In critical care, a prediction model can only learn the outcomes someone bothered to look for. I study what happens to clinical AI when the label itself depends on who got tested.
I am a physician-scientist focused on critical care and clinical artificial intelligence. My research examines how AI can support high-consequence decisions in acute illness, particularly when the data used to train a model are shaped by who gets tested, what clinicians choose to measure, and which outcomes become visible in the medical record.
I study outcome ascertainment, selective testing, missing clinical information, interpretability, and treatment decisions: factors that determine whether a model is learning clinically meaningful biology or simply reproducing patterns of observation and care.
Critical care and acute illness are the central clinical setting for this work, including sepsis, severe infection, and acute illness-associated cardiovascular outcomes. Collaborative work in pulmonary medicine, kidney disease, transplantation, and surgery extends these questions into other high-risk clinical populations.
My research is organized around the reliability of clinical evidence when observation, testing, treatment, and documentation are themselves selective.
My primary research focus is the development, evaluation, and critical appraisal of clinical AI for high-acuity care. Work in sepsis and severe infection examines model discrimination, interpretability, bias, generalizability, and the gap between retrospective performance and bedside clinical utility.
A prediction model can only learn the outcomes someone bothered to look for. I study what happens when clinical AI is trained on labels shaped by surveillance intensity, selective testing, missing outcomes, and clinical workflow rather than biology alone.
Risk prediction asks who is likely to have an outcome. A more consequential question is which patient is likely to benefit from a specific intervention. This direction connects clinical AI with causal inference, treatment-effect heterogeneity, and individualized decision support in acute care.
Clinical outcomes research in sepsis, severe infection, and acute illness-associated atrial fibrillation provides real-world settings for studying prediction, treatment selection, competing risks, and clinically actionable decision support.
Selected collaborative work in transplantation, nephrology, pulmonary medicine, and surgery extends the same methodological questions into complex populations where surveillance, selection, and longitudinal outcomes are especially important.
Recent scientific work includes multicenter outcomes research in acute illness-associated atrial fibrillation presented at Heart Rhythm 2026, alongside research in clinical AI, transplantation, kidney disease, and acute-care medicine.
Heart Rhythm 2026
Selected work reflecting the clinical and methodological themes of the research program.
International Journal of Medical Informatics · Clinical AI / acute illness
Heart Rhythm 2026 · Cardiovascular outcomes / acute illness
International Urology and Nephrology · Kidney disease / clinical outcomes
International Urology and Nephrology · Transplantation
Respiratory and multidisciplinary clinical literature
Book chapter · Contemporary General Surgery · Ain Shams University
Surgical training and multidisciplinary clinical research provide the clinical context for a program focused on high-consequence decisions rather than algorithms in isolation.
Research collaboration across critical care, pulmonary medicine, nephrology, transplantation, acute illness outcomes, and clinical AI.
Graduate surgical training and scholarship, including work on surgical and non-surgical management of intussusception.
Training and practice across general, gastrointestinal, trauma, emergency, and perioperative surgical care.
For research collaboration, scholarly correspondence, and scientific inquiries.