Decoding mental health crises in youth with intellectual and developmental disabilities through ehr network models and genomics: the mhc-idd study

Approximately 7 million youth in the US have intellectual and developmental disabilities (IDDs) and about 10% of youth with IDDs are admitted to hospitals due to a mental health crisis (MHC) each year2. In preadolescents with IDDs, elopement (88%) and self-injury (81%) are the most frequent behaviors that contribute to crisis whereas physical (60%) and verbal (42%) aggression are more common in adolescents. With insufficient community services to support youth with IDDs and less than half of US mental health facilities providing services for them, emergency departments (EDs) have boarding rates for children with IDDs 2-3 times higher than peers without IDDs. While psychiatric comorbidities are common in youth with IDDs, not all youth with IDDs and psychiatric comorbidities have MHCs. There is a critical need to understand which biopsychosocial risk factors beyond psychiatric comorbidities put youth with IDDs at risk for MHCs so that at-risk patients can be identified early and improved models of psychiatric care can be developed to help mitigate this national crisis. As a first step, we must demonstrate that including youth with IDDs and MHCs in clinical genomics research is feasible. First, we are developing a neural network model that can accurately classify youth with IDDs by psychiatric outcomes. I am extracting demographic, clinical, socioeconomic, and service use data from Rady Children's Hospital San Diego (RCHSD) EHR for ∼2000 youth with IDDs who have presented to the ED for MHCs (CRISIS) over the past decade. We are comparing them to two groups of youth from the RCHSD EHR: 1) ∼4000 youth with IDDs and psychiatric comorbidities but no history of MHCs (PSYCH) and 2) ∼4000 youth with IDDs without any psychiatric history. We are building a machine learning-based neural network model to predict psychiatric outcomes for youth with IDDs and then validate the classification accuracy using 10-fold cross validation as well as on a dataset of 150 prospectively recruited youth with IDDs and MHCs. Second, we are performing genome sequencing (GS) in 150 prospectively recruited youth with MHCs and IDDs. We will use an unmatched case-control study design comparing the effect sizes of neuropsychiatric PGS between CRISIS, from two groups of youth with IDDs from the Simons Simplex Collection: 1) ∼500 youth with IDDs and severe psychiatric comorbidities and 2) ∼1000 youth with IDDs and minimal or no psychiatric comorbidities. Lastly, we will assess the burden of pathogenic rare variants in youth with IDDs and MHCs from GS. To our knowledge, this is the first EHR/genomics study to recruit patients with IDDs during an MHC directly from an emergency department setting. Although there are significant challenges, including patient identification, parental consent, and safe sample collection, we demonstrate that research on this historically excluded population is feasible with proper planning. A multi-disciplinary team-based, approach allows for successful inclusion of these high acuity patients in clinical research. Preliminary GS results from the first five enrolled patients demonstrate feasibility of GS in an acute clinical setting. The construction of the machine learning-based neural network model to predict psychiatric outcomes for youth with IDDs is ongoing. Severely affected youth with IDDs can be successfully included in clinical genomics research studies even when presenting in an acute mental health crisis. Proper multi-disciplinary planning is essential.