Scholarlink

Multiple myeloma (MM) is a heterogenous plasma cell malignancy with rising worldwide incidence. While traditional staging systems (e.g., R-ISS) are not genomics-aware, machine learning models (e.g., SCOPE and IRMMa) predict outcomes such as overall survival (OS) and treatment response on multi-modal data. This narrative review (1) contrasts 85 studies (2015–2025) to emphasize challenges to clinical utility, e.g., data heterogeneity and limited sample sizes, and (2) proposes federated learning as a novel solution to privacy-preserving model training. The main findings are that transformer-based models improve AUC by 15–20% compared to traditional methods but require validation in prospective studies. By resolving these issues, predictive modeling can streamline CAR-T therapy selection and mitigate healthcare disparities, an important step toward precision medicine in MM.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Challenges in Predictive Modeling in Multiple Myeloma: A Narrative Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sohaib Irfan,
  • Rao Nouman Ali,
  • Nausheen Ahmed,
  • Bilal Hassan

摘要

Multiple myeloma (MM) is a heterogenous plasma cell malignancy with rising worldwide incidence. While traditional staging systems (e.g., R-ISS) are not genomics-aware, machine learning models (e.g., SCOPE and IRMMa) predict outcomes such as overall survival (OS) and treatment response on multi-modal data. This narrative review (1) contrasts 85 studies (2015–2025) to emphasize challenges to clinical utility, e.g., data heterogeneity and limited sample sizes, and (2) proposes federated learning as a novel solution to privacy-preserving model training. The main findings are that transformer-based models improve AUC by 15–20% compared to traditional methods but require validation in prospective studies. By resolving these issues, predictive modeling can streamline CAR-T therapy selection and mitigate healthcare disparities, an important step toward precision medicine in MM.