No Black Box Anymore: Demystifying Clinical Predictive Modeling with Temporal-Feature Cross Attention Mechanism

Presentation Time: 02:00 PM - 02:12 PM

Abstract Keywords: Deep Learning, Clinical Decision Support, Artificial Intelligence, Machine Learning
Primary Track: Applications

Despite the outstanding performance of deep learning models in clinical prediction tasks, explainability remains a significant challenge. Inspired by transformer architectures, we introduce the Temporal-Feature Cross Attention Mechanism (TFCAM), a novel deep learning framework designed to capture dynamic interactions among clinical features across time, enhancing both predictive accuracy and interpretability. In an experiment with 1,422 patients with Chronic Kidney Disease, predicting progression to End-Stage Renal Disease, TFCAM outperformed LSTM and RETAIN baselines, achieving an AUROC of 0.95 and an F1-score of 0.69. Beyond performance gains, TFCAM provides multi-level explainability by identifying critical temporal periods, ranking feature importance, and quantifying how features influence each other across time before affecting predictions. Our approach addresses the "black box" limitations of deep learning in healthcare, offering clinicians transparent insights into disease progression mechanisms while maintaining state-of-the-art predictive performance.

Speaker:
Yubo Li, M.S. in Information Systems
Carnegie Mellon University

Authors:
Yubo Li, M.S. in Information Systems - Carnegie Mellon University; Xinyu Yao, Master - Carnegie Mellon University; Rema Padman, PhD - Carnegie Mellon University;

---

Temporal Harmonization: Improved Detection of Mild Cognitive Impairment from Temporal Language Markers using Subject-invariant Learning

Presentation Time: 02:12 PM - 02:24 PM

Abstract Keywords: Machine Learning, Natural Language Processing, Fairness and elimination of bias, Deep Learning, Diagnostic Systems, Artificial Intelligence, Bioinformatics
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Mild Cognitive Impairment (MCI) is an early stage of dementia characterized by cognitive decline and behavioral changes. Early detection is crucial for timely interventions, improved clinical trial cohort selection, and the development of targeted therapies. Linguistic markers have recently emerged as a non-invasive, cost-effective method for MCI detection. This study analyzes linguistic markers from conversations between participants and healthcare professionals to distinguish MCI from cognitively normal (NL) individuals. The dynamics of multiple conversations of a subject carry fine-granular linguistic change over time and expect to greatly enhance detection accuracy. However, individual variations in speaking styles pose challenges for learning cognitive characteristics from temporal sequences of conversations. To address this, we propose a temporal harmonization method to mitigate distributional differences in linguistic features across subjects, improving model generalization. Our results show that machine learning models leveraging subject-invariant harmonized temporal features greatly improve the prediction performance of MCI detection from multiple conversations.

Speaker:
Jiayu Zhou, Ph.D.
University of Michigan

Authors:
Bao Hoang, B.S. - Michigan State University; Siqi Liang, M.S. - University of Michigan; Yijiang Pang, M.S. - Michigan State University; Hiroko Dodge, Ph.D. - Harvard Medical School; Jiayu Zhou, Ph.D. - University of Michigan;

---

Rare Disease Patient Stratification Based on Temporal Phenotype Risk Scores Using Unstructured EHR Data: A Study on Ciliopathies

Presentation Time: 02:24 PM - 02:36 PM

Abstract Keywords: Artificial Intelligence, Machine Learning, Precision Medicine, Information Extraction, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Understanding disease progression in rare disorders is crucial for effective patient stratification and personalized management. However, prospective studies are often constrained by small patient cohorts and delayed diagnoses, limiting insights into disease trajectories. In this study, we developed a temporal analysis framework using Phenotype Risk Scores derived from unstructured Electronic Health Records to characterize disease trajectories in ciliopathies, a genetically and clinically heterogeneous group of rare diseases. We extracted phenotypic information using Natural Language Processing, computed disease likelihood at each patient visit, modeled its evolution over time, and stratified patients based on their phenotypic progression patterns. Our results highlight the heterogeneity in ciliopathy trajectories and reveal distinct progression subgroups. This framework provides a data-driven approach leveraging knowledge bases for rare disease stratification, paving the way for improved patient monitoring, risk assessment, and personalized therapeutic strategies. Future work will focus on multicentric validation and predictive modeling to refine disease progression insights.

Speaker:
Xiaoyi Chen, PhD
Imagine Institute of Genetic Diseases, Paris, France | University of Minnesota, Twin Cities, MN, USA

Authors:
Xiaoyi Chen, PhD - Institut Imagine; Xiaomeng Wang, PhD - Imagine Institute of Genetic Diseases, Paris, France; Carole Faviez, PhD - Imagine Institute of Genetic Diseases, Paris, France; Nicolas Garcelon, PhD - Imagine Institute of Genetic Diseases, Paris, France; Sophie Saunier, PhD - Imagine Institute of Genetic Diseases, Paris, France; Rui Zhang, PhD, FAMIA, FACMI - University of Minnesota, Twin Cities; Anita Burgun - INSERM DR PA06;

---

Benchmarking Waitlist Mortality Prediction Through Time-to-Event Modeling using New UNOS Dataset

Presentation Time: 02:36 PM - 02:48 PM

Abstract Keywords: Clinical Decision Support, Machine Learning, Data Mining
Primary Track: Applications

Decisions about managing patients on the heart transplant waitlist are currently made by committees of doctors who consider multiple factors, but the process remains largely ad-hoc. With the growing volume of longitudinal patient, donor, and organ data collected by the United Network for Organ Sharing (UNOS) since 2018, there is increasing interest in model-based approaches to support clinical decision-making. In this study, we benchmark machine learning models that leverage dynamic waitlist history data for time-dependent, time-to-event modeling of waitlist mortality. We train on 23,807 patient records with 71 variables and evaluate both survival prediction and discrimination at a 1-year horizon. Our best model achieves a C-index of 0.94 and AUC of 0.89, significantly outperforming previous static models. Key predictors align with known clinical risk factors while also revealing novel associations. Our findings can support urgency assessment and policy refinement in heart transplant allocation.

Speaker:
Yingtao Luo, Ph.D.
Carnegie Mellon University

Authors:
Yingtao Luo, Ph.D. - Carnegie Mellon University; Reza Skandari, Ph.D. - Imperial College, London; Carlos Martinez - United Network for Organ Sharing; Arman Kilic, M.D. - Medical University of South Carolina; Rema Padman, PhD - Carnegie Mellon University;

---

An Interactive Information Visualization System for Temporal Queries in a Large-scale COVID-19 EHR Dataset (COVID-SPHERE): development and qualitative evaluation

Presentation Time: 02:48 PM - 03:00 PM

Abstract Keywords: Informatics Implementation, Information Visualization, Usability, User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

COVID-SPHERE is a self-service web application designed to advance clinical research informatics by facilitating secondary use of Electronic Health Records (EHR) for COVID-19 research. The system employs a flexible EHR concept framework that defines hierarchical concepts and ontologies, enabling clinical researchers to build complex temporal queries through an intuitive, single-click interface without requiring database expertise. Our method dynamically generates MongoDB queries in real-time and offers interactive, faceted visualizations to analyze longitudinal patient activities and integrated health records, supporting both individual patient analysis and population-level research. Hosted on a server managing over 5 TB of data encompassing 30 billion health records spanning 15 years from more than 8.8 million patients, this work demonstrated its generalizability by supporting 10 published research studies investigating various COVID-related research topics on epidemiology, treatment outcomes, and long-term sequelae since November 2020. By simplifying the cohort discovery process, COVID-SPHERE reduces the informatics barriers between researchers and EHR data, enhancing the efficiency of clinical and translational research while promoting data-driven insights for COVID-19 surveillance and intervention. Its architecture is applicable to other large-scale clinical research data warehouses, offering a model for future public health informatics systems.

Speaker:
Yan Huang, Ph.D
UT Health Science Center

Authors:
Yan Huang, Ph.D - UT Health Science Center; Shiqiang Tao, PhD - The University of Texas Health Science Center at Houston; Wei-Chun Chou, M.S; Licong Cui, PhD - The University of Texas Health Science Center at Houston (UTHealth Houston); GQ Zhang, PhD - The University of Texas Health Science Center at Houston; Xiaojin Li, Ph.D. - University of Texas Health Science Center at Houston;

---

Spatio-Temporal Data Integration for Liver Cell Repair Following Acetaminophen-Induced Injury

Presentation Time: 03:00 PM - 03:12 PM

Abstract Keywords: Bioinformatics, Artificial Intelligence, Omics (genomics, metabolomics, proteomics, transcriptomics, etc.) and Integrative Analyses
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics

We integrated spatial transcriptomics and single-cell RNA-seq to reconstruct the spatio-temporal landscape of liver regeneration after acetaminophen-induced injury. Using graph-based models and Scanorama, we identified dynamic interactions among hepatocytes, macrophages, and endothelial cells. Key regenerative pathways included Wnt/β-catenin signaling and macrophage polarization. Our findings reveal zonal differences in repair and highlight novel therapeutic targets to enhance liver regeneration.

Speaker:
Hui Li, Phd
University of Texas Health Science Center at Houston

Authors:
jinlian wang, PhD - UTHealth; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;

---