Int J Legal Med. 2026 May 19. doi: 10.1007/s00414-026-03831-z. Online ahead of print.
ABSTRACT
BACKGROUND AND OBJECTIVE: A Likelihood ratio (LR) is a numerical measure of evidential value. Our objective was to use LRs to express the patient-specific evidential values of medical findings that best differentiate abusive versus non-abusive head trauma (AHT). We hypothesized that the evidential values of patients’ AHT-related medical findings would be highly variable.
METHODS: We analyzed existing, uniform, prospective, de-identified data regarding 973 acutely head-injured children < 3 years hospitalized for intensive care across 18 sites between 2011 and 2021; applied two different proxies for AHT and non-AHT ground truth; trained and validated statistical models that differentiate AHT versus non-AHT; and analyzed patient-specific LRs in a log10 (LLR) format that facilitated assessment of evidential values (where LLR values > 0 and < 0 supported hypotheses of AHT and non-AHT, respectively).
RESULTS: The two best performing statistical models revealed evidential (LLR) values for patient-specific, AHT-related medical findings that varied from modest (-1 to + 1) to relatively large (-2.5 to -1 and + 1 to + 3.5), and values that were misleading (AHT patients with LLR values indicative of non-AHT, and vice versa). A few non-AHT patients presented with misleading evidence that was moderately strong, with LLRs approaching + 3.
CONCLUSIONS: Reasonable medical certainty of AHT and non-AHT can be enhanced or limited by the highly variable evidential values of patients’ most discriminating medical findings. Physicians can use available LRs to inform their AHT-related diagnostic reasoning, opinions, and testimony.
PMID:42151630 | DOI:10.1007/s00414-026-03831-z
