Explainable deep learning for tumor dynamic modeling and overall survival prediction using Neural-ODE

DN-ODE: Data-driven neural-ODE modeling for breast cancer tumor dynamics and progression-free survivals

Pharmacokinetic/Pharmacodynamic (PK/PD) modeling is crucial in the development of new drugs. However, traditional population-based PK/PD models encounter challenges when modeling for individual patients. We aim to explore the potential of constructing a pharmacodynamic model for individual breast cancer pharmacodynamics leveraging only limited data from early clinical trial phases. While previous studies on Neural Ordinary Differential Equations (ODEs) suggest promising results in clinical trial practices, they primarily focused on theoretical applications or independent PK/PD modeling. PD modeling from complex and irregular clinical trial data, especially when interacting with PK parameters, is still unclear. To achieve that, we introduce a Data-driven Neural Ordinary Differential Equation (DN-ODE) modeling for breast cancer tumor dynamics and progression-free survival data. To validate this approach, experiments are conducted with early-phase clinical trial data from the Amcenestrant (an oral treatment for breast cancer) dataset (AMEERA 1-2), aiming to predict pharmacodynamics in the later phase (AMEERA 3). DN-ODE model achieves RMSE scores of 8.78 and 0.21 in tumor size and progression-free survival, respectively, with R2 scores over 0.9 for each task. Compared to PK/PD methodologies, DN-ODE is able to predict robust individual tumor dynamics with only limited cycle data. We also introduce Principal Component Analysis visualizations for encoder results, demonstrating the DN-ODE's capability to discern individual distributions and diverse tumor growth patterns. Therefore, DN-ODE facilitates comprehensive drug efficacy assessments, pinpoints potential responders, and aids in trial design.

Bridging pharmacology and neural networks: A deep dive into neural ordinary differential equations

The advent of machine learning has led to innovative approaches in dealing with clinical data. Among these, Neural Ordinary Differential Equations (Neural ODEs), hybrid models merging mechanistic with deep learning models have shown promise in accurately modeling continuous dynamical systems. Although initial applications of Neural ODEs in the field of model-informed drug development and clinical pharmacology are becoming evident, applying these models to actual clinical trial datasets-characterized by sparse and irregularly timed measurements-poses several challenges. Traditional models often have limitations with sparse data, highlighting the urgent need to address this issue, potentially through the use of assumptions. This review examines the fundamentals of Neural ODEs, their ability to handle sparse and irregular data, and their applications in model-informed drug development.

Predicting progression-free survival from measurable residual disease in chronic lymphocytic leukemia

Association between measurable residual disease (MRD) and survival outcomes in chronic lymphocytic leukemia (CLL) has often been reported. However, limited quantitative analyses over large datasets have been undertaken to establish the predictive power of MRD. Here, we provide a comprehensive assessment of published MRD data to explore the utility of MRD in the prediction of progression-free survival (PFS). We undertook two independent analyses, which leveraged available published data to address two complimentary questions. In the first, data from eight clinical trials was modeled via a meta-regression approach, showing that median PFS can be predicted from undetectable MRD rates at 3-6 months of post-treatment. The resulting model can be used to predict the probability of technical success of a planned clinical trial in chemotherapy. In the second, we investigated the evidence for predicting PFS from competing MRD metrics, for example baseline value and instantaneous MRD value, via a joint modeling approach. Using data from four small studies, we found strong evidence that including MRD metrics in joint models improves predictions of PFS compared with not including them. This analysis suggests that incorporating MRD is likely to better inform individual progression predictions. It is therefore proposed that systematic MRD collection should be accompanied by modeling to generate algorithms that inform patients' progression.

Predicting overall survival from tumor dynamics metrics using parametric statistical and machine learning models: application to patients with RET-altered solid tumors

In oncology drug development, tumor dynamics modeling is widely applied to predict patients' overall survival (OS) via parametric models. However, the current modeling paradigm, which assumes a disease-specific link between tumor dynamics and survival, has its limitations. This is particularly evident in drug development scenarios where the clinical trial under consideration contains patients with tumor types for which there is little to no prior institutional data. In this work, we propose the use of a pan-indication solid tumor machine learning (ML) approach whereby all three tumor metrics (tumor shrinkage rate, tumor regrowth rate and time to tumor growth) are simultaneously used to predict patients' OS in a tumor type independent manner. We demonstrate the utility of this approach in a clinical trial of cancer patients treated with the tyrosine kinase inhibitor, pralsetinib. We compared the parametric and ML models and the results showed that the proposed ML approach is able to adequately predict patient OS across RET-altered solid tumors, including non-small cell lung cancer, medullary thyroid cancer as well as other solid tumors. While the findings of this study are promising, further research is needed for evaluating the generalizability of the ML model to other solid tumor types.

Current Status and Future Directions: The Application of Artificial Intelligence/Machine Learning for Precision Medicine

Technological innovations, such as artificial intelligence (AI) and machine learning (ML), have the potential to expedite the goal of precision medicine, especially when combined with increased capacity for voluminous data from multiple sources and expanded therapeutic modalities; however, they also present several challenges. In this communication, we first discuss the goals of precision medicine, and contextualize the use of AI in precision medicine by showcasing innovative applications (e.g., prediction of tumor growth and overall survival, biomarker identification using biomedical images, and identification of patient population for clinical practice) which were presented during the February 2023 virtual public workshop entitled "Application of Artificial Intelligence and Machine Learning for Precision Medicine," hosted by the US Food and Drug Administration (FDA) and University of Maryland Center of Excellence in Regulatory Science and Innovation (M-CERSI). Next, we put forward challenges brought about by the multidisciplinary nature of AI, particularly highlighting the need for AI to be trustworthy. To address such challenges, we subsequently note practical approaches, viz., differential privacy, synthetic data generation, and federated learning. The proposed strategies - some of which are highlighted presentations from the workshop - are for the protection of personal information and intellectual property. In addition, methods such as the risk-based management approach and the need for an agile regulatory ecosystem are discussed. Finally, we lay out a call for action that includes sharing of data and algorithms, development of regulatory guidance documents, and pooling of expertise from a broad-spectrum of stakeholders to enhance the application of AI in precision medicine.