Becker's Clinical Quality & Infection Control

44 DATA ANALYTICS & AI AI can diagnose diseases on par with physicians, study finds By Mackenzie Bean A rtificial intelligence models can diagnose dis- eases just as well as human clinicians, suggest the findings of a study published in The Lancet Digital Health. For the study, researchers analyzed 14 studies pub- lished between 2012 and 2019. The studies compared the diagnostic performance of AI models to healthcare professionals based on a review of medical images. On average, AI algorithms offered correct diagnoses 86.4 percent of the time, compared to 87 percent for clinicians. AI models also accurately identified patients who did not have the disease 92.5 percent of the time. This rate was 90.5 percent for clinicians. Despite this finding, researchers noted many studies comparing diagnostic ability contain methodological deficiencies and poor reporting. "These issues are pertinent for ensuring studies of deep learning diagnostics … are of sufficient quality to evaluate the performance of these algorithms in a way that can benefit patients and health systems in clinical practice," the study authors concluded. n AI system interprets chest X-rays in 10 seconds By Anuja Vaidya A n automated chest X-ray interpretation system that uses artificial intelligence was able to identify key findings in X-rays accurately in about 10 seconds, according to researchers from Intermountain Healthcare in Salt Lake City and Stanford (Calif.) University. The researchers detailed their findings at the European Respiratory Society's International Congress in Madrid, Spain, Sept. 28-Oct. 2. The system, called CheXpert, was developed by the Stanford Machine Learning Group. For the study, the system reviewed X-ray images taken at several emergency departments at Intermountain hospitals in Utah. Radiologists categorized chest images from 461 Inter- mountain patients as being "likely," "likely-uncertain," "unlikely-uncertain," or "unlikely" to have pneumonia. They also pinpointed patients whose X-ray images showed pneumonia in multiple parts of the lungs. The system was then used to review the chest images and performed comparably to the radiologists. The CheXpert system also created radiology reports for all key pneumonia findings in about 10 seconds. n Geisinger partners with IBM for AI-based sepsis prevention By Andrea Park D anville, Pa.-based Geisinger Health System and the IBM Data Science Elite team announced Sept. 10 a collaboration to develop an artificial intelligence model to detect sepsis mortality risk. Geisinger first developed the project before tapping IBM's Data Science and AI Elite teams to construct the predictive model and an accompanying tool to gath- er the latest sepsis research. The model used machine learning to analyze a decade's worth of de-identified EHR data comprising thousands of sepsis patients in order to identify clinical biomarkers associated with higher rates of mortality due to sepsis. Using that analysis, the model will be able to predict mortality in sepsis patients during hospitalization or the subsequent 90 days. Geisinger clinicians can also use the model's analysis to develop personalized care plans for at-risk sepsis patients, in hopes of increasing their chances of recovery. n WakeMed deploys predictive analytics software to detect patient health decline By Jackie Drees W akeMed Health & Hospitals implement- ed predictive analytics software from Pera Health, a software-as-a-service company, that sorts through EHR data to identify potential predictors for decline in a patient's health. The Raleigh, N.C.-based health system deployed the Rothman Index software in September at WakeMed Cary (N.C.) Hospital with a 47-day implementation process. WakeMed plans to integrate the software throughout the rest of the 941-bed health system later this year. The Rothman Index uses an algorithm that automat- ically integrates a patient's EHR data, including age, disease and care setting, with the individual's current vitals to detect problems in advance using predictive analytics. It can identify changes in patient condition hours or days earlier than existing vitals-based algo- rithms, according to the news release. n

• Cover