the Periodic Risk Evaluation: A Tool to Identify Risk for Unwanted Outcomes Among Adults with Autism

Background: Clinicians and providers who deliver services have a limited array of valid and reliable assessment tools to assess service needs of adults with autism. The lack of tools is especially concerning with respect to identifying those with the most complex needs that, if unidentified and unaddressed, escalate to crises such as hospitalization, homelessness, and police interactions. The Periodic Risk Evaluation (PRE) was designed to detect adults with autism having complex needs in order to flag them for additional services and intervention.

Objectives: This presentation will define the process used to design the PRE and demonstrate initial results, including sensitivity and specificity, of the PRE in identifying adults with autism with complex needs.

Methods: The PRE was established by clinical experts who identified conceptual clinical domains predicting characteristics of adults with autism who were complex cases in crisis, and/or need of additional services and support. Domains were organized into discrete question areas with scoring criteria and feedback from service providers on PRE content and design was solicited. The PRE was implemented by two Medicaid-funded programs administered by the Bureau of Autism Services in Pennsylvania, the Adult Autism Waiver and Adult Community Autism Program, which included 794 adults over age 21 with autism. The PRE questionnaire was completed by 530 individuals from January-July of 2017. ROC curves were used to determine optimal scoring cutoffs for identifying complex cases in a training sample and tested in a single validation sample. Analyses utilizing random forests to a best possible subset of classifying items are underway.

Results: The PRE was completed on a training sample of 86 adults with ASD. Gold-standard outcome designation (complex or non-complex service needs) was completed by the Clinical Team at the Bureau of Autism Services, who operate with individual-level knowledge of each program participant and were blind to PRE score. An ROC curve indicated a score of 13 yielded a sensitivity of 86% and a specificity of 84.8%. Using the initial cut point of 13 in a validation sample of n=530, the PRE with the initial cut point of 13, yielded sensitivity of 54.2%, specificity of 86.8%, positive predictive value of 43.3%, and negative predictive value of 91.1%. Additional analysis with approximately 300 additional participants and added time points employing random forest models with cross-validation are underway to determine if there is a better performing classification approach.

Conclusions: The PRE represents the first effort to establish a tool for adults with autism to identify individuals with complex service needs. The relatively large sample size in a government-funded program provides a natural laboratory to develop, test, and implement the PRE. Once optimal scoring for the PRE is established, testing of the PRE in additional programs, including those not funded by Medicaid, and expanding the age range are immediate priorities and would support generalization of the tool.