Diagnostic accuracy of machine learning approaches to identify psoriatic arthritis: a meta-analysis

While machine learning (ML) approaches are commonly utilized in medical diagnostics, the accuracy of these methods in identifying psoriatic arthritis (PsA) remains uncertain. To evaluate the accuracy of ML approaches in the medical diagnosis of PsA. As a result, we thoroughly searched PubMed, Web of Science (WoS), Embase, Scopus, Cochrane Library, Wanfang, and the Chinese National Knowledge Infrastructure (CNKI) between their inception and October 1, 2024. The overall test performance of ML approaches was evaluated using the following metrics: pooled sensitivity, pooled specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), the area under the curve (AUC), and Fagan plot analysis. Additionally, we assessed the publication bias using the asymmetry test of the Deeks funnel plot. Six studies were included. The combined diagnostic data showed sensitivity of 0.72 (95% CI 0.60-0.81), specificity of 0.81 (95% CI 0.61-0.92), PLR of 4.00 (95% CI 3.06-5.23), NLR of 0.41 (95% CI 0.34-0.49), DOR of 11.06 (95% CI 6.42-19.06), and AUC of 0.81 (95% CI 0.78-0.84). The Fagan plot showed that the positive probability is 48% and the negative probability is 8%. Meta-regression identified country and sample size (all P < 0.05) as key sources of heterogeneity. The Deek funnel plot suggested that publication bias has no statistical significance (P = 0.99). The study suggests a promising accuracy of ML approaches in diagnosing PsA.