An increasing demand for real-world evidence (RWE) to support regulatory submissions and payer negotiation has accelerated the big-data revolution in health care. Major strides have been made in improving the availability and usability of real-world data (RWD), and the application of machine learning (ML) in RWD is an essential step forward to augment meaningful insights into the care pathway. This symposium will introduce several creative ML approaches to establish consistency across data sources, harness the research value of RWD, and discuss the application of transparent and interpretable ML methods.

First, a case study will illustrate how ML methods are used to improve the consistency, transparency, and traceability of RWD across hospitals in China, where RWE research has traditionally faced many challenges. In a large-scale multi-center hematology study, researchers established a disease model to incorporate both consensus-based decision logic and ML-based data-driven optimization to unify data standards and definitions. This tool was applied to mechanize information integration across top hematology centers in China to enable result synchronization, which is essential in the generation of high-quality evidence to support China’s National Reimbursement Drug List’s negotiation and post-launch activities.

The second presentation will demonstrate how ML tools help to empower data transformation while ensuring relevance and validity. The raw RWD captured by hospital information systems (HIS) are fragmentary and lack an integrated picture of the care pathway. To close this gap, researchers developed and validated progression algorithms, drawn from China’s National Longitudinal Cohort of Hematological Diseases (NICHE), to reconstruct the complex patient journey through a hematologic condition. The progression algorithm was then incorporated into an automated data capture system to obtain, filter, and process traceable data, allowing researchers to supply important information that is not readily available from the HIS, including patients' treatment response by line of therapy.

Lastly, the presenters will demonstrate new transparent and interpretable ML approaches using case studies. Generalized linear models (GLM) are often preferred over ML models because of concerns regarding the lack of interpretability of complex ML algorithms. State-of-the-art interpretable ML methodologies will be discussed, along with their applications to a wide-ranging set of ML models. The objective of these approaches is to describe both general model behavior and the logic behind individual data unit predictions. These methods provide an improved level of interpretability and transparency that is both informative and, to some extent, more granular than GLM.

We hope this symposium can introduce these recent creative examples in the intersection of ML and RWD to the audience and stimulate discussions to further advance RWE research and methodology.