Sequential Pattern Mining of Electronic Health Record for Early Diagnosis of Amyotrophic Lateral Sclerosis

• Activity Number: 90 - Novel Statistical Methods for COVID Pandemic and Other Current Health Policy Issues
• Type: Contributed
• Date/Time: Monday, August 9, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Ying Liu* and Cindy Liang and Lily Sun and Jeffery Zhang
• Companies: Princeton Pharmatech and Texas Academy of Mathematics and Science and Stanford OHS and Princeton Pharmatech
• Keywords: machine learning; ALS diagnosis; sequential pattern mining; big data; bioinformatics; epidemiology

Abstract

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease primarily affecting the upper and lower motor neurons. Average survival time for ALS patients is 19 months from the time of diagnosis and 30 months from symptom onset. Diagnosis is primarily based on clinical evaluation along with a series of tests to rule out other mimicking diseases but remains challenging with an average diagnostic delay of up to more than 12 months after the onset of symptoms. Thus, early diagnosis of ALS is critical to prolonging survival and improving quality of life. A possible tool that could be used for early detection of ALS is sequential pattern mining: a technique used to identify patterns of ordered events. In this study, we use sequential pattern mining to predict ALS based on electronic health records. Our objective is to determine whether sequential pattern mining is effective for early ALS diagnosis based on electronic health records (EHR). We use cSpade, a version of SPADE (Sequential Pattern Discovery using Equivalence classes), to mine patients’ history of prior ALS diagnosis to observe frequent patterns and evaluate how useful these patterns are when making a prediction of ALS.