Using a latent class model to refine risk stratification in multiple myeloma

Area of Science:

• Oncology
• Genomics
• Biostatistics

Background:

• The GEP70 model uses 70 gene expression levels to predict high-risk multiple myeloma patients.
• Current GEP70 risk stratification relies on a continuous gene score and a binary high/low-risk classification.
• The continuous nature of the GEP70 score and its binary cutoff present interpretation challenges and sensitivity issues.

Purpose of the Study:

• To develop a more interpretable risk prediction method for multiple myeloma.
• To refine the existing GEP70 model for improved patient risk stratification.
• To convert the continuous GEP70 gene score into a probabilistic output.

Main Methods:

• Latent class modeling was employed to address the continuous score issue.
• A novel grey zone model was proposed to enhance risk stratification.
• Model robustness was assessed using a simulation study.

Main Results:

• The proposed grey zone model offers a superior refinement of the GEP70 risk stratification.
• Latent class modeling facilitates the conversion of continuous scores into interpretable probabilities.
• The grey zone model demonstrates robustness in risk stratification.

Conclusions:

• The developed grey zone model improves upon the GEP70 model for multiple myeloma prognosis.
• Probabilistic outputs enhance patient and clinician understanding of disease progression risk.
• Advanced statistical modeling can refine existing genomic biomarkers for clinical utility.