Testing for Abuse in Children With Sentinel Injuries

Discussion

Among young children found to have putative sentinel injuries without a motor vehicle collision at leading children’s hospitals, rates of diagnosed abuse were high. Rib fractures were by far the most concerning for abuse, with intracranial hemorrhage and abdominal injury also associated with abuse in more than 20% of cases. Children with burns and isolated skull fractures had the lowest rates of abuse diagnoses, though rates were still high relative to the baseline risk for children without putative sentinel injuries.

The American Academy of Pediatrics considers the radiographic SS to be “mandatory” for children with concern for abuse; this recommendation is echoed by the American College of Radiology.^29,30 In our population where all children were <24 months of age, the decision to obtain SS might be taken as 1 surrogate measure of whether there was a concern for abuse that rose to the level of evaluation. The observed correlation between rates of abuse diagnosis and rates of SS is consistent with the ideas that (1) SS is more likely to be ordered when concern for abuse is higher and (2) concern for abuse is likely to increase if SS identifies additional, occult fractures in a child with a sentinel injury. However, the likelihood of abuse does not account for all the variability in SS use, either between centers, or from injury to injury. These data suggest that SS use should be increased for children with abdominal injury, subconjunctival hemorrhage, or genital injury.

Overall rates of abuse diagnosis were also highly variable (range, 0.04%–0.46%) between centers. We suggest several reasons for this variability. First, it is possible that true rates of abuse are actually highly variable between centers on the basis of characteristics of the population the hospital serves or on referral patterns. Alternatively, it is possible that the recognition of abuse differed between hospitals. Finally, it is possible that recognition of abuse was similar between centers, but ICD-9 coding of abuse suspicion differed systematically between centers.³¹ Systematic undercoding for abuse by some centers would result in underestimating the significance of putative sentinel injuries and overestimating the variability between centers.

Children who were noted to have been transferred from another institution were significantly more likely to undergo SS and to be diagnosed with abuse. This is consistent with the hypothesis that children with traumatic injuries are more likely to be transferred to the pediatric referral centers that participate in the PHIS database when there is some concern for abuse. It is also possible that abused children are more likely to present initially to nonpediatric centers, or that the difference could be explained by other characteristics of transferred children (age, injury type) that correlate with abuse likelihood. Children may also have had testing that we did not identify if they were transferred out from a PHIS center to another institution. Because all centers are tertiary referral centers, we suspect that the number of such subjects is small.

Our study is subject to several limitations. ICD-9 codes have been shown to have variable and limited sensitivity for abuse.^31,32 Because a final determination of any traumatic injury’s etiology may come after a child is discharged from the hospital, it is possible that many children ultimately determined to have been abused were not identified by our search. The relatively lower rate of abuse diagnosis in young infants with bruises is consistent with this possibility. Sugar et al²⁶ demonstrated that <1% of healthy infants <6 months old had any bruising, and when such bruised infants are evaluated for abuse, rates of additional injuries and abuse diagnoses have been reported at 50%.²⁷ However, the rate of abuse that we identified in children with burns or isolated skull fractures seems more consistent with previous work.^33–37 This limitation might be especially relevant for children without any identified putative sentinel injuries who might only be finally diagnosed as abused after hospital discharge. If this is true, rates of abuse would increase for children with and without putative sentinel injuries, but our estimate of the relative increase in abuse prevalence among children with putative sentinel injuries would be inflated.

Furthermore, these administrative data may have poor sensitivity to detect all diagnostic procedures that were performed in each subject. For example, among subjects coded as having intracranial hemorrhage, neuroimaging was identified in only 89.3%, whereas it would seem that neuroimaging would be needed to establish the diagnosis for 100% of cases. This could be explained if ICD-9 codes for intracranial hemorrhage were incorrectly applied to children with subgaleal hematoma or other injuries that could be diagnosed without neuroimaging, or if we failed to identify children who were transferred from another institution after neuroimaging had been obtained. This latter possibility seems especially likely given the high rate of children coded as other or unknown for the source of admission. These limits of our administrative data imply that true rates of testing may be higher than our estimates, but would not affect the variability we identify between sites and injuries unless there is also systematic regional variability in the testing practices of referring hospitals. Because children with concern for abuse are more likely than other children with sentinel injuries to be transferred to PHIS centers, inaccuracies in coding transfers could artificially inflate the rates of abuse for children with putative sentinel injuries.

Although CTC procedure codes are relatively consistent for SS and neuroimaging, it is likely that there is more variability in coding for hepatic transaminases. Although some hospitals may include AST and ALT in a “14-test chem profile” (CTC code 310114), others may only include them in “critical care panel” (CTC code 300171) or another such panel. Further, it is possible that some hospitals obtained hepatic transaminases as part of a panel that was obtained to check other laboratory data, or from routine ordering, and not to test for additional abusive injuries. We attempted to include all CTC codes that might have included AST/ALT to use the maximum estimate of testing prevalence. If CTC codes were not accurate, and if this inaccuracy was systematically associated with different centers, our estimates of the variability in ordering of hepatic transaminases would be inflated.

Although SS is primarily done for occult injury evaluation, neuroimaging and hepatic transaminase testing may be undertaken either to identify occult injuries or those with clear or subtle signs. We are not able to determine the reasons that testing was undertaken. If rates of clinically apparent injuries differed systematically between centers, this would account for some of the variability in testing that we identified.

Whereas our use of the term sentinel injury applied to children in whom abuse was diagnosed on the index visit, other authors have used the term to refer to children who are subsequently diagnosed with recurrent abuse on a future visit. Whether the risk of abuse is in the immediate, near, or long-term, we feel that the core attribute of a sentinel injury is that it should prompt the clinician to consider the possibility of physical abuse, and in most cases to undertake testing for additional occult injuries.

True sentinel injuries would present without other clinically apparent injuries or history of abuse. We were not able to determine the presenting injury for any given child, nor whether a child’s sentinel injury was accompanied by other obvious signs of abuse, such as a history of witnessed assault. It is therefore possible that the putative sentinel injuries that are most likely to have been identified in the course of an abuse evaluation (eg, rib fractures) would have inflated rates of abuse. However, for the large majority of patient visits, only a single putative sentinel injury was identified, suggesting at least that these injuries were not only the least noticeable injuries in children with obvious polytrauma. We therefore believe that the high rates of abuse among children with the listed injuries create a strong presumption that abuse should be carefully considered when these injuries are identified.

Similarly, because we are not able to identify the reason for presentation, children who present with clear evidence of a sentinel injury are grouped with children whose injuries were only identified after testing prompted for less specific complaints. It has been well-described that abused children frequently present with subtle signs and symptoms, and that the history may be incomplete or misleading.^6,22,23 Therefore, although rates of abuse diagnoses were vanishingly low in patients without a putative sentinel injury, our results should not be interpreted to discourage abuse evaluations when sentinel injuries are not obvious.

Several methods could be used to increase testing and decrease variability for children with putative sentinel injuries. These and other data could be included in physician training and in the development of clinical guidelines and pathways that address abuse evaluations. In centers with access to subspecialty child abuse pediatricians, the presence of a putative sentinel injury could be considered an indication for subspecialty consultation to determine the plan for occult injury testing. Finally, these data could form the basis for electronic clinical decision support prompting clinicians to consider occult injury testing for children with putative sentinel injuries.