Bayesian Diagnosis of Amyotrophic Lateral Sclerosis ( ALS ) to improve Clinical Trial Eligibility

Amyotrophic Lateral Sclerosis (ALS) is unique among the neuromuscular diseases (NMD) in that it’s diagnosis is clinical, and cannot be made with genetic tests, biopsies, or imaging. Current ALS diagnostic criteria reflect what “the experts” believe ALS is, not necessarily what ALS really is. Consequently, some 10 to 20% of ALS patients do not meet current diagnostic criteria. Bayesian diagnosis ignores expert opinion in favor of data; in this case we used a dataset of over 1000 patients, half with ALS, and half with conditions from which ALS must be distinguished. Using a logistic regression model, “Bayes Factors” (BF or odds ratios), were determined for the key clinical signs that distinguish ALS from other NMD; tongue atrophy, tongue fasciculations, deep tendon reflexes, Babinski sign, and muscle tone. The BF of a clinical sign is the odds ratio that a patient with that sign has ALS. The BF’s provide a relative weighting for each clinical sign, and can be combined when more than one sign is present, to provide a final “Odds of ALS” for that patient. When applied to a new dataset of patients with progressive weakness and motor disability suggestive of ALS, the Bayesian method shows that even people who, by current criteria, do not have ALS or who have “suspected or possible “ ALS, have extremely high “odds of ALS”. As new drugs become available for ALS trials, this Bayesian diagnosis will allow the inclusion of more subjects, especially those early in the course of the disease.