To develop and validate machine learning models for predicting the incidence of cervical spondylotic myelopathy (CSM) using electronic health records (EHRs), highlighting the significance of early diagnosis.
Key Findings:
The clmbr-t-5k-csm model achieved superior performance at most prediction horizons, with AUPRC ranging from 0.12 to 0.163, indicating its potential for clinical application.
The best-performing models showed a 5.07- to 6.9-fold improvement over a non-informative classifier, suggesting significant advancements in predictive accuracy.
The CEHBERT and CoreBEHRT models underperformed across all tested horizons, highlighting the need for further refinement.
Interpretation:
The study demonstrates the potential of machine learning models, particularly the clmbr-t-5k-csm, to enhance early diagnosis of CSM using EHR data, which could lead to timely interventions and improved patient outcomes.
Limitations:
The study's findings may not generalize across all healthcare systems due to the specific datasets used, and performance metrics may vary based on clinical context and patient demographics, such as age and comorbidities.
Conclusion:
Machine learning models, especially foundation models, can significantly improve the prediction of CSM, facilitating earlier diagnosis and intervention, which is crucial for preventing irreversible neurological damage.
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
