PROJECT SUMMARY
Up to 5% of hospitalized adult 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 improve the care of high-risk ward patients to decrease preventable in-hospital deaths.
The current paradigm for attempting to decrease mortality from deterioration has several limitations. First,
most early warning scores designed to identify high-risk patients are based only on vital signs and have limited
accuracy. Clinical notes are an underutilized, rich source of information comprising nearly 80% of electronic
health record (EHR) data. Natural language processing (NLP) can extract important risk factors from clinical
notes for machine learning models to improve accuracy over existing tools. Second, current early warning scores
only tell clinicians that a patient is at high risk but provide no information regarding what clinical condition is
causing a patient’s deterioration. This leads to diagnostic and treatment errors, which results in worse patient
outcomes. Developing tools to enhance diagnostic accuracy for high-risk ward patients could lead to fewer
medical errors, decreased costs, and improved outcomes. Third, the initial treatment decisions for deteriorating
patients are made by clinicians with limited experience caring for critically ill patients, which can result in delays
of potentially life-saving therapies. By utilizing a large, granular, multicenter dataset, algorithms to predict the
treatments a patient should receive can be developed, resulting in early, targeted, potentially life-saving therapy.
The long-term goal is to develop and implement clinically useful decision support tools to decrease
preventable death from deterioration. The overall objective of this project is to develop a clinical decision support
tool for the identification, diagnosis, and treatment of patients at high risk of deterioration. This objective will be
pursued in the following three specific aims: 1) Develop machine learning models to identify patients at high risk
of deterioration using both structured data and unstructured clinical notes; 2) Develop models to predict the
diagnosis that is causing the deterioration event and the potentially life-saving treatments that should be provided
to high-risk patients; 3) Develop a clinical decision support tool with a graphical user interface incorporating the
models from Aims 1 and 2 via user-centered design principles and then test its effectiveness, efficiency, and
user satisfaction in a case-based simulation study. This research is innovative because it will utilize NLP,
reinforcement learning, interpretable machine learning, and multi-task transfer learning approaches. The
proposed research is significant because it will provide clinicians with powerful new tools that can be
implemented in the EHR to identify, diagnose, and make treatment recommendations for high-risk patients. This
will result in the delivery of early, personalized care to decrease preventable death from deterioration.
