-social contexts Country: China	1. ASD group a Mage (SD) = 69.96 (9.72) months, N = 28, %male = 82.14 2. TD group Mage (SD) = 66.36 (5.04) months, N = 28, %male = 82.14	Laptop computer (specifications not provided)	Task description: Gamified tasks based on distrust and deception tasks. Token hidden in one of three boxes. Player incorrectly indicated to opponent the box with the hidden token. Two conditions: non-social – computer as opponent, and social – computer-controlled avatar as opponent that children believed were real people. Games included (1) recognising and avoiding misleading cues (distrust) and (2) providing misleading cues (deception). Child response: Distrust task: correct response was to choose any of the two boxes not indicated by the computer (computer always indicated the wrong box). Deception task: correct response was to indicate a box other than the one in which token was hidden (Computer always chose the indicated box). #trials: 10 each Setting: Not specified Duration: Not specified	1. Number of correct trials: Distrust task – Significant difference (ASD < TD) in both social and non-social conditions. Deception task – Significant difference (ASD < TD) in social, but not in non-social condition. 2. Speed of learning: Significant difference (ASD < TD) in both tasks in social condition. No difference in non-social condition.
Nakai et al. (2014) Specific measure: Speech intonation Country: Japan	1. ASD group a Mage (SD) = 87.23 (6.76) months, N = 26, %male = 76.92 2. TD group Mage (SD) = 83.48 (9.13) months, N = 37, %male = 54.05	Lavalier Microphone (Sony, ECM-77B/9X), Audio Capture (ROLAND, EDIROL UA-25EX), Mobile Note PC (TOSHIBA, Dynabook SS RX2/T7G)	Task description: 50 picture cards of animals and objects displayed on the screen. Child had to name them. Child response: Microphone fixed to child’s clothing recorded verbal responses. Audio data of correct responses isolated for analysis. #trials: 2 Setting: Not specified Duration: Not specified	1. Pitch coefficient of variation (each word): No difference in preschool children (4–6 years). Significant difference in school-age children (7–9 years). 2. Pitch range and SD (each word): No difference 3. Whole speech pitch metrics: No difference.
Wijesinghe et al. (2019) Specific measure: Speech characteristic Country: Sri Lanka	1. ASD group Age range in both groups: 18–36 months Mage (SD) = NA (NA) months, N = 8, %male = NA 2. TD group Mage (SD) = NA (NA), N = 9, %male = NA	Voice recorder (specifications not provided)	Task description: Voice recorder placed either within a pocket in the child’s clothing or within a metre from the child for periods varying from 2 to 10 h. Recording of conversations between the index child with a familiar adult in a familiar environment was captured. Child response: Audio signals were segmented as ‘silent’ and ‘non-silent’, and further to ‘vocal’ and ‘noise’ segments from non-silent segments. Child utterance data measured from vocal segments only. #trials: Not specified Setting: Child’s familiar environment Duration: 2–10 h	1. ML accuracy: ML model with feature set (duration of each utterance category (child uttering a meaningful word, meaningless word, vegetative sounds, adult utterances, noises, silences, total child duration and total audio duration per 10 min) not effective in classifying children).
Gyori et al. (2008) Specific measure: Social-emotional reciprocity/facial expressions Country: Hungary	1. ASD group Mage (SD) = 58.38 (8.45) months, N = 13, %male = 69.2 2. TD group Mage (SD) = 57.15 (6.74) months, N = 13, %male = 46.2	Desktop computer. Webcam below the monitor captured emotional expressions. Noldus FaceReader (v5.1, Noldus Information Technology) for emotional states analysis	Task description: Gamified task to assess ability to use deception and sabotage as social strategies in competitive and co-operative contexts. Game included tasks to evoke emotional, behavioural and gaze responses. Child response: Webcam captured child’s video; analysed by the Noldus FaceReader #trials: Not specified Setting: Laboratory Duration: ~24 min	1. Mean intensities of emotions: No difference
Lin et al. (2013) Specific measure: Language Country: Taiwan	1. ASD group a Mage (SD) = 66.11 (8.90) months, N = 35, %male = NA 2. TD group Mage (SD) = 60 (12) months, N = 300, %male = 49	Computer. Online (worldwide web) language assessment tool connected to backend server. ‘Offline’ version available which allows temporary storage of data on device for later upload.	Task description: Six language tests presented in auditory and visual formats – (1) decoding (DE), (2) homographs (HOM), (3) visual vocabulary comprehension (VVC), (4) auditory vocabulary comprehension (AVC), (5) visual sentence comprehension (VSC) and (6) auditory sentence comprehension (ASC). Child response: Test administrator recorded correct/incorrect responses using keyboard key presses or mouse clicks. #trials: DE = 50, HOM = 14, VVC = 53, AVC = 38, VSC = 15, ASC = 15. However, 104/186 items retained in the final test. Setting: Quiet room (clinic or school) Duration: ~35 min	1. Accuracy on each sub-test: Significant difference in DE, HOM, VVC, VSC (ASD > TD) and ASC (ASD < TD). No difference in AVC. In DE and VVC, largest difference seen at 4 years (ASD > TD), reduces at 6 years. For ASC, differences at 4 years (ASD < TD) further enhanced at 6 years.
Chaminade et al. (2015) Specific measure: Biological motion/anthropomorphic bias Country: France	1. ASD group Mage (SD) = 61 (25) months, N = 12, %male = 91.6% 2. TD group Mage (SD) = 39 (7) months. In total, 12 of the youngest TD children mentally age-matched to ASD sample, N = 24, %male = 58.3%	Computer touchscreen (specifications not provided)	Task description: Gamified task where four characters (two humans and two cartoons) are associated with two kinds of motions: biological and artificial. From these set of combinations, two videos were presented simultaneously on the left and right of the touchscreen. Child response: Children touched the video they liked more. The touched video grew in size while the other disappeared. #trials: 16/session, up to 4 sessions/participant. Setting: Laboratory Duration: Not specified	1. Proportion of choices to biological motion (human): Significant difference (ASD < TD) 2. Proportion of choices to biological motion (cartoon): No difference. Results similar for mentally and chronologically age-matched samples
Deschamps et al. (2014) Specific measure: Empathy and prosocial behaviour Country: The Netherlands	1. ASD group a Mage (SD) = 81.59 (6.95) months, N = 22, %male = 81.82 2. TD group Mage (SD) = 86.39 (6.71) months, N = 29, %male = 82.76	Computer screen (specifications not provided)	Task description: Ball-throwing computer game against two computer-controlled players who gave rewards when ball was passed to them. When rewards ran out (final round), one player showed progressively distressed facial expressions each time the ball was not passed to them. Two conditions – girl and boy as the distressed player. Child response: Choose player to pass ball. #trials: 20 in final round, 10 in previous rounds Setting: Quiet school room Duration: Not specified	1. Number of ball throws to distressed player: No difference
Aresti-Bartolome et al. (2015) Specific measure: Social interaction and eye contact Country: Spain	1. ASD group a Mage (SD) = NA. Age range: 36–96 months, N = 20, %male = NA 2. TD group Mage (SD) = NA. Age range: 36–96 months, N = 20, %male = NA Gender-matched to the ASD group	21″ touch screen. Tactile pointer of 40 cm. Games configured by APNABI Association	Task description: Gamified task with three levels of difficulty involving collecting as many pre-specified items within a 3-min period as possible by touching the screen with a pointer. Game stopped every 30 s or when an error was made. Child response: For game to continue, child had to interact with test administrator, who recorded the latency of the interaction through key presses on the keyboard (separate keys for interactions with and without eye contact). #trials: 1 per level Setting: Classroom Duration: 12 min	1. Latency of interaction with administrator: Significant difference (ASD > TD). 2. Number of interactions with eye contact: Significant difference (ASD < TD) 3. Task completion (% not completing levels): 10, 14, 15% in ASD versus 0, 0, 5% in TD 4. Number of pre-specified items collected per level: Significant difference (ASD < TD)
P. Li et al. (2016) Specific measure: Selective trust Country: China	1. ASD group a Mage (SD) = 73.55 (9.71) months, N = 30, %male = 86.67 2. TD group Mage (SD) = 70.31 (7.79) months, N = 30, %male = 86.67	Computer screen (specifications not provided)	Task description: Virtual candy hidden in one of the two boxes. Boxes had pictures of faces with contrasting themes: own versus other race, attractive versus unattractive, trustworthy versus not trustworthy Child response: Tap on box with preferred face characteristic. #trials: 30 trials, 10 per condition. Setting: Not specified Duration: Not specified	1. Accuracy of choosing own race, more attractive, or more trustworthy faces: (a) Race: No difference; (b) Attractiveness: Significant difference (ASD > TD) for more attractive face; (c) Trustworthiness: No difference
Borsos and Gyori (2017) Specific measure: Social-emotional reciprocity/facial expressions Country: Hungary	1. ASD group Mage (SD) = 58.38 (8.45) months, N = 13, %male = 69.2 2. TD group Mage (SD) = 57.15 (6.74) months, N = 13, %male = 46.2	Desktop computer. Webcam below the monitor captured emotional expressions. Noldus FaceReader (v5.1, Noldus Information Technology) for emotional states analysis Standard desktop-mounted eye-tracking device (EyeFollower 2 by LC Technologies) to detect attention to screen.	Task description and child response: Same as Gyori et al. (2008). For each frame, data were classified as invalid if the Noldus FaceReader was unable to identify the face or unable to assign an emotion to the frame. #trials: 3 game sections played one time each Setting: Laboratory Duration: ~24 min	1. Mean of static (i.e. frame-by-frame) intensities of emotional states: Significant difference in scared and surprised emotions (ASD > TD). 2. Speed of emotion expression change: Significant difference in ‘surprised’ emotion (ASD > TD). 3. Valid data ratio (ratio of valid vs invalid data): No significant difference.
Martin et al. (2018) Specific measure: Social preference and head postural response Country: USA	1. ASD group Mage (SD) = 60.8 (16.52) months, N = 21, %male = 80.95 2. TD group Mage (SD) = 51.23 (15.35) months, N = 21, %male = 66.67	19″ video monitor with a camera mounted on the top edge	Task description: Six videos containing social and non-social stimuli presented on a screen. One video each was purely non-social and purely social. Remaining four had a mix of both. Child response: Camera recorded children’s videos while they observed the stimuli. A computer vision algorithm (Zface) tracked pitch, yaw and roll of head movement frame-by-frame. #trials: 1 Setting: Laboratory Duration: 16 min	1. Proportion of successfully tracked frames: No significant difference 2. Number and duration of epochs (successfully tracked consecutive  frames) per video: No significant difference 3. Angular displacement: Significant difference in yaw (ASD > TD) in social video. No difference for pitch and roll. No difference between groups in non-social video. 4. Angular velocity: Significant difference in yaw and roll in social video (ASD > TD), but no difference in pitch. No difference between groups in non-social video.
Li et al. (2020) Specific measure: False-belief understanding Country: China	1. ASD group a Mage (SD) = 77.04 (9) months, N = 17, %male = 82.4 2. TD group Mage (SD) = 77.04 (9) months, N = 17, %male = 58.82	Notebook computer (specifications not provided)	Task description (1): Truth value judgement (TVJ) task: One puppet made statements about pictures shown on a screen. Statements contained factive (‘knows’), counter-factive (‘pretend’) and non-factive (‘thinks’) verbs. Child response: Judge the truth of the puppet’s statement by pressing one of the three buttons – ‘yes’, ‘no’ and ‘maybe’. # trials: 10 ‘pretend’ and 15 (‘knows’, ‘thinks’) Task description (2): First- and second-order false-belief tasks presented as series of pictures with voiceovers. Child response: Answer questions to demonstrate FB understanding. Each correct answer awarded one point. #trials: 8 Setting: Kindergarten/primary school/training centre Duration: Not specified	1. Mean (SD) of TVJ task score: Significant difference (ASD < TD) for ‘pretend’ condition; approaching significance (ASD < TD) in ‘know’ condition. No difference in ‘think’ condition. 2a. Mean (SD) of first-order FB tasks: Significantly different (ASD < TD) 2b. Mean (SD) of second-order FB tasks: Significantly different (ASD < TD)
Li et al. (2019) Specific measure: Social gaze Country: China	1. ASD group Mage (SD) = NA (NA) months; Range: 48–84 months, N = 136, %male = NA 2. TD group Mage (SD) = NA (NA) months; Range: 72–96 months, N = 136, %male = NA	Computer screen (specifications not provided). Head movements recorded using 360 intelligent camera	Task description: Children seated in front of a screen with their mother’s picture displayed. Child response: Camera recorded children’s videos while they observed the stimuli. ML approach used to track trajectory of eye movements in the first 2001 frames of the video. Trajectory data divided into angle and length information and used to classify children as ASD or TD. #trials: 1 Setting: Primary and special education schools Duration: 10 min	1. No. of frames with eyes not visible: Significant difference (ASD > TD) 2. Accuracy of ML algorithm to classify children into diagnostic groups: 92.6% using all features (both length and angle information)
Shahab et al. (2017) Specific measure: Joint attention and imitation Country: Iran	1. ASD group a Mage (SD) = 58.8 (9.96) months, N = 14, %male = NA 2. TD group Mage (SD) = 60 (10.8) months, N = 21, %male = NA	Virtual-reality (VR) setup	Task description: VR setup to play the xylophone and drums, led by robots controlled remotely via an operator. Step 1: Virtual robot showed child how to play a virtual drum and the xylophone, using VR controllers as mallets. Step 2: Children asked to describe what they saw in the virtual room. In case of unable to name objects present in the room, robots pointed to objects to direct the child’s attention. Midway through task, children requested to wear VR headset. Child response: Imitate robot’s actions. Child behaviour recorded by two video cameras. One point awarded for naming each picture in virtual room (4 total). #trials: 1 Setting: Laboratory Duration: ~ 10 min	1. Score in the drum and xylophone imitation tasks: Significant difference (ASD < TD). 2. Picture naming: Significant difference (ASD < TD) 3. Task engagement (defined as (1) duration engaged with the game and (2) duration for which the child wore the VR headset): Significant difference (ASD < TD) for both metrics.
Jyoti et al. (2020) Specific measure: Joint attention Country: India	1. ASD group a Mage (SD) = 76.8 (14.76) months, N = 20, %male = 65 2. TD group Mage (SD) = 80.4 (10.38) months, N = 20, %male = 65	VR-enabled HCI-based task platform. Touch-sensitive monitor to record child responses.	Task description: Joint attention (JA) task with 3D avatar within a VR setup. The avatar provided cues through (1) eye gaze alone; (2) head turn with gaze; (3) finger pointing, head turn and gaze and (4) all of the above and sparkling of the target contingent with JA cues. JA cues offered randomly. Child response: Touching the target object indicated by JA cue on a touch-sensitive monitor. Performance recorded on confirmation of choice. #trials: 8 (2 per cue) Setting: Laboratory Duration: 20 min	1. Average performance score on JA task: Significant difference (ASD < TD) in two cues – eye gaze alone and eye gaze with head turn. Ceiling effect (100%) for TD group irrespective of JA cue type. ASD group performance improved with increasing information in the JA cue. 2. Average reaction time: Significant difference (ASD > TD). Approximately three times longer in ASD. Average response time decreases as cue information increases. 3. Effectiveness index (EI): EI = PI (scaled performance score) + (1-RTI) (scaled reaction time). Significant difference (TD > ASD). EI consistently > 1.5 across all cues for TD. For ASD group, EI progressively increases from 0.5 to 1.5 for JA cues with increasing information.
DSM-5 criteria: Stereotypical, repetitive or restricted behaviours or interests
Moradi et al. (2017) Specific measure: Patterns of play movements Country: Iran	2.ASD group a Mage (SD) = 57.24 (12.36) months, N = 25, %male = 84% 2. TD group Mage (SD) = 60.24 (8.76) months, N = 25, %male = 64%	Wii remote (with a 3-axis Micro-Electro-Mechanical System (MEMS) ADXL330 accelerometer) embedded into a toy car	Task description: Children played with a toy car embedded with an accelerometer. Child response: Accelerometer recorded car’s movement in 3D. Data transferred using Bluetooth or Wi-Fi. Typical data file consisted of ~5000 samples of time and acceleration data in 3D representing child’s play with the car. #trials: 1 Setting: quiet room Duration: ~5 min	1. Accuracy, sensitivity, specificity of ML (SVM) methods: The full feature set (44 – see select examples below) discriminated between groups with reasonable accuracy (62%), sensitivity (65%) and specificity (61%). Feature examples: Play time; In each three dimension – correlation of acceleration between two of the three axes, mean and variance of acceleration, dominant frequencies of acceleration direction, total acceleration signal energy, number of jolts in the forward direction.
Motor (not covered in DSM-5 criteria)
Rafique et al. (2019) Specific measure: Fine-motor drawing movements Country: Pakistan	1. ASD group Mage (SD) = NA (NA) months, Age range in both groups: 60–144 months, M: 89 months, N = 22, %male = 77.2 2. TD group Mage (SD) = NA (NA) months, N = 22, %male = 54.5	Android phone (version 6.0.1)	Task description: Trace and colour a dotted square shape. Smartphone was placed on a flat table, so that the sensor values recorded force applied on the phone. Smartphone recorded four inertial and six touch data. Child response: Perform drawing task. #trials: Not specified Setting: School room Duration: Not specified	1. Accuracy of ML methods to classify children into diagnostic groups using top 10 features of the creativity game reported by Anzulewicz et al. (2016): Accuracy > 85%.
Mahmoudi-Nejad et al. (2017) Specific measure: Fine-motor tracing movements Country: Iran	1. ASD group Mage (SD) = NA (NA) months. Range: 48–84 months, N = 5, %male = 100% 2. TD group Mage (SD) = NA (NA) months. Age-matched with ASD group, N = 7, %male = 57.14%	Tablet computer and smartphone (specifications not provided)	Task description: Follow a pre-specified path marked with pink flowers to take a bee to its hive. Pink flower turns green (win) when bee touches it. Untouched flowers turned red (fail). Haptic feedback provided if defined trajectory is not followed. Child response: Trace path to take bee to hive. Touch data recorded on the tablet. Children awarded 0–4 stars/trial depending on their accuracy (win/total). Data analysed only from Level 1 since ASD group could not play beyond this level. #trials: 8 sub-levels within 4 difficulty levels. Setting: Not specified Duration: Not specified	45 features extracted from touch data in two categories – point-based and progress-based. Points-based: Calculated using two adjacent points the child touched (e.g. distance, velocity, acceleration, time, curvature, error) Progress-based: Indicator of attempts made (e.g. score, levels played, completion time). ML methods (Linear SVM) used to classify children into their respective groups. Three features – Total score, Average velocity and Average curvature – discriminated between ASD and TD groups.
Dawson et al. (2018) Specific measure: Rate of head movements Country: USA	1. ASD group Mage (SD) = 26.19 (4.07) months, N = 22, %male = 77.27 2. TD group Mage (SD) = 21.91 (3.78) months, N = 82, %male = 58.54 Includes eight children with a diagnosis of language delay or developmental delay sufficient to qualify for speech or developmental therapy.	Tablet computer (iPad). Front camera recorded video at 1280 × 720 resolution and 30 frames/s	Task description: Same as Campbell et al. (2019). Child response: Camera captured child video while they viewed stimuli on tablet screen. ML algorithm detected and tracked head position. Three head pose angles – yaw (left-right), pitch (up-down) and roll (left-right tilt) calculated per frame. Child was considered to be looking at the screen when yaw ⩽ 20° #trials: 1 Setting: Clinic Duration: 5 min	1. Rate ratio of head movement of ASD with TD as reference: Rate ratio calculated from the association between head movement rate (for every one-third second across the movie) and ASD diagnosis. Robust differences observed in four out of five movies (except bubbles) (ASD > TD). These four movies had complex stimuli compared to bubbles video. Associations unchanged with eight DD participants removed from the TD group during sensitivity analysis.
Fleury et al. (2013) Specific measure: Fine-motor drawing movements Country: Canada	1. ASD group a Mage (SD) = 81.6 (19.08) months, N = 15, %male = 86.96 2. TD group Mage (SD) = 63 (12) months, N = 19, %male = 85	Wacom Cintiq 15X digitising tablet and pen	Task description: Draw circles on a touchscreen using a stylus. Six conditions based on hand-use (dominant vs non-dominant) and three circle drawing styles (continuous drawing without pausing: comfortable vs fast pace; and discontinuous by pausing at the top before starting again). Child response: Draw circles as per the six specified conditions. #trials: 1 each (6 total) Setting: Not specified Duration: 14 min	1. Mean circle drawing time: No difference 2. Coefficient of variation (CV) of circle drawing time: Significant difference in discontinuous condition (ASD > TD) 3. Power spectral density (PSD), root mean squared (RMS) fluctuations during drawing and statistical persistence (Hurst coefficient): No difference
Dowd et al. (2012) Specific measure: Motor planning and execution Country: Australia	1. ASD group a Mage (SD) = 74.39 (16.8) months, N = 11, %male = 72.73 2. TD group Mage (SD) = 79.19 (18) months, N = 12, %male = 75	17″ LCD touch screen (MicroTouch 3MM170), connected to a HP Compaq 6910p laptop	Task description: Point-to-point movement task. Join two points presented on a vertical plane using a stylus. Start position at the bottom centre of the screen. End position at the top of the screen either left, centre or right from start. Some trials had a visual distractor near the target. Child response: Trace a line from start to end. Stylus movement was sampled at 125 Hz. #trials: 25 Setting: Laboratory Duration: Not specified	1. Various kinematic variables: a. Significant difference in variability of movement preparation time (ASD > TD). No difference in any other variable b. Interaction effects observed for distractor condition: longer and more variable total movement time in TD. No difference in ASD.
Crippa et al. (2013) Specific measure: Hand-eye coordination Country: Italy	1. ASD group a Mage (SD) = 74.4 (25.2) months, N = 14, %male = 85.7% 2. TD group Mage (SD) = 75.6 (27.6) months, N = 14, Age- and gender-matched	100 Hz 17″ LCD touch screen (Elo TouchSystem) at 1024 × 768 pixel resolution. 50X Tobii for eye-tracking, AB Danderyd, Sweden, with 5-point calibration. Video recorder	Task description: Gap overlap task. Child response: Two conditions – (1) ‘press’: participants indicated the target’s left/right position on screen using a button box; and (2) ‘touch’: participants touched the targets on the screen. Movements started from a set position by lifting the hand off a pad on the table. #trials: 3 sessions of 16 trials each. Setting: Laboratory Duration: Not specified	1. Reaction time for button press in ‘press’ condition: No difference 2. Accuracy of touching target in ‘touch’ condition: No difference 3. Eye-hand coordination (Pearson correlation between eye fixation latency on target and hand response (key presses and touch times): Strong and significant correlation in TD. Weak (‘press’) or no (‘touch’) correlation in ASD.
				4. Differences in gap and overlap conditions: No difference in ‘press’ condition. Significant gap effect in TD but not ASD in ‘touch’ condition.
Jung et al. (2006) Specific measure: Visuomotor coordination Country: South Korea	1. ASD group a Mage (SD) = 72 (0) months, N = 12, %male = 83.33 2. TD group Age range: 60–72 months, N = 20, %male = NA, Mage (SD) = NA	VR setup: Pentium IV PC, one projector, one screen (200 X150 cm), one infrared reflector, one digital camera and tangible devices (stick)	Task description: Burst virtual balloons in a VR setup. Auditory and visual reinforcements provided when successful. Number of balloons and the type of reinforcement changed by level. Child response: Burst virtual balloons by moving a real stick. #trials: 10 sessions Setting: Laboratory Duration: Not specified	1. Accuracy: No difference (may be due to high variability in ASD group). 2. Reaction time: Significant difference (ASD > TD) 3. Movement of stick to pop balloons: Significant difference (ASD < TD) 4. PCA index: Significant difference indicating TD more efficient in popping balloons using combination of three variables.
Alcañiz Raya et al. (2020) Specific measure: Gross-motor movements Country: Spain	1. ASD group a Mage (SD) = 61.56 (16.2) months, N = 24, %male = 87.5 2. TD group Mage (SD) = 58.32 (10.92) months, N = 25, %male = 64	RGB-D camera (includes depth information in video recordings): RealSense – camera D435 (FRAMOS, Munich, Germany) and Intel RealSense SDK 2.0 (Intel RealSense Technology, Santa Clara, CA, USA) VR setup: CAVE-Automatic Virtual Environment (CAVETM): semi-immersive room with 3–6 rear-projected surfaces	Task description: VR setup showing street intersection with three types of stimuli conditions: visual (V): avatar walks into the scene and waves to participant; visual–auditory (VA): avatar dances to a song for 10 s; visual–auditory–olfactory (VAO): two avatars bite into a buttered muffin, while participant is able to smell, see and hear actions. Child response: Imitate avatars while a camera recorded movements. OpenPose algorithm used to detect body joints in each frame. Joint displacement computed across two consecutive frames and then averaged across a condition. #trials: 9 (3 per condition) Setting: Laboratory Duration: 14 min	1. Average movement of each joint in each condition: Significant difference (ASD > TD) in all three conditions. Condition V: leg, head and trunk Condition VA: leg and head Condition VAO: head ML models used to discriminate between TD and ASD groups. 2. ASD classification accuracy: Head metrics performed the best in all conditions. Highest accuracies observed when using head metrics in VAO condition and all joints in V condition (89.36%). Least accuracy (70.2%) in model including all joints and all conditions.
Cognitive (not covered in DSM-5 criteria)
Chen et al. (2019) Specific measure: Executive functioning Country: China	1. ASD group a Mage (SD) = 54 (11) months, N = 40, %male = NA 2. TD group Mage (SD) = 55 (7) months, N = 51, %male = NA. Groups were gender-matched.	Tablet computer (PC or PAD)	Task description: Series of gamified tasks presented on a screen. Included joint attention tasks, responding to social requests, matching shapes, categorisation, visual search and visuomotor coordination. Child responses: Tap on target objects on the screen during gameplay. Responses stored on the tablet. One point awarded for each correct answer. #trials: 1 Setting: Therapeutic centre and kindergarten Duration: 15–20 min	1. Completion rate (proportion of participants completing game): Significant difference (ASD < TD for children > 48 months) 2. Efficiency (ratio of average score to average completion time): Significant difference (ASD < TD; visual search and visuomotor coordination in ⩽ 48 months only; Categorisation and matching shapes in > 48 months)
Hetzroni et al. (2019) Specific measure: Relational (abstract) thinking Country: Israel	1. ASD group a Mage (SD) = 78.84 (10.44) months, N = 24, %male = 75 2. TD group Mage (SD) = 67.56 (3.24) months, N = 24, %male = 41.66 3. IDD group Mage (SD) = 142.56 (27.48) months, N = 24, %male = 62.5	Portable computer (specifications not provided)	Task description: Pictures of animals presented in various relational configurations (example – two koalas as mirror images). In a subset of trials, another panel with the same configuration, but with a different animal, was shown to strengthen concept. Influence of level of familiarity determined using local (known), foreign (partially known) and made-up (unknown) animal pictures. Child response: Tap on one of the two options that match the configuration of the target panel. Correct responses awarded one point. #trials: 8 Setting: quiet room Duration: 30–40 min	1. Accuracy (proportion of correct choices): Significant difference (IDD < ASD < TD in single panel condition). No benefit of second panel (to strengthen concept) to ASD group. 2. Influence of familiarity: Performance unaffected by familiarity in ASD group. In the TD and IDD groups, better performance with known target images.
Veenstra et al. (2012) Specific measure: Executive functioning Country: The Netherlands	1. ASD group a Mage (SD) = 61.2 (9.6) months, N = 13, %male = NA 2. TD group Mage (SD) = 45.6 (1.68) months, N = 5, %male = NA	Computer screen (specifications not provided)	Task description: Web-based gamified Go/No-go task (www.samenslim.nl). Child response: Through mouse clicks during gameplay. #trials: 2–3 sessions, max of 7 games/session. Setting: quiet room in a medical day-care centre or playgroup Duration: Not specified	Significant differences observed in all metrics: 1. Accuracy: ASD < TD 2: No-go (Number of clicks when no clicks should have been made): ASD < TD 3. Missing go (No clicks during clicking moments): ASD > TD 4. Go (Number of clicks during clicking moments): ASD < TD 5. Reaction time: ASD > TD 6. Repeated clicks (Repeated clicks on the same objects): ASD > TD 7. Variability across levels: ASD < TD
Gardiner et al. (2017) Specific measure: Executive functioning Country: Canada	1. ASD group Mage (SD) = 66.88 (13.41) months, N = 24, %male = 83.33 2. TD group Mage (SD) = 58.47 (15.87) months, N = 19, %male = 57.89	Touchscreen monitor (specification not provided)	Task description: Computerised tasks assessing executive functioning: 1. spatial working memory (Boxes task) 2. cognitive flexibility, inhibition and working memory (Go/No-Go, Preschool-Continuous Performance Test (PCPT)) 3. Multicomponent planning task (Monkey Tower) – adaptation of Tower of Hanoi task Child response: Tap on screen during gameplay #trials: 1 Setting: Laboratory Duration: 90–120 min	1. Accuracy on Boxes, Go/No-Go, PCPT: No difference 2. Number of correct trials in multicomponent planning task: Approaching significance (ASD < TD; p = 0.036).

ASD: autism spectrum disorder; TD: typically developing; AUC: area under the ROC curve; ROI: region of interest; RT: reaction time; DD: developmental delay; ROC: Receiver Operating Characteristic; FB: false belief understanding; PI: performance index (scaled); RTI: reaction time index (scaled); PCA: Principal Component Analysis; IDD: Intellectual and Developmental Disabilities.

The following colour coding has been used in the column named ‘Device specifications’ to indicate the feasibility of the device for use in low-resource settings: Green = most feasible; Red = least feasible.

Colour coding

Figure added to Table 1 legend: Colour coding to indicate feasibility of administering the identified digital tools in low-resource settings.

The time taken to complete the assessment, when specified in the article, is included in the column titled ‘Experimental setup’ in red font colour, along with the number of trials of the assessment and the experimental setting (laboratory, clinic, school, home, etc).

^aTo maintain consistency, age when reported in years was converted to months by multiplying by 12.