Wednesday, July 31, 2019

Natural Language Processing Detects High-Risk Diabetic Patients


A natural language processing tool was able to identify diabetic patients at high risk of low blood sugar.
July 29, 2019 - Using natural language processing tools could help providers better detect low blood sugar in patients with diabetes, leading to improved chronic disease management, revealed a study published in the journal Current Medical Research and Opinion
Low blood sugar, or hypoglycemia, occurs in 20 percent to 60 percent of patients with diabetes, researchers said. Many individuals with diabetes are unaware when hypoglycemia occurs, especially if they have recurring episodes of low blood sugar. 
Intervening and preventing hypoglycemia is critical for patients, as the condition can lead to adverse events like cognitive impairment, coma, and even death. 
Researchers from Regenstrief Institute, Indiana University School of Medicine, and Merck gathered EHR data across ten years covering nearly 39,000 patients with diabetes. The team used laboratory tests, diagnostic codes, and natural language processing (NLP) technology to identify episodes of hypoglycemia among individuals with diabetes. 
The group found that NLP was a viable way to identify hypoglycemia, because there weren’t always lab tests to confirm the episode. Often, hypoglycemic episodes were only recorded in narrative clinical notes. 
Researchers also found that the strongest predictors of hypoglycemia are recent infections, using insulin other than long-acting insulin, recent occurrences of hypoglycemia, and dementia. Variables associated with the lowest risk of low blood sugar were using long-acting insulin in combination with other drugs, as well as being 75 years of age or older. 
“Knowledge of these factors could assist clinicians in identifying patients with higher risk of hypoglycemia, allowing them to intervene to help their patients in lowering that risk,” said Michael Weiner, MD, MPH, director of the Regenstrief Institute William M. Tierney Center for Health Services Research and the senior author of the study. 
“Some factors influencing hypoglycemia may not be immediately obvious. In addition, reassessing hypoglycemia risk as a patient's health status changes may be important as new factors are identified.”
NLP tools have demonstrated their value in a wide range of areas in healthcare. The technology has previously shown its ability to detect social determinant search terms in the EHRs of high-risk patients, to help patients better understand the data in their EHRs, and to accurately record medical conversations between patients and providers. 
Market analysts have also noted the strong impact that NLP and other artificial intelligence technologies are expected to make on clinical decision support tools. In 2017, a series of market reports discussed the growing demand for clinical decision support tools that feed on unstructured data, and the important role NLP will play in advancing these tools. 
“Used as a part of artificial intelligence systems, applications of NLP technologies are being deployed for predictive analysis and clinical decision support systems,” one report said. “The global healthcare natural language processing market is expected to receive an impetus from the uptake of these technologies by several companies for extracting knowledge from several clinic documents via machine learning or deep learning applications.”
“The growing volume of unstructured clinical data and the unstoppable penetration of EHR systems are expected to fuel the growth of this market in the coming years.”
The researchers on the hypoglycemia study said they will examine the development of a clinical decision support tool that examines EHR data to tell clinicians when their patients have hypoglycemic risk factors. 
The team will conduct an outpatient study that uses wearable devices to monitor and record the actions and continuous glucose levels of people with diabetes. 
The information collected includes physical activity, diet and adherence to medication regimens, all of which is not typically included in medical records. The goal is to identify patterns that enable providers to detect hypoglycemia earlier. 
“This study has implications for clinical support,” said Weiner. “The predictive model could lead to changes in practice as well as new strategies to help patients lower their risk of hypoglycemia.”

No comments:

Post a Comment