Engineering Features From Advanced Medical Technology Initiative Submissions to Enable Predictive Modeling for...

Area of Science:

• Military Medicine
• Biomedical Informatics
• Artificial Intelligence

Background:

• The U.S. Army's Advanced Medical Technology Initiative (AMTI) funds military medicine research through a proposal-driven process.
• Proposal selection involves a rigorous peer review assessing innovation, military relevance, success metrics, and return on investment.
• Current review processes are time-intensive and can be subject to bias.

Purpose of the Study:

• To explore the viability of artificial intelligence and machine learning (ML) for predicting proposal ratings.
• To develop a model-based approach that assists human experts in evaluating research proposals.
• To identify key proposal features that correlate with successful funding outcomes.

Main Methods:

• Conducted a literature review to identify potential predictive features for ML models.
• Extracted features from 824 AMTI proposals submitted between 2010 and 2022.
• Features included requested funds, word count, readability scores, citation/partner data, and TF-IDF word vectors.

Main Results:

• Identified high-ranking keywords such as 'data,' 'health,' 'injury,' 'device,' 'treatment,' and 'technology' in proposal abstracts and key sections.
• Assessed proposal text readability using the Flesch scale, finding most fields required a college graduate reading level.
• Analyzed citations and partners as indicators of proposal success.

Conclusions:

• This initial research is the first step toward an ML-powered system to support proposal review.
• The project aims to provide automated support, reduce reviewer bias, and streamline the AMTI administrative process.
• Insights gained will benefit the military health system by improving research funding efficiency and fairness.