Artificial Intelligence to Customize Participation-Focused Pediatric Re/habilitation Interventions

Objectives: This thesis aimed at: 1) examining factors related to school participation attendance and involvement among children and youth with craniofacial microsomia (CFM) and other childhood-onset disabilities; 2) identifying the current scope of participation-focused interventions using artificial intelligence (AI) in pediatric re/habilitation and to examine how participation is currently operationalized in this area of work; and 3) developing a predictive model using natural language processing that classifies participation-focused caregiver strategies to four key predictors of participation . Methods: Aim 1 applied structural equation modeling using secondary analyses of a subset of data including 260 families of children and youth with CFM and other childhood-onset disabilities. Aim 2 included two scoping reviews, which together mapped research according to the type(s) of AI and extent of customization used in participation-focused pediatric re/habilitation interventions and ways of operationalizing participation in pediatric re/habilitation research using AI. Aim 3 employed narrative data on 1,576 participation-focused strategies from 236 families of the Aim 1 dataset. A series of binary and multinomial logistic regression, naïve Bayes, and support vector machines (SVM) models were developed and compared. Results: For Aim 1, results revealed a significant positive association between school environmental supports and school participation attendance and involvement, a significant negative association between physical functioning problems and school participation involvement and a significant effect of the number of participation-focused caregiver strategies on the relationship between school environmental support and school participation attendance. For Aim 2, most AI-based participation-focused pediatric re/habilitation interventions used robotics and were not customized to individually reported needs of children and youth with childhood-onset disabilities. Operationalizations of participation in AI-based participation-focused pediatric rehabilitation research were not aligned with contemporary definitions of child and youth participation. For Aim 3, SVM models revealed accuracy scores from 78.10% - 94.92% and macro-averaged F1-scores from 0.58 – 0.83. Conclusion: Results yield evidence of the use of AI to customize the use of information about participation-focused caregiver strategies when delivering pediatric re/habilitation services for a diversified group of children and youth with childhood-onset disabilities.