Artificial intelligence for predicting transformation of monoclonal gammopathy of undetermined significance to multiple myeloma: A systematic review and meta-analysis

Abstract Objectives The purpose of this study was to assess the empirical evidence of the efficacy of AI in predicting the transformation of Monoclonal Gammopathy of Undetermined Significance (MGUS) to Multiple Myeloma (MM). Methods A comprehensive and systematic electronic database search was performed in Scopus, PubMed, Cochrane Library, ScienceDirect, and Google Scholar. Modified PICOS criteria were used to screen and select the eligible studies from the potential articles retrieved from the database search. Studies were considered if they included patients with MGUS whose progression was monitored using AI approaches. The selected studies were assessed for risk of bias using the Newcastle-Ottawa Scale (NOS). Data was then procedurally extracted and analyzed. Results The study selection process identified nine studies, including 42,853 patients. Ensemble methods (ElasticNet, GBM, Random Forest) consistently outperformed traditional risk stratification systems, with AI models achieving C-statistics of 0.692-0.879 compared to 0.533-0.670 for conventional IMWG/2-20-20 criteria. The meta-analysis demonstrated the favourable predictive performance of AI models for predicting MGUS to MM Transformation, with a pooled AUC of 0.824 (95% CI: 0.785-0.858, p< 0.001). The multi- modal integration of clinical parameters, genomic profiles, and cytogenetic markers enhanced the discriminative capacity. Conclusion AI models demonstrated high prediction accuracy for the transformation of MGUS to MM. In addition, various AI models integrate multimodal biological data, transforming complex genomic, cytogenetic, and clinical information into actionable risk assessments influencing surveillance intensity and intervention timing. KEYWORDS Artificial Intelligence; Monoclonal Gammopathy of Undetermined Significance; Multiple Myeloma.