Survival prediction models in motor neuron disease

Background and purpose: This study aimed to assess the predictive value of multimodal brain magnetic resonance imaging (MRI) on survival in a large cohort of patients with motor neuron disease (MND), in combination with clinical and cognitive features. Methods: Two hundred MND patients were followed up prospectively for a median of 4.13 years. At baseline, subjects underwent neurological examination, cognitive assessment and brain MRI. Grey matter volumes of cortical and subcortical structures and diffusion tensor MRI metrics of white matter tracts were obtained. A multivariable Royston–Parmar survival model was created using clinical and cognitive variables. The increase of survival prediction accuracy provided by MRI variables was assessed. Results: The multivariable clinical model included predominant upper or lower motor neuron presentations and diagnostic delay as significant prognostic predictors, reaching an area under the receiver operating characteristic curve (AUC) of a 4-year survival prediction of 0.79. The combined clinical and MRI model including selected grey matter fronto-temporal volumes and diffusion tensor MRI metrics of the corticospinal and extra-motor tracts reached an AUC of 0.89. Considering amyotrophic lateral sclerosis patients only, the clinical model including diagnostic delay and semantic fluency scores provided an AUC of 0.62, whereas the combined clinical and MRI model reached an AUC of 0.77. Conclusion: Our study demonstrated that brain MRI measures of motor and extra-motor structural damage, when combined with clinical and cognitive features, are useful predictors of survival in patients with MND, particularly when a diagnosis of amyotrophic lateral sclerosis is made.