Critical care medicine is a data-intensive medical specialty that deals with a large amount of heterogeneous data on a daily basis. Although the volume and variety of critical care data pose many challenges, they also present interesting opportunities for health data science research. Our over-arching objective is to uncover useful information and knowledge from critical care data by applying advanced data analytics.
We primarily utilize the public critical care database called the Medical Information Mart for Intensive Care (MIMIC) which contains rich clinical data from approximately 60,000 intensive care unit admissions at Beth Israel Deaconess Medical Center in Boston, Massachusetts, U.S.A. MIMIC was created and is maintained by the Laboratory for Computational Physiology at Massachusetts Institute of Technology (MIT).
We have used MIMIC data to study:
- Hypotension (Interrogating a clinical database to study treatment of hypotension in the critically ill)
- Acute kidney injury (Severity of Acute Kidney Injury and Two-Year Outcomes in Critically Ill Patients; Empirical relationships among oliguria, creatinine, mortality, and renal replacement therapy in the critically ill; Outcome of Critically ill Patients with Acute Kidney Injury using the AKIN Criteria)
- Novel mortality predictors (Red cell distribution width improves the simplified acute physiology score for risk prediction in unselected critically ill patients)
- Fluid balance (Association between fluid balance and survival in critically ill patients)
- Adverse effects of medications (Proton-pump inhibitor use is associated with low serum magnesium concentrations)
We have also developed data access tools for MIMIC-II (Accessing the public MIMIC-II intensive care relational database for clinical research) as well as a machine-learning-based algorithm that can predict impending hypotension (An investigation of patterns in hemodynamic data indicative of impending hypotension in intensive care).
More recently, our focus has been on developing personalized patient outcome prediction algorithms that utilize patient similarity metrics (Personalized Mortality Prediction Driven by Electronic Medical Data and a Patient Similarity Metric; Personalized mortality prediction for the critically ill using a patient similarity metric and bagging).