Using machine learning to identify features associated with different types of self-injurious behaviors in autistic youth.

BACKGROUND

Self-injurious behaviors (SIB) are common in autistic people. SIB is mainly studied as a broad category, rather than by specific SIB types. We aimed to determine associations of distinct SIB types with common psychiatric, emotional, medical, and socio-demographic factors.

METHODS

Participants included 323 autistic youth (~50% non-/minimally-speaking) with high-confidence autism diagnoses ages 4-21 years. Data were collected by the Autism Inpatient Collection during admission to a specialized psychiatric inpatient unit (www.sfari.org/resource/autism-inpatient-collection/). Caregivers completed questionnaires about their child, including SIB type and severity. The youth completed assessments with clinicians. Elastic net regressions identified associations between SIB types and factors.

RESULTS

No single factor relates to all SIB types. SIB types have unique sets of associations. Consistent with previous work, more repetitive motor movements and lower adaptive skills are associated with most types of SIB; female sex is associated with hair/skin pulling and self-rubbing/scratching. More attention-deficit/hyperactivity disorder symptoms are associated with self-rubbing/scratching, skin picking, hair/skin pulling, and inserts finger/object. Inserts finger/object has the most medical condition associations. Self-hitting against surface/object has the most emotion dysregulation associations.

CONCLUSIONS

Specific SIB types have unique sets of associations. Future work can develop clinical likelihood scores for specific SIB types in inpatient settings, which can be tested with large community samples. Current approaches for SIB focus on the behavior functions, but there is an opportunity to further develop interventions by considering the specific SIB type in assessment and treatment. Identifying factors associated with specific SIB types may aid with screening, prevention, and treatment of these often-impairing behaviors.