Witnessing abusive head trauma: Accidents show higher rates of intracranial pathologies than shaking
Abstract
Objectives
This study aims to determine whether intracranial injuries, such as seizures, encephalopathy, bilateral subdural hematoma (SDH), and severe bilateral retinal hemorrhage (RH), are indicators of abusive head trauma (AHT), particularly in cases involving shaking.
Methods
Data comprising 54 witnessed shaking cases were drawn from two studies in the literature. Data of 100 witnessed accidents comes from the pediBIRN collaboration. Rates of intracranial injuries in cases of unconflicted witnessed accidents are compared to rates in cases of witnessed shaking and also to cases of unconflicted witnessed shaking. Unconflicted is defined as observed by an independent, unbiased witness, or by a potentially biased witness (such as partner) if reported prior to medical examinations.
Results
When all witnessed shaking cases were considered, including potentially biased witnesses, there are higher rates of findings commonly associated with AHT in witnessed accidents than in cases of witnessed shaking, although the difference is only statistically significant for seizures and encephalopathy. When restricted to cases when the witness was unconflicted, the rates of all findings are significantly more common in accidents than in shaking.
Interpretation
Accidents result in more severe intracranial pathologies than shaking, aligning with biomechanical studies that have shown that impact exerts greater force on the brain than violent shaking.
Introduction
Certain intracranial findings are considered highly suggestive of abusive head trauma (AHT) and in particular of violent shaking.1 For example, certain severe retinal hemorrhages (RHs) are considered highly suggestive of AHT.1-4 Encephalopathy and bilateral subdural hemorrhages (SDHs) are also considered “far more common” in AHT than in non-AHT.1, 5 Additionally, seizures are claimed to be suggestive of abuse.3, 4 When combinations of such findings are present, the diagnosis of AHT involves ruling out other known causes of such findings, such as metabolic and bleeding disorders.
However, an alternative theory posits that these findings are not suggestive of violent shaking but are instead markers of the degree of intracranial pathology.6-8 According to this theory, the rates of such findings in any selected category of cases will be determined by the distribution of the degrees of intracranial pathologies within that category, which will depend on how cases are selected. Under this theory, claims that certain findings are more common in diagnosed or admitted AHT, or cases where caregivers deny trauma, than in cases involving witnessed or reported accidents (non-AHT), are a result of selection effects.8
To avoid such selection effects, categorization can be made using witnessed cases for both AHT (in particular shaking events) and non-AHT (accidents). Using witnessed events as a reference standard allows for a more equitable comparison of findings between shaking and accidental cases, and also circumvents the circular reasoning that has long plagued the field.7 However, there has long been a paucity of data involving witnessed shaking cases.
This study uses data from witnessed cases to re-examine the rates of findings of intracranial pathologies in cases of shaking, compared with rates in cases of accidents.
Materials and Methods
This case referent study includes 100 cases of unbiased independently witnessed accidents and 54 cases of witnessed shaking, of which 50 were deemed unconflicted witnesses.
Clinical findings
The focus is on four clinical findings that are recorded in all data sets and that are often associated with AHT: seizures, encephalopathy (defined as loss of consciousness), bilateral and/or interhemispheric SDHs, and extensive bilateral RHs.
Data
The data come from the following data sets:
1. pediBIRN data
Witnessed accidental (non-AHT) cases were taken from the pediBIRN (https://www.pedibirn.com/home.php) data set of 937 acutely symptomatic infants (aged <3 years), who presented in one of 18 academic medical centers during the period of February 2011 to March 2021. The data were collected prospectively. There are 100 unbiased independently witnessed accident cases in the data, defined as events “witnessed and described thoroughly by an unbiased, independent observer” and “described by this observer as an ‘accidental’ or ‘non-abusive’ head injury event.”4 All 100 were imaged with magnetic resonance imaging (MRI), while 38 also had ophthalmological exams. Data for severe RHs were recorded, defined as RHs that were “described by an ophthalmologist as dense, extensive, covering a large surface area, and/or extending to the ora serrata.” Records of SDH included whether they were unliteral, bilateral, or interhemispheric. Data on seizures were also provided.
2. Thiblin data
In Thiblin et al.9 (Thiblin20 hereafter), cases were retrospectively identified from those under suspicion of maltreatment within the Swedish National Patient Registry using the International Classification of Diseases, covering the period 1997–2014. The medical records were searched for cases of witnessed or admitted AHT by shaking.
Witnessed shaking was defined as “shaking having been observed directly or filmed.” There were 33 cases of witnessed shaking identified. A requirement was that the shaking behavior was reported in the initial phase of seeking care, before clinical, radiological, and ophthalmological examination. Also identified were three cases where a history of shaking was spontaneously admitted prior to diagnostic procedures. Six of the 36 cases involved some type of impact as well as shaking. An appendix to the published study provides details of each case, including the reported physical abuse, who reported the abuse, and the clinical findings.
Of the 36 identified cases, five cases had no brain imaging, while 31 had both imaging (computed tomography or MRI) and fundoscopy. The five cases with no brain imaging were examined and found to have no requirement for further tests. The 31 cases with imaging and fundoscopy are included in this study as unconflicted witnessed shaking cases. The main findings recorded were SDH, RH, fractures, and skin lesions. Details of the extent of SDH and RH, when present, were provided, including whether they were bilateral and extensive.
3. ExSTRA data
ExSTRA10 was a retrospective study of cases referred for child abuse evaluations at 20 US medical centers between January 2010 and April 2011. Witnessed cases of shaking were selected from the ExSTRA data and presented in a study by Feldman et al. in 202211 (Feldman22 hereafter). These cases were selected if a witness reported observing abusive shaking at the time of presentation. In detail, the cases were identified if:
(i) the physician indicated on a form that the injury event was witnessed by an independent observer;
(ii) further cases were identified by a full-text search of the database for records containing the terms “shak*” or “witness*”; and
(iii) a full-text search for children coded as having been injured in an assault was included if the history revealed a witnessed shaking incident.
Twenty-one such cases were identified, 19 of which had neuroimaging. Two cases were excluded due to signs of impact, that is, a skull fracture in one and scalp soft tissue swelling in the other. This left 17 cases of witnessed shaking, which are included in this study as witnessed shaking cases.
4. Helfer Society data
Feldman22 also identified six cases of witnessed shaking that were gathered from members of the Ray E. Helfer Society, a medical professional society of 625 physicians with expertise in child maltreatment. The data were collected through a series of list-serve queries of society members, without limits on the time of the events. This was not a systematic or controlled search of data, and the cases come from the thousands of cases seen by the members of the society. Cases identified with a history of shaking without cranial impact were included in this study as witnessed shaking cases, bringing the total number of such cases to 54. Some of the witnesses cannot be described as unbiased or independent (see below).
Feldman22 provided clinical findings of the ExSTRA and Helfer cases, including seizures, encephalopathy, SDH and its extent, and RH and its extent. Full details including the reported physical abuse, who reported the abuse, and detailed clinical findings were provided for the 10 cases with neurological signs or symptoms.
Unconflicted witnesses
An important issue is how to define unconflicted for this study. Cases caught on video would be considered unconflicted: there is just one such videotaped case of shaking in the data. People with no relation to the caregiver would be considered as independent, unbiased witnesses and thus unconflicted. However, when one partner accuses the other, care must be taken because there are circumstances that can lead to bias. In this study, such cases will be considered unconflicted only if the statement is made prior to medical examination and prior to investigations being made, so prior to suspicion falling on one or the other, or on both. Future studies with more data may apply more strict criteria, requiring unbiased, independent witnesses.
There are multiple reasons why witnesses related to the accused and/or to the infant may be considered potentially biased. First, if both caregivers/parents are initially under suspicion of shaking the infant, then there is an incentive to claim that the other caregiver was the one doing the shaking, in order for the one making the accusations to avoid being incriminated.
Second, if the doctors or the police have informed both caregivers that the infant was shaken, and one or both partners come to believe that the infant was shaken, then one partner may know for sure that it was not them who shook the infant and logically conclude that their partner must have shaken it. If one partner comes to believe that the other partner shook the infant, then they may want to ensure that justice is served by incriminating the person that they believe to be guilty of injuring or killing their child.
Third, many cases involve child removal. This is highly distressing for parents as well as traumatic for the removed children. Parents will do anything to get the children back, and this can include one parent making false admissions in order for the infant to be returned to the care of the other parent.12 The return of children may also incentivize one partner to bear false witness against the other partner, especially if they came to believe that the other partner must have shaken the infant.
Thus, cases where doctors already suspect shaking has occurred based on medical findings, and where an investigation is underway, have a significant risk of subsequent witness statements being biased. Like spontaneous confessions, cases of partners accusing each other should only be considered unbiased and independent if the accusation is made and recorded prior to any medical evaluation.
In the Feldman22 data, there are several cases where there is a significant risk of bias regarding the witness accounts.
For example, in case 2 from Feldman22, a methamphetamine-using mother was seen shaking an infant by an independent unbiased neighbor. However, this was 48 h prior to its collapse. Between the time of the witnessed shaking and the time of collapse, the infant had been in the care of the mother's boyfriend. The boyfriend reported that the infant had been “unwell” during the period between the shaking and collapse. However, if the baby was instead reported to be well after the shaking, and in the care of the boyfriend, then he would have come under suspicion. The link between the witnessed shaking and the collapse 48 h later is therefore not supported by any unbiased witness statement.
Feldman22 case 3 also involved the use of methamphetamines by the mother, as well as domestic violence. The mother and father both claimed to have witnessed the other shake the infant. Both parents were arrested by police. With both parents under suspicion and no indication that any shaking was reported prior to medical examination, this case cannot be considered to involve an independent unbiased witnessed shaking event.
Feldman22 case 4 also involved allegations of both parents shaking the infant, with the father confessing to shaking but claiming to have also witnessed the mother shake the infant “more violently” than he had shaken her.
In Feldman22 case 5, a mother reported seeing the father shake the infant but no information was provided as to the sequence of events, the timing of the witness report, whether the mother was ever under suspicion, or whether any children had been removed.
In this study, these four potentially biased witnessed shaking events are included in the category of witnessed shaking but excluded from the category of unconflicted witnessed shaking.
Note that case 1 of Feldman22 is included in the category of unconflicted witnessed shaking in this study. This case involved the mother witnessing her boyfriend shaking the infant, and the boyfriend also making admissions. However, there is no information as to whether the witness statement and admissions were made before or after accusations of shaking were made, whether the mother was ever a suspect, whether any children were removed, or whether any plea bargain was made. It is thus far from clear that the case truly involved an unconflicted witness or true admissions.
In the Thiblin20 data used in this study, the witness reports were made prior to medical evaluations. These are all treated as examples of unconflicted witnessed shaking in this study.
Statistical testing
Comparisons are made between categories of cases, and chi2_contingency and fisher_exact from the python scipy.stats13 module are both used to calculate p values. Odds ratios and their 95% confidence levels are also provided. p values < 0.05 are referred to as being “significant” and p values < 0.01 as “highly significant,” but p values are shown in each case, so the degree of significance is clear.
Results
Table 1 compares cases of independent, unbiased unconflicted witnessed accidents (n = 100, of which 38 had ophthalmological exams) with cases of witnessed shaking (n = 54, column 2) and unconflicted witnessed shaking (n = 50, column 2). p values using the chi-squared test (and Fisher's exact test) and odds ratios (with 95% confidence interval) show the comparison between findings in unconflicted witnessed accidents and findings in the other two categories of cases.
| Unconflicted witnessed accident | Witnessed shaking | Unconflicted witnessed shaking | |
|---|---|---|---|
| Finding | Rate | Rate p value (Fisher's exact test) OR (95% CI) | Rate p value (Fisher's exact test) OR (95% CI) |
| Seizure(s) | 16/100 (16%) | 2/54 (3%) 0.023 (0.033) 5.0 (1.1–22.4) |
1/50 (2%) 0.011 (0.012) 9.3 (1.2–72.6) |
| Encephalopathy | 47/100 (47%) | 2/54 (3%) <0.001 (<0.001) 23.1 (5.3–99.9) |
2/50 (4%) <0.001 (<0.001) 21.3 (4.9–92.4) |
| Bilateral/IH SDH | 20/100 (20%) | 6/54 (11%) 0.160 (0.183) 2.0 (0.8–5.3) |
2/50 (4%) 0.009 (0.008) 6.0 (1.3–26.8) |
| Severe RH | 7/38 (18%) | 6/54 (11%) 0.322 (0.37) 1.8 (0.6–5.9) |
2/50 (4%) 0.027 (0.036) 5.4 (1.1–27.8) |
- Note: Findings of unconflicted witnessed accidents (n = 100, of which 38 had ophthalmological examinations) compared with findings of witnessed shaking (n = 54) and unconflicted witnessed shaking (n = 50). Also shown are p values and ORs with their 95% CIs (in brackets) for the comparisons. Abbreviations: CI, confidence interval; IH, interhemispheric; OR, odds ratio; RH, retinal hemorrhage; SDH, subdural hemorrhage.
Table 1 shows that seizures, encephalopathy, bilateral SDHs, and severe bilateral RHs have higher rates in cases of unconflicted witnessed accidents than in cases of witnessed shaking, with the difference being statistically significant for seizures and highly significant for encephalopathy.
When restricted to cases involving unconflicted witnesses, all findings are significantly more common in cases of unconflicted accidents than in cases of unconflicted shaking.
Discussion
Witnessed events are the most reliable reference standard for classifying cases as involving accidents or shaking. Categorization using witnessed cases avoids circular reasoning and selection effects by using uniform criteria for selecting cases in both categories.
In this study, rates of findings that are often believed to be characteristic of AHT, and in particular of shaking, are compared between accidents and cases of shaking, using witnessed events as a reference standard. Criteria are set for witnessed events being unconflicted and thus more reliable.
Unbiased, independent unconflicted witnessed accidents are first compared with cases of witnessed shaking. Rates of seizures, encephalopathy, bilateral SDH, and extensive bilateral RH are all higher in accidents than in shaking, although the difference is only statistically significant for seizures and encephalopathy.
When cases of unconflicted witnessed accidents are compared with cases of unconflicted witnessed shaking, then all findings are significantly more common in accidents than in shaking. These results are consistent with well-established findings from biomechanics studies: impact, even from falls lower than 1 m, imparts significantly more force on the brain than violent shaking.14, 15 The higher rates of intracranial pathologies in impact than in shaking, as shown in this study, are naturally attributed to the greater force imparted on the brain.
Indeed, it is generally accepted that most witnessed shaking events lead to little or no intracranial pathologies. As stated in Feldman22, “Few witnessed shaken infants have signs and symptoms of AHT.”
One suggestion is that there is a minimal threshold of force required before shaking will cause injury, with Feldman22 stating that “perhaps only the most violent shaking events cause the full spectrum of AHT.” This is consistent with what can a priori be expected with all trauma, whether it be abusive impact trauma, shaking, or accidental trauma: cases with a higher degree of force imparted on the brain will tend to result in greater intracranial pathologies.
It should be noted that if adherence is made to reasonable requirements of witnessed cases being unbiased and independent, then there are no documented unbiased independent witnessed shaking cases that have resulted in the supposed “full spectrum of AHT.” For its part, Feldman22 found only five cases of witnessed shaking with abnormal neuroimaging among the thousands of cases to which they had access. None of the witnesses in those five cases can be considered unbiased and independent based on the information provided.
One case of unconflicted witnessed shaking from Thiblin20 did result in acute SDH and bilateral, extensive RHs. First, however, this case did not involve an unbiased independent witness, meeting the criteria of unconflicted used in this study because the reports of abusive acts were made prior to medical evaluation. Second, the case also involved impact. Third, the infant recovered rapidly and had no lasting sequelae. In other words, the case was not severe. Finally, there were underlying medical conditions that may have made this infant more susceptible to the effects of shaking.
If stricter criteria were used, that of videotaped or unbiased independently, witnessed shaking events, then none of the eight such cases from the data sets had any intracranial findings.
Again, these outcomes are consistent with what is found by biomechanical studies that have consistently shown that impact trauma can attain thresholds of force required to cause SDH, while, by contrast, no such study has been able to demonstrate that shaking can exceed such a threshold.14-17
Strengths and weaknesses
A major strength of this study is that the same reference standard is employed to categorize cases as accidents and shaking: witnessed events. Using independent unbiased witnessed events is the most rigorous reference standard available. This strength ensures that the study does not involve circular reasoning nor the selection effects that occur when comparing categories that have been selected using different reference standards, such as comparing witnessed accidents to diagnosed or admitted shaking.
A study weakness is that the data sets are not uniform, with one data set used for witnessed accidents and three different data sets for witnessed shaking, one of which was not systematic but instead was compiled via a survey of members of the Helfer Society. The use of different data sets may lead to some selection effects stemming from the different collections and selections employed in the different studies. A further issue is that no details were provided about the nature of each “unbiased, independent” witness to the non-AHT events in the pediBIRN data. It was assumed that child abuse pediatricians would not describe a person suspected of abuse, or potentially suspected of abuse, or with some other biasing circumstance as an “unbiased, independent” witness to an accident.
Further, witnesses to shaking were included as unconflicted witnessed cases if the abuse was reported prior to medical examination and prior investigations being made, even if the witness was a partner of the accused. If more strict criteria for being independent and unbiased witnesses were used for the shaking cases, then there would only be eight such cases, none of which had any intracranial findings.
It is hoped that researchers will attempt to reproduce the findings of this study using a single high-quality data set and that more independently witnessed shaking events will be identified.
Conclusions
When cases of shaking and cases of accidental trauma are selected using the same criteria, that of unbiased independent witnesses, findings that are often claimed to be indicators of AHT and of shaking are significantly more common in accidental trauma than in shaking.
The results support the theory that findings often associated with AHT are instead markers of the degree of intracranial pathology rather than signifying abusive trauma or shaking6-8 and indicate that the degree of intracranial pathology in cases of head trauma is related to the magnitude of the force exerted on the brain.
Author Contributions
Chris Brook: Conceptualization; formal analysis; methodology; writing—original draft.
Conflict of Interest Statement
The author declares no conflict of interest.
