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.

This content is only available as a PDF.
2025
Sign in via your Institution