From Clinical Questions to Testable Research

International Pharmacovigilance & Heterogeneity-Aware Signal Detection

Moving drug safety from average signals to "for whom, under what conditions."

Using the global spontaneous-reporting database (VigiBase), I examine not only the average association between drugs and adverse events but also how reporting patterns shift with age, co-medication, region, and clinical setting. Machine learning generates testable safety hypotheses about "which patients and conditions warrant caution," which I aim to confirm in independent clinical data.

JSPS Early-Career grant 2026–2029 ｜ 3 lead-author + 6 related collaborative papers

Privacy-Preserving Medical AI & International Data Sharing

Can we train and validate medical AI across borders without moving patient data?

This project implements and validates federated learning in real hospital settings, where patient-level data never leave each site and only model updates are shared. Beyond predictive accuracy, it addresses cross-site differences in variable definitions, missingness, operations, and reproducibility, aiming for a sustainable international research infrastructure.

SECOM Foundation grant 2026–2029 ｜ Japan–France (Nagoya & Caen)

An ongoing project launched in 2026. See the official site CANAL-AI ↗ — a Japan–France collaboration centered on Nagoya University and the University of Caen (France).

Evidence Synthesis & Real-World Data Analysis in Critical Care

How can we layer multiple forms of evidence for critical-care questions that trials alone cannot answer?

An ongoing collaboration on sepsis, mechanical ventilation, ECMO, and hemodynamics that pairs systematic reviews/meta-analyses with observational and large-scale data analysis. Sharing clinical questions and refining design, analysis, and interpretation over time, I served as principal investigator for the work on P0.1-based ventilator weaning.

18 related papers (9 meta-analyses / 9 observational, RWD, ML) ｜ PI project: P0.1 (JSPS Start-up)

Severity Stratification & Physiological Monitoring in Emergency and Critical Care

How far can we predict outcomes in critically ill patients from limited early information?

Using multicenter registries and observational data, this body of work studies severity stratification and physiological monitoring — targeted temperature management after cardiac arrest, outcome prediction, early emergency assessment, and ICU circulation, coagulation, and care systems. It addresses questions tied directly to bedside decisions, such as early triage indicators and the diagnostic accuracy of neurological examination.

20 related papers ｜ Funding: targeted temperature management, sepsis coagulopathy, START triage

Multicenter Clinical Research & Personalized Treatment in Cerebrovascular Disease

For the same treatment, how does the response differ across patients with cerebrovascular disease?

Centered on clazosentan therapy for subarachnoid hemorrhage, I contribute continuously to the design and analysis of multicenter clinical studies. Combining causal machine learning with external validation, I pursue personalized-treatment research that identifies "which patients gain most benefit and which face higher complication risk."

8 related papers ｜ Core: 5 subarachnoid-hemorrhage / clazosentan studies