Predicting Risk of Low Back Pain Using a Super Learner Algorithm

• Activity Number: 442 - Disease Prediction, Statistical Methods for Genetic Epidemiology and Mis
• Type: Contributed
• Date/Time: Thursday, August 12, 2021 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Hannah Michelle Brown* and Elizabeth Malloy and Jay Kapellusch
• Companies: American University and American University and University of Wisconsin–Milwaukee
• Keywords: Cox regression; cross validation; machine learning; occupational health; super learner

Abstract

Low back pain (LBP) can contribute to early retirement, increased disability compensation claims, and decreased quality of life for workers in labor-intensive industries. Employers can mitigate the frequency and severity of LBP through enhanced understanding of the impact of biomechanical, physiological, and psychophysical stressors. We examine a subset of 466 workers from a multi-centered prospective cohort study conducted in 4 different U.S. states. Data on demographic, medical history, psychosocial factors, pastime, physical activity, and pre-existing musculoskeletal disorders information was assessed. Due to the complexity of existing associations between risk factors and the response in this occupational health data, new analytic methods are necessary to improve the current estimation capabilities. Data-driven methods such as a super learner algorithm provide a solution by creating a weighted combination of candidate learners using cross-validation. We present an application of the super learner in the Cox proportional hazards regression setting to predict the risk of developing LBP based on several candidate measures.