The added prognostic value of magnetic resonance imaging in traumatic brain injury: The importance of traumatic axonal injury when performing ordinal logistic regression

Background and purposeThis study was performed to investigate the prognostic value of traumatic axonal injury (TAI) in severe head trauma.MethodsWe attempted to determine whether any MR imaging findings of TAI could be related to prognosis in 264 patients with severe head trauma. We performed an ordinal logistic regression, adjusted for the prognostic factors according to the IMPACT studies, adding each MR feature related to prognosis one at a time. A new prognostic model was described by adding these MR features to the classic prognostic factors. The model was externally validated in a prospective series. Harrel's c-statistic and ordinal c-index (ORC) were calculated to measure its predictive accuracy.ResultsWe found 178 patients with TAI lesions. Lesions in the basal ganglia/thalamus, corpus callosum (CC) and brain stem were associated with poor outcome (P < 0.01). The highest OR was for TAI lesions in the splenium (OR: 2.6) and brain stem dorsal lesions (OR: 3.1). We only found significant differences in outcome between haemorrhagic and non-haemorrhagic TAI lesions in the subgroup of patients with white matter and basal ganglia/thalamus lesions (P = 0.01). We obtained a superior discriminatory capacity by adding these MR findings to the previous prognostic model (Harrel's c-statistic 0.72 and ORC 0.7) in a prospective series of 93 patients.ConclusionsThe prognostic model including MR findings maintained a superior discriminatory capacity than that obtained for the model with the classic prognostic factors alone.