Becker's Clinical Quality & Infection Control

27 DATA INFORMATICS AND ANALYTICS Northwell Rolls Out One-of-a-Kind Flu Tracker for Long Island, NYC By Brian Zimmerman N ew Hyde Park, N.Y.-based Northwell Health established a one- of-a-kind biosurveillance system to track the large number of flu patients presenting at Northwell hospitals in Long Island and New York City. The biosurrveillance dashboard allows Northwell to track flu cases at integrated hospitals to more efficiently manage the distribution of resources, such as antiviral medications, rapid-flu tests, masks and gloves. The dashboard also allows the health system to make informed decisions about where to funnel additional staff members and where to house flu patients throughout the hospital. "The new flu surveillance dashboard gives Northwell the ability to see what's going on in near real-time," said Mark Swensen, an emergency management coordinator at Northwell who helped develop the sys- tem. "This is an important tool that helps us in terms of planning and managing patient care." The surveillance system allows Northwell to track both the number of positive flu deaths and the number of patients who present at hospi- tals with flu-like symptoms. In the future, the surveillance system will be used to track the spread of infectious disease beyond flu, enabling Northwell to better respond to public health crises and coordinate efforts with state, local and federal health officials. As of Feb. 6, 12 Northwell hospitals adopted the tracking system and treated 10,500 patients with flu-like symptoms. Five more of the health system's acute care hospitals are poised to link up with the tracking system in the coming months. n Emory Researchers Use AI to Predict Sepsis Onset in ICUs By Jessica Kim Cohen A team of researchers from Atlanta-based Emory University developed an artificial intelligence algorithm to predict a pa tient's likelihood of developing sepsis in real time, according to a study published in the journal Critical Care Medicine. To create the algorithm, the researchers considered vital sign and EMR data from 27,000 intensive care unit admissions at two Emory University hospitals. The algorithm analyzes 65 key variables to pre- dict a patient's likelihood of developing sepsis after 12, eight, six and four hours. The researchers validated the algorithm with data from 42,000 patients held in the publicly available Medical Information Mart for Intensive Care-III database. Although the study authors acknowledged a prospective study is needed to determine the clinical utility of the prediction model, they concluded, "Using data available in the ICU in real-time, [the algo- rithm] can accurately predict the onset of sepsis in an ICU patient four to 12 hours prior to clinical recognition." n Google Predicts Medical Outcomes With 46B Data Points, Artificial Intelligence: 5 Things to Know By Jessica Kim Cohen A team of Google researchers developed an artifi- cial intelligence system that predicts medical outcomes like mortality and readmissions based on patient data held in EHRs, Quartz reported. e team's results were published in a research paper submitted to the scientific preprint website arXiv.org Jan. 24. e research paper has not been peer reviewed. Here are five things to know about the research project. 1. To construct a predictive model, researchers typically extract select variables from standardized data. How- ever, to avoid what Google researchers referred to as a "labor-intensive process," they attempted to use deep learning, an advanced machine learning technique that doesn't require standardized data. 2. e researchers developed a deep learning approach to analyze a patient's raw EHR record. e deep learn- ing system evaluated data from thousands of patients to determine words and events associated with certain outcomes, while also identifying which data could be ignored. 3. To validate the deep learning approach, research- ers used de-identified EHR data from 216,221 adult patients who were hospitalized for 24-plus hours at UC San Francisco Medical Center and University of Chi- cago Medicine. is volume of EHR data encompassed 46,864,534,945 data points, including clinical notes. 4. e deep learning models achieved high accuracy for predicting in-hospital mortality, 30-day unplanned readmissions, prolonged length of stay and all of a patient's final discharge diagnoses, according to the research paper. 5. e study authors concluded their deep learning approach outperformed existing "state-of-the-art tradi- tional predictive models." "We believe that this approach can be used to create accurate and scalable predictions for a variety of clini- cal scenarios, complete with explanations that directly highlight evidence in the patient's chart," the study authors wrote. n

• Cover