844P - Progression -free survival prediction of multiple myeloma patients in five European countries using machine learning models

Background

The objective of this study is to apply machine learning (ML) models to multiple myeloma(MM) patient-level data to evaluate if ML models can identify high-risk features linked to faster disease progression.

Methods

Data for 15,931 MM patients in France, Spain, UK, Germany, and Italy were extracted from Oncology Dynamics (OD), a large cross-sectional survey that collects drug-treated patient-level data through a panel of cancer specialists. The analyses include 6 years (2017-2022) of records. Progression-free survival (PFS) was measured as the time from start of therapy to date of progression. The underlying data included treatment regime, line of therapy (LOT), ECOG status, stem cell transplant (SCT) eligibility, Durie-Salmon stage, cytogenetic risk, and age and gender. Machine Learning-based survival models (Gradient Boosted Survival Trees, Random Survival Forest) were applied to predict outcomes. Additionally, Shapley values were calculated to assess the model’s decision making and identify key risk drivers.

Results

Data was split into Training (80%) and Test Data (20%). The output of the model was the ML derived risk score: higher risk score was associated with shorter PFS. Model performance was good (Harrell’s C-Index = 0.741). Data shows that higher LOTs and not being eligible for SCT are associated with increased risk of disease progression. Inversely, being in the 1st LOT, ECOG asymptomatic and certain drug regimens (e.g., Lenalidomide or Dexamethasone/Lenalidomide) are associated with longer PFS.

Conclusions

This is a first successful step in leveraging OD data for the development of a prognostic score for MM patients using machine learning algorithms. ML models applied to secondary market research can support optimal disease management and planning of clinical trials. Also, given the heterogeneity of treatment plans across Europe, further investigation into country-specific differences in PFS may be studied.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

IQVIA.

Funding

Has not received any funding.

Disclosure

M.L. Pleguezuelo Witte, M. Muller, J. Hartmann, C. Ricarte, T. Moehler, M. Lopez, K. Jindal: Financial Interests, Personal, Full or part-time Employment: IQVIA.