Methods

We conducted an all language literature search from 1970–2008, searching 20 databases, websites, references and bibliographies, using over 100 keyword combinations (box 1, table 1). This yielded 320 studies for review by reviewers, drawn from paediatricians, paediatric neurologists, neuro-radiologists and ophthalmologists, designated and named doctors/nurses in child protection. All reviewers underwent standardised critical appraisal training, based on the NHS Centre for Reviews and Dissemination critical appraisal standards,12 supported by a dedicated electronic critical appraisal module. All studies underwent two independent reviews and a third if there was disagreement (fig 1).

Inclusion/exclusion criteria and quality standards

We defined the term “brain injury” as extra-axial haemorrhage (subdural, extradural, subarachnoid) and/or injury to the brain (hypoxic ischaemic injury, parenchymal injury, contusion, diffuse axonal injury, cerebral oedema). Inclusion/exclusion criteria are listed in box 2. Inflicted brain injury (iBI), the item of interest, refers to children who had sustained brain injury from physical child abuse. Those with non-inflicted brain injury (niBI) had other causative mechanisms including accidental trauma, and medical causes of intracranial lesions.

Given the nature of this research field, our optimal study type was population based studies that compared the clinical features of iBI and niBI with consecutive case ascertainment. As only studies with both niBI and iBI cases could be used for any analysis of predictive power of the features identified, many oft cited (non-comparative) studies were not eligible for inclusion. The included studies detailed the clinical features of interest but did not rely solely on these for the determination of diagnosis (iBI vs niBI). Only studies after 1970 were considered for inclusion, as radiological techniques prior to this would not be relevant to current practice. Included studies had a high surety of diagnosis of iBI13 and confirmation of cause in niBI. To determine a high surety of abuse, we implemented our previously published “ranking of abuse”, that is, children stated as “abused” in the study had had that outcome confirmed by multi-agency child protection teams, legal decision, witnessed abuse or perpetrator admission (rank 1–2)13 (table 2). Studies where the decision of abuse had relied solely on clinical features were excluded to minimise selection bias and circularity. Where studies included a category “aetiology indeterminate”, these cases were excluded from the final analysis.

Statistical analysis

The analysis was limited by the items that the authors chose to report and we were able to analyse the following features:

• apnoea

• retinal haemorrhages

• rib fractures

• long bone fractures

• bruising to the head and/or neck

• seizures

• skull fractures.

We were dependent on the primary authors’ definitions of the above terms. “Apnoea” reflects either recorded or reported apnoea in the two studies that recorded this item.14 15 “Seizures” were not always defined, but some authors did record that some children presented in status epilepticus.16 “Long bone fractures” included any fracture to the femur, tibia, fibula, radius, ulna or humerus. Likewise, “skull” or “rib fractures” included any type or location of fracture, respectively.

We included only data from studies where a key feature was explicitly mentioned by the authors. Further, even when a feature was included in the study, not all children in each group were examined for the feature in question. For example, not all children in the niBI group had a fundoscopy examination to look for retinal haemorrhages. We have elected to adopt an extremely conservative imputation approach to account for this missing information.17 For example, if we are analysing a feature that we suspect may be associated with iBI, for instance, retinal haemorrhages: in the group of children with iBI, if only 6/10 iBI children underwent fundoscopy, we made the assumption that the four children who were not examined would not have had the feature.

By contrast, in the group of children with niBI, if an investigation for this feature was not performed, for example, if only 2/10 niBI children underwent fundoscopy, we made the assumption that the eight children who were not examined would have had the feature. Only in the case of skull fractures, a feature whose presence we suspect to be associated with niBI, was the opposite imputation performed. We believe that by approaching the missing data in this way, we err on the side of caution, and risk underestimating, rather than overestimating, the strength of the discrimination provided by this feature. We have specifically employed this technique to avoid reinforcing any prior assumptions, but rather to determine a valid estimate of the significance of each of the features under analysis.

These imputations being given, we then conducted a multilevel logistic regression analysis,18 allowing not only the prevalence of abuse to vary between studies, but also the odds ratios (OR) for the features in question. By allowing the odds ratio to differ between studies, we again aim to minimise the risk of circularity, where an individual study may have overly relied upon a particular feature in order to arrive at the diagnosis of abuse. For each feature, we report the estimated odds ratio for the feature in discriminating between iBI and niBI, together with a 95% confidence interval (95% CI). We also provide a p value to determine if this odds ratio is significantly different from 1, and a positive predictive value (PPV), the estimated probability of abuse given the presence of this feature in a child with brain injury, together with a 95% CI.

Results

Of 320 reviewed studies, 14 met the inclusion criteria.14 15 16 19 20 21 22 23 24 25 26 27 28 All were published after 1995. Six were cross-sectional studies,19 22 24 27 28 29 five case series,14 16 20 23 25 two case–control studies15 26 and one a longitudinal cohort study.21 Eight studies involved prospective enrolment of cases.14 16 20 23 25 The six retrospective studies achieved full case ascertainment from hospital inpatient coding and/or radiology databases. Full study details are given in online appendix A.

Of the 306 studies excluded, the most common reasons for exclusion were the lack of comparative data or a rank of abuse of less than 1 or 2. A small number (11 studies) were excluded due to study design (expert opinion/review article, etc) and a further 10 included mixed adult and child data. Many studies were excluded on more than one basis (eg, rank of abuse/non-comparative/lack of detail on features for analysis).

Studies represented combined data on 1655 children, 779 of whom had suffered iBI and 876 niBI. Eleven of 14 studies included children less than 3 years of age.14 15 16 19 20 22 24 25 27 28 29 Seven studies ascertained all children who were hospitalised with traumatic head injuries,19 20 21 22 25 26 29 six studies included all children with subdural haemorrhage of any aetiology (trauma, coagulopathy, metabolic disorder or post infection, etc)14 15 16 23 25 28 and one study included children with subdural or extradural haemorrhage.27 The majority (seven) were conducted in the USA,19 20 21 22 24 26 27 five were European14 15 23 25 29 (three UK14 15 23) and one each was from Australia28 and Hong Kong.16 In nine studies, children were included solely on the basis of age and confirmation of cause. However, five had specific exclusion criteria (table 3).

Gender was recorded in 980 cases and was not a distinguishing feature. More boys than girls sustained brain injuries, regardless of aetiology (PPV of iBI if the child is a boy 0.397 (95% CI 0.277 to 0.530), OR 0.697, p>0.2).

It was not possible to undertake a formal analysis of age as a discriminating feature as each study reported it differently. Eight studies confirmed that children with iBI were significantly younger than those who sustained niBI.15 16 20 21 23 24 26 28 Of the eight studies addressing children less than 3 years of age, where mean ages were available,15 16 19 20 22 24 28 29 three studies found no significant difference between iBI and niBI.19 22 29 Five studies15 16 20 24 28 stated that children with iBI were significantly younger than those with niBI. Four of these studies15 16 24 28 gave the mean ages of children with iBI as less than or equal to 6 months, while the niBI groups ranged between 7.5 and 12.4 months.15 16 24 28

Combined feature analysis

It was not possible to identify which children had specific combinations of features, as this level of detail was not given.

Apnoea

Apnoea appeared to be a highly discriminatory finding, with a PPV of 93% (97.5% CI 0.733 to 0.986) and an OR of 17.062 (97.5% CI 5.018 to 58.011, p<0.001). However, it was one of the least recorded clinical items, only being analysed by two authors (405 children), perhaps reflecting that its significance has only been recognised more recently.15 30

Retinal haemorrhage

Although 1283 children were included in studies that considered this feature, only 998 had fundoscopy (670 iBI and 328 niBI). Many children presenting with witnessed trauma did not have formal fundoscopy, whereas the majority of children with iBI did. Therefore, to account for studies where this was not explicit,20 21 we used the conservative approach described above. Retinal haemorrhages were strongly associated with iBI, with a PPV of 71% (97.5% CI 0.483 to 0.868) and an OR of 3.504 (97.5% CI 1.088 to 11.280, p = 0.03). A child with an intracranial injury who has co-existent retinal haemorrhages is significantly more likely to have iBI than niBI.

Rib fractures

Rib fractures were considered in studies representing 1002 children, and data were available on 903 children. Rib fractures were associated with inflicted injury, with a PPV of 73% (97.5% CI 0.500 to 0.882) and an OR of 3.027 (97.5% CI 0.716 to 12.799). The wide confidence interval reflects the fact that relatively few children had data recorded on this item. Our conservative analysis therefore results in lower significance.

Seizures

Seizures were considered in studies involving 760 cases, and results were documented in 758. Seizures were more associated with iBI than niBI, but not significantly so, with a PPV of 66% (97.5% CI 0.454 to 0.821) and an OR of 2.924 (97.5% CI 0.731 to 11.694, p = 0.13).

Long bone fractures

One of the difficulties in analysing fractures as recorded in studies, is knowing whether the author is counting individual fractures, or individual children who have sustained fractures. In studies where this was not explicit,20 we made the assumption that the number of children with fractures was recorded, a conservative approach. Studies considering long bone fractures represent 1020 children, of whom 921 underwent skeletal imaging. Overall, the PPV of long bone fractures for iBI was 59% (97.5% CI 0.48 to 0.69), OR 1.722 (97.5% CI 0.824 to 3.601, p = 0.14). Therefore a long bone fracture in association with an intracranial injury was a weak predictor for iBI, but not significantly so.

Skull fractures and bruising to the head and neck

Skull fractures were considered in studies representing 1014 children, and documented as present or absent in 916 children. Skull fractures were more strongly associated with niBI than iBI, with a PPV for iBI of 44% (97.5% CI 0.223 to 0.678), OR 0.852 (97.5% CI 0.316 to 2.301, p>0.2). Likewise, bruising to the head and neck was more common in the niBI children, but it was the least recorded item, weakening its significance. Only 212 cases had this feature considered, and overall the PPV of head and/or neck bruising for iBI was 37% (97.5% CI 0.035 to 0.906), OR 0.811(97.5% CI 0.070 to 9.410, p>0.2). No specific mention was made of the use of 3D reconstruction techniques to define such fractures; the authors simply recorded their presence or absence when looked for.

Discussion

Making the clinical distinction between iBI and niBI in children is always going to be challenging. By producing a multilevel logistic regression of specific clinical features on over 1600 children, we have shown that there is scientific evidence to support the distinction between iBI and niBI, and we are able to offer positive predictive values and odds ratios for individual clinical indicators in children hospitalised with brain injury. The published literature does not allow an analysis of combined features.

This review is the largest of its kind, and offers for the first time a valid statistical probability of iBI when certain key features are present (eg, retinal haemorrhages). Although in some instances, for example, rib fractures, the sample size limits our statistical power to detect the full effect of this feature, as with any systematic review, we are of course limited by the amount of high quality data available in the published literature. However, even with this limited data set, the presence of rib fractures (PPV 73%, OR 3.027) is still indicates an association with iBI rather than niBI, and emphasises how important it is to conduct appropriate imaging to determine if such fractures are present.

It appears that the younger the child, the greater the likelihood that an intracranial injury is due to abuse. However, children as old as 10 years with severe iBI have also been described in isolated case reports.31 32 33 34

Apnoea appears to be a critical distinguishing feature (PPV for abuse 93%, OR 17.06, p<0.001). There is increasing evidence that hypoxic ischaemic injury and its attendant complications are an integral part of the cascade in iBI.4 The presence or absence of apnoea must be recorded in all cases of head injury in children.

Retinal haemorrhages are a valuable discriminator, reinforcing the need for fundoscopy on any child with unexplained intracranial injury, or suspected iBI. Indirect ophthalmoscopy should be conducted by an ophthalmologist, as studies have shown that other clinicians are either unable to conduct an accurate examination, or when they do, may miss up to 13% of retinal haemorrhages present.35

Fractures in physical abuse are often occult.36 37 Therefore, in line with the Royal College of Paediatrics and Child Health/Royal College of Radiologists standards for the radiological investigation of suspected physical abuse, any child aged less than 2 years in whom an inflicted injury is considered, should have a full skeletal survey, including oblique views of the ribs.38 Seizures were not a valuable discriminatory finding. Perhaps an analysis of specific patterns of seizures (eg, status epilepticus, complex, focal) would be more informative.

It was not possible to analyse the value of history given as a predictor, as a number of studies had used absence of a history of trauma or inconsistent explanation as part of their definition of abuse.20 21 Only one large study set out to estimate the value of the initial history in predicting whether the head injury was inflicted or non-inflicted.22 Study design ensured that the history did not contribute to the decision as to whether the child belonged to the iBI or niBI group. The study concluded that having no history of trauma had a high specificity (0.97) and high PPV (0.92) for inflicted injury.

Although many studies addressed encephalopathic features, the differing classifications that were used (altered consciousness, coma, irritability, Glasgow Coma Score (GCS)) precluded any combined analysis of this information. It is recognised that performing a GCS on young infants is imprecise; a study24 that noted an increased relative risk of this item for iBI (RR 1.7, 95% CI 0.9–3.2) did not attach weight to this feature.

We did not address social features which are more culturally specific, and less suited to meta-analysis. There is also the risk of introducing bias by undue reliance on social features, as shown by Jenny,3 who noted that white infants from two parent families were more likely to have their iBI missed than ethnic minority children or those living with one parent.

This review did not address biomechanical features. It is recognised that iBI can occur as a consequence of shaking,39 40 impact,41 shaking and impact,39 42 water intoxication,43 44 suffocation45 or strangulation.46 In our review skull fracture and bruising to the head/neck were more strongly associated with niBI than iBI, suggesting blunt trauma as a common mechanism of niBI. Clinical evidence of impact may only be found on post mortem,42 so conclusions as to whether an impact injury has occurred or not in iBI cannot reliably be drawn from the clinical features alone.

It is recognised that subdural haemorrhage (SDH) can occur in infancy from other causes, and consideration must be given to possible organic disease, such as metabolic disorders (eg, glutaric aciduria),6 albeit this has characteristic MRI findings. While coagulopathy may cause intracerebral haemorrhage,7 8 9 it may also be secondary to an iBI,47 highlighting the need for thorough investigations in this group. It is recognised that SDH in particular, may occur as a consequence of birth, particularly following instrumental delivery,48 49 50 although a neonatal SDH without skull fracture has also been described where the mother was assaulted ante-natally.51

While this review has highlighted certain clinical features that in association with intracranial injury have a high positive predictive value and odds ratio for iBI, such as retinal haemorrhage, apnoea or rib fractures, none of these features are exclusive to iBI. It is essential that the diagnosis of iBI is only made after taking into consideration full details on history (birth, medical, family, developmental, social, details of any reported trauma) and a detailed examination, and having actively excluded other relevant organic diseases. It is vital that future work should endeavour to collect data on all the features above, with detailed recording of precise abnormalities found, for example, on ophthalmology, if we are to refine these statistical correlations.

Baroness Kennedy has made it abundantly clear that clinicians must be able to cite scientific evidence to support their clinical opinions when providing evidence to the courts,52 and it is challenging for any clinician to keep up to date with all the current evidence (in the year 2007 alone 242 studies relating to iBI were published). Therefore, we feel that this unique, rigorously conducted systematic review will be a valuable resource for clinicians and other professionals in the field, providing, as it does, odds ratios of abuse for seven easily documented clinical features.