Clinicians are routinely making decisions about which diagnostic procedures to perform (e.g., biopsy, blood test, advanced imaging) to assess a patient’s condition. However, the process of reaching an accurate diagnosis is currently error prone and inefficient. Clinicians are tasked with reviewing a growing amount of clinical information that is often drawn from multiple sources. As the variety of diagnostic procedures increases (e.g., new molecular diagnostic testing and improved imaging protocols), it is critical to select the right tests, both to provide relevant and accurate information and to diagnose patients in a timely, cost-efficient manner.
This project, funded by the National Science Foundation through its Smart & Connected Health program, harnesses the growing amount of data that is captured in the electronic health record to discover the optimal diagnostic pathway for an individual patient. A multidisciplinary team led by Mihaela van der Schaar (Electrical Engineering) and I will investigate approaches that transform data into actionable knowledge, enabled by a new class of clinical decision support algorithms that actively learn from available clinical data. The objective of this project is to develop and evaluate a data-driven framework for decision support that helps clinicians to deliver individualized patient care by discovering optimal sequences of actions and to diagnose patients in a timely, accurate, and cost-effective manner. Towards this goal, the project addresses challenges related to finding relevant information from large, longitudinal patient data; learning sequences of actions from past patient cases; and handling uncertainty that is inherent to the practice of medicine.
This project will lead to the creation of a decision support tool called Smart Diagnostics (SmartDx) for ordering diagnostic tests. Two commonly problematic diagnostic scenarios are initially targeted:
  1. whether a patient presenting with chest pains has a pulmonary embolism (PE); or
  2. whether a patient with diabetes, high blood pressure, and a history of smoking has coronary artery disease (CAD).

In 2014, imaging exams for diagnosing PE accounted for $72 million in total payments (2% of all claims) while CAD represented $433 million (14% of all claims), as reported by the Chronic Conditions Data Warehouse. We will demonstrate the feasibility and advantages of our work by evaluating SmartDx’s ability to address these high impact areas. The deployment of our learning algorithms will improve how observational clinical data can be used to generate evidence that improves healthcare delivery, efficiency, and ultimately, realizes precision medicine and improves patient outcomes. Importantly, this project will support the involvement of a diverse group of graduate students who will be trained in an interdisciplinary manner to translate algorithms and data science concepts into applications that have real-world clinical utility, and with a clear understanding of the myriad technical and cognitive challenges of such implementation.