Identifying injuries suggestive of child physical abuse: An innovative application of the Trauma Quality Improvement Program

Abstract

BACKGROUND Child physical abuse (CPA) carries high risk of morbidity and mortality. Screening for CPA may be limited by subjective risk criteria and racial and socioeconomic biases. This study derived, validated, and compared age-stratified International Classification of Diseases, 10th revision (ICD-10) diagnosis codes indicating high risk of CPA. METHODS Injured children younger than 6 years from the Trauma Quality Improvement Program (TQIP) database were included; years 2017 to 2018 were used for derivation and 2019 for validation. Confirmed CPA was defined as a report of abuse plus discharge with alternate caregiver. Patients were classified as high vs. low CPA risk by three methods: (1) abuse-specific ICD-10 codes, (2) previously validated high-risk ICD-9 codes crosswalked to equivalent ICD-10 codes, and (3) empirically-derived ICD-10 codes from TQIP. These methods were compared with respect to sensitivity, specificity, area under the receiver-operator curve (AUROC), and uniformity across race and insurance strata. RESULTS A total of 122,867 children were included (81,347 derivation cohort, 41,520 validation cohort). Age-stratified high-risk diagnoses derived from TQIP consisted of 40 unique codes for ages 0 year to 2 years, 30 codes for ages 3 years to 4 years, and 20 codes for ages 5 years to 6 years. In the validation cohort, 890 children (2.1%) had confirmed CPA. On comparison with abuse-specific and crosswalked ICD-9 codes, TQIP-derived codes had the highest sensitivity (70% vs. 19% vs. 54%) and the highest AUROC (0.74 vs. 0.59 vs. 0.68, p < 0.0001) for confirmed abuse across all age groups. Age-based risk stratification using TQIP-derived codes demonstrated low variability by race (25% White vs. 25% Hispanic vs. 28% Black patients considered high-risk) and insurance status (23% privately insured vs. 26% uninsured). CONCLUSION High-risk CPA injury codes empirically derived from TQIP produced the best diagnostic characteristics and minimized some disparities. This approach, while requiring further validation, has the potential to improve CPA injury surveillance and decrease bias in screening protocols. LEVEL OF EVIDENCE Diagnostic Test/Criteria; Level III.