Model predicts hypoglycemia in diabetics using EHRs, NLP

Researchers have developed a model leveraging electronic health records and natural language processing to better identify patients with diabetes who are at risk for low blood sugar.

Hypoglycemia occurs in 20 percent to 60 percent of diabetics and can result in serious adverse events, including cognitive impairment, coma and death.

However, researchers from the Indiana University School of Medicine, Merck and the Regenstrief Institute contend that their hypoglycemia prediction model identifies the patient risk factors for the condition that can potentially lead to improved interventions.

Also See: FDA approves patch-based glucose monitoring system

Using EHR data from nearly 39,000 diabetic patients over a 10-year period, investigators conducted a retrospective cohort study in which laboratory tests, diagnostic codes and NLP were used to identify hypoglycemia.

According to researchers, the strongest predictors of low blood sugar are infections, non-long-acting insulin, dementia and recent hypoglycemia.

"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,” says senior author Michael Weiner, MD, director of the Regenstrief Institute William M. Tierney Center for Health Services Research.

“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,” Weiner adds.

In the study, published in the journal Current Medical Research and Opinion, NLP was found to be useful in identifying hypoglycemia, given that there were not always laboratory tests to confirm episodes and it was often only documented in EHR clinical notes.

In fact, the study showed that hypoglycemia occurred in 8,128 patients, of which 539 were identified only by NLP, which improved identification of the condition.

“This study has implications for clinical support,” Weiner concludes. “The predictive model could lead to changes in practice as well as new strategies to help patients lower their risk of hypoglycemia.”

Going forward, researchers plan to study how a clinical decision support tool leveraging EHR data could be used to alert clinicians when their patients are at risk for hypoglycemia. They are also conducting an outpatient study in which wearable devices will collect diet, physical activity and medication adherence information for diabetic patients—as well as continuous glucose levels—to better predict hypoglycemia.