Table A2.

Main characteristics of the studies reviewed: main findings, study limitations, risk of biases (n = 19).

Reference	Main Findings	Study Limitations	Risk of Biases
[14]	The application was deemed able to train children’s focus with a high percentage of agreement among expert therapists.	Limited generalizability due to small sample size; shortcomings of participants’ demographic information.	Sampling bias due to lack of sampling criteria explication; bias due to unspecified blinding.
[15]	Participants with ASD were more satisfied with performance and showed relatively deep interest in the game. Mean playtime decreased, mean collaborative operations efficiency increased. Control group had a higher collaborative efficiency, both groups had a similar increase trend in level of communication.	Limited generalizability due to small sample size; the majority of sample were male.	Sampling bias due to lack of sampling criteria explication; bias due to tool technical issues.
[16]	Cooperation performance and communication improved in the experimental group. Participants with ASD spoke more words per minute. Offline spontaneous communication was encouraged.	Limited generalizability due to small sample size.	Sampling bias due to lack of sampling criteria explication.
[23]	Total performance scores had a moderate positive correlation with three validated cognitive screening tools (Abbreviated Mental Test, Mini-Mental State Examination, and MoCA). A moderately significant relationship was found between total performance scores and presence of cognitive impairment.	Limited generalizability due to small sample size; significant difference in the education level between the groups; limited representativeness of the population at risk of cognitive impairment due to the sample recruitment carried out only at one location with multiple exclusion criteria and classification based only on MoCA scores.	Sampling bias due to the classification of the study population relying solely on MoCA scores, which is not considered diagnostic of cognitive impairment; Measurement bias due to using an assessment module that was originally used for rehabilitation and then adapted for cognitive screening.
[34]	Children’s relaxation, motivation, and concentration improved. Average time of less than 20 min was equivalent to 10.67% improvement in both groups.	Limited generalizability due to small sample size.	Sampling bias due to lack of sampling criteria explication; bias due to unspecified blinding.
[35]	A functional relationship was found between the gesture-based instruction via Leap-Motion-aided VR technology and the response accuracy and task engagement of students with ASD. Maintenance of the acquired skills was found at a high level up to 12 weeks.	Limited generalizability due to small sample, segregate setting of intervention (individual training room), and lack of female participants.	Sampling bias due to recruitment of a male-only sample.
[36]	CAI and TII were both effective in teaching visual matching skills, but CAI was more effective for the two students with ASD. CAI was more efficient than TII, since it required a lower number of prompts and a shorter instructional time. CAI promoted more task engagement than TII. Generalization to similar untaught skills and maintenance were found at a high level for up to 5 weeks under both CAI and TII.	Limited generalizability due to small sample size; the vast majority of sample was male; exclusion of younger students due to sensibility issues of small hand-gestures recognition.	Bias due to exclusion of younger students for technical issues.
[37]	Participant’s recognition and fine motor skills improved considerably, reaching performance accuracy of 100%. Skills such as looking at the hands and objects and moving the gaze with them were increased by the game. Transfer of learned rules and skills was found.	Limited generalizability due to small sample size and use of a single subject research AB design; participants’ improvement may be a mixed result of various factors (i.e., better emotional control that affected concentration; better understanding of the rules of the games; better skills of operating Leap Motion controller; improvement due to rote learning); participants’ different level of experience in using technology devices.	Bias due to use of single subject research design.
[38]	Fine motors skills and cognition of colors and fruits were improved, reaching accuracy of 100%.	Limited generalizability due to small sample size.	Bias due to use of single subject research design.
[39]	High levels of engagement, sustained attention, and independent manipulation were found in children. High satisfaction was found in families.	Limited generalizability due to small sample size and lack of female participants.	Sampling bias due to recruitment of a male-only sample and lack of sampling criteria explication; bias due to no blinding of participants and nonprobability sampling techniques.
[40]	High levels of engagement and sustained attention were found in children. High satisfaction was found in families. Children’s independence and natural manipulation increased.	Limited generalizability due to small sample size.	Sampling bias due to lack of sampling criteria explication; bias due to no blinding of participants and nonprobability sampling techniques.
[41]	Participants’ accuracy increased and time needed to complete the task decreased. The eye gazing game confused the children because of item distances issues.	Limited generalizability due to small sample size and lack of inclusion of sample with target diagnosis; shortcomings of participants’ demographic information.	Sampling bias due to lack of sampling criteria explication and lack of inclusion of sample with target diagnosis; bias due to unspecified blinding.
[42]	High level of satisfaction was found. Learning curve stabilized around an average response time of 20–30 s for the first training session. Immersive Virtual Reality interface showed efficacy in improving communication performance.	Limited generalizability due to lack of clinical sample; shortcomings of participants’ demographic information.	Sampling bias due to lack of clinical sample.
[43]	Clusters formed by using acceleration data seemed reasonably analogous to performance measures (i.e., type and number of occurred errors).	Limited generalizability due to small sample size and including non-clinical sample; shortcomings of participants’ demographic information.	Sampling bias due to lack of clinical sample; bias due to unspecified blinding.
[44]	Participants’ satisfaction was shown for LEAP Motion. Natural movements were represented well by LEAP Motion. The tool exhibited promising results for virtual rehabilitation.	Limited generalizability due to small sample size; the majority of sample were male.	Sampling bias due to lack of sampling criteria explication; bias due to lack of sample diagnosis specification.
[45]	LEAP Motion lost accuracy and was influenced by light.	NA	NA
[46]	LEAP-Motion-aided VR technology measures of functional abilities showed consistent functional impairment in mild Alzheimer’s disease, amnestic single and multiple domain MCI in comparison with healthy subjects. Total performance scores showed significant discrimination power.	Limited generalizability due to exclusion of elderly with technophobia.	Sampling bias due to exclusion of technophobic participants; statistical bias due to the use of statistical models with a limited number of covariates.
[47]	LEAP-Motion-aided VR technology measures of functional abilities was strongly correlated with standard cognitive and functional measurements as Mini-Mental State Examination and Bristol Activities of Daily Living scale scores. Total virtual measures of functional abilities showed consistent functional impairment in mild Alzheimer’s disease and amnestic MCI in comparison with healthy participants. Assessment module showed moderately good psychometric properties in discriminating healthy from pre-dementia and mild dementia patients.	Limited generalizability due to exclusion of elderly with technophobia.	Sampling bias due to exclusion of technophobic participants; statistical bias due to use of statistical models with limited number of covariates.
[48]	Users’ learning caused by the system and the interface obtained a considerably high punctuation. Meaningful correlations were found between interface and learning outcomes and information and learning outcomes. The hand–eye coordination exercise helped improve attention.	Limited generalizability due to small sample size and unspecified diagnosis of participants	Sampling bias due to lack of sampling criteria explication; bias due to lack of sample diagnosis specification; bias due to unspecified blinding.

Note: NA = not available given the type of study.