Identifying Children at Risk for Maltreatment Using Emergency Medical Services' Data: An Exploratory Study

Area of Science:

• Public Health
• Emergency Medicine
• Data Science

Background:

• Child maltreatment poses a significant public health concern.
• Identifying at-risk populations early is crucial for intervention.
• Emergency Medical Services (EMS) encounters provide a unique data source.

Purpose of the Study:

• To utilize natural language processing (NLP) to analyze EMS electronic health records (EHRs).
• To identify specific variables within EMS EHRs associated with child maltreatment.
• To test the hypothesis that EMS encounters correlate with the risk of child maltreatment reports.

Main Methods:

• Retrospective cohort study of pediatric EMS encounters (2011-2018).
• NLP applied to EMS EHRs to extract single words, bigrams, and trigrams.
• Probabilistic linkage of EMS encounters to child maltreatment reports.
• Univariable and multivariable logistic regression analyses to identify predictors.

Main Results:

• Eleven variables demonstrated significant associations with child maltreatment reports in multivariable modeling.
• Positive associations included: sexual abuse, chronic conditions, developmental delay, unconsciousness on arrival, police involvement, exposure, and age <2 years.
• Negative associations included: refusal of care and death on arrival (DOA/PEA/asystole).

Conclusions:

• EMS EHRs contain identifiable risk factors for child maltreatment.
• NLP is an effective tool for extracting this information from unstructured clinical notes.
• Future work could involve developing screening tools for high-risk households based on EMS data.