Description
Recent work in machine learning and modeling from healthcare data has shown much promise in leveraging large datasets to infer signal and predict outcomes of interest. With increasing numbers of such data sources available via collaborative efforts such as those created by the Observational Health Data Science and Informatics (OHDSI) network, it is now possible conduct clinical studies at an unprecedented scale. In this didactic panel, we will focus on another contribution of such data networks beyond the increased power of larger sample size, namely learning across multiple data sources as a way to improve the robustness of the learned models and increase the reliability of derived evidence. Our panel participants are junior and mid-career researchers in the field of machine learning for healthcare. We will (1) review the OHDSI network as one example of such a data network, (2) describe use cases in predictive analytics and an on-demand informatics consult service to distill clinical evidence from such data networks, where increased reliability is achieved through learning across different types of observational data, (3) describe current efforts in facilitating external model validation in OHDSI, and (4) illustrate current efforts in machine learning and specifically transfer learning to increase model robustness.
Learning Objective:
- Understand the landscape of healthcare data sources available to informatics and data science researchers, along with the OHDSI common data model across such data sources;
- Appreciate the need for model robustness and increased evidence reliability;
- Enumerate current efforts in increasing reliability of evidence through external validation across the OHDSI network;
- Learn about current efforts in machine learning for improving robustness and reliability of learned models across different datasets;
Authors:
Noemie Elhadad (Presenter)
Columbia University
Iñigo Urteaga (Presenter)
Columbia University
Alison Callahan (Presenter)
Stanford University
Jenna Reps (Presenter)
Janssen Research and Development
Patrick Ryan (Presenter)
Janssen Research and Development
Presentation Materials: