Search Results

DBLG2, a mobile application with algorithm technology, is intended for use with compatible integrated continuous glucose monitors (iCGM) and alternate controller enabled (ACE) insulin infusion pumps to automatically increase, decrease, and suspend delivery of basal insulin based on iCGM readings and predicted glucose values. It can also deliver correction boluses when glucose values are predicted to exceed a predefined threshold. To do this, the DBLG2 software takes into account the patient's profile, glycemia (current and predicted), announced meals and physical activities.

DBLG2 is intended for the management of type 1 diabetes mellitus in persons 12 years of age and greater.

DBLG2 is intended for single patient use.

DBLG2 is Rx - For Prescription Use Only.

Diabeloop DBLG2 is an Android application installed on patient personal mobile phone, intended for managing glucose levels in people with type 1 diabetes, using a hybrid closed loop approach (automated insulin delivery). It is designed to be connected with a compatible Automated Controller Enabled (ACE) insulin pump and integrated Continuous Glucose Monitors (iCGM).

DBLG2 has a regulation algorithm to automatically manage the patient's blood glucose level. It takes as input glycemia value received from the CGM, personal patient medical parameters and patient input related to meals and physical activities, and it calculates every 5 minutes the amount of insulin to deliver in order to keep the patient in the normoglycemia bounds. It sends this information to the pump that automatically delivers this quantity of insulin.

The software can ask the pump to deliver:

• A meal bolus
• A correction bolus (small amount of insulin)
• A basal rate over a given period of time.

The software can also ask the patient to take a calculated amount of carbohydrates if the system determines that the patient would go into hypoglycemia even if the insulin basal rate is brought down to zero.

DBLG2 acts mostly by modulating the basal rate of insulin delivery, but in some cases can deliver, automatically, correction boluses. It includes a patient-confirmed meal bolus calculator that simplifies meal dosing by allowing the patient to enter their meal carbohydrate amount while the system retrieves the patient's personalized insulin dosing parameters from their profile. The system calculates and displays a recommended meal bolus dose, which the patient must review and confirm before delivery is initiated.

In addition, DBLG2 has a self-learning module that applies improvements to the patient's algorithm parameters, based on estimated glycemia history and insulin delivery quantities, from the patient's history.

DBLG2 is designed to be secure, with by-design and structural security mechanisms that prevent from both hypoglycemia and hyperglycemia:

• Hypoglycemia: by detecting an existing or upcoming hypoglycemia:

  • The algorithm cuts down insulin delivery if a risk of hypoglycemia exists within the next fifteen minutes, the system will ask the user to take carbohydrates (by an alert).
  • If the patient is in hypoglycemia below 55mg/dL (= 3.1 mmol/L), an alarm is triggered.

• Hyperglycemia: the algorithm orders the delivery of insulin correction bolus to reduce the glycemia.

The software also guarantees patient's safety by returning to the pre-programmed basal pattern if the automatic regulation cannot be done for any reason (including loss of communication with the CGM or pump) or if the patient wishes to return to manual control by stopping the automatic regulation.

The algorithm includes appropriate alerts/alarms in case of any malfunction of one of the components.

Here's an analysis of the acceptance criteria and the study proving the device meets them, based on the provided FDA 510(k) clearance letter for Diabeloop DBLG2:

1. Table of Acceptance Criteria and Reported Device Performance:

The document explicitly states that the DBLG2 software's clinical performance has been demonstrated through "consistent results in 6 prospective clinical trials." While specific numerical acceptance criteria (e.g., minimum Time in Range percentage, maximum hypoglycemia duration) are not clearly detailed in the provided text as a formal table of acceptance criteria, the document implies that the device is deemed safe and effective based on improvements in several glycemic outcomes compared to current treatment or a predicate device.

2. Sample Size Used for the Test Set and Data Provenance:

• Test Set Description: The clinical performance is supported by "6 prospective clinical trials."
• Sample Size:
  • Adults: Total of 15,325 patient-weeks of data collected.
  • Adolescents (with T1D): Total of 1,594 patient-weeks of data collected.
• Data Provenance: The trials were "prospective clinical trials." The provided text doesn't explicitly state the country of origin for the data collection, but the submitter is based in Grenoble, France, which suggests these studies could have been conducted in Europe or internationally.

3. Number of Experts Used to Establish the Ground Truth for the Test Set and Qualifications:

The document does not provide details on the number of experts used or their specific qualifications (e.g., "radiologist with 10 years of experience") for establishing ground truth within the clinical trials. The phrase "ground truth" generally refers to definitive, objectively measured outcomes. For a glycemic control system, the primary ground truth would be the continuous glucose monitoring (CGM) data itself, along with lab-measured HbA1c, and recorded hypoglycemic/hyperglycemic events, which are direct physiological measurements rather than expert interpretations.

4. Adjudication Method for the Test Set:

The document does not specify any adjudication method (like 2+1, 3+1, none) for the test set. Given the nature of objective physiological measurements (CGM, HbA1c), manual adjudication of "ground truth" by multiple experts is less common than in, for example, image interpretation studies. The "ground truth" is largely the direct physiological data collected.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was Done:

No, an MRMC comparative effectiveness study was not done. MRMC studies are typically relevant for diagnostic aids where human readers interpret medical images or data, and the AI's assistance is being evaluated regarding its impact on their diagnostic accuracy. This device is an automated glycemic controller, not a diagnostic interpretation tool, so an MRMC study is not applicable. Therefore, there is no mention of an effect size for human reader improvement with AI assistance.

6. If a Standalone (Algorithm Only Without Human-in-the-Loop Performance) was Done:

Yes, the studies evaluate the performance of the DBLG2 software, which is an "algorithm technology" designed to "automatically increase, decrease, and suspend delivery of basal insulin" and "deliver correction boluses." While it integrates with user input (meal announcements, physical activities) and requires patient confirmation for meal boluses, the core "automatic regulation" of insulin delivery based on CGM readings and predictions represents its standalone algorithmic performance within the hybrid closed-loop system. The clinical trials assess the overall system performance, which inherently includes the algorithm's performance without constant human override of every automated decision.

7. The Type of Ground Truth Used:

The ground truth used is primarily physiological outcomes data, specifically:

• Continuous Glucose Monitoring (CGM) values, from which Time in Range (TIR), time in hypoglycemia, time in hyperglycemia, and mean CGM values are derived.
• HbA1c measurements.
• Documented occurrence of severe metabolic episodes.
• Patient feedback on satisfaction.

8. The Sample Size for the Training Set:

The document does not explicitly state the sample size used for the training set for the DBLG2 algorithm. The provided information focuses on the data used for clinical validation/testing. The DBLG2 system also includes a "self-learning module that applies improvements to the patient's algorithm parameters, based on estimated glycemia history and insulin delivery quantities, from the patient's history," which implies ongoing, personalized "training" within each patient's use. However, the initial development and training data for the core algorithm are not detailed here.

9. How the Ground Truth for the Training Set was Established:

The document does not detail how the ground truth for the training set was established. Given the nature of the device, it is highly likely that for the initial algorithm development and training, retrospective or simulated physiological data, potentially from a variety of sources (e.g., historical patient data, simulated diabetes models), would have been used. The "self-learning module" uses the patient's own glycemia history and insulin delivery quantities to refine parameters, establishing its internal "ground truth" from these real-time individual patient data points.

Ask a specific question about this device