Combining Clinical, Genetic and Protein Markers Using Machine Learning Models Discriminates Psoriatic Arthritis...

Area of Science:

• Immunology
• Rheumatology
• Genetics

Background:

• Psoriatic arthritis (PsA) is an immune-mediated inflammatory condition affecting psoriasis patients.
• Early diagnosis of PsA is challenging, hindering timely treatment.
• Biomarker-based tests could improve early PsA detection.

Purpose of the Study:

• To identify clinical, genetic, and protein markers for distinguishing PsA from psoriasis without arthritis (PsC).
• To evaluate the combined diagnostic power of these markers using machine learning.

Main Methods:

• Collected demographic and clinical data from PsA and PsC patients.
• Genotyped 19 "PsA weighted" single-nucleotide polymorphisms (SNPs).
• Assessed 15 serum protein markers using ELISA and applied machine learning for discrimination.

Main Results:

• Clinical features alone had limited predictive value (AUC=0.607).
• SNP and protein panels showed moderate discrimination (AUCs ≈ 0.69).
• A combined model (Random Forest) integrating all markers achieved the best discrimination (AUC=0.733).

Conclusions:

• A combination of clinical, genetic, and protein markers offers fair diagnostic ability for PsA.
• Further research is needed to discover more effective diagnostic signatures for PsA.