Combining Clinical, Genetic and Protein Markers Using Machine Learning Models Discriminates Psoriatic Arthritis Patients From Those With Psoriasis.

BACKGROUND: Psoriatic Arthritis (PsA), an immune mediated inflammatory arthritis, affects a quarter of patients with cutaneous psoriasis, usually after psoriasis onset. Early diagnosis of PsA is challenging. A biomarker-based diagnostic test may facilitate early diagnosis. OBJECTIVES: We aimed to determine whether specific clinical features or genetic and protein markers, alone or in combination, can distinguish patients with PsA from those with psoriasis without PsA (PsC). METHODS: Patients with PsA and PsC were identified from a database of patients with psoriatic disease. Detailed demographic and clinical information were collected at time of assessment. Single-nucleotide polymorphisms (SNPs) of 19 "PsA weighted" genes were genotyped. Serum samples were used to assess 15 protein markers by ELISA. Association between clinical, genetic and protein markers and PsA were determined, and models were developed to discriminate PsA from PsC using machine learning algorithms. RESULTS: Demographic and clinical information had low predictive value in distinguishing PsA from PsC (AUC - 0.607, P < .01). SNP and protein panels also had low value in discriminating PsA from PsC (AUC - 0.691, P < .001 and AUC - 0.694, P < .001, respectively). Combining protein, SNPs and clinical features provided better discriminatory value (best performing model: Random Forest, AUC - 0.733, P < .001). CONCLUSION: Combining previously identified clinical, genetic and protein markers have a fair ability to differentiate PsA from PsC. Further studies are required for identifying better diagnostic signatures.