Researchers from Johns Hopkins University have created ML algorithms that can recognize the early indications of delirium and predict which patients are at a high risk of developing delirium during their stay in an intensive care unit (ICU).
Delirium is a sudden confusion and severe disturbance in mental functioning. It is most commonly seen in elderly individuals, but can affect people of any age. Delirium can cause a wide range of symptoms, including disorientation, memory loss, restlessness, agitation, hallucinations, speech difficulties, and changes in behavior. It is usually caused by an underlying medical condition or medication, and can be a sign of a more serious problem. According to the study’s press release, Approximately one third or more of those admitted to the hospital, and up to 80% of individuals in an intensive care unit (ICU), have delirium during their hospitalization. Treating delirium typically entails resolving the root cause, in addition to providing supportive care. Care bundles, early physical and occupational therapy, and alterations in medication may constitute successful anti-delirium interventions; however, due to the unpredictable needs of delirium patients and a lack of time and resources, they are not always utilized.
According to Robert Stevens, M.D., associate professor of anesthesiology and critical care medicine at the Johns Hopkins University School of Medicine and senior author of the results published in the Dec. 20 issue of the journal Anesthesiology, “it is essential to be able to distinguish between patients at low and high risk of delirium in the ICU, as this allows us to allocate more resources to the high-risk population.” Stevens is also the director of precision medicine and informatics, and co-director of the Johns Hopkins Precision Medicine Center of Excellence in Neurocritical Care.
Undergraduate and master’s level engineering students, under the teaching of Stevens, developed the new AI program to address this problem. Two computerized models were created to predict delirium risk from a dataset of over 200,000 ICU stays from 208 hospitals around the country. The static model evaluates a single snapshot of patient info, including age, severity of illness, diagnoses, physiological variables and medications, to forecast delirium risk. In contrast, the dynamic model tracks data during the hospital stay, including blood pressure, pulse and temperature readings, to give a consistently updated delirium risk prediction over a period of 12 hours.The researchers tested their AI models on two data sets from a Boston hospital, containing over 100,000 ICU stays. The accuracy of the first 24-hour model in predicting delirium was 0.785 (95% CI), while the dynamic model was even more successful, achieving a maximum of 90%.
“The underlying idea was that this routinely collected data stored in patients’ electronic health records contains signatures that are associated with delirium risk,” says Kirby Gong, a recent master’s degree graduate from the Johns Hopkins Department of Biomedical Engineering and first author of the new technology.
AI is proving to be a valuable asset in precision medicine, providing researchers with powerful tools to treat and predict illnesses. The algorithms developed by Stevens and his lab are significantly advancing translational and clinical research and are being applied to historical patient data from Johns Hopkins Medicine ICUs. He and his lab are collaborating with engineering students and faculty to apply AI approaches to predicting stroke, heart failure, pulmonary embolisms, and other emergent conditions in critical care medicine. Stevens believes that many physiological changes may show subtle early warning signs that could be detected with the use of artificial intelligence-assisted pattern analysis and that these signs may not be visible to a clinician. Should the upcoming clinical trial be successful, AI will prove to be a powerful force in improving patient outcomes and a more efficient medical system.