To optimize the management of diabetes, the long-term risks of associated complications must be balanced with risks for hyperglycemia and hypoglycemia. “Because the short-term consequences of diabetes can result in immediate disability and death, the fear of hypoglycemia often leads to less aggressive insulin therapy,” says Madhav Erraguntla, PhD. “This increases patients’ long-term exposure to hyperglycemia and risks for debilitating complications later in life.”
Real-time continuous glucose monitoring (CGM) devices allow for frequent, automated glucose readings using interstitial fluid in subcutaneous tissue. “Early detection of impending hypoglycemic events based on CGM readings—with high sensitivity and specificity and low false alarm rates—can help patients with diabetes better manage hypoglycemia and their overall health,” says Dr. Erraguntla.
Researchers studying predictive alerts for hypoglycemia based on CGM readings have reported achieving sensitivity and specificity rates of about 90%, according to Dr. Erraguntla. “However, due to small numbers of hypoglycemic events compared with non-hypoglycemic events, even this relatively high sensitivity and specificity can lead to false alert rates of around 80%,” he says. “Improving specificity in such highly imbalanced class situations may reduce the false alert rate and therefore improve user experience and trust in these alerts. This may ultimately facilitate persuasive adoption.”
“It’s exceedingly important to reduce false alert rates when it comes to wearable devices for health monitoring in general, given the exponential increase in data curation to patients today,” says Balakrishna Haridas, PhD. “Patient compliance with medical care modifications initiated by alerts will be much greater if false alert rates are diminished or, ideally, eliminated.”
Low False Alert Rate
For a study published in JMIR Diabetes, Dr. Erraguntla, Dr. Haridas, Daniel DeSalvo, MD, and colleagues aimed to develop a prediction model for hypoglycemic events with low false alert rates, high sensitivity and specificity, and good generalizability to new patients and time periods. The authors evaluated performance improvement by focusing on sustained hypoglycemic events, which was defined as glucose values less than 70 mg/dL for at least 15 minutes. The researchers used patient-based and time-based validation to address the generalizability and robustness of the model. In total, the study analyzed data from 110 patients during 30-90 days, comprising 1.6 million CGM values under activities of daily living (ADL) conditions.
“In our study, we discovered that focusing predictive models on sustained hypoglycemic events instead of all hypoglycemic events reduced the false alert rate and improved sensitivity and specificity,” says Dr. Erraguntla (Table). “It also resulted in models that have better generalizability to new patients and time periods. In addition, the model accurately predicted sustained events with higher than 97% sensitivity and specificity for both 30- and 60-minute prediction horizons. Furthermore, the false alert rate was reduced to below 25%.”
The study team noted that their robust prediction model provides high-quality alerts for sustained hypoglycemic risk in patients with type 1 diabetes. Since the false alert rate was kept low, this may lead to improved user trust in, and adoption of, CGM-based alerts. Having an accurate and actionable hypoglycemia prediction model with low false alert rates is essential to the durability of CGM in diabetes management.
Optimizing CGM Use in Diabetes Management
Collectively, findings from the study have important implications for sustaining CGM use and optimizing glycemic control with fewer hypoglycemic events, improved confidence, and potentially improved glucose levels. “Utilizing alerts generated from predictive models that accurately predict low glucose levels will help reduce hypoglycemic events,” says Dr. DeSalvo. “It will also give people with diabetes the confidence they need to safely and effectively manage their condition.”
Dr. DeSalvo says the current study was based on pediatric patients younger than 20, but future research will expand to non-pediatric patients with type 1 diabetes. The predictive model presented in the study will also be implemented in a smartphone app in an upcoming clinical pilot study at Texas Children’s Hospital and Baylor College of Medicine. “Furthermore, we’re developing non-invasive sensor alternatives to CGM devices for real-time detection of blood glucose levels,” Dr. Erraguntla adds. “With more research, we hope to find ways to better optimize use of CGM in the management of diabetes.”
