Estimating the Probability of Abusive Head Trauma: A Pooled Analysis

TLDR

The study aims to identify combinations of clinical features that differentiate abusive head trauma from nonabusive head trauma and to develop a refined predictive tool for clinical decision‑making. The authors pooled individual patient data from six comparative studies of children under three with intracranial injury, analyzed associations between AHT and six clinical features, and applied a bespoke hot‑deck multiple‑imputation strategy to handle missing data. The analysis of 1,053 children showed that combinations of specific clinical features yield highly variable PPVs for AHT, with apnea alone OR 6.89, rib fracture or retinal hemorrhage plus any other feature OR > 100 (PPV > 85 %), and any three or more of the six significant features OR > 100 (PPV > 85 %).

Abstract

CONTEXT AND OBJECTIVE: To determine which combinations of clinical features assist in distinguishing abusive head trauma (AHT) from nonabusive head trauma. METHODS: Individual patient data from 6 comparative studies of children younger than 3 years with intracranial injury were analyzed to determine the association between AHT and combinations of apnea; retinal hemorrhage; rib, skull, and long-bone fractures; seizures; and head and/or neck bruising. An aggregate analysis of data from these studies used multiple imputation of combined clinical features using a bespoke hotdeck imputation strategy, which accounted for uncertainty arising from missing information. RESULTS: Analyzing 1053 children (348 had AHT), excluding nonsignificant variables (gender, age, skull fractures), for a child with an intracranial injury and 1 or 2 of the 6 features, the positive predictive value (PPV) of AHT varies from 4% to 97% according to the different combinations. Although rarely recorded, apnea is significantly associated with AHT (odds ratio [OR]: 6.89 [confidence interval: 2.08–22.86]). When rib fracture or retinal hemorrhage was present with any 1 of the other features, the OR for AHT is >100 (PPV > 85%). Any combination of 3 or more of the 6 significant features yielded an OR of >100 (PPV for AHT > 85%). CONCLUSIONS: Probabilities of AHT can be estimated on the basis of different combinations of clinical features. The model could be further developed in a prospective large-scale study, with an expanded clinical data set, to contribute to a more refined tool to inform clinical decisions about the likelihood of AHT.