Modelling learning trajectories of facial emotion recognition training in autistic children

BackgroundEmotion recognition (ER) skills are important in social interaction. ER difficulties are associated with social difficulties and may lead to poorer educational outcomes and wellbeing. There is therefore considerable interest in tools that support development of ER skills. In addition, given that autism is associated with ER difficulties, such tools may be particularly beneficial for autistic children. In this study we examined how autistic childrens ER skills changed over 8 sessions of using an ER training task. MethodsAutistic schoolchildren or those considered to be autistic by parents (N=88; 79 with complete data) were recruited from school and community settings to complete 8 sessions of the digital ER task. We used mixed-effects regression models to estimate learning growth across all 8 sessions. We also examined how participant characteristics influenced results. Finally, we identified the optimal number of sessions needed for this task in terms of ER improvement. ResultsWe found overall improvements in ER with mean accuracy scores increasing from 58 to 73 (out of 100), representing a 25% improvement. Age, verbal age, gender, parental education and income influenced improvements in ER. The optimal number of sessions was 6. ConclusionsOur study demonstrated that completing multiple sessions of the ER training task was associated with improvements in ER. Improvements were seen across all children, but some characteristics influenced the pattern and degree of this change. Our study utilised mixed-effects smoothing spline models, bringing novel value to this context. These findings support the view that ER is a tractable target and inform how future interventions/tools may be implemented.