NIH Data Sharing and Reuse Seminar- Sex Differences in Acute Kidney Injury Risk and Outcomes: Insights from MIMIC-III and MIMIC-IV

Here is the information from NIH ODSS:

"September Data Sharing and Reuse Seminar- Sex Differences in Acute Kidney Injury Risk and Outcomes: Insights from MIMIC-III and MIMIC-IV

Date: September 11, 2026

Time: 12 pm - 1 pm ET

Acute kidney injury (AKI) affects 13-18% of U.S. hospitalizations and increases the risks of chronic kidney disease (CKD) and death. However, which clinical factors are associated with AKI and its sequelae, and whether these associations differ between women and men, remain poorly understood. This study combines machine learning and classical regression models to examine predictors across the continuum from before ICU admission, for AKI onset, post-AKI CKD and death.

We analyzed adult ICU admissions from MIMIC-III (2001–2012) and MIMIC-IV (2008–2019). Six machine-learning models were developed in MIMIC-IV and validated in MIMIC-III using demographic, clinical, laboratory, treatment, and clinical-text features.

Performance was assessed through discrimination, diagnostic accuracy, and calibration. Sex differences in predictor associations were evaluated using regression interaction terms and Friedman’s H-statistics. Progression from AKI to CKD and death was evaluated using competing-risk regression and DeepSurv.

XGBoost achieved the strongest discrimination for AKI, and DeepSurv performed best in predicting CKD and death. Creatinine, bicarbonate, and hemoglobin showed differential associations with AKI risk between women and men. Sex differences involving baseline creatinine and potassium were also observed for CKD and mortality after AKI. These findings help explain heterogeneity in AKI risk and progression and may inform the development and evaluation of more individualized risk-assessment and surveillance strategies.

The talk will also illustrate how human researchers and AI collaborated throughout the project, from refining research questions to evaluating models and interpreting results.

Speaker: Rui Feng, Ph.D., is an Associate Professor in the Department of Biostatistics, Epidemiology, and Informatics at the University of Pennsylvania Perelman School of Medicine. Her research focuses on statistical and machine-learning methods for high-dimensional biomedical data, with particular interests in deep learning, explainable AI, and modeling complex interactions that influence health outcomes.

Registration: https://events.teams.microsoft.com/event/7d68563c-2fb0-40d4-a646-8dd02004cdd9@d0b80d06-1f2d-409c-9cad-ccda77eb7381?source=copyLinkLegacyShareLinkDialog"

Source and more information: https://datascience.nih.gov/news/september-data-sharing-and-reuse-seminar-2026