Predicting Progression Events in Multiple Myeloma from Routine Blood Work

Background

Multiple Myeloma (MM) is a complex disease with varied survival outcomes, traditionally assessed using scoring systems like the International Staging System (ISS) and its revisions, which incorporate clinical, demographic, and genomic data. Despite these advancements, existing prognostic models, including machine learning (ML)-based ones, rely on initial assessments and do not account for the longitudinal dynamics observed in routine follow-ups. Neural networks (NN) offer the potential of leveraging these longitudinal data to predict disease trajectories and anticipate disease progression, yet their application in MM remains unexplored.Methods We utilized data from 875 newly diagnosed MM patients from the CoMMpass study, focusing on routine blood work parameters relevant to MM progression. Our predictive framework encompassed a hybrid NN combining Long Short-Term Memory (LSTM) networks and Conditional Restricted Boltzmann Machines (CRBM) to capture the complex temporal dynamics and heterogeneity of MM disease trajectories. Model performance was validated using area under the receiver operating characteristic (AUROC) and precision-recall curve (AUPRC) alongside traditional statistical measures, with uncertainty quantification achieved through 5-fold cross-validation and sensitivity analysis of forecasting horizons and data history.ResultsOur predictive model demonstrated high performance in reproducing correlations and autocorrelations (0.95 ± 0.01 ≥ R² ≥ 0.83 ± 0.03) in blood work time series data. Specifically, the model effectively replicated clinically relevant correlations, such as between serum creatinine and beta-2-microglobulin levels. The individual features in the model's forecasted blood work showed strong correlations to actual patient data (0.92 ± 0.02 ≥ r ≥ 0.52 ± 0.09). The identification of progression events achieved an AUROC of 0.88 ± 0.01 and an AUPRC (Chance = 0.07 ± 0.01) of 0.41 ± 0.02 at baseline, with an optimized decision threshold yielding a sensitivity of 0.92 ± 0.02 and a specificity of 0.65 ± 0.01. The prediction of future progression events started with an AUROC of 0.78 ± 0.02 at the 3 months forecast interval and gradually declined to 0.65 ± 0.01 at the 12 months forecast interval. The length of prior patient observation had a minimal impact on short-term forecasts, but was crucial for long-term forecasts. Overall, our models demonstrate robust potential for early detection of progression events in MM, offering a significant tool for enhancing patient management and treatment planning.ConclusionsOur study highlights the potential of using NNs to leverage the wealth of longitudinal blood work data accumulating in routine clinical practice for predicting individual disease progression events in MM patients. Our predictive framework demonstrates robust performance across diverse patient subsets and treatment regimens. The modular design of our prediction system not only enhances its interoperability but also allows for the integration of additional predictive modules for various clinical endpoints, promising broader applications in patient care. By leveraging routine blood work data and advanced predictive modeling, our system mitigates the need for resource- or labor-intensive tests and thus offers a scalable, cost-effective tool for enhancing clinical decision-making and optimizing patient management in MM care.

Platzbecker:Amgen: Consultancy, Research Funding; BMS: Consultancy, Membership on an entity's Board of Directors or advisory committees, Other: Travel support, Research Funding; MDS Foundation: Membership on an entity's Board of Directors or advisory committees; Abbvie: Consultancy, Research Funding; Curis: Consultancy, Honoraria, Research Funding; Geron: Consultancy; Janssen: Consultancy, Honoraria, Research Funding; Merck: Research Funding; Novartis: Consultancy, Research Funding. Oeser:Janssen-Cilag: Research Funding. Merz:Amgen, BMS, Celgene, Gilead, Jannsen, Stemline, SpringWorks and Takeda: Honoraria, Membership on an entity's Board of Directors or advisory committees, Research Funding.