Dietary patterns and psoriasis severity in Thai patients: a machine learning approach for small sample data.

Area of Science:

• Dermatology and Computational Biology
• Nutritional Science and Machine Learning

Background:

• Psoriasis severity is influenced by various factors, including diet.
• Machine learning (ML) offers advanced tools for analyzing complex clinical datasets.
• Understanding dietary impacts on psoriasis is crucial for patient management.

Purpose of the Study:

• To investigate the relationship between dietary patterns and psoriasis severity using ML.
• To identify specific dietary triggers and protective factors for psoriasis.
• To develop predictive models for psoriasis severity based on clinical and dietary data.

Main Methods:

• Utilized a dataset of 142 Thai psoriasis patients with 37 demographic, clinical, and dietary features.
• Employed a hybrid resampling strategy (bootstrapping with K-fold Cross-Validation) to address small sample size.
• Evaluated Random Forest (RF) and eXtreme Gradient Boosting (XGB) models with feature selection (LASSO, MDA, MDI).
• Applied SHapley Additive exPlanations (SHAP) for feature importance analysis.

Main Results:

• Optimal RF and XGB models achieved >90% sensitivity, specificity, and F1-scores, with AUC >95%.
• Key dietary triggers identified: high-sodium foods, processed meats, alcohol, red meats, fermented products, and dark-colored vegetables.
• Body Mass Index (BMI) was the strongest predictor of psoriasis severity.
• Certain Thai dishes like Tom Yum Kung showed potential benefits, while others like Som Tum and Pad Thai were potential triggers when consumed excessively.

Conclusions:

• ML models effectively predict psoriasis severity based on dietary and clinical factors.
• Dietary moderation and personalized nutritional guidance are essential for managing psoriasis.
• Identifying specific food triggers and promoting weight management can improve patient outcomes and healthcare innovation.