AI Model Could Help Diagnose Spinal Cord Disease Up To 30 Months Earlier

Cervical spondylotic myelopathy (CSM) is the leading cause of spinal cord dysfunction in older adults and occurs when arthritis in the neck compresses the spinal cord. The condition is chronic and progressive, often causing neck pain, muscle weakness, walking difficulties and other disabling symptoms. Diagnosis can take years because early signs are frequently overlooked, and by the time it is recognized, treatment options may be limited. Researchers have now developed an artificial intelligence (AI)-based method that can identify individuals at risk of CSM up to 30 months before clinical diagnosis, potentially enabling earlier intervention and improved outcomes.

A multidisciplinary team of surgeon-scientists and computer scientists at Washington University in St. Louis (WashU, St. Louis, MO, USA) created and evaluated seven different AI models to analyze electronic health record data from more than two million individuals with and without CSM. The models examined patterns of health care interactions, including diagnostic tests and recorded conditions, to identify patients whose medical histories resemble those already diagnosed with CSM. Both large foundation models pretrained on extensive clinical datasets and smaller, clinically informed models focusing on relevant variables, were trained and tested to predict risk well before formal diagnosis.

The models were trained using a large external dataset and a smaller dataset from a St. Louis–based health system to assess prediction accuracy across different time horizons. The foundation models performed best during internal validation on a large, heterogeneous dataset, while the smaller, clinically derived model demonstrated stronger generalizability and more consistent results across external systems. Two mid-scale models underperformed across the evaluated time points. The findings, published in npj Digital Medicine, show that risk for CSM can be predicted as early as 30 months before a clinical diagnosis, with simpler models incorporating clinical insight achieving comparable or superior performance.

Earlier identification of CSM risk could allow clinicians to intervene at a stage when treatment may prevent further neurological decline. By screening electronic health records for subtle warning patterns, the AI system may help flag patients who would otherwise remain undiagnosed until advanced stages. The researchers suggest that combining deep learning with established clinical knowledge may improve AI applications across health care. Future work will focus on refining the models and evaluating their integration into real-world clinical workflows.

“We were able to achieve at least comparable, if not superior performance with a much, much simpler model by focusing on existing clinical knowledge while still using a deep learning model,” said Assistant Professor Jacob Greenberg, MD, co-senior author of the study. “AI clearly has emerging opportunities in medicine, but we often focus only on the areas where purely data-driven solutions excel. There’s still an important role for clinical knowledge, which is going to be true for a lot of applications in health care.”

Related Links:
WashU