Âé¶¹´«Ã½Ó³»­

Âé¶¹´«Ã½Ó³»­

Explore the latest content from across our publications

Log In

Forgot Password?
Create New Account

Loading... please wait

Abstract Details

Use of Artificial Intelligence to Identify Risk Factors Predictive of Wheelchair Use in Patients with Fascioscapulohumeral Muscular Dystrophy (FSHD)
Neuromuscular and Clinical Neurophysiology (EMG)
Neuromuscular and Clinical Neurophysiology (EMG) Posters (7:00 AM-5:00 PM)
018

To identify risk factors for wheelchair use (WC) in people with facioscapulohumeral muscular dystrophy (FSHD) participating in the United States FSHD Registry.

FSHD is a dominantly inherited, slowly progressive muscular dystrophy caused by deletion of repeats in the D4Z4 region on chromosome 4. Normal individuals have >10 repeats; patients with FSHD type 1 have between 1-10 repeats. There is limited data about predictors of functional burden.

De-identified data from 578 participants with FSHD type 1 with an average of 9 years of follow-up reports were analyzed using epidemiological methods and artificial intelligence (AI) to assess interactions between characteristics including: age, gender, genetics (# of D4Z4 repeats), age of symptom onset and diagnosis, use of assistive devices, wheelchair use, job loss due to FSHD, and progressive functional burden/disability. These data were also used to develop AI random forest algorithms to identify risk factors that were predictive of WC. 

Small allele size (1-3 D4Z4 repeat units) was associated with earlier diagnosis (median 14 years, 95% confidence limit 11, 17), facial weakness as the initial symptom (53.7%), and higher risk of WC.  Our AI final model showed accuracy 0.79 and AUC 0.85, and revealed only a small contribution of genetics (lower risk for WC with 8-10 D4Z4 repeat units).  The most significant predictors of WC were: disease duration, number of medications, age at diagnosis or symptom onset, and medical comorbidity (e.g. breathing difficulties, pneumonia, or arthritis).  When looking at classes of medications, the model predicated that all classes influenced towards (+) WC except for amino acids.

Early AI modeling identified several features associated with WC use in FSHD, including number of medications and medical comorbidities, which might suggest aggressive medical management could be protective, but would require confirmation in additional data sets.
Authors/Disclosures
Natalie Katz, MD, PhD (Duke University)
PRESENTER
Dr. Katz has nothing to disclose.
No disclosure on file
No disclosure on file
No disclosure on file
No disclosure on file