Machine Learning Approaches for Predicting Psoriatic Arthritis Risk Using Electronic Medical Records:...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Rheumatology

Background:

• Psoriasis (PsO) is a chronic, systemic, immune-mediated disease.
• Psoriatic arthritis (PsA) affects 6%-42% of PsO patients, with 15% undiagnosed.
• Early PsA detection is crucial to prevent irreversible disease progression.

Purpose of the Study:

• Develop and validate a machine learning prediction model for PsA risk.
• Utilize chronological, multidimensional electronic medical records for prediction.
• Identify PsO patients at high risk for PsA for early intervention.

Main Methods:

• Case-control study using Taiwan's National Health Insurance Research Database (1999-2013).
• A convolutional neural network model trained on 2.5-year patient records.
• Model predicts PsA risk within 6 months using sequential diagnostic and prescription data.

Main Results:

• The model achieved an area under the receiver operating characteristic curve of 0.70.
• Mean sensitivity was 0.80, mean specificity 0.60, and mean negative predictive value 0.93.
• Occlusion sensitivity analysis identified key predictive features.

Conclusions:

• The developed risk prediction model can identify PsO patients at high risk of developing PsA.
• This tool can assist healthcare professionals in prioritizing high-risk patients for timely treatment.
• Early intervention guided by the model may prevent irreversible disease progression and functional loss.