Systematic Review of Currently Available Predictive Machine Learning Models for Multiple Myeloma

Introduction: Multiple myeloma (MM) is a complex hematological malignancy characterized by the proliferation of clonal plasma cells in the bone marrow. Predicting disease progression, patient outcomes, and response to treatment in MM is crucial for improving clinical decision-making and patient management. With the advent of machine learning (ML) and artificial intelligence (AI), predictive models have been developed to enhance the accuracy of these predictions. This systematic review aims to evaluate the current landscape of ML and AI models used in MM, focusing on their methodologies, predictive performance, and clinical applicability.

Methods: A comprehensive search strategy was employed to identify relevant studies published up to 2024 in PubMed. The search terms included “multiple myeloma,” “machine learning,” “artificial intelligence,” “predictive models,” “prognosis,” and “survival.” Inclusion criteria encompassed studies that developed or validated ML models for predicting outcomes in MM patients. Data extracted included author, year, link, training type, focus of the study, and performance metrics such as Area Under the Curve (AUC), specificity, sensitivity, and F-score.

Results: A total of 26 studies met the inclusion criteria. The models focused on various aspects of MM, including:

1. Survival Prediction: Models predicting progression-free survival (PFS) and overall survival (OS) in MM patients.

2. Response to Treatment: Models assessing the effect of different treatment strategies and predicting patient response.

3. Disease Progression: Models forecasting disease biomarkers and minimal residual disease (MRD).

4. Diagnosis and Screening: Models aiding in the rapid screening and early diagnosis of MM.

5. Prognostic Factors: Models identifying prognostic factors and risk stratification.

Key Findings:

• Hussain et al. (2024) developed a deep learning transformer-based architecture that predicted PFS, OS, and adverse events (AEs), with AUCs ranging from 0.85 to 0.90.

• Ren et al. (2023) utilized Cox regression analysis and LASSO to select ubiquitin proteasome pathway-associated genes correlated with OS, achieving an AUC of 0.82.

• Peng et al. (2023) created an infection prediction model for newly diagnosed MM (NDMM) patients with an AUC of 0.78.

• Zhong et al. (2023) identified 5 PET and 4 CT-based features, along with 6 clinical features, significantly related to PFS, reporting an AUC of 0.84.

• Wennmann et al. (2023) used deep learning and radiomics to predict bone marrow biopsy results from MRI, with an AUC of 0.86.

• Cai et al. (2024) proposed an RF model for rapid screening of MM patients, achieving a sensitivity of 0.87 and specificity of 0.81.

The reported performance metrics for the models varied, with AUCs ranging from 0.78 to 0.90, sensitivities from 0.75 to 0.87, specificities from 0.70 to 0.85, and F-scores between 0.75 and 0.85.

Conclusion: The application of ML and AI in MM has shown promising results in improving the prediction of patient outcomes, response to treatment, and disease progression. The diverse range of models and methodologies highlights the potential of these technologies in enhancing clinical decision-making. Future research should focus on the integration of multi-omics data, longitudinal patient monitoring, and the development of interpretable models to facilitate their adoption in clinical practice.

No relevant conflicts of interest to declare.