Automating Construction of Machine Learning Models With Clinical Big Data: Proposal Rationale and Methods.

Area of Science:

• Clinical Informatics
• Machine Learning in Healthcare
• Health Data Science

Background:

• Healthcare systems face challenges in utilizing clinical big data for prediction and classification due to a lack of machine learning expertise among researchers and a shortage of data scientists.
• The manual iteration process for building machine learning models is resource-intensive, hindering the adoption of these technologies in healthcare settings.
• Limited budgets and the high demand for data scientists create significant barriers to leveraging clinical big data for improved patient outcomes.

Purpose of the Study:

• To enable healthcare researchers to directly utilize clinical big data for machine learning tasks.
• To make machine learning feasible for healthcare systems with limited budgets and data scientist resources.
• To facilitate the realization of value from clinical big data for enhanced patient care and cost reduction.

Main Methods:

• Developing and validating Automated Machine Learning (Auto-ML) software designed to automate model selection for clinical big data.
• Applying Auto-ML and novel methodologies to critical care management allocation problems, including piloting a model with care managers.
• Conducting simulations to estimate the potential impact of Auto-ML adoption on patient outcomes in the United States.

Main Results:

• The design document for Auto-ML is currently in progress.
• The study is anticipated to conclude around the year 2022.

Conclusions:

• Auto-ML is expected to generalize across diverse clinical prediction and classification challenges.
• Healthcare researchers will be able to develop high-quality machine learning models with minimal data scientist assistance using Auto-ML.
• The adoption of Auto-ML is poised to increase the utilization of machine learning in healthcare, ultimately leading to improved patient outcomes.