A Deep Learning Model to Predict Knee Osteoarthritis Based on Nonimage Longitudinal Medical Record

Area of Science:

• Medical informatics
• Artificial intelligence in healthcare
• Osteoarthritis research

Background:

• Knee osteoarthritis (KOA) is a prevalent degenerative joint disease.
• Early prediction of KOA risk is crucial for timely intervention.
• Non-image-based electronic medical record (EMR) data offers a rich source for predictive modeling.

Purpose of the Study:

• To develop and validate a deep learning model, Deep-KOA, for predicting incident knee osteoarthritis (KOA) risk.
• To utilize three years of non-image-based EMR data for KOA risk prediction.
• To enable early identification of individuals at high risk for KOA.

Main Methods:

• A deep learning model (Deep-KOA) was developed using convolutional neural network (CNN) and artificial neural network (ANN).
• Data from two million patients in the Taiwan National Health Insurance Research Database (NHIRD) from 1999-2013 was analyzed.
• A feature matrix was constructed using three-year sequential diagnoses, drug prescriptions, age, and sex.

Main Results:

• Deep-KOA achieved high performance with an AUROC of 0.97, sensitivity of 0.89, specificity of 0.93, and precision of 0.80.
• Key predictive features included diagnoses of eye, respiratory, metabolic, and musculoskeletal disorders, as well as medications like antacids and cough suppressants.
• Age and sex were less discriminative compared to specific diagnoses and medications.

Conclusions:

• Deep-KOA effectively predicts the risk of knee osteoarthritis (KOA) one year in advance.
• The model can serve as a clinical decision support tool for identifying high-risk patients.
• This facilitates targeted precision prevention programs for knee osteoarthritis.