Inpatient Continuous Glucose Monitor Alerts Show Mixed Results in Trial

Continuous glucose monitoring (CGM) threshold alerts modestly improved inpatient glycemic control, but adding predictive alerts provided no further benefit, a three-arm trial showed.

Among 533 adults with type 1 or type 2 diabetes admitted to endocrinology wards, the average time spent in target glucose range (TIR; 70-180 mg/dL) was significantly higher among patients with threshold alerts turned on compared with patients who had all alerts turned off (adjusted mean difference 3.9%, 95% CI 0.1%-7.8%, Holm-adjusted P=0.04).

However, the mean TIR did not differ significantly between patients with both threshold and predictive alerts turned on versus those with all alerts off (3.1%, 95% CI -0.8% to 6.9%, Holm-adjusted P=0.11).

The results were consistent across sensitivity analyses, Jian Zhou, MD, PhD, of Shanghai Jiao Tong University School of Medicine in China, and colleagues reported in JAMA Network Open.

International consensus on CGM metrics for clinical trials considers a minimum 3% change in mean TIR clinically meaningful for a treatment group difference, the authors pointed out.

“Therefore, the observed 3.9% difference … may represent a meaningful change in the distribution of TIR, indicating that the threshold alerts of the CGM system had a beneficial effect on glucose control in the hospital,” Zhou’s group wrote.

At the individual level, a difference of 5% or more is generally considered the threshold for being clinically meaningful, “which has been associated with clinically significant benefits in outpatient populations with type 1 and type 2 diabetes,” they added.

No consensus has been established for hospitalized patients, but the American Diabetes Association recommends continuation of personal CGM use during hospitalization among people with diabetes. “No studies have evaluated the effect of CGM alerts in inpatient settings, where CGM systems are managed under clinical supervision rather than solely for patient self-care,” Zhou’s group noted.

The researchers suggested several factors that might explain why enabling threshold and predictive alerts simultaneously failed to improve time in range. Current predictive algorithms may not fully account for exogenous insulin administration, they noted. Staffing constraints and high alert frequency in hospital settings also may have created barriers to prompt clinical response. Alarm fatigue was not evaluated, but the threshold-plus-predictive group logged nearly 2,000 more alerts than the threshold-on/predictive-off group, they added.

In an accompanying commentary, Eileen Faulds, PhD, RN, of the Ohio State University Wexner Medical Center in Columbus, emphasized that both the study setting and patient population “differ substantially from typical inpatient practice” and must be considered when interpreting generalizability.

“Unlike pharmacologic interventions, the effectiveness of alarms depends not only on the technology itself but also on the clinical systems through which alarms are delivered, interpreted, and acted upon,” Faulds noted.

While threshold alerts followed a standardized response pathway with confirmatory point-of-care testing and protocol-driven treatment, predictive alerts were instead managed by clinical judgment.

“It remains unclear who received each alert, who held primary responsibility for responding, what actions were expected following a predictive alert, and how alerts were escalated or incorporated into the clinical workflow,” added Faulds. “Future studies should also routinely evaluate and report implementation fidelity, that is, whether alarms were delivered, acknowledged, and acted upon as intended.”

The single-center, open-label SIGNAL trial randomized patients in China 1:1:1 to one of three alert strategies. All participants were admitted for suboptimal glucose control. Most (62.1%) were male, average age was 62, and mean HbA1c was 9.2%. The trial ran from April to December 2024.

Two secondary outcomes favored the threshold-only group over the no-alert group: reduced time spent above range (>180 mg/dL) and a lower glycemia risk index. Other secondary outcomes — including time spent below range (<70 mg/dL and <54 mg/dL) — did not differ across groups.

None of the secondary outcomes showed significant differences when comparing the threshold-plus-predictive group to either cohort.

In subgroup analyses, patients not on intensive insulin therapy, those 65 and older, and those with type 2 diabetes achieved higher TIR in the threshold-only group compared to the alerts-off group. No significant treatment effect was observed among patients on intensive insulin therapy, those under 65, individuals with type 1 diabetes, or when stratified by baseline HbA1c (<9% vs ≥9%) or overall insulin use.

Two severe hypoglycemia events occurred in the no-alerts group, one in the threshold-plus-predictive group, and none in the threshold-only group. No cases of diabetic ketoacidosis, hyperosmolar hyperglycemic syndrome, or serious adverse events occurred.