The research station
AI/ML for Healthcare Informatics
I work on applying artificial intelligence and machine learning to healthcare informatics — making health data more usable, interoperable, and intelligent. The broad goal: help clinical and biomedical data actually talk to the systems (and models) that could learn from it.
ML on health data
Applying machine-learning methods to biomedical and clinical datasets, where messiness, sparsity, and privacy constraints are the norm rather than the exception.
Health-data interoperability
Working with the standards and structures (terminologies, metadata, FAIR principles) that decide whether health data can move between systems without losing meaning.
AI-assisted informatics tools
Exploring how modern AI — including LLMs — can lower the barrier between clinicians/researchers and the data infrastructure they depend on.
My contributions
- Literature review and experiment support across active lab projects.
- Data preparation and model-evaluation tooling in Python.
- Write-ups and figures for lab presentations.
Publications & posters
Tap a publication to read the abstract and open the publisher page.
Abstract
Clinician burnout poses a substantial threat to patient safety, particularly in high-acuity intensive care units (ICUs). Existing research predominantly relies on retrospective survey tools or broad electronic health record (EHR) metadata, often overlooking the valuable narrative information embedded in clinical notes. In this study, we analyze 10,000 ICU discharge summaries from MIMIC-IV, a publicly available database derived from the electronic health records of Beth Israel Deaconess Medical Center. We introduce a weakly supervised framework for provider-level surveillance of burnout risk, combining BioBERT sentiment embeddings fine-tuned for clinical narratives, a lexical stress lexicon, topic modeling, workload proxies, and temporal tracking. Using quantile-based ordinal classification, a provider-level logistic regression model achieves an F1 score of 0.84 for identifying high-risk cases, surpassing metadata-only baselines. Specialty-specific analysis indicates elevated stress signals among providers in Radiology, Psychiatry, and Neurology. Our findings demonstrate that ICU clinical narratives encode actionable, longitudinal signals for scalable burnout surveillance.