A predictive model for amyotrophic lateral sclerosis (ALS) diagnosis

Abstract

Objective The clinical diagnosis of amyotrophic lateral sclerosis (ALS) usually takes several months. The delay in diagnosis compromises the effective therapeutic interventions. Therefore, the present study was aimed to develop a statistical model for predicting the risk of ALS at earlier stages for better management of ALS patients. Methods The study recruited 44 sporadic ALS patients and 29 normal controls. Thirteen different independent variables (predictors) which were believed to be associated with ALS were included in the study. Forward stepwise (likelihood ratio) binary logistic regression was used to find significant variables and probability of disease prediction. Results The Hosmer–Lemeshow goodness of fit statistic (χ 2= 4.379, df=8, p=0.821) indicate the appropriateness of forward stepwise (likelihood ratio) binary logistic regression model. Serum chemokine ligand-2, chemokine ligand-2 mRNA, vascular endothelial growth factor-A mRNA, smoking and alcohol consumption are the independent variables found significant to predict risk of ALS (p<0.05). The current model yielded 93.2% sensitivity and 86.2% specificity with 90.4% overall validity of correct ALS prediction. Conclusion Forward stepwise (likelihood ratio) binary logistic regression model is an accurate method to predict ALS in the presence of serum CCL2, CCL2 mRNA, VEGFA mRNA, smoking and alcohol consumption with high sensitivity and specificity. However, bed side diagnostic utility of these variables needs to be validated further in larger ALS cohorts.