AI-supported Residential Care Placement Decision Making for Individuals with Autism Spectrum Disorder and Complex Medical Issues

Finding an appropriate residential setting for individuals with autism spectrum disorder and co-occurring medical conditions is challenging due to the heterogenous needs of the individuals, but also because residential programs utilize different approaches. As such, the decision making process of which individual is a good fit for a particular residential setting is both important and challenging. The decision to accept an individual’s placement in a particular program is generally made based upon the medical history, an intake medical examination, and prior experience of the program’s staff with residents with similar needs. As such, the decision is strongly influenced by the experience of the staff. This work builds an artificial intelligence model to support this decision making process for screening potential residents for one particular program, The Center for Discovery (TCFD). The model is based upon a nonlinear classifier and trained with data from a cohort of current and past residents to predict if an individual will respond well to the type of care offered by TCFD. The classifier is able to predict a successful placement with over 80% balanced accuracy when using information about past medical diagnoses of a resident and sleep data from a resident’s baseline after admission. While this work focuses on one particular residential program, it demonstrates the validity of utilizing AI models to support admissions decisions in these settings and the approach can be generalized to other settings, assuming data from successful and unsuccessful placements of past residents is available.