Emergency Department Child Abuse Evaluations During COVID-19: A Multicenter Study

We conducted a multicenter retrospective study using data from the PECARN Registry from January 1, 2018 through March 31, 2021.³⁴  PECARN Registry participation was approved by the institutional review boards of all study sites and the Data Coordinating Center.

The PECARN Registry comprises electronic medical health record data from every ED patient encounter at participating institutions, harmonized into a deidentified, central repository.³⁴  Variables include standardized demographics (race, ethnicity, age, sex, insurance type), clinical care orders, laboratory and radiology results, International Classification of Diseases, 10^th Revision, Clinical Modification (ICD-10-CM) coded diagnoses and dispositions. This study included 9 participating institutions.

Three methods were used to identify ED encounters related to child physical abuse: (1) diagnosis codes: child abuse diagnosis by at least 1 ICD-10-CM code indicative of suspected or confirmed abuse (Supplemental Table 4), (2) injury codes: an ICD-10-CM code for injury associated with a diagnosis of abuse in certain young age groups (Supplemental Table 5), (3) skeletal survey cohort: evaluation of potential child physical abuse by skeletal survey radiographic study in children <24 months of age.^35–38  Visits identified by the 3 methods were treated as separate subgroups in analyses; an encounter could be included in >1 subgroup.

For the injury code cohort, International Classification of Diseases, 9^th Revision, Clinical Modification injury codes associated with high specificity for child physical abuse were identified from the literature and converted to ICD-10-CM codes by using general equivalence mapping.³⁵  Using a consensus process, 3 authors (Drs Chaiyachati, Wood, and Lindberg) reviewed the ICD-10-CM codes to ensure they represented injuries associated with a high risk of abuse. To increase specificity, exclusionary ICD-10-CM codes representing motor vehicle crashes, birth injuries, metabolic bone diseases, bleeding disorders, or follow-up visits were applied (Supplemental Table 6).³⁹ 

Demographic characteristics included age at encounter, insurance type (private, Medicaid, self-pay), and composite race and ethnicity. Race and ethnicity were hierarchically categorized by ethnicity (Hispanic or non-Hispanic) and then by race within the non-Hispanic group. Due to limited sample sizes, non-Hispanic racial categories were analyzed as Black, White, and additional racial group (including American Indian or Alaska Native, Asian, Native Hawaiian or Other Pacific Islander, Multiple Races, and other). Race and ethnicity were included in analyses due to documented differential COVID-19 pandemic experiences by groups.⁴⁰  In addition, the likelihood to seek emergency health care and for clinicians to consider abuse may differ by racial and ethnic group.^41–43  Exposure to pandemic-related changes was defined by date: before the pandemic (January 1, 2018–March 31, 2020) or during the pandemic (April 1, 2020–March 31, 2021).

Clinical severity was assessed by injury severity measures and clinical disposition. Injury severity measures included triage acuity defined by emergency severity index (ESI), injury severity scores, and disposition.⁴⁴  Injury severity scores were tabulated from ICD-10-CM codes with 2 methods: maximum abbreviated injury score (MAIS; 0: none, 1: minor, 2: moderate, 3: serious, 4: severe, 5: critical, 6: maximum; categories analyzed as 0–2, 3–6) and injury severity scale (ISS; range 0–75, the sum of squares of highest AIS scores for 3 most severely injured body regions; categories analyzed as 0–8, 9–15, >16).^45–48  Clinical disposition included ED visit disposition of discharged, admitted, transferred, or died and ED or hospital disposition vital status of alive or deceased.

Average daily rates of child abuse encounters and all PECARN Registry encounters were calculated for each calendar month and pandemic time period and plotted over time to visualize absolute and relative temporal patterns.

We described the number and rate per day of child abuse encounters overall as well as for each demographic and clinical characteristic of interest. For each characteristic, a “partially adjusted” multivariable Poisson regression model including an interaction between pandemic time period and the characteristic was fit to estimate rate ratios comparing pandemic rates to prepandemic rates while controlling for calendar month and clinical site for each of the 3 cohorts. The fully adjusted daily encounter rate models were generated with age, sex, race/ethnicity, primary payer, calendar month, site, ISS as severity measure, and an interaction term between ISS and pandemic time period, as well as interaction terms between pandemic time period and demographic covariates if the test for the interaction in the partially adjusted model had P <.10. Fully adjusted rate ratios and 99% confidence intervals (CI) were estimated for overall encounters and each severity level by using marginal estimation methods to average over other model covariates.³⁴ 

Because of strong correlations between clinical severity measures, separate, fully adjusted models were fit for each measure of clinical severity (MAIS, triage ESI level, ED disposition, and vital status) with covariates as listed above.

Because of the large sample size, we used a significance level of 0.10 for tests of interactions and 0.01 for all other tests of statistical significance. No adjustments were made for multiple comparisons. Encounters with missing data were excluded. We performed all analyses using SAS/STAT software version 9.4 (SAS Institute Inc., Cary, NC, USA).

After the application of exclusion criteria, a total of 1 579 014 ED encounters were identified in the PECARN Registry, including 10 270 (0.7%) as possible child physical abuse (Fig 1). Because of missing data, 315 (3.1%) of the abuse encounters were excluded.

Encounter rates as identified by child abuse diagnosis codes (diagnosis code cohort) decreased from 4.6 encounters per day prepandemic to 3.8 encounters per day during the pandemic, with the month and site-adjusted (“partially adjusted”) rate ratio (RR) of 0.82 (99% CI: 0.73–0.92; P <.001, Table 1). Differences were noted by age in partially adjusted analysis with decreased encounter rates for children aged 2 to <6 years (RR: 0.70 [CI: 0.58–0.85], P <.001) and 6 to <13 years (RR: 0.79 [CI: 0.64–0.98], P = .004), but no significant change for children <2 or >13 years of age. There was also a decrease for children reported as Black non-Hispanic (RR: 0.78 [CI: 0.64–0.96], P = .002) and White non-Hispanic (RR: 0.80 [CI: 0.65–0.99], P = .006) with no change for children recorded as Hispanic. Encounters with public insurance decreased significantly (RR: 0.81 [CI: 0.70–0.93], P <.001) without changes in rates for self-pay or private insurance encounters. Encounter trends were also variable by site with 2 of 9 institutions having significantly reduced daily diagnosis rates. After full adjustment with marginal estimation to average over model covariates, including ISS, the overall reduction in encounter rate during the pandemic in the diagnosis-code cohort was 19% (adjusted rate ratio [ARR]: 0.81 [CI: 0.75–0.88], P <.001).

Encounter rates for high-risk injuries (injury code cohort) also had a statistically significant reduction by partial adjustment (3.28 to 2.95 encounters per day, RR: 0.90 [CI: 0.81–0.99], P = .007, Table 2). There were no significant differences by specific demographic characteristic categories between the prepandemic and pandemic periods. After full adjustment, the reduction in encounter rate was 10% (ARR: 0.90 [CI: 0.82–0.98], P = .002).

Encounter rates for child physical abuse evaluation by skeletal survey completion (skeletal survey cohort) did not have a statistically significant reduction by partial adjustment (3.8 to 3.5 encounters per day, RR: 0.92 [CI: 0.84–1.01], P = .03, Table 3). There were no significant differences in demographic factors between the prepandemic and pandemic periods. After full adjustment, there was not a significant reduction in encounter rate for the skeletal survey cohort (ARR: 0.92 [CI: 0.84–1.00], P = .013).

For all methods of identifying suspected child abuse, the magnitude of reduction for abuse encounter rates was smaller than the decrease in total encounters with the most marked reduction in abuse encounter rates during the first months of the pandemic (Fig 2).

In the diagnosis-code cohort, low-severity presentations by ISS (0–8) were reduced in partially adjusted analysis (RR: 0.80 [CI: 0.72–0.90], P <.001), whereas higher ISS category presentations did not change significantly (ISS 9–15: P = .42, ISS 16+: P = .93, Table 1). After full adjustment, including interaction with ISS, reduction in low-severity ISS presentation was 20% (ISS: 0–8; ARR: 0.80 [CI: 0.73–0.87], P <.001). Other independently modeled clinical severity measures revealed similar reductions in low-severity encounters in the diagnosis code cohort, including MAIS (0–2, ARR: 0.80 [CI: 0.73–0.88], P <.001), ED discharges (ARR: 0.78 [CI: 0.70–0.87], P <.001), and hospital survival (ARR: 0.80 [CI: 0.74–0.88], P <.001). As with ISS, there were no significant changes in the rate for encounters with higher severity for each measure (Table 1).

Patterns for clinical severity were consistent in diagnosis code, injury code, and skeletal survey cohorts. After full adjustment, reductions in low-severity presentations by ISS were significant for both the injury code cohort (ARR: 0.85 [CI: 0.75–0.96], P <.001; Table 2) and the skeletal survey cohort (ARR: 0.90 [CI: 0.81–0.99], P = .006; Table 3); however, there were no detected changes in daily encounter rates with higher ISS scores. A similar reduction in low-severity encounters was observed when modeling low MAIS (injury code cohort: ARR: 0.85 [CI: 0.75–0.96], P <.001; skeletal survey cohort: ARR: 0.90 [CI: 0.81–0.99], P = .006) and hospital survival (injury code cohort: ARR: 0.89 [CI: 0.80–0.98], P = .002; skeletal survey cohort: ARR: 0.91 [CI: 0.83–1.0], P = .007) without changes in higher-severity encounter rates. ED discharges also decreased during the pandemic period in the injury code cohort (ARR: 0.86 [CI: 0.75–0.99], P = .005); there was no detected change in ED discharges when identified by skeletal survey. There were no changes in encounters with severe disposition (ie, admitted/transferred/died) for either the injury code or skeletal survey code cohort.

Our results reveal that pediatric ED encounters concerning child physical abuse by ICD-10-CM codes decreased by 19% during the pandemic when assessed across all ages within a multicenter pediatric ED data registry. Rates of encounters among children <2 years old with high-risk injuries were reduced by 10%, whereas children with a potential concern for abuse as indicated by skeletal survey completion did not have a significant reduction. The decrease in high-risk injury identification, but not in rates of skeletal survey, implies that decreases were not due to decreased likelihood of clinicians to evaluate or identify abuse. Additionally, our data support our hypothesis regarding clinical severity and presentation to medical care: encounter rates with lower clinical severity decreased during the pandemic whereas encounter rates with higher clinical severity were unchanged. This was largely consistent across identification methods and measures of severity.

Although reduced or stable emergency health care encounters related to child physical abuse may be reassuring, critical assessment is required to further understand and contextualize these results. With regard to reduced low-severity clinical encounters without change in high-severity clinical encounters, these results can be interpreted in at least 2 ways. First, if a reduction in low-severity clinical encounters is driven by decreased recognition whereas high-severity encounters continue to present for care due to medical need, our results would suggest that the actual population rate of abuse may not have decreased during the pandemic. Moreover, if any proportion of severe cases were undetected within medical care related to pandemic shifts, these results may suggest that overall abuse increased. A second interpretation is that the reduction in encounter rates represents a true decrease in the population rate of child physical abuse. A true decrease could result from novel protective factors within the pandemic, such as having more caregivers at home, for instance, older children engaged in virtual schooling or parents who became unemployed or remained home related to social distancing.²⁸  To explain the specific reduction within the lower-severity cases of abuse, the presence of additional caregivers would have to reduce children’s risk of less-severe physical abuse while failing to impact higher-severity occurrences. Our results are unable to distinguish between these interpretations, and additional delineation will require additional, multimodal data. These findings add information to previous studies which did not reveal differences in severity of child abuse by several markers in the medical record.^27,28  Our results may be related to using more sensitive assessments of clinical severity.

Second, these results suggest that children experienced different risks during the COVID-19 pandemic-related to their age. Children <2 years of age had a smaller reduction in the rate of emergency care for concerns for physical abuse. Specifically, the significant reductions in the diagnosis code cohort were concentrated among preschool (2 to <6 years; 30% reduction) and school-aged (6 to <13 years; 20% reduction) children, whereas the reduction among youngest children (<6 months) was only 5% to 10% across identification methods. This pattern supports the importance of school encounters for the recognition of child physical abuse and may be consistent with the importance of mandatory reporter or alternative caregiver exposures for detection in preschool-aged children as well.⁴⁹  Several assessments support that a driving force for decreased reports to CPS was decreased attendance at in-person school.^11,30,33  Similarly, decreased in-person health care use may contribute to the reduced diagnosis of physical abuse.^41,50–54  We included race and ethnicity in our models, recognizing that these social constructs may associate with the differential impact of the pandemic and evaluation patterns for child abuse.^40–43  Our findings indicate that there was a differential reduction by race and ethnicity within the diagnosis code cohort, but not for other cohorts.

Third, our results reveal institutional variability. Specifically, in the diagnosis code cohort, 2 of 9 sites had significant reductions whereas 2 other sites revealed nonsignificant trends toward increases. Variability by site is consistent with previous literature that includes multiple single-site reports revealing increases in child maltreatment and emphasizes the need for contextualization of results by granularity of inquiry.^22–26  Local policy and practice such as virtual versus in-person school may influence variability by site.^11,30,33  Thus, aggregate assessment across institutions is valuable for trends, yet may obscure significant variation evident at higher granularity, providing 1 potential unifying theory for the existing variability in the literature.

Two assumptions are worth reviewing to contextualize our results. First, the identification of child physical abuse-related encounters within EDs assumes this is a relevant way to understand harm experienced by children in retrospective study designs. Diagnosis codes do not represent a reference standard for diagnosis of abuse in isolation yet are frequently used to measure population trends, and, although skeletal surveys are predominantly performed in EDs related to physical abuse evaluation, other medical indications are possible.^55,56  Thus, differences across 3 encounter identification methods may be reflective of variable sensitivity and specificity. For example, using ICD-10-CM child abuse codes seems the most restrictive by the number of encounters identified. This is consistent with previous studies that have suggested low sensitivity for health care-assigned child abuse diagnosis codes and reiterates the importance of complementary encounter identification schema.^38,57  Second, the comparison of prepandemic to pandemic periods relies on the assumption that hospital catchment with regard to child physical abuse evaluation and diagnosis was not specifically impacted by the COVID-19 pandemic. For example, this includes the assumption that transfer patterns to participating institutions were not impacted by pandemic-related behavior changes.

Our findings are also subject to at least 2 limitations. First, this analysis focused on physical abuse and does not provide insight into neglect, sexual abuse, or other forms of child maltreatment, although it is likely that some cases of sexual abuse were included in general abuse diagnosis codes (eg, “child maltreatment” [T76.92XA]). Second, the collection of patients and hospitals within the PECARN Registry may not be representative of the community or other care settings. Results require replication and expansion in other datasets to clarify generalizability.

In conclusion, our results suggest pediatric emergency medical encounters related to physical abuse were reduced during the COVID-19 pandemic compared with previous years. In addition, these reductions were driven by decreases in lower-severity encounters and among preschool and school-aged children. Although a reduction could be reassuring, there was no evidence of a reduction in the rates of higher-severity injuries. This pattern calls for additional critical assessment to clarify the role of decreased recognition and the associated potential for unrecognized harm experienced by children during the COVID-19 pandemic versus true reduction related to novel protective factors.²¹ 

Members of the PECARN Registry Study Group and PECARN Child Abuse Special Interest Group include: Kathleen M. Adelgais, MD MPH, Section of Pediatric Emergency Medicine, Department of Pediatrics, University of Colorado School of Medicine, Aurora, Colorado; Lynn Babcock, MD, MS, Department of Pediatrics, University of Cincinnati, Cincinnati Children's Hospital Medical Center, Cincinnati, Ohio; James M. Chamberlain, MD, Children's National Hospital, George Washington University School of Medicine and Health Sciences, Washington, District of Columbia; Susan Duffy, MD, MPH, Division of Pediatric Emergency Medicine, Departments of Emergency Medicine and Pediatrics, Hasbro Children's Hospital, Warren Alpert Medical School of Medicine at Brown University, Providence, Rhode Island; Robert W. Grundmeier, MD, Division of General Pediatrics, Department of Pediatrics, Children's Hospital of Philadelphia, Perelman School of Medicine, and the Center for Center for Pediatric Clinical Effectiveness, Children's Hospital of Philadelphia, Philadelphia, Pennsylvania; Sadiqa Kendi, MD, Division of Pediatric Emergency Medicine, Boston Medical Center, Boston University School of Medicine, Boston, Massachusetts; E. Brooke Lerner, PhD, Department of Emergency Medicine, University at Buffalo, Buffalo, New York; Julia N. Magana, MD, Department of Emergency Medicine, University of California Davis Health, University of California Davis School of Medicine, Sacramento, California; Prashant Mahajan, MD, MPH, MBA, Department of Emergency Medicine, University of Michigan, Ann Arbor, Michigan; Stephanie M. Ruest, MD, MPH, Division of Pediatric Emergency Medicine, Departments of Emergency Medicine and Pediatrics, Hasbro Children's Hospital, Warren Alpert Medical School of Medicine at Brown University, Providence, Rhode Island; Bashar S. Shihabuddin, MD, MS, Section of Emergency Medicine, Division of Pediatrics, Nationwide Children's Hospital, The Ohio State University College of Medicine, Columbus, Ohio; Norma-Jean E Simon, MPH, MPA, Division of Emergency Medicine, Department of Pediatrics, Lurie Children's Hospital of Chicago, Northwestern University Feinberg School of Medicine, Chicago, Illinois; Asha Tharayil, MD, Division of Pediatrics, Emergency Medicine, UT Southwestern Medical Center, Dallas, Texas; Danny G. Thomas, MD, MPH, Section of Pediatric Emergency Medicine, Medical College of Wisconsin, Milwaukee, Wisconsin; and Joseph J. Zorc, MD, MSCE, Division of Emergency Medicine, Department of Pediatrics, Children's Hospital of Philadelphia, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania.

Additionally, we wish to acknowledge Cara Elsholz, BS, and Cody Olsen, MS, for their contributions to project planning and organization.

ARR

adjusted rate ratio

CI

99% confidence interval

COVID-19

coronavirus disease 2019

CPS

child protective services

ED

emergency department

ESI

emergency severity index

ICD-10-CM

International Classification of Diseases, 10^th Revision, Clinical Modification

ISS

injury severity scale

MAIS

maximum abbreviated injury score

PECARN

Pediatric Emergency Care Applied Research Network

RR

rate ratio