Samantha Kleinberg, associate professor in the Department of Computer Science and leader of the Health and AI Laboratory, recently received a grant of $1,139,620 from the National Institutes of Health (NIH) for her project “BIG DATA: Causal Inference in Large-Scale Time Series.”
Large datasets generated by hospitals could have a transformative effect on medical knowledge and patient care. Yet currently the volume of data is more likely to overwhelm clinicians, and the challenges of the data can overwhelm machine learning algorithms.
Intensive care units (ICUs) are filled with monitors that generate data at a resolution of seconds, for the entirety of a patient’s stay, yet these data are not always stored. In ICUs that keep the data, researchers quickly find that many popular methods from artificial intelligence and machine learning do not work.
While there may be hundreds of thousands of data points on each patient, the number of patients is more likely to number in the dozens. That's a problem for neural networks, which need many thousands of examples to find patterns. Even more challenging, different variables are measured for different patients, as these recordings depend on a patient's prognosis and disease severity. Further, clinicians need interpretable models that get at causation – not just correlation – to guide their choices.
Over the next four years, Kleinberg's group, comprised of collaborators from Columbia and McGill Universities, will be developing methods that can piece together data from different patients to identify a complete causal model. This work will turn a weakness of the data – that patients have different sets of only partly overlapping variables – into a strength.
One of the major challenges of AI for health is a lack of training data with ground truth. This makes it difficult for researchers to know when their methods work and under what conditions. Simulated data can help, but it is difficult to create simulations that represent real-world performance. Conducted in collaboration with computer science colleague Professor Giuseppe Ateniese, this project will develop modular, data-driven simulations that preserve privacy. The first test of these methods will be understanding consciousness and why it changes in neurological ICU patients, and neurological status in neonatal ICU patients.
Her group will focus in particular on understanding consciousness in adults, and neurologic status in neonates. While 7% of ICU admissions are due to loss of consciousness — and degree of consciousness is critical to evaluating prognosis, making difficult choices such as when to withdraw care, and providing early interventions to improve quality of life — there are no objective or automated assessments for consciousness (adults) or neurologic status (neonates). Research has shown that unresponsive patients with brain activation were twice as likely to regain the ability to follow commands compared to unresponsive patients without such activation, yet these assessments are too time-consuming for regular clinical use. However, Kleinberg’s work has also shown that physiological data routinely collected in ICUs can be used as a proxy to classify consciousness. Her work in this area earned the Homer R. Warner Award at the 2019 AMIA Annual Symposium.
Her long-term goal is to turn this data into actionable knowledge, like risk factors for a disease, early intervention targets, and real-time information to support clinical decisions. This is a broad problem, but particularly important in ICUs, which involve high stakes decisions being made in a complex environment under time pressure.
Kleinberg said, “It was surprising that we were able to use physiological data to classify each patient's degree of consciousness, as prior work has relied mainly on fMRI and EEG. Using this data, though, lets us develop methods that can be used continuously around the clock. I'm excited to build on this to figure out why consciousness changes and to expand our work to a new population (neonates) with McGill.”
Learn more about computer science at Stevens: