Assessment of Predictive Factors That Shorten Duration of Treatment in Patients With Multiple Myeloma Using AI:...

Area of Science:

• Oncology
• Data Science
• Pharmacoeconomics

Background:

• Treatment duration for Japanese multiple myeloma (MM) patients shortens with each subsequent line of therapy.
• Understanding factors influencing treatment duration is crucial for optimizing patient management.

Purpose of the Study:

• To identify factors associated with shortened duration of therapy (DoT) in Japanese MM patients.
• To leverage machine learning (ML) for analyzing real-world data and predicting DoT.

Main Methods:

• A nationwide, retrospective observational study using the Medical Data Vision (MDV) claims database (2003-2022).
• An explainable deep learning model (point-wise linear, PWL) was developed using 647 variables to predict DoT.
• Model performance was validated against elastic net and extreme gradient boosting, with clustering analysis (k-means) applied to patient samples.

Main Results:

• The PWL model achieved area under the curve scores of 0.61, 0.64, and 0.66 for predicting 3, 6, and 12-month DoT, respectively.
• Patients in clusters associated with shorter DoT exhibited higher pretreatment Charlson Comorbidity Index scores.
• Increased use of immunomodulatory drugs and aspirin was significantly associated with shorter DoT in specific patient clusters.

Conclusions:

• Machine learning, particularly the PWL model, effectively identified trends and patient characteristics linked to shortened DoT in Japanese MM patients.
• Disease status and management factors, including immunomodulatory drug use and thromboprophylaxis, are associated with DoT length.
• These findings can inform clinical practice and treatment strategies for MM.