PARAMO: a PARAllel predictive MOdeling platform for healthcare analytic research using electronic health records

Area of Science:

• Health Informatics
• Computational Biology
• Data Science

Background:

• Electronic Health Records (EHRs) are increasingly used for predictive modeling in healthcare analytics.
• Developing predictive models requires a complex pipeline including cohort and feature construction, cross-validation, feature selection, and classification.
• Comparing diverse models across various patient cohorts, features, and statistical frameworks is crucial for appropriate model development.

Purpose of the Study:

• To develop and evaluate a predictive modeling platform to simplify and expedite the process of building disease prediction models from health data.
• To enable efficient comparison and refinement of models derived from varied patient cohorts, features, and statistical frameworks.

Main Methods:

• Developed the PARAllel predictive MOdeling (PARAMO) platform.
• PARAMO constructs a dependency graph of predictive modeling tasks.
• Tasks are scheduled topologically and executed in parallel using Map-Reduce in a cluster computing environment.

Main Results:

• Demonstrated significant computational efficiency gains compared to standard approaches.
• PARAMO can build 800 models on a 300,000-patient dataset in 3 hours, versus 9 days sequentially.
• Performance was assessed using datasets from Geisinger Health System, Vanderbilt University Medical Center, and a claims database.

Conclusions:

• An efficient parallel predictive modeling platform for EHR data has been successfully developed.
• The platform facilitates large-scale modeling, accelerates research workflows, and promotes health information reuse.
• This foundational platform supports the ultimate goal of specialized analytic pipelines for health data researchers.