Using Artificial Intelligence to Improve Hospital Inpatient Care
By Daniel B. Neill
NOTE: This is an overview of the entire article, which appeared in the March/April 2013 issue of the IEEE Intelligent Systems magazine.
Click here to read the entire article.
Recent advances in machine learning and Artificial Intelligence (AI) build predictive models and make real-time inferences from a large patient population. Several of these approaches focus on critical care, using physiological data that are routinely recorded in intensive care units. They have shown that the use of temporal information representing the evolution of each patient’s health state over time leads to improved classification accuracy, as compared to classifiers that only consider the patient’s current state.
These advances demonstrate the potential of AI and machine learning, but they are limited in scope to specific diseases or diagnoses or only applicable to a small subset of the patient population. The next great challenge for AI in healthcare might be to develop approaches that can be applied to the entire population of patients, monitoring huge quantities of data to automatically detect problems and threats to patient safety (including patterns of suboptimal care, as well as outbreaks of hospital acquired illness), and to discover new best practices of patient care.
Two different AI approaches are respectively based on question-answering (QA) and on large-scale anomalous pattern detection. The first is the DeepQA architecture by IBM Research in collaboration with Carnegie-Mellon University. IBM is currently partnering with the Sloan-Kettering Cancer Center to enable patient-specific diagnostic test recommendations for various types of cancer. Many of Watson’s features are also relevant to the healthcare domain. The second AI approach being developed by the author seeks to efficiently identify anomalous data in massive numbers of healthcare records. His system will automatically detect significant variations in care between groups, where the variations have significant (positive or negative) impacts of patient outcomes.
The article includes a more thorough description of the anomalous pattern detection techniques being developed, and the remaining challenges to applying them in both the hospital and broader healthcare settings.
ABOUT THE AUTHOR
Daniel B. Neill (neill@cs.cmu.edu) is the Dean’s Career Development Professor and associate professor of information systems at Carnegie Mellon University’s Heinz College, where he directs the Event and Pattern Detection Laboratory. His research interests include machine learning, data mining, and event detection in massive datasets. Neill has a PhD in computer science from Carnegie Mellon University.
