Background Abusive head trauma (AHT) is the most common cause of death and disability in abused children, and presents significant diagnostic challenges. Previous research identified six individual features (retinal haemorrhage, rib and long bone fractures, facial bruising, apnoea and seizures) associated with AHT to create a statistical model to determine the probability of AHT based upon different combinations of these features in a child with intracranial injury.

Aims The primary aim was to independently validate the statistical model on a novel dataset. The secondary aim was to look for association between AHT and the original six features, and further features not included in the original model, to suggest areas for refinement.

Methods Retrospective, notes-based review of 44 cases of children aged less than 36 months admitted with intracranial head injury (20 AHT), identified at neuroimaging (01/01/2007–31/02/2012). Sensitivity, Specificity, Positive Predictive Value (PPV) and Negative Predictive Value (NPV) were calculated to determine the model’s accuracy. Fisher’s Exact Test and logistic regression were used to test for association between individual features and AHT.

Significant association was found between AHT and retinal haemorrhage (p < 0.001), seizures (p < 0.02). Strong but not significant association was found between AHT and apnoea (p < 0.08), and between non-AHT and skull fracture (p < 0.25). Subdural haemorrhage, not included in the original model, was significantly associated with AHT (p < 0.04) On sub-analysis of retinal features, too numerous to count retinal haemorrhage was significantly associated with AHT (p < 0.04). Retinal haemorrhages were more likely to be multi-layered and bilateral in AHT cases.

Conclusions When tested on this dataset the model had similar sensitivity and specificity to the original study, although imputing data caused variation. Type of intracranial injury and specific retinal features were identified as areas for refinement. The high sensitivity suggests that the tool has the potential to identify cases of suspected AHT that Warrant further detailed assessment, and could be useful for clinical practise.