Research Summary: Still funded, no longer counted: how NIH's 2025 award reviews changed what the government counts as minority health research

The U.S. National Institutes of Health (NIH) implemented a policy change in 2025 requiring the removal of Diversity, Equity, and Inclusion (DEI) language from award applications not directly supporting DEI activities. This policy has altered how NIH classifies and counts minority health research, as classification relies on text analysis of award descriptions. An analysis of 37,790 continuing awards revealed that population names in award titles were highly predictive of enrolled minority shares, and text features determined the survival of these names, impacting what the government now counts as minority health research.

Why it matters

This development is strategically important because it demonstrates how subtle changes in policy and language requirements can significantly impact the classification and reporting of critical research areas like minority health. It highlights the potential for official metrics to diverge from actual research activities, which can misrepresent funding priorities, resource allocation, and progress in addressing health disparities. This could have long-term consequences for public health strategies and equity initiatives.

Key insights

• NIH's 2025 policy mandates the removal of DEI language from award applications unless the award specifically supports DEI activities.
• The classification of NIH research spending, including minority health, is heavily reliant on automated text analysis of award descriptions.
• The removal of DEI language impacts the words available for classification, thereby changing what NIH counts as minority health research.
• Analysis of continuing awards demonstrated that explicit naming of minority populations in award titles strongly correlated with higher enrolled shares of those populations (e.g., Black populations predicted a 60-percentage-point higher enrolled share).
• Text features within award applications were critical in determining whether population names survived the review process, influencing classification outcomes.
• The policy change creates a discrepancy between the actual practice or focus of research and its official categorization by the funding body.
• Funders typically identify portfolio content through software-based text classification, making language choices critical for reporting and resource allocation.
• The study indicates that the applicant's choice of language in proposals directly influences how the funding agency categorizes and reports on its portfolio.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2610.03443