During her high school years, Siru Liu volunteered in a large hospital in southwestern China and immediately noticed a problem: Doctors could only spend two to three minutes with patients who sometimes traveled long distances to get care.
Access for rural patients was particularly strained. The doctors, who at the time were switching to electronic records, needed help making accurate decisions with imperfect technology and incomplete information.
“At that moment I started thinking, can we use AI to solve this situation and create a more efficient health care environment to let more people have access?” said Liu, now a tenure-track assistant professor at Vanderbilt University Medical Center.
She has pursued the answer to that question with singular intensity. Liu’s research focuses on using artificial intelligence to help predict which patients might become dangerously ill and reduce disparities in outcomes. In that field, she has authored or co-authored 32 papers published in influential journals, such as the Journal of the American Medical Informatics Association — and her work is already having an impact on care practices.
At Vanderbilt, she developed an AI tool to predict delirium in ICU patients, using hundreds of data variables to deliver a score to help physicians identify patients at risk of developing the condition. In more recent work, Liu has also begun exploring the power of generative AI models, finding in one recent study that ChatGPT could help suggest ways to improve the relevance of automated alerts in electronic health records.
But she also is not yet sold on the technology’s ability to reliably improve care out of the box. “How do we teach it to learn how to do a task and put clinical knowledge into generative AI, so that ChatGPT can become DoctorGPT, and not just a chat?” Liu asked. That is the question she will address next.
— Casey Ross