AIM-AHEAD Research Spotlight Series: Showcasing Innovative Research Across the Consortium

The AIM-AHEAD Research Spotlight Series highlights the work of program participants across the consortium, including awardees, fellows, and trainees. Each showcase features AIM-AHEAD–supported research that uses artificial intelligence and machine learning (AI/ML) to address pressing healthcare challenges and drive meaningful impact.

Featured Program: AIM-AHEAD Consortium Development Program (Year 2)

The AIM-AHEAD Consortium Development Program supports multidisciplinary research initiatives that apply AI/ML to address pressing healthcare challenges. Through these projects, awardees develop and refine novel algorithms and analytic approaches using new and existing data sources, including electronic health records, clinical imaging, and other health-related datasets. This work is designed to advance methods for identifying emerging health concerns, supporting earlier detection, and informing strategies to improve health outcomes and access to care.

Across the program, investigators engage in interdisciplinary collaborations that integrate clinical expertise, data science, and community-informed perspectives to strengthen the development and application of AI/ML in healthcare settings.

The Year 2 cohort included 21 awardees whose projects reflected a range of research priorities aligned with AIM-AHEAD North Star III: Use AI/ML to improve behavioral health, cardiometabolic health and cancer outcomes for all.

Program Director: Harlan Jones, PhD

Predicting Kidney Disease Risk in Low-Resource Populations Using AI

Katherine Tuttle, MD
Providence Sacred Heart Medical Center & Children’s Hospital
North/Midwest Hub

Chronic kidney disease (CKD) is one of the most pressing public health challenges in the United States, affecting more than 37 million Americans and significantly increasing the risk of cardiovascular death. Diabetes is among the leading risk factors for CKD, and certain populations experience a disproportionately high burden of both conditions. Through AIM-AHEAD’s Consortium Development Program, Dr. Katherine Tuttle is working to improve how kidney disease risk is predicted and understood in these populations by harnessing artificial intelligence and machine learning (AI/ML).

Study Design and Methods

Dr. Tuttle’s study, Major Adverse Kidney Events in Populations with Diabetes, focuses on improving risk prediction for what researchers term a “Major Adverse Kidney Event” (MAKE), a composite outcome that includes all-cause death, significant kidney function decline, kidney failure, dialysis, or transplant. The study draws on data from the CURE-CKD Registry, a real-world dataset containing electronic health records for more than 8 million patients spanning nearly two decades.

The approach combines traditional statistical methods with multiple AI/ML models to identify factors that best predict kidney decline and to evaluate whether machine learning offers improvements over conventional analyses.

Key Findings

Across study cohorts, the findings were notable. Among individuals with diabetes, the risk of MAKE varied across patient groups, even after accounting for demographic, clinical, and residential characteristics. Nearly one in four individuals experienced a major adverse kidney event. Some groups were also more likely to be younger, live in rural areas, and have higher social vulnerability scores.

Across both cohorts, machine learning models modestly outperformed traditional statistical approaches while also serving a complementary role. Although both approaches reached similar conclusions, they differed in how they ranked contributing risk factors, offering a more nuanced understanding of kidney disease progression.

Implications for Research and Clinical Practice

By validating AI/ML tools alongside traditional methods and leveraging real-world clinical data to analyze outcomes across demographic subgroups, this work lays the foundation for more precise, informed clinical decision-making in CKD care.

Dr. Tuttle's research has resulted in four poster presentations at various national conferences, including the 2024 and 2025 AIM-AHEAD Annual Meetings and ASN Kidney Week 2025. Two manuscripts are expected in 2026, with findings also informing a forthcoming NIA/NIH grant submission focused on CKD complications and aging across demographic groups.

Modeling Non-Medical Determinants of Health Through Human-Centered AI

Suresh K. Bhavnani, PhD
The University of Texas Medical Branch
South Central Hub

Non-medical determinants of health (NMDoH), such as access to transportation, education, and employment, account for an estimated 30-55% of health outcomes and often co-occur, shaping patient risk. However, most research examines these factors independently. Through AIM-AHEAD’s Consortium Development Program, Dr. Suresh Bhavnani is applying human-centered artificial intelligence (AI) to better understand NMDoH clusters and their influence on patient outcomes in cancer care.

Study Design and Methods

Dr. Bhavnani’s study uses data from the All of Us Research Program, including 44,792 cancer participants. The analysis incorporates 110 self-reported NMDoH variables, aggregated into 18 factors across five domains, with outcomes including depression, delayed medical care, and emergency room utilization.

Given substantial and non-random missingness in survey responses, inverse probability weighting was used in place of standard imputation. The study integrates clustering methods to identify NMDoH subtypes with predictive modeling approaches, including logistic regression, random forest, and XGBoost, to evaluate risk across outcomes.

Key Findings

Distinct clusters of co-occurring NMDoH factors were identified, each associated with different outcome profiles. These clusters reflect varying combinations of barriers that correspond to differences in depression, delayed care, and emergency room use.

Predictive performance was strong, with the random forest model achieving an AUC of 0.80 for depression, outperforming prior NMDoH-based models. Model performance varied across clusters, with one subgroup consistently underperforming, suggesting the need for additional variables or subgroup-specific modeling approaches. While traditional and machine learning methods produced consistent overall findings, the clustering approach provided added insight into how combinations of factors contribute to risk.

Implications for Research and Practice

This work highlights how combining AI with real-world data can improve risk stratification by accounting for non-medical factors that influence health. By identifying patterns in patient needs, the study supports more targeted care and better resource alignment.

Findings from this work are also informing ongoing efforts, including collaboration with the Houston Health Department, to translate these insights into practical strategies for care navigation and resource allocation.

Dr. Bhavnani’s work has received national recognition, including selection as a 2025 Presidential Leadership Scholar. He has presented findings at national and international meetings and continues to collaborate with AIM-AHEAD investigators across institutions.