An Evaluation of Child Maltreatment Coding within a Large Health Care System in Rural Pennsylvania

Background: Over 500,000 children were abused or neglected in the United States in 2023, and over 2,000 children died from abuse-related injuries. A decline in child abuse reporting was observed both nationally and in Pennsylvania during the COVID-19 pandemic, likely due to limited interactions between victims and mandated reporters including health care providers. Inconsistencies have been observed when comparing prevalence rates of child maltreatment obtained through electronic health record (EHR) documentation to those obtained from child welfare data. We aimed to determine whether the prevalence of child abuse coded within EHRs mirrored the national underreporting trends observed during the COVID-19 pandemic and to assess the reliability of EHR documentation of child maltreatment for public health surveillance. Methods: Medical records for pediatric patients were extracted from a large, integrated health care system. Fourteen child maltreatment diagnosis codes were used to identify EHR encounters involving abuse in the emergency department, primary care, well-child, specialist, and hospital settings prior to and during the COVID-19 pandemic. Over 5,300 patients were identified with more than 7,600 encounters since Jan. 1, 2000. Systematic chart reviews were conducted in representative samples to determine the validity of child abuse coding for each encounter. The Pennsylvania Department of Human Services reports child abuse and market share data and estimates of the percent of visits affiliated with this health care system per county were obtained to approximate the number of child abuse reports expected from health care providers within the system. Values were compared to the actual number of cases documented with maltreatment-associated diagnosis codes in the EHRs. Results: Preliminary analyses identified trends inconsistent with the anticipated declines in reporting during the pandemic period. Three potential sources of data bias were investigated via targeted review of 83 charts: the use of adult abuse codes for pediatric patients, the overuse of a single sexual abuse code, and the repeating of certain abuse codes over time. This chart review suggested that overlap in the coding criteria for pediatric and adult patients and coding error contributes to the first source of bias. Outside laboratory testing and lab-specific protocol for coding pediatric sexual abuse were responsible for the second source of bias. Of charts reviewed with repeated abuse codes, less than 5% contained evidence of new or continued maltreatment. Just over 50 substantiated reports of child maltreatment were expected from system-affiliated health care providers. Approximately 96 new cases of child maltreatment were documented in EHRs using maltreatment-associated diagnosis codes. Conclusion: Documentation of child maltreatment within the EHR did not mirror trends observed nationally during the COVID-19 pandemic. While EHR data may improve monitoring of health care utilization and aid in the development of primary prevention strategies, epidemiological investigations are limited by potentially inflationary or inconsistent coding practices. Data biases that decrease reliability include overlap in pediatric and adult coding, laboratory-specific coding protocol, and inappropriate coding of follow-up visits. As an adjunct to diagnosis codes, EHR-based surveillance may be improved by artificial intelligence-based algorithms programmed to detect or predict pediatric abuse from free text within charts.