Examining factors associated with HRQOL among people with MS who use wheelchairs/scooters: Machine learning approach

Research Objectives: To examine factors associated with health-related quality of life (HRQOL) among people living with multiple sclerosis (pwMS) who use wheelchairs or scooters full-time and to compare the statistical predictive accuracy between traditional statistical modeling; linear regression (LR), and the machine learning algorithm; decision tree (DT).

Design: Cross-sectional study

Setting: University-based and clinical-based research laboratories

Participants: Forty-nine pwMS who use wheelchairs (n=41) or scooters (n=8) full-time (PDDS 6 -7): average age 60.3 ± 11.4 years (40 male) average MS duration 23.5 ± 12.5 years, average wheelchair/scooter use duration 8.2 ± 4.9 years.

Interventions: Not applicable

Main Outcome Measures: The dependent variable, the participant’s HRQOL, is indexed by the Multiple Sclerosis Quality of Life-54 (MSQOL-54). The independent variables contained 36 features, which were comprised of demographic information and validated questionnaires measuring cognitions, falls, fear of falling, and community participation.

Results: The model that the machine learning DT algorithm created (RMSE= 7.299, R2= 0.975) has significantly higher statistical predictive accuracy compared to the statistical model using the LR (RMSE= 45.939, R2= 0.879) as indicated by higher root mean square error. The DT analysis identified the five factors associated with MSQOL-54 among pwMS who use wheelchairs or scooters full-time: community participation, cognition, weight of the participant, fear of falling, and wheelchair/scooter usage hours per week.

Conclusions: This pilot study indicates the applicability of machine learning, which has the potential to improve the accuracy of statistical prediction in clinical research. Machine learning may help identify the most relevant factors associated with HRQOL among pwMS who use wheelchairs/scooters full-time, which clinicians need to consider in their practice.

Author(s) Disclosures: The authors have no conflicts of interest to disclose.