Predictive analytics for clinical decision support
Sepsis is an overwhelming immune response to infection, which damages its own tissues and organs. This process can happen at any age, regardless of health condition, and many times from seemingly benign incidents. Severe sepsis, sepsis with acute organ dysfunction, strikes about 18 million people annually (750,000 cases in the United States) and has a high short-term mortality risk (28% to 50%). We trained machine learning models with the Electronic Health Records of 1497 patients that were collected from the ICU of the UC Davis Medical Center. We tailored Partially Observable Markov Decision Processes (POMDP) to identify the optimal policy (actions), given the data, as well as regressors and classifiers to predict mortality and length-of-stay.
Publications:
- Gultepe E, Green JP, Nguyen H, Adams J, Albertson T, Tagkopoulos I. From vital signs to clinical outcomes for sepsis patients: A clinical decision support system based on discriminative classification. JAMIA2014 Mar-Apr;21(2):315-25. PubMed PMID: 23959843.(link)
- Tsoukalas A., Albertson T., Tagkopoulos, I. A data-driven, probabilistic machine learning approach to decision support for patients with Sepsis,,JMIR Med Inform.2015 Feb 24;3(1):e11. PubMed PMID: 25710907.(link)
Collaborators:
Tim Albertson M.D. (Chair, Internal Medicine), Hien Nguyen M.D., Jeff Green M.D., Jason Adams M.D.
Trainees involved:
Athanasios Tsoukalas (Postdoc), Eren Gultepe (MS student).
Funding:
CITRIS and CTSC (Seed funding).