BIG DATA ANALYTICS FOR EARLY DIAGNOSIS OF AMYOTROPHIC LATERAL SCLEROSIS (ALS)

OBJECTIVES: Analyze a large claims database to explore if early predictors of ALS can be identified and potentially shorten the diagnosis timeline. The average delay in ALS diagnosis is one year after the appearance of first symptom, which can be detrimental as it delays initiating approved treatments and may preclude patients from enrolling in clinical trials. METHODS: The Truven MarketScanli database, containing patient-level claims for 170+ million patients, was used without any code pre-selection for this analysis. A mutual information (MI) measure was used to quantify the statistical relevance of every code in Market-Scan to a future ALS diagnosis in four US states. Codes included: diagnosis codes, procedure codes, medications, provider types, and care facility types. An ensemble of classifiers developed employing machine learning techniques was applied to optimize the selection and ranking of ALS diagnosis predictors. We looked for predictors within the following time brackets: 3,6,9,12,18,24,36,48, and 60 months before the initial ALS diagnosis. RESULTS: The ALS ICD-9 diagnosis code identified 12,332 ALS patients with an average of 4.4 years of claims history. ALS patients: average age 60 years ± 14 years; 58% male, and 25% had a prescription claim for riluzole. The top differentiating diagnoses more common in ALS group compared to overall population were: non- traumatic joint disorders (-60 months), connective tissue diseases (-60 months), skin disorders (-48 months), fatigue (-36 months), lower respiratory diseases (-24 months), gastrointestinal disorders (-18 months) and other nervous system disorders (-12 months). CONCLUSIONS: This study suggests 5 years before ALS diagnosis, patients may be presenting with symptoms suggestive of connective tissue disorders, skin disorders, and nonspecific neurological complaints. Next steps of this project are to validate these findings in national dataset, optimize the algorithm differentiating ALS patients prior to diagnosis, and further characterize early predictors of ALS.