Research

The overarching goal of my research is developing methods for automatic semantic analysis of texts. My work spans such areas of computer science as natural language processing (NLP), machine learning, and data mining. Most recently, my research has focused on semantic analysis of clinical texts. I work both on method development and applications.

Prior to to joining Loyola, I was a researcher at Boston Children's Hospital and Harvard Medical School. I received my PhD in computer science from the University of Colorado Boulder, my MS in computer science from the State University of New York at Buffalo, and my BS in computer science from Loyola University Chicago.

I am involved in many active collaborations with major research centers across the US including University of Wisconsin, Boston Children's Hospital, Harvard Medical School, University of Chicago, University of Arizona, Rush University, University of Colorado, and others.

My research is currently funded by the following NIH awards:

  • Temporal Relation Discovery for Clinical Text (THYME), in which we are designing machine learning algorithms to extract timelines from clinical text and integrate those with structured data from the electronic medical record (NIH/NLM R01LM010090).

  • Learning Universal Patient Representations from Clinical Text with Hierarchical Recurrent Neural Networks. We develop neural network methods to make more efficient use of human and data resources, applying deep neural networks to millions of patient records for learning universal, task independent patient representations. These universal patient representations can be applied to thousands of clinical research tasks, including EHR-based phenotyping. To learn these representations, we train across entire health care systems using EHR events from later dates as a source of supervision. We extract patient representations from these networks, and make them available to clinical investigators for streamlining clinical research tasks (NIH/NLM R01LM012973).

  • Integrating citywide data to identify, differentiate, and prioritize hospitalized patients with substance misuse. We provide novel and critically important AI tools for the detection of substance misuse from the electronic health record (EHR). Development and validation of the substance misuse classifier would enable a standardized approach to perform screening on all patient encounters on a daily basis in health systems. We will rigorously develop and test substance misuse classifier retrospectively and then examine its performance prospectively in both a naïve and mature screening program. This will serve as the first step towards a comprehensive universal screener that leverages available data in the EHR (NIH/NLM 1R01DA051464).

  • Developing a clinical decision support tool for the identification, diagnosis, and treatment of critical illness in hospitalized patients. Up to 5% of adult hospitalized patients on the medical-surgical wards develop clinical deterioration requiring intensive care. Medical errors are common before deterioration events, including delays and misjudgments in identification, diagnosis, and treatment, and these errors lead to increased morbidity and mortality. Therefore, it is critically important to optimize the care of high-risk ward patients to decrease preventable in-hospital deaths. The goal of this project is to develop a clinically useful clinical decision support tool for the identification, diagnosis, and treatment of hospitalized patients at high risk of clinical deterioration (NIH/NHLBI 1R01HL157262)

  • Using Machine Learning for Early Recognition and Personalized Treatment of Acute Kidney Injury. Acute kidney injury (AKI) occurs in up to 20% of hospitalized patients and is associated with increased risk of readmission, morbidity, and mortality. The estimated annual cost of AKI care in the US is over 10 billion dollars, and, with the incidence rising, these costs will continue to increase. The objective of this project is to develop novel tools to improve the identification and treatment of patients at high risk of AKI using a large, multi-center cohort (NIH/NIDDK 1R01DK126933).