Clinical Scorecard: Comparing Clinically-Guided and Foundation Models for Predicting Cervical Spondylotic Myelopathy Using Electronic Health Records
At a Glance
Category
Detail
Condition
Cervical spondylotic myelopathy (CSM), a progressive spinal cord dysfunction caused by degenerative cervical stenosis
Key Mechanisms
Degenerative spinal changes causing cervical stenosis leading to neurological deficits such as impaired hand dexterity, gait disturbances, bladder/bowel dysfunction, and paralysis
Target Population
Older adults, particularly those over 60 years with spinal cord compression or neurological symptoms
Care Setting
Primary care, spine surgery practices, and healthcare systems utilizing electronic health records (EHRs)
Key Highlights
CSM diagnosis is often delayed 2 to 6 years after symptom onset, leading to worse outcomes.
Approximately one-third of asymptomatic individuals over 60 show spinal cord compression; about 22% become symptomatic.
Machine learning models using large-scale EHR data can predict CSM incidence up to 30 months before diagnosis, enabling earlier intervention.
Guideline-Based Recommendations
Diagnosis
Early recognition of clinical features such as unsteady gait and neurological deficits is critical.
Use of targeted neurological assessments and advanced imaging is recommended for suspected cases.
Screening programs are currently limited to spine surgery practices due to practical constraints.
Management
Timely intervention is essential to prevent irreversible neurological damage.
Clinical decision support systems leveraging EHR-based prediction models may guide earlier diagnostic workups.
Monitoring & Follow-up
Monitor patients with risk factors or early symptoms closely to detect progression.
Utilize EHR data to track clinical encounters and relevant diagnostic studies over time.
Risks
Delayed diagnosis is associated with significant disease progression and poorer outcomes.
Limited awareness among primary care providers contributes to diagnostic delays.
Patient & Prescribing Data
Over 1.9 million patients across multiple datasets including 47,306 with CSM and over 1.9 million controls
Machine learning models demonstrated 5- to 7-fold improvement in predictive performance over non-informative classifiers, supporting their potential utility in early detection and management strategies.
Clinical Best Practices
Increase awareness of CSM clinical features among primary care providers to reduce diagnostic delays.
Incorporate machine learning-based clinical decision support tools into EHR systems to identify high-risk patients.
Prioritize early neurological assessments and imaging within 6 to 30 months of predicted risk to optimize outcomes.
by Salim Yakdan, Ben Warner, Zoher Ghogawala, Wilson Z. Ray, Mohamad Bydon, Michael P. Steinmetz, Richard T. Griffey, Randi Foraker, Adam Wilcox, Chenyang Lu, Jacob K. Greenberg
