P3A157: Identification of Child-maltreatment-related Emergency Department Visits from Electronic Health Records

Background: The prevalence of child abuse in the US has been mainly determined by studies using child welfare data or victim reports, leading to wide-ranging prevalence estimates. Recently, the International Classification of Diseases and Related Health Problems, 10th Revision, Clinical Modification (ICD-10-CM) included for the first time codes for both confirmed and suspected child maltreatment. However, its utility in capturing accurate prevalence rates of child abuse has not been well-studied. The purpose of this study was to compare search strategies using ICD 10 codes, keywords, and chief complaints in the electronic medical records to 1) determine rates of child maltreatment concerns during ED visits to children’s hospitals in Connecticut; and 2) determine sensitivity and specificity of different search strategies for child maltreatment.

Methods: We conducted this study at Yale-New Haven Children’s Hospital (YNHCH) and Connecticut Children’s Medical Center (CCMC). Institutional Review Boards at each location approved this study. We examined ED visits related to confirmed or suspected child maltreatment from January 1, 2019, to August 31, 2020, for children and adolescents younger than 18 years old. We identified cases by a pre-determined set of inclusion/exclusion criteria based on: (1) chief complaints, (2) keywords in provider notes, and (3) ICD-10-CM codes (Table 1). Cases that met one or more of these three criteria were included in the study sample set for chart review. Clinical teams at each site performed a chart review of cases using ED provider note(s) from each encounter of interest. Cases were categorized as either confirmed/suspected maltreatment or unrelated to maltreatment. Sensitivity and specificity to identify maltreatment were calculated for ICD-10, keywords in the EHR, and chief complaints. Categorization after chart review was used as the standard for true positive and true negative. The sample identified through all search strategies combined was used as our full population. Therefore, our estimates for specificity are low compared to the overall sensitivity of identifying cases from the full ED population.

Results: We reviewed 3841 ED records; 1248 were identified as confirmed/suspected maltreatment. Overall, using ICD codes alone had a 47% sensitivity and 65% specificity for identifying maltreatment cases; provider notes (65%) and chief complaints (55%) had higher sensitivities than ICD codes. When ICD codes were combined with provider notes and chief complaints, the sensitivity was 82% and 77%, respectively. Using keywords in provider notes and chief complaints had the highest sensitivity (91%) and lowest specificity (28%; Table 2).

Conclusion: The use of ICD 10 codes alone may miss 53% of maltreatment cases. ED notes and chief complaints have a better sensitivity (91%) to identify maltreatment cases, although the relatively low specificity may require confirmatory chart review or alternative methods. Future studies should focus on using these search strategies to determine the prevalence of child maltreatment.