Soleil Research | Evidence Review
Continuous glucose monitoring can provide far more glucose information than intermittent bedside testing, but in the hospital the value of that information depends partly on how alerts are configured and acted on. The SIGNAL randomized clinical trial tested whether threshold and predictive CGM alerts improve glycemic control in hospitalized adults with diabetes.
Plain-Language Summary
In 533 hospitalized adults with diabetes, turning on simple high- and low-glucose threshold alerts modestly increased the percentage of time patients spent in the target glucose range of 70–180 mg/dL compared with using the same real-time CGM system with alerts turned off. Adding predictive alerts designed to warn about glucose excursions 15 minutes in advance did not improve time in range further.
This does not show that more alerts are always better, nor does it establish that CGM should replace conventional inpatient glucose monitoring. Every study group used real-time CGM, frequent point-of-care glucose testing and a structured specialist diabetes-care workflow. The trial specifically tested the incremental value of different alert configurations.
Why This Study Matters
Hospital glucose management requires clinicians to detect hyperglycemia and hypoglycemia quickly while avoiding unnecessary interruptions. CGM can continuously display glucose trends, but excessive alerts may increase workload without improving outcomes. SIGNAL is useful because it randomized patients to different alert strategies while keeping the underlying CGM platform and inpatient care structure broadly consistent across groups.
What the Study Tested
SIGNAL was a single-center, open-label randomized clinical trial conducted at Shanghai Sixth People’s Hospital from April 15 to December 9, 2024. A total of 533 adults with diabetes were randomized 1:1:1 to one of three real-time CGM strategies:
- Alerts off: 178 participants.
- Threshold alerts only: 177 participants.
- Threshold plus predictive alerts: 178 participants.
All participants used the Guardian Connect system with the Guardian Sensor 3. Point-of-care glucose testing was performed five times daily, and insulin adjustments incorporated both CGM and bedside glucose information. A dedicated advanced-practice nurse monitored CGM alerts and glucose graphs. The prespecified primary outcome was the percentage of CGM time spent between 70 and 180 mg/dL.
What the Study Found
Threshold alerts produced a modest improvement in time in range
Mean time in range was 71.2% with alerts off, 75.1% with threshold alerts only and 74.7% with threshold plus predictive alerts. Compared with alerts off, threshold alerts increased time in range by an adjusted 3.9 percentage points (95% CI, 0.1 to 7.8; Holm-adjusted P = .04).
Predictive alerts did not add measurable glycemic benefit
The threshold-plus-predictive strategy did not significantly improve time in range compared with alerts off, and it did not outperform threshold alerts alone. The difference between the predictive-plus-threshold group and the threshold-only group was -0.8 percentage points (95% CI, -4.7 to 3.0).
Hyperglycemia improved, but hypoglycemia did not clearly differ
Threshold alerts reduced time above 180 mg/dL by 3.9 percentage points compared with alerts off. Time below 70 mg/dL did not differ significantly among groups.
More alerts did not translate into better glucose control
The threshold-plus-predictive arm generated 7,012 alerts among 149 participants, with a median of 18 alerts per participant and 2.6 alerts per patient-day. The threshold-only arm generated 5,163 alerts among 117 participants. The study authors discussed alarm fatigue as a possible explanation for why additional predictive alerts did not improve outcomes, but the trial did not directly measure alarm burden, usability or clinician experience. Therefore, alarm fatigue should be considered a hypothesis rather than a demonstrated causal mechanism.
Methodological Quality
Overall assessment: strong randomized comparative evidence for alert configuration within a specialized inpatient diabetes workflow, with important limits on generalizability and clinical-outcome inference.
The study was prospectively registered, followed CONSORT reporting, used computer-generated randomized allocation prepared by an independent statistician, and employed an objective CGM-derived primary endpoint. The design is particularly informative because all groups used the same real-time CGM platform, allowing the investigators to isolate the incremental effect of alert strategy rather than comparing CGM with conventional monitoring.
However, the clinical effect was modest and the lower confidence limit for the primary comparison was close to zero. The trial was not designed to establish effects on major complications, mortality, readmission or other hard clinical outcomes.
Strengths
- Prospective randomized design with 533 participants.
- Prospective registration and CONSORT-based reporting.
- Independent computer-generated randomization sequence.
- Objective CGM-based primary outcome.
- Active-comparator design that isolates the incremental value of alert settings.
- Holm adjustment for the primary pairwise comparisons.
- Detailed reporting of alert frequency and glycemic metrics.
Limitations and Cautions
- The trial was conducted at a single tertiary center on endocrinology wards with a specialized diabetes team, limiting generalizability to other inpatient settings.
- The study was open label, so clinicians and participants knew the assigned alert strategy.
- A high proportion of participants received continuous subcutaneous insulin infusion during hospitalization, which may not reflect routine practice elsewhere.
- The intervention period was relatively short, and the primary outcome was a sensor-derived glycemic metric rather than a hard clinical endpoint.
- The 3.9-percentage-point improvement in time in range was statistically significant but modest, with a confidence interval extending from 0.1 to 7.8 percentage points.
- The trial was not powered to establish differences in uncommon safety outcomes such as severe hypoglycemia.
- Alarm burden, usability and clinician experience were not directly assessed, so alarm fatigue cannot be inferred as the cause of the predictive-alert result.
- Detailed timing and dosing of temporary insulin adjustments and hypoglycemia treatment were not collected.
Funding and Conflicts of Interest
The investigators reported support from public and institutional grant programs and stated that the funders had no role in study design, data collection, analysis, interpretation, manuscript preparation or the decision to publish. The authors reported no conflicts of interest. Robert Vigersky of Medtronic reviewed the final manuscript and provided suggestions without compensation. Because the study used a Medtronic CGM platform, that involvement is worth disclosing, but the published report does not describe Medtronic as a study funder or analytical decision-maker.
Soleil Insight
More alerts are not automatically better. In SIGNAL, simple threshold alerts produced a small improvement in time in range, while adding predictive alerts increased the alert volume without producing additional glycemic benefit. The clinically important question is therefore not only whether an algorithm can generate an earlier warning, but whether that warning changes care enough to justify the interruption and response workload it creates.
That conclusion should remain bounded by the study design. SIGNAL does not establish that predictive alerts are ineffective in every hospital, nor does it prove that alarm fatigue explains the result. It shows that within this specialist inpatient workflow, the additional predictive-alert layer did not measurably improve glycemic control beyond threshold alerts alone.
Original Publication
Wang Y, Cao L, Lu J, et al. Threshold and Predictive Alerts of Continuous Glucose Monitoring and Glycemic Control in Hospitalized Adults With Diabetes: A Randomized Clinical Trial. JAMA Network Open. 2026;9(8):e2629291. Published August 20, 2026. doi:10.1001/jamanetworkopen.2026.29291.
View the original peer-reviewed publication
ClinicalTrials.gov: NCT05941286
Featured Image Attribution
Featured image: MailariX / Wikimedia Commons, licensed under CC BY-SA 4.0. The image shows a Medtronic Guardian Sensor 3 glucose sensor and Guardian Link 3 transmitter attached to skin. It illustrates the device type used in the study and does not depict a SIGNAL trial participant. Image source.
Advancing Science. Empowering Humanity.
Soleil Research provides evidence-focused scientific education and does not replace clinical judgment, institutional policy or professional medical guidance.