AIM-AHEAD Report

AIM-AHEAD Request for Information Responses: Full Report

The National Institutes of Health (NIH) released a Request for Information (RFI) from June to September, 2021 to understand the needs, interests, and opportunities for building and advancing artificial intelligence/machine learning (AI/ML) approaches to redress health disparities and advance health equity using electronic health records (EHR) and other types of data. Inputs from the RFI and Stakeholder Engagement Forum are summarized in this report.

Download the Report

Executive Summary

In June 2021, the National Institutes of Health (NIH) released a Request for Information (RFI) to better understand the needs, interests, and opportunities for building and advancing Artificial Intelligence/Machine Learning (AI/ML) approaches using electronic health record (EHR) and other types of data (e.g., genomics, imaging, social determinants of health) to redress health disparities and advance health equity. The NIH received 76 responses to the RFI, of which 2 are considered not responsive to the RFI. In addition, NIH conducted a stakeholder engagement forum (SEF) that included academia, federal agencies, healthcare providers, and the data science/technology industry in order to provide an overview of the NIH AI/ML for health disparities research initiative and engage attendees in listening sessions. As part of the registration process, attendees were offered the opportunity to provide input on essential research, infrastructure, training, and partnership needs to advance the field. Westat conducted a qualitative analysis that highlighted key themes in the responses to the RFI and SEF.

Key findings include the following:

  1. Research - Respondents suggested a wide variety of research topics in the areas of health disparities, improving data usefulness, and improving AI/ML methodologies to mitigate bias in data sources and models. Respondents suggested ways to link and use a variety of data sources (e.g., EHR, self-reported patient information, genomics, social determinants of health (SDoH), biomarker, wearable sensor, geospatial, and mobile health data) to study a variety of diseases and conditions (e.g., cancer, mental health, infectious diseases [e.g., Covid-19, HIV, and Ebola], dementia and neurological disorders, maternal health, pediatric health, heart disease, and diabetes).
     
  2. Infrastructure - While some organizations in academia and industry have invested heavily in AI/ML and can marshal robust infrastructure and resources, other respondents indicated minimal current infrastructure for AI/ML. Respondents indicated a priority need for access to high-quality, federated data integrated across organizations, cloud computing resources, and data management platforms.
     
  3. Partnerships - Respondents expressed readiness to partner in both multi-organizational and multi-disciplinary partnerships. Particularly strong is the need for partnerships that span data science, clinical research and health care delivery, and community-based organizations that serve minority or underrepresented populations. Respondents stressed the importance of community engagement in building trust and reducing harm and bias.
     
  4. Training - A small number of RFI respondents noted training resources or opportunities currently available, while many more indicated training needs at all levels. Specific training needs include: AI/ML methods and data science, statistical methods to reduce bias, health disparities, ethics in the use of AI/ML, uses of AI/ML in healthcare, and training to increase the diversity of the AI/ML workforce.
     
  5. Opportunities and Challenges - RFI respondents expressed various options for using or improving the usability of AI/ML, including bias detection methods, use of natural language processing, and creating open-source AI systems. However, respondents also identified many challenges or limitations to the use of AI/ML in healthcare including privacy concerns, equitable access to data and technology, data completeness and data biases, model biases, barriers to data sharing, and financial sustainability. There was also concern about applying AI/ML methods to rare outcomes or to ration the availability of health care services due to its limitations.
     
  6. Prioritizing Under-resourced Institutions - Several respondents specifically suggested that NIH prioritize under-resourced or minority-serving institutions in funding decisions.


The remainder of this report is organized as follows:

  • Introduction
  • Part 1: Narrative Analysis
  • Part 2: Chartbook
  • Appendix A: RFI Questions
  • Appendix B: Qualitative Analysis Methodology
  • Appendix C: Codebook

This page last reviewed on March 23, 2023