Disparities in Detection of Suspected Child Abuse

Abstract

Background:

Child abuse is a significant cause of injury and death among children, but accurate identification is often challenging. This study aims to assess whether racial disparities exist in the identification of child abuse.

Methods:

The 2010-2014 and 2016-2017 National Trauma Data Bank was queried for trauma patients ages 1-17. Using ICD-9CM and ICD-10CM codes, children with injuries consistent with child abuse were identified and analyzed by race.

Results:

Between 2010-2014 and 2016-2017, 798,353 patients were included in NTDB. Suspected child abuse victims (SCA) accounted for 7903 (1%) patients. Of these, 51% were White, 33% Black, 1% Asian, 0.3% Native Hawaiian/Other Pacific Islander, 2% American Indian, and 12% other race. Black patients were disproportionately overrepresented, composing 12% of the US population, but 33% of SCA patients (p<0.001). Although White SCA patients were more severely injured (ISS 16-24: 20% vs 16%, p<0.01) and had higher in-hospital mortality (9% vs. 6%, p=0.01), Black SCA patients were hospitalized longer (7.2±31.4 vs. 6.2±9.9 days, p<0.01) despite controlling for ISS (1-15: 4. 5.7±35.7 vs. 4.2±6.2 days, p<0.01). In multivariate regression, Black children continued to have longer lengths of stay despite controlling for ISS and insurance type.

Conclusions:

Utilizing a nationally representative dataset, Black children were disproportionately identified as potential victims of abuse. They were also subjected to longer hospitalizations, despite milder injuries. Further studies are needed to better understand the etiology of the observed trends and whether they reflect potential underlying unconscious or conscious biases of mandated reporters.

INTRODUCTION

In 2020, over 1700 children died of abuse and neglect in the United States at a rate of 2.38 per 100,000 children in the population with 43% of fatalities due to physical abuse.[1] Child abuse, however, is often underreported with the true prevalence likely higher.[2] Previous studies have suggested child abuse may be more prevalent among families of low socioeconomic status and among minority populations.[3–8] In addition, studies have also shown that minority children are disproportionately identified for child abuse evaluation and work-up.[9–14] Minority children also constitute a larger proportion of the child welfare system and have higher child abuse case substantiation rates than their White counterparts.[15,16] Whether race is an independent risk factor for child abuse or reflective of other socioeconomic factors (such as poverty) has been disputed widely in the literature.[15–24]

The inherent potential for subjectivity in evaluating child abuse creates additional complexity in studying the effect of race on child abuse. Consequently, findings have often been contradictory. Many studies which reported increased rates of child abuse among minority populations have largely been from single institutions or relied on regional data that is not representative of the general United States (US) population.[3,4,6,12,18,25] The four waves of the National Incidence Study of Child Abuse and Neglect (NIS), the largest federally-sponsored attempt to study child abuse incidence, have produced conflicting results.[26–29] Data from the first three waves suggested no differences in the incidence of child abuse among races; however, the incidence among races diverged in the fourth wave. [15]

To better assess child abuse identification patterns, we used a national registry of injured patients with data compiled using standardized data abstraction techniques. We sought to characterize the racial composition of cases identified as child abuse and determine whether certain races were disproportionately represented nationwide.

METHODS

This retrospective study used pediatric data from the National Trauma Data Bank (NTDB) available through the Trauma Quality Programs Participant Use File (PUF). The NTBD is the world’s largest trauma data repository with over 7 million records collected from more than 950 trauma centers in the United States. Patient information submitted to NTDB follows standardized guidelines and includes demographic data such as age, gender, race, as well as hospitalization and injury data (i.e. injury severity score (ISS), mortality, intensive care admission, insurance status, diagnosis codes, etc). [30]

We used data from 2010-2014 and 2016-2017 from NTDB for children ages 1-17 years. The year 2015 was excluded from analysis given International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9CM) was transitioning to ICD-10CM. During this transition period, coding was inconsistent. Patients lacking ages or year of birth as well as duplicate IDs were excluded. Cases of suspected or confirmed child abuse (SCA) were identified using ICD-9CM codes 995.50-995.55, 995.59, 967.0-967.9 and ICD-10CM codes T74.02XA, T74.1, T74.12XA, T74.22XA T74.32XA, T74.4XA, T74.9A, T74.92XA, T76.02XA, T76.1, T76.12XA, T76.22XA, T76.9A, T76.92XA, Y07-Y07.59, and Y07.9.

The primary outcome was to characterize racial demographic differences between the general pediatric trauma population and the suspected child abuse population. Racial demographics were also compared to the overall U.S. population using the 2010 U.S. Census Data. [31] Secondary analyses compared hospitalization characteristics by race, including injury severity score (ISS), hospital length of stay (LOS), intensive care unit (ICU) LOS, primary method of payment, overall mortality, emergency department (ED) mortality, and in-hospital mortality. ISS was categorized as mild/moderate (1-15), serious (16-24), and severe (>25). Further subgroup analysis was performed comparing White children with suspected abuse to Black children with suspected abuse because these two racial groups represented over 80% of the data. The remaining racial sub-groups were too small for statistically meaningful comparisons.

Patient characteristics and outcomes are described using frequencies and percentages for categorical variables. Categorical data were analyzed using Chi-Square analysis and ANOVA. Statistical significance was determined with p-value <0.05. Statistical analysis was performed using Python (Scotts Valley, CA).

This study was reviewed by the institutional review board of Stanford University and determined to be an exempt study.

RESULTS

We identified 798, 353 pediatric trauma incidents for patients 1-17 years old from the NTDB registry for the study years 2010-2014, 2016-2017. Among these patients, we identified 7903 (1%) incidents of child abuse.

Overall comparison of trauma patients and suspected child abuse patients

When compared with the trauma population, children with suspected abuse were younger overall (median age 10 years vs 2 years) and a smaller difference was observed between genders (59% male SCA patients vs 65% male trauma patients, p<0.01). Among pediatric trauma patients, the population was 66% White, 18% Black, 2% Asian, 0.3% Native Hawaiian/Other Pacific Islander, 1% American Indian, and 13% other race. The SCA group was 51% White, 33% Black, 1% Asian, 0.3% Native Hawaiian/Other Pacific Islander, 2% American Indian, and 12% other race (Figure 1). Hispanic or Latino patients comprised 19% of both SCA and injured pediatric trauma patients and 16% of the US population in 2010. According to the 2010 US census, 72% of the population were White, 13% Black, 5% Asian, 0.2% Native Hawaiian/Other Pacific Islander, and 0.9% American Indian. Black patients were disproportionately overrepresented in the SCA group, composing 33% of SCA patients compared with 18% of the NTDB pediatric population and 13% of the overall US population (p<0.001, Figure 1). Government insurance payor-type accounted for 77% of SCA patients and 43% of trauma patient (p<0.01, Table 1).

Fig. 1. Distribution of races among SCA and pediatric trauma patients compared to 2010 US population.

Black children constitute 12.6% of the United States population and 18% of the pediatric trauma population, however, represent 33% of SCA patients (p<0.001).

Table 1.

Patient demographics and characteristics of SCA patients and pediatric trauma patients in NTDB 2010-2014 and 2016-17.

	Number of SCA patients [%] Total n=7903	Number of Pediatric Trauma patients [%] Total n=790450	p-value	Odds Ratio
Gender			p<0.01
Female	3243 [41]	272906 [35]
Male	4656 [59]	517242 [65]
Age Group (years)			p<0.01
1-3yrs	5810 [74]	138526 [18]	p<0.01
4-7yrs	1050 [13]	171192 [22]	p<0.01
8-11yrs	294 [4]	139272 [18]	p<0.01
12-17yrs	749 [10]	341460 [43]	p<0.01
Median age (years)	2.0	10.0	p<0.01
Race			p<0.01
White	3843 [51]	493769 [66]	p<0.01
Black	2454 [33]	133070 [18]	p<0.01
Asian	82 [1]	14548 [2]	p<0.01
American Indian	158 [2]	7530[1]	p<0.01
Native Hawaiian or Other Pacific Islander	23 [0.3]	2387 [0.3]	p=0.94
Other Race	927 [12]	96688 [13]	p=0.17
Ethnicity			p=0.89
Hispanic or Latino	1298 [19]	125365 [19]
Not Hispanic or Latino	5406 [81]	524621 [88]
ISS			p<0.01
Mild/Moderate (1-15)	5394 [69]	688939 [88]	p<0.01
Serious (16-24)	1519 [19]	66244 [8]	p<0.01
Severe (>25)	949 [12]	30596 [4]	p<0.01
Median ISS	6.0	4.0
Payment			p<0.01
Government*	5662 [77]	310296 [43]	p<0.01
Private/Commercialǂ	1111 [15]	356341 [49]	p<0.01
Self-Pay§	563 [8]	54613 [8]	p=0.76
Mortality
Overall Mortality	757 [9.6]	11437 [1.5]	p<0.01	7.2 [6.7-7.8]
ED Mortality	127 [1.6]	5062 [0.65]	p<0.01	2.5 [2.1-3.0]
Hospital Mortality	630 [8.2]	6375 [0.84]	p<0.01	10.7 [9.8-11.6]
ICU Admission	2879 [68.1]	134824 [48.3]	p<0.01	2.3 [2.2-2.4]
Length of Stay (mean±SD days)
Hospital (days)	6.8±20	2.8±5.1	p<0.01
ICU	6.2±9.0	3.9±6.1	p<0.01

Abbreviations: SCA, suspected child abuse patients; SD, standard deviation; ISS, injury severity score; ED, emergency department; ICU, intensive care unit

^*Government payment- Medicare, Medicaid, other government

^ǂPrivate/Commercial payment- private/commercial and Blue Cross/Blue Shield

^§Uninsured- self pay

Overall, SCA patients had a significantly higher mean ISS (10.5±10.1 vs 6.9±7.7, p<0.01). SCA patients had more serious (ISS 16-24; 19% vs 8%, p<0.01) and severe (ISS ≥25; 12% vs. 4%, p<0.01) injuries. Approximately 68% of SCA patients were admitted to the intensive care unit compared with 48% of the general pediatric trauma population (OR 2.3; 95%CI 2.2-2.4) and had longer ICU stays (6.2±9.0 days among SCA vs. 3.8±6.1 days among trauma patients, p <0.01). SCA patients had statistically significant higher overall mortality compared to pediatric trauma patients (9.6% vs 1.5%, p<0.01), representing a seven-fold increase in mortality (OR 7.2; 95% CI 6.7-7.8). The odds ratio for in-hospital mortality was 10.7 (95% CI 9.8-11.6, p<0.01) among SCA patients compared with trauma patients, and OR 2.5 for ED-mortality (OR 2.5; 95% CI 2.1-3.0, p<0.01). (Table 1)

Suspected child abuse subgroup analysis

Patient characteristics

Of the 7903 incidents of child abuse identified from the NTDB registry, Black children comprised 2454 incidents, of which 60% were male. White children comprised 3843 incidents, of which 59% were male. (p=0.83, Table 2). While the majority of SCA patients were between the ages of 1-3 years old in both groups, a greater percentage of White SCA patients fell into the 1-3 year old age group (77% vs. 68%, p<0.01). Older children were more common among Black SCA patients compared with their White counterparts (Table 2). Eighty-one percent (81%) of Black SCA patients had government insurance compared with 74% of White SCA patients (p<0.01).

Table 2.

Demographics of 2010-2014, 2016-2017 NTDB child abuse patients

	White SCA Number of patients [%] Total n=3843	Black SCA Number of patients [%] Total n=2454	p-value
Gender			p=0.83
Female	1561 [41]	989 [40]
Male	2284 [59]	1466 [60]
Ethnicity
Hispanic or Latino	454 [14]	20 [1]	p<0.01
Not Hispanic or Latino	2856 [86]	2055 [99]
Age Group (years)			p<0.01
1-3	2968 [77]	1662 [67]	p<0.01
4-7	471 [12]	369 [15]	p<0.01
8-11	125 [3]	115 [4]	p<0.01
12-17	285 [7]	309 [12]	p<0.01
Median age (years)	2.0	2.0
ICU admission	1422 [69]	811 [63]	p<0.01
Mortality
ED Mortality	34 [1]	68 [3]	p<0.01
Hospital Mortality	337 [9]	149 [6]	p<0.01
Mortality	371 [9]	217 [9]	p=0.31
Payment
Government	2687.0 [74]	1840.0 [81]	p<0.01
Private/Commercial	666.0 [18]	260.0 [12]	p<0.01
Self-Pay	274.0 [8]	159.0 [7]	p=0.49
ISS			p<0.01
Minor/Moderate (1-15)	2575 [67]	1784 [73]	p<0.01
Serious (16-24)	784 [20]	384 [16]	p<0.01
Severe (>25)	477 [12]	268 [11]	p=0.095
Median ISS (iqr)
Length of Stay (mean±SD days)
ICU	5.8±8.0	6.2±8.0	p=0.12
Hospital	6.2±9.9	7.2±31.4	p<0.01
Hospital LOS by ISS (mean±SD days)
Mild/Moderate (1-15)	4.2±6.2	5.7±35.7	p<0.01
Serious (16-24)	9.5±12.5	9.8±11.1	p=0.05
Severe (>25)	12.0±16.3	13.5±17.7	p=0.11
Average ISS	1.1±2.3	1.5±3.5

Abbreviations: SCA, suspected child abuse patients; SD, standard deviation; LOS, length of stay; ICU, intensive care unit; ED, emergency department; ISS, injury severity score

Injury Severity

Black SCA patients predominantly presented with a mild-ISS (73%) compared to White SCA patients (67%, p< 0.01) (Table 2). White SCA patients were more likely to present with both serious (ISS 16-24; 20%) and severe ISS (ISS≥25; 12%) than Black SCA patients (16% and 11%, respectively, p<0.01). Overall, Black children had a lower mean ISS (9.7±10.2 vs 10.8±10.0, p < 0.01). White children were more likely to require an ICU admission than Black children were (69% vs 63%, p<0.01), however, there was no significant difference in ICU LOS (White: 6.2±9.9 vs Black: 7.2±31.4 days, p=0.12).

No differences were observed in overall mortality between Black and White patients (p=0.31). ED mortality was higher among Black SCA patients (3% vs. 1%, p<0.01) while hospital mortality was higher among White SCA patients (9% vs. 6%, p <0.01). Logistic regression was performed to determine predictors of mortality based on age, race, gender, insurance status and ISS. When controlling for these factors, race was not an independent predictor of mortality.

Black children experienced longer average hospital LOS (7.2±31.4 days) than White children (6.2±9.9 days) (p<0.01, Table 2). Among SCA patients with mild injuries (ISS 1-15), Black children on average spent 1.5 days longer in the hospital than their White counterparts (5.7±35.7 days among Black patients compared with 4.2±6.2 days among White children; p<0.01). Among severely injured SCA patients (ISS≶25), White children continued to have a shorter hospital average LOS however this was not significant (9.8±11.1 vs 13.5±17.7 days, p=0.11). A multiple linear regression modeling was constructed to predict length of stay based on age group, race, gender, insurance status, and ISS. After controlling for the effects of all other independent variables, we found that race remained a significant predictor of length of stay (p<0.01).

Univariate and Multivariate Analysis

In a univariate analysis, Black race was an independent variable for a predictor of child abuse. On a multivariate analysis controlling for payment type, hospital bed size, hospital type, and teaching status, race remained an independent variable for child abuse.

DISCUSSION

In this study using a nationally representative database of injured children, we observed that Black children were significantly overrepresented among suspected child abuse victims at a rate of 2.5 times their proportion in the overall US population. We found this trend to be persistent, even when controlling for socioeconomic factors using payer type. While Black SCA patients had less severe injuries overall (lower mean ISS), our study noted that Black SCA patients had longer hospital length of stay as compared to White SCA patients. This was particularly true among those patients with the mildest injuries. White SCA patients more often presented with serious and severe injuries, but their hospital stays were similar or shorter when compared to Black children with comparable injury severity.

Few studies have evaluated racial trends in child abuse identification using large, nationally representative dataset over time. The over-identification of children of color in cases of abuse has been supported in the literature, but continues to be debated whether race is a surrogate of SES. Our study found that when controlling for SES factors, Black children remained at greater odds of being identified as victims of abuse. Management of suspected child abuse patients also varied by race. When controlling for both SES and injury severity, we found Black children had longer lengths of stay compared to their White counterparts. The increased additional inpatient hospital evaluation among Black children was similarly demonstrated by Wood et al. [17]. In their retrospective study of infants who had sustained non-motor vehicle associated TBI at 39 pediatric hospitals, the authors found that when adjusting for age, type and severity of TBI, and other injuries, Black infants were more likely to receive a skeletal survey than White infants. Lane et al., [12] found similar results for increased rate of skeletal surveys among Black children when controlling for insurance status, likelihood of abuse determined by review from an independent expert, and appropriateness of performing skeletal survey. Despite these findings, neither study addressed the underlying disparities nor performed a detailed evaluation of hospital course for these children.

The overrepresentation of Black SCA patients in NTDB may be due, in part, to identification based on provider suspicion, but not necessarily verification. In 2018, Child Protective Services (CPS) agencies received over 4 million referrals alleging abuse of which 2.4 million were investigated.[32] In total, 16.8% of children were classified as victims of abuse with substantiated or indicated dispositions. Child abuse victims were predominantly White, composing approximately 45% of victims while 21% were Black. Despite this, Black children were 2-3 times more likely to be investigated by CPS than White children and twice as likely to be placed in the foster care system. [33–35] Within their study, Lane et al. [12] also found that Black children were more likely to be reported to CPS compared to White toddlers even after having an independent expert review cases for possible abuse and controlling for insurance. Unsubstantiated referrals – those where little to no evidence of child abuse was found – made up 56% of 2018 CPS referrals. [36] On the opposite spectrum, Jenny et al. and Escobar et al. [11,16] have demonstrated providers’ reluctance to report child abuse in children from affluent, white families. Such discrepancies in reporting can lead not only to disproportionately higher numbers of Black families being falsely suspected of child abuse, but can also place White children at risk for suffering continued, unidentified child abuse. [11] Standardization of processes for which cases are opened and investigated, may better elucidate risk factors for child abuse and decrease racial disproportionalities.

Our findings indicate that despite the severity of injuries that child abuse victims present with, Black SCA patients have extended hospitalizations compared to White SCA patients. This is most prominent among those who present with a mild (1-15) ISS. It is paradoxical that milder injuries resulted in longer average length of stay among Black children. The decision for hospital discharge can often be subjective, based on the providers’ assessment of a patient’s clinical stability and post-discharge needs. It is plausible that medical providers are more reluctant to release children home based on the provider’s assessment of a family’s socioeconomic status or perceived stability of home environment. However, we found LOS continued to be significantly longer with Black SCA patients even after controlling for insurance status as a proxy for socioeconomic status since income was not a metric that was available through NTDB. It is difficult to determine if an element of implicit bias is influencing providers’ decisions regarding hospital discharge. In healthcare, implicit bias has been suggested as one of the main factors to contributing to the stark differences in morbidity and mortality between White and Black patients. [37–39]. The limitations of the dataset do not allow us to infer whether implicit bias played a role.

Examining implicit bias in healthcare has been challenging. Medical providers make-up 10.5% of all mandated reporters who submit a referrals to CPS.[32] While steps have been taken to increase diversity in the medical profession as a way mitigate implicit bias, bias persists among adult and pediatric providers alike.[40,41] Given the high morbidity and mortality associated with child abuse, prevention is key. Implicit bias not only incorrectly places the burden on one population, but may also ignore a subset of children who are at higher risk. The child abuse literature is replete with studies evaluating screening guidelines for child abuse. These studies have largely focused on using specific signs and symptoms when a child presents to the emergency room that are consistent with signs of child abuse.[11,24,42–46] However, the threshold for initiating investigations for abuse may be different for children in minority racial groups. [47] There have been increasing calls to action to address structural racism throughout medicine, including in the evaluation of child abuse. [48,49] It has been suggested that universal screening applied to all children are necessary to detect child abuse and eliminate unconscious reporter bias in order to more accurately identify those at risk for child abuse.

There are a number of hospital-based child abuse screening tools, however, the content and administration vary widely.[50–52] Ideally, a screening tool should be simple, accessible, and not interrupt clinical workflow.[50,53] Gonzalez and Deans [50] suggest that screening should include automated notes and checklists, referrals to multidisciplinary teams, and rapid rule-out. There is evidence that instituting a systemic screening protocol increases the detection of child abuse cases. Louwers et al. [54] developed a screening checklist for child abuse in Dutch emergency departments and trained nurses to administer it. After implementation, they saw an over 40% increase in the screening rate as well as a child abuse detection rate 5 times higher in children who were screened compared to those who were not. This same screening tool had an 80% sensitivity and 98% specificity.[55]

Early detection of child abuse is key to prevent repeated and often escalating abuse. Quiroz et al., demonstrated that 10% of patients admitted for child abuse had a prior admission. [56] In addition, 37% of prior admissions occurred at a different hospital. Our results demonstrated that White SCA patients presented with more severe injuries than Black SCA patients did. It is difficult to discern whether the severity of injury is a consequence of a progression of abuse or if White SCA patients experience abuse that is overall more severe compared to children of other races. One alternative hypothesis for lower ISS among Black SCA patients could be that providers have a lower threshold to initiate investigations on Black patients. The lower threshold for suspicion can be multifactorial, including the potential implicit bias amongst medical providers. Those children with mild injuries that may otherwise not trigger detailed investigations may be more commonly captured among Black patients because the baseline suspicion is higher. Concurrently, White children who are victims of child abuse may be more frequently missed and therefore return with more severe injuries. Among both racial groups, the over and under-reporting of potential child abuse can lead to significant adverse consequences among each population. Sonderman and colleagues [57] have demonstrated that while Black patients are overrepresented in child abuse data, this does not translate to increased mortality. Our study found a strikingly higher in-hospital mortality rate among White children of abuse (9%) compared with Black children (6%), a potential consequence of missed earlier opportunities for child abuse detection.

There are limitations to this study. The data available for analysis are limited to trauma centers that contribute to the NTDB dataset. In addition, not all elements from the National Trauma Data Standard Data Dictionary were available for review and analysis as part of the Participant Use File, which is designated for research studies. For the years reviewed in this study, data were not available for infants less than one year of age. However, data captured by the NTDB is still representative of young children, which comprise a significant portion of the abused population. Data to indicate patient socioeconomic status, such as zip codes or regions, were also not available for analysis. Thus, we used insurance payer type and hospital type as our closest surrogate for socioeconomic status. We acknowledge the fact that not every case is represented equally in NTDB. When evaluating the validity of Medicaid codes in identifying maltreatment, Raghavan et. al found that in their sample of children only 15% had a relevant ICD-9CM code in their Medicaid file which identified them as being maltreated even though they had been identified by caseworkers as such. [58] Despite this, aside from institutional studies, there is not a database that accurately gathers all data. This study serves as a foundation for future studies in evaluating disparities in child abuse identification.

In conclusion, child abuse is a major cause of morbidity and mortality in young children. Early detection is key to preventing long-term physical and emotional sequelae of child abuse. We have found in our nationally representative dataset that Black children are disproportionately identified as potential victims of abuse and are subject to longer hospital stays, despite milder injuries. Whether these trends reflect underlying unconscious or conscious biases of mandated reporters requires further detailed investigation. However, universal screening protocols may be a first step to reduce unconscious biases and more precisely identify children who are at risk for abuse. More objective protocols for screening can serve to reduce the consequences of unconscious biases both in reducing the number of minority children who may be subject to overidentification while simultaneously reducing the underidentification of White children. We believe that identification of such reporting disparities identified in this study is critical in the dialogue of improving child abuse detection and decreasing racial inequities in health care.