Prediction and Prevention of Nocturnal Hypoglycemia in Persons With Type 1 Diabetes Using Machine Learning Techniques

Hospital Clinic of Barcelona10 enrolled

Overview

The objective is to develop a novel system to predict and prevent nocturnal hypoglycemia in type 1 diabetic (T1D) patients, focused in patients with multiple daily injections (MDI) therapy. The general idea is to make use of previous-day information in the moment when patients go to sleep, and then predict if in the next following hours any hypoglycemic event will occur. If the system will have predicted any hypoglycemic event in that moment, it is expected that it will be able to warn the patient to take some action: such as reduce basal insulin dose or to consume a snack before sleep. 10 patients with T1D for more than five years will be included. It is a longitudinal, prospective, interventional study in which every patient will use intermittently scanned Continuous Glucose Monitoring (isCGM) and a physical activity tracker during 12 weeks. Moreover, during this period, patients will store in a mobile application (Freestyle LibreLink) or in a reader information regarding their diabetes management activities, such as insulin delivery doses and meal consumption.

Study Type

INTERVENTIONAL

Allocation

Purpose

OTHER

Masking

NONE

Enrollment

Conditions

Type1diabetes Hypoglycemia

Eligibility

Sex: ALLMin age: 18 Years

Medical Language ↔ Plain English

Inclusion Criteria: * Patients \> 18 years with Type 1 Diabetes: * \> 4 hypoglycemia / week (\< 70 mg/dl, including day and night), last 2 weeks and / or * One severe hypoglycemia during the last year and / or * Hypoglycemia unawareness (Clarke Test \>3) * Disease duration \> 5 years * On multiple doses of insulin (MDI) therapy using a rapid acting insulin analogue as prandial insulin (lispro, aspart or glulisine) and any basal analogue as basal insulin. * A1c 6.5 - 9.5 % * Able to use an intermittently scanned continuous glucose monitoring (isCGM) system. * Performing \>4 self-monitoring blood glucose (SMBG) per day * Using carb-counting * Providing an informed consent * No CGM user previously (during the last 3 months). Exclusion Criteria: * Patients with a previous Diabetic Ketoacidosis (DKA) episode in the previous 6 months. * Patients with a severe hypoglycemia in the previous 6 months. * Severe diabetic complications or comorbidities: eye, renal, cardiovascular...from the clinicians point of view. * Pregnancy and breastfeeding. * History of drug or alcohol abuse. * Scheduled surgery during the study period. * Mental conditions that prevent the subject to understand the nature, purpose and possible consequences of the study. * Subjects those are unlikely to meet the clinical study protocol, eg uncooperative attitude, inability to return for follow-up visits, or poor probability of completing the study. * Using an experimental drug or device during the past 30 days.

Outcomes

Primary Outcomes

Sensitivity (SE) and specificity (SP) to predict Level 1 hypoglycemia

Primary outcome will be sensitivity (SE) and specificity (SP) to predict Level 1 hypoglycemia. Level 1: a hypoglycemia alert glucose value between 54-70 mg/dL (3.0-3.9 mmol/L) with or without symptoms. A hypoglycemic event should be defined as follows. Beginning of a CGM event: readings below the threshold for at least 15 min are considered an event. For example, at least 15min, \<70 mg/dL (3.9 mmol/L) to define a significant hypoglycemic event. End of a CGM event: readings for 15 min at \>70 mg/dL (3.9 mmol/L).

Time frame: 90 days

Secondary Outcomes

Sensitivity (SE) and specificity (SP) to predict Level 2 hypoglycemia

Secondary outcome will be sensitivity (SE) and specificity (SP) to predict Level 2 hypoglycemia1. Level 2: a glucose level of \<54mg/dL (3.0 mmol/L) with or without symptoms. A hypoglycemic event should be defined as follows. Beginning of a CGM event: readings below the threshold for at least 15 min are considered an event. For example, at least 15min, \<54 mg/dL (3.0 mmol/L) to define a clinically significant hypoglycemic event. End of a CGM event: readings for 15 min at \>70 mg/dL (3.9 mmol/L). A second hypoglycemic event outcome of prolonged hypoglycemia is considered when CGM levels are \< 54 mg/dL (3.0 mmol/L) for consecutive 120 min or more.

Time frame: 90 days

Predicted HbA1c from the sensor data

Time frame: 90 days

standard deviation (SD)

Standard deviation (SD)

Time frame: 90 days

Mean glucose

Time frame: 90 days

Level 3 hypoglycaemia

Number of Severe hypoglycemia Clinical diagnosis: event requiring assistance (level 3)

Time frame: 90 days

Percentage of time in hypoglycemic ranges

Percentage of time in hypoglycemic ranges, mg/dL (mmol/L), %: * Clinically significant/very low/immediate action required \<54 (\<3.0) (level 2) * Alert/low/monitor 70-54 (3.9-3.0) (level 1)

Time frame: 90 days

Percentage of time in target range

Percentage of time in target range, mg/dL (mmol/L), %: * Default 70-180 (3.9-10.0) * Secondary 70-140 (3.9-7.8)

Time frame: 90 days

Percentage of time in hyperglycemic range >180

Percentage of time in hyperglycemic ranges, mg/dL (mmol/L), % Alert/elevated/monitor \> 180 (\>10)

Time frame: 90 days

Percentage of time in hyperglycemic range >250

Percentage of time in hyperglycemic ranges, mg/dL (mmol/L), % Clinically significant/very elevated \> 250 (\>13.9)

Time frame: 90 days

Glucose variability LBGI

Low Blood Glucose Index (LBGI)

Time frame: 90 days

Glucose variability HBGI

High Blood Glucose Index (HBGI)

Time frame: 90 days

Number of Level 3: severe hypoglycemia

Number of Level 3: severe hypoglycemia. This denotes cognitive impairment requiring external assistance for recovery but is not defined by a specific glucose value.

Time frame: 90 days

Prediction and Prevention of Nocturnal Hypoglycemia in Persons With Type 1 Diabetes Using Machine Learning Techniques

Overview

Conditions

Interventions

Eligibility

Locations (1)

Outcomes

Primary Outcomes

Secondary Outcomes