A Predictive Model for Prognosis in Motor Neuron Disease

Abstract

Objectives: There have been attempts to establish biomarkers for motor neuron disease, without success. The study aim to seek possible markers to be used in the clinical routine evaluation, to optimize timing for palliative interventions.Methods: A cohort study evaluated clinical, respiratory and neurophysiological variables every 3-4 months across 20 months in 101 patients with motor neuron disease using riluzole. Primary endpoint was death or tracheostomy. The most significant parameters in cox regression analysis created a predictive model.Results: There were 58 men and 43 women with a mean age of 57.2 ± 11.7 years. 77 patients (76.2%) had spinal onset and 24 (23.8%) had bulbar onset. The mean survival time was 43.5 ± 5.7 months ). The variables related to worse prognosis were: age > 65 years (HR=2.50, CI 95% (1.23-5.08); involvement of a second site in less than six months (HR=2.02, CI 95% (1.04 -3.94); supine Forced Vital Capacity <63% (HR=2.78, CI 95% 1.03-7.48), neck weakness (HR=2.28, CI 95% (1.03-5.05) and presence of pyramidal syndrome (HR= 2.36, CI 95% (1.05-5.33). Conclusion:It was created a five-factors set that predicts evolution to death or tracheostomy within one year.

Applications of machine learning to diagnosis and treatment of neurodegenerative diseases

“…Such an approach might prove particularly useful in diseases characterized by aggressive decline and poor survival such as MND, for which the average survival after diagnosis is 3-5 years 160 . Existing statistical models of MND use data from routine clinical assessments to predict patient prognosis for up to one year 161 , and recent improvements to this approach can make personalized predictions for up to 120 months depending on the disease severity 162 . These models, however, do not take into consideration an individual's previous medical history, which could be informative, and missing data can result in biases.…”