OBJECTIVE: To evaluate in a national database the association of race and socioeconomic status with radiographic evaluation and subsequent diagnosis of child abuse after traumatic brain injury (TBI) in infants.
METHODS: We conducted a retrospective study of infants with non–motor vehicle–associated TBI who were admitted to 39 pediatric hospitals from January 2004 to June 2008. Logistic regression controlling for age, type, and severity of TBI and the presence of other injuries was performed to examine the association of race and socioeconomic status with the principal outcomes of radiographic evaluation for suspected abuse and diagnosis of abuse. Regression coefficients were transformed to probabilities.
RESULTS: After adjustment for type and severity of TBI, age, and other injuries, publicly insured/uninsured infants were more likely to have had skeletal surveys performed than were privately insured infants (81% vs 59%). The difference in skeletal survey performance for infants with public or no insurance versus private insurance was greater among white (82% vs 53%) infants than among black (85% vs 75%) or Hispanic (72% vs 55%) infants (P = .022). Although skeletal surveys were performed in a smaller proportion of white than black or Hispanic infants, the adjusted probability for diagnosis of abuse among infants evaluated with a skeletal survey was higher among white infants (61%) than among black (51%) or Hispanic (53%) infants (P = .009).
CONCLUSIONS: National data suggest continued biases in the evaluation for abusive head trauma. The conflicting observations of fewer skeletal surveys among white infants and higher rates of diagnosis among those screened elicit concern for overevaluation in some infants (black or publicly insured/uninsured) or underevaluation in others (white or privately insured).
WHAT'S KNOWN ON THIS SUBJECT:
Results of small regional studies have suggested that racial and SES biases exist in the evaluation and diagnosis of AHT, a leading cause of morbidity and mortality in young children.
WHAT THIS STUDY ADDS:
Data from 39 pediatric hospitals suggest continued racial and SES biases in the evaluation for and diagnosis of AHT among infants with TBI.
In the United States, infants younger than 1 year have the highest risk of experiencing and dying from child abuse and neglect. Child protective services (CPS) investigations determined that 1 in 45 infants in the United States were victims of child abuse and neglect in 2007.1 Among the more than 1700 cases of fatal child abuse and neglect reported to CPS in 2007, 42% occurred in infants.1 However, CPS data have been shown to capture only a fraction of the cases; the true incidence of nonfatal and fatal abuse and neglect may be several times higher.2,3 Despite recognition that abusive head trauma (AHT) is the leading cause of mortality and morbidity among physically abused infants,4,5 research results have suggested that medical professionals sometimes fail to recognize cases of AHT; as a result, the abused children do not receive additional evaluation, are not appropriately diagnosed, and suffer additional injury.6
Because of the high rate of occult fractures in young victims of AHT and other forms of physical abuse, the American Academy of Pediatrics recommends that all suspected infant victims of abuse undergo a skeletal survey to identify occult fractures.4,7,–,10 However, the degree to which such recommendations are followed systematically remains uncertain. Of particular concern are previous studies in which is it suggested that clinicians are more likely to evaluate for and diagnose child abuse among black infants and infants of lower socioeconomic status (SES) and less likely to evaluate white infants and infants of higher SES.11,–,15 However, these studies were restricted to small regions of the country or conducted solely in the outpatient setting. Thus, we sought to examine racial and SES differences in the frequency of radiographic evaluation for occult injuries and diagnosis of suspected physical abuse in a nationally representative cohort of infants with traumatic brain injury (TBI).
We used the Pediatric Health Information System (PHIS) database, an administrative database maintained by Child Health Corporation of America (CHCA; Shawnee Mission, KS). Forty hospitals that are located in 17 of the 20 major metropolitan areas in the United States and that account for more than 70% of all freestanding children's hospitals in the United States currently contribute patient-level data to the PHIS database (data from the National Association of Children's Hospitals and Related Institutions, Alexandria, VA). The participating hospitals electronically submit detailed patient data, including demographics (age, gender, race/ethnicity), payer source, episode of care information (admission date, disposition, repeat hospitalization), up to 21 International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes, and resource utilization information (including imaging procedure codes) to the database. Maintaining and validating the quality of the PHIS data is a joint effort among CHCA, the participating hospitals, and Thomson Reuters (the data warehouse vendor for PHIS). Validity and reliability checks of the data are performed, and data are included in the database only when classified errors occur in <2% of a hospital's quarterly data.
Infants younger than 1 year who were admitted to a PHIS hospital between January 1, 2004, and June 30, 2008, and were assigned an ICD-9 diagnosis code for TBI (800–801, 803–804, and 851–853) in any of the 21 diagnosis fields in the PHIS database were included, with the exception of infants with ICD-9-CM codes only for unspecified intracranial injury, isolated skull fractures, or concussions without an identified hemorrhage, contusion, or laceration (800.0, 800.5, 801.0 801.5, 803.0, 803.5, 804.0, 804.5, 850, 854). Infants with isolated skull fractures or concussions were excluded because the likelihood of abuse is lower among this group than among infants with more severe head injuries.5,12,16 To exclude repeat admissions for medical complications that resulted from the injury only, the first admission for TBI during the study time frame was included, and infants who were admitted with TBI during the year before the study period were excluded. Because we were interested only in the group of infants who were potential victims of abuse, we excluded infants with the following characteristics: (1) an external cause of injury code (E-code) for transportation accidents (E800–E807, E810–E829, E830–E838, E840–E845, E48), gunshot, or dog bite; (2) hospitalization pertaining to birth; or (3) previous admissions with ICD-9-CM diagnosis codes for coagulation disorders (286.0–286.5, 776.0, 286.7, 284.0–284.9, 287.3, 287.1). In an effort to exclude infants who may have had a skeletal survey at another hospital or the medical examiner's office, we excluded children who were transferred to or from another hospital inpatient unit and infants who died on the day of admission. Thirty-nine of the PHIS hospitals had patients who met the inclusion and exclusion criteria during the study period. Data were excluded for patients admitted to 1 of the 39 hospitals during a 16-month time period because of data quality concerns that were identified by CHCA.
The primary outcomes were performance of a skeletal survey and diagnosis of child abuse. Because some hospitals use bone scans or “babygrams” (single radiographic images of infants) instead of skeletal surveys to assess for occult fractures, we included these imaging modalities in our definition of skeletal survey, although they are not the recommended standard of care. Infants were categorized as receiving a diagnosis of abuse on the basis of the presence of an ICD-9-CM or E-code for child abuse or assault (ICD-9-CM 995.5 and E960–E968). Infants for whom there was an E-code for an accidental fall, other accidental mechanism of injury (E880–E888, E916–E928), or an undetermined mechanism of injury (E987–E989) and no E-code for abuse were categorized as not receiving a diagnosis of abuse.
Predictor variables included 2 categories: patient demographics and injury characteristics. Demographic variables included age, race/ethnicity (black, non-Hispanic white, Hispanic, or other/missing), gender, and insurance status (private insurance, public/none, or other/missing). Age was categorized as younger than 1 month, 1 to 6 months, and 7 months or older because of the a priori hypothesis that infants younger than 1 month and infants 7 months or older who have TBI may be less likely to be evaluated for and diagnosed with abuse than are infants aged 1 to 6 months. Some intracranial injuries in infants younger than 1 month may be attributed to birth injuries and in infants 7 months or older may be attributed to accidental falls related to early mobility. Injury characteristics included type of TBI (TBI with skull fracture versus TBI without skull fracture), injury severity category, and other injuries (burns, bruises, lacerations, and injuries to internal organs). Retinal hemorrhages and fractures were categorized separately from other injuries because they may have been discovered during an evaluation for suspected abuse (such as an ophthalmology examination or skeletal survey, respectively). Bruises on the head and skull fractures were considered part of the TBI and were not counted as additional injuries. The overall injury severity score (ICD/ISS) was calculated on the basis of ICD-9-CM codes by using the ICDMAP-90 injury coding software (Johns Hopkins University and Tri-Analytics Inc, Baltimore, MD). Because it was possible that the ICD/ISS scores were influenced by the presence of injuries discovered on a skeletal survey, we were concerned about adjusting for ICD/ISS scores in the models that predicted skeletal survey as the outcome. Thus we also calculated and used in a sensitivity analysis an ICD-9-based abbreviated injury score (ICD/AIS) that only considered injuries for the head region. The results of that analysis mirrored the use of ICD/ISS scores, so we chose to present only the results that used ICD/AIS scores in this manuscript.
In univariate analysis, χ2 tests were conducted to assess the association between each of our binary responses and each covariate of interest. Predictor variables were included in a multivariable logistic regression model if univariate analysis indicated an association when a criterion of P < .10 was used. A robust variance estimator that used the generalized estimating equations approach accounted for clustering by hospital. Because of the small number of infants in the mild-severity category, the mild- and moderate-severity categories were combined for the analysis. Fractures and retinal hemorrhages were not included in the model that predicted evaluation for abuse with a skeletal survey because these injuries would likely have been noted during the evaluation for abuse. In previous literature it has been suggested that there are demographic differences in child-abuse evaluations, and therefore, tests for interaction were conducted between injury severity and race, injury severity and insurance status, and race and insurance status.6,12,–,15 Results are presented as marginal estimates calculated from predictions from a fitted model at fixed values of specified covariates and averaging or integrating over remaining covariates.17,18 Marginal estimates are presented as opposed to odds ratios because the baseline prevalence of the outcomes was high enough to render odds ratios less meaningful.
This study was approved by the Children's Hospital of Philadelphia institutional review board. Data analysis was conducted by using SAS 9.0 (SAS Institute, Inc, Cary, NC) and Stata 11.0 (Stata Corp, College Station, TX).
Among the 3063 infants included in the study 46% (1410) were non-Hispanic white, 20% (626) were black, 18% (544) were Hispanic, 61% (1860) had public insurance or were uninsured, and 21% (646) had private insurance (Table 1). Black and Hispanic infants were more likely than were non-Hispanic white infants to have public insurance or be uninsured (74% vs 73% vs 50%, respectively; P < .001). The age of the infants ranged from 2 days to 364 days with a median of 113 days (3.8 months) and an interquartile rage of 58 to 207 days. Because this was an in-hospital sample that required intracranial injury as an inclusion criterion, the majority of the infants had moderate (58%) or severe (41%) injuries by AIS scores, and only a few had mild injuries (1.3%).
Skeletal Survey Performance
Skeletal surveys were performed in 72% of the infants. Univariate analysis indicated that skeletal survey performance was more prevalent among children with more severe TBI than among those with other injuries (contusions, lacerations, burns, internal injuries), and those with an accompanying skull fracture (all P < .001). Black race, public insurance/no insurance, male gender, and age older than 1 month were also associated with an increased performance of skeletal survey (all P ≤ .011) (Table 2).
The final model that predicted skeletal survey performance included insurance, age, race, TBI severity (ICD/AIS category), TBI type, and a dichotomous variable for other injuries (that indicated whether the child had contusions, burns, lacerations, or internal organ injuries). After adjusting for other variables, gender was not a significant predictor of skeletal survey performance and was not included in the final model. In that model, uninsured infants and infants with public insurance/no insurance were more likely to receive skeletal surveys than were infants with private insurance (81% vs 59%). This effect was modified by race, such that the difference in performance of skeletal survey for infants with public insurance/no insurance versus private insurance was greater among white infants (82% vs 53%) than among black (85% vs 75%) or Hispanic (72% vs 55%) infants (P = .022). Fig 1 shows the adjusted proportions of skeletal survey by injury severity, race, and insurance type. Overall, 71% (95% confidence interval: 65%–77%) of white infants underwent a skeletal survey, but if the white infants had received testing at the same rate as black infants, the proportion who received a skeletal survey would be predicted to increase by 14% (95% confidence interval: 3%–26%).
Diagnosis of Child Abuse
A total of 37% of the infants received a diagnosis of abuse. The remainder received an E-code for an accidental injury mechanism (46%), an E-code for an undetermined mechanism (3.2%), or no E-code for mechanism of injury (13%). Among the 2650 infants for whom a mechanism of injury was noted, the following characteristics were significantly associated with a diagnosis of abuse univariately: more severe TBI, fractures, retinal hemorrhages, other injuries (contusions, lacerations, burns, internal injuries), and presence of a skull fracture (all P < .001) (Table 3). Black race, public insurance/no insurance, male gender, and age older than 1 month were also associated with an increased diagnosis of child abuse (P = .021 for gender; all other P < .01) (Table 3). Almost all of the infants (95%) diagnosed as abused were evaluated with a skeletal survey.
Of the 2205 (72%) of infants who were evaluated for abuse with a skeletal survey, 89% received an ICD-9-CM or E-code for the mechanism of injury. A total of 55% of the infants were diagnosed as abused, and 45% were reported to have injuries from accidental or undetermined mechanisms.
The final model that predicted diagnosis of child abuse accounted for clustering by hospital and included insurance, age, race, TBI severity, TBI type, retinal hemorrhages, fractures, and other injuries (contusions, burns, lacerations, and internal organ injuries). After adjusting for other variables, gender was not a significant predictor of diagnosis of abuse and was not included in the final model. In that model, white infants evaluated with a skeletal survey were more likely to be diagnosed as abused (61%) than were non-Hispanic white (53%) and black (51%) infants (P = .009). Infants with public insurance or no insurance who were evaluated for abuse were more likely than were infants with private insurance to be diagnosed with abuse, but the effect of insurance was modified by the TBI severity, such that the effect of insurance was larger among infants with mild/moderate TBI than among infants with severe TBI (P = .011 for interaction of insurance and severity). Fig 2 shows the adjusted proportions of child abuse diagnosis by injury severity, race, and insurance type.
Our results demonstrate a few important trends in the use and diagnostic yield of skeletal survey for the evaluation of young infants with TBI for abusive injury. First, black infants and publicly insured/uninsured infants were more likely to undergo skeletal surveys. Second, the disparity in performance of skeletal surveys according to type of insurance (public/none versus private) was greatest among white infants. However, most concerning is the finding that once a skeletal survey is ordered, the diagnostic contribution of the imaging study toward the diagnosis of abuse was not the same among children of different races. In fact, the highest proportion of children who received the diagnosis of abuse once a skeletal survey was obtained was among the children who were white.
Results from previous research studies of young children with injuries have suggested that racial and SES biases may influence medical professionals' threshold for evaluating and reporting physical abuse. However, with the exception of a large multicenter outpatient study, the previous studies have been primarily small, single-center studies.12,13,15,19 In a retrospective study of young victims of AHT at 1 hospital, almost one-third of the children were evaluated by a physician after the initial trauma occurred, and the diagnosis of AHT was missed. The failure of the physicians to recognize and report the abuse resulted in additional inflicted injury in 28% and medical complications in 40% of the missed cases.6 Studies at another pediatric center also found that racial- and SES-based disparities occurred in the evaluation and reporting of abuse in infants and young children with fractures.12,13 Our results support these findings and suggest that the disparities are not limited to a few hospitals but exist across a wide network of pediatric hospitals.
Our findings may be limited by the fact that we could not examine individual records of children to determine the true likelihood of abuse. However, it can be argued that if the differences in the use of skeletal survey were the result of baseline unmeasured differences in likelihood of abuse (eg, unmeasured history of presentation, caretaker capacity, time to presentation, or supervision during a child's injury), then once a skeletal survey was obtained, its contribution toward the diagnosis of child abuse should have been roughly equivalent between the different groups of children. The observation that white children were much more likely to receive the diagnosis of abuse if a skeletal survey was obtained suggests that this group of children had to reach a higher threshold for suspicion for abuse before a clinician ordered a skeletal survey.
The results, therefore, raise concerns that different thresholds for suspicion of child abuse were being used for children of different races. The use of different thresholds may lead to overevaluation in some infants (eg, those who are black or publicly insured/uninsured) and underevaluation in others (eg, those who are white or privately insured). Overevaluation for abuse among black or publicly insured and uninsured infants may result in infants being unnecessarily subjected to a skeletal series of 20 or more radiographs10,20 and may increase medical-care costs, hospitalization length, and caregiver anxiety. Equally concerning are the potential consequences of underevaluation for suspected abuse in some populations. Failure of physicians to appropriately evaluate and diagnose AHT can lead to infants suffering additional injuries. If white infants had been treated as black infants, an additional 14% of white infants would have been evaluated with a skeletal survey that may have revealed occult fractures and a diagnosis of child abuse.
The use of administrative data enabled us to study a large population of children across a national network of pediatric hospitals but had limitations. As mentioned above, some patient-level characteristics that may have affected the decision to evaluate for and diagnose abuse including history of presentation were not available. As a retrospective study it was prone to misclassification bias because of inaccuracies in the medical chart and administrative data-abstraction process. Race and ethnicity were not available for every patient. Finally, insurance status was used as a measure of SES but may not accurately reflect SES in all cases.
The recognition that racial and SES disparities in child-abuse evaluation are not limited to a few hospitals but occur nationwide requires the design, implementation, and evaluation of interventions to decrease such disparities. A recent study at a pediatric hospital demonstrated that the introduction of a screening guideline for children with TBI reduced disparities in skeletal survey performance,19 which raises hope that such disparities can be addressed.
- Accepted May 27, 2010.
- Address correspondence to Joanne N. Wood, MD, MSHP, General Pediatrics, Children's Hospital of Philadelphia, 3535 Market Street, Room 1517, Philadelphia, PA 19104. E-mail:
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
- CPS =
- child protective services •
- AHT =
- abusive head trauma •
- SES =
- socioeconomic status •
- TBI =
- traumatic brain injury •
- PHIS =
- Pediatric Health Information System •
- CHCA =
- Child Health Corporation of America •
- ICD-9-CM =
- International Classification of Diseases, Ninth Revision, Clinical Modification •
- ISS =
- injury severity score •
- AIS =
- abbreviated injury score
- 1.↵US Department of Health and Human Services, Administration on Children, Youth, and Families. Child Maltreatment 2007. Washington, DC: US Government Printing Office; 2009
- Theodore AD,
- Chang JJ,
- Runyan DK,
- Hunter WM,
- Bangdiwala SI,
- Agans R
- Rubin DM,
- Christian CW,
- Bilaniuk LT,
- Zazyczny KA,
- Dennis R
- Billmire M,
- Myers P
- Duhaime AC,
- Alario AJ,
- Lewander WJ,
- et al
- 10.↵American Academy of Pediatrics, Section on Radiology. Diagnostic imaging of child abuse. Pediatrics. 2009;123(5):1430–1435
- Sege RD,
- Flaherty EG
- Wood JN,
- Christian CW,
- Adams CM,
- Rubin DM
- Flaherty EG,
- Sege RD,
- Griffith J,
- et al
- 18.↵Stata Corp [Stata Base reference manual]. Release 11. Vol 2. College Station, TX: Stata Corp LP; 2009:957–1009
- 20.↵American College of Radiology. ACR practice guideline for skeletal surveys in children. Available at: www.acr.org/SecondaryMainMenuCategories/quality_safety/guidelines/pediatric/skeletal_surveys.aspx. Accessed July 15, 2010
- Copyright © 2010 by the American Academy of Pediatrics