Diabetes Mellitus, Hypoglycemia
Conditions
Keywords
Hypoglycemia, Voice, Speech, Detection, Artificial intelligence (AI)
Brief summary
The HypoVoice study aims at identifying potential vocal biomarkers associated with hypoglycemia to pave the way towards a voice-based hypoglycemia detection approach.
Detailed description
While hypoglycemia has been widely studied in medical research, studies assessing vocal changes associated with this state are limited. This study aims at collecting a data set labelled with the gold standard (blood glucose) to provide a solid basis for the identification of vocal biomarkers using machine learning. Additionally, physiological data are collected using wearable sensors to assess whether additional integration of vital signs (e.g. heart rate) enhances the performance of hypoglycemia detection.
Interventions
OTHERControlled hypoglycemic state
Voice sampling is performed in different glycemic states (euglycemia and hypoglycemia).
Sponsors
Idiap Research Institute
CSEM Centre Suisse d'Electronique et de Microtechnique SA
Ludwig-Maximilians - University of Munich
Insel Gruppe AG, University Hospital Bern
Study design
Allocation
NA
Intervention model
SINGLE_GROUP
Primary purpose
OTHER
Masking
NONE
Masking description
Participants are aware that hypoglycemia will be induced during the study but they are blinded to their blood glucose levels throughout the hypoglycemia procedure.
Eligibility
Sex/Gender
ALL
Age
18 Years to 60 Years
Healthy volunteers
No
Inclusion criteria
* Written informed consent * Type 1 Diabetes mellitus as defined by WHO for at least 6 months * Aged 18 - 60 years * HbA1c ≤9.0 % * Functional insulin treatment with good knowledge of insulin self-management * Native language German or Swiss German * Use of continuous glucose monitoring (CGM) or flash glucose monitoring (FGM)
Exclusion criteria
* Incapacity to give informed consent * Contraindications to insulin aspart (NovoRapid®) * Total daily insulin dose \>2 IU/kg/day * Pregnancy, breast-feeding or lack of safe contraception * Active heart, lung, liver, gastrointestinal, renal or psychiatric disease * Pacemaker or implantable cardioverter defibrillator (ICD) * Epilepsy or history of seizure * Chronic neurological or ear-nose-and-throat (ENT) disease influencing voice or history of voice disorder * Illiteracy or dyslexia * Active smoking * Active drug or alcohol abuse * Medication known to interfere with voice or to induce listlessness (e.g. opioids, benzodiazepines, etc.)
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Diagnostic accuracy of the HypoVoice approach to detect hypoglycemia based on voice data quantified as area under the receiver operating characteristic curve (AUROC) | 4 hours | Voice data will be collected in eu- and hypoglycemia |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Diagnostic accuracy of the HypoVoice approach to detect hypoglycemia based on voice and physiological data quantified as area under the receiver operating characteristic curve (AUROC) | 4 hours | Voice and physiological data will be collected in eu- and hypoglycemia |
| Voice parameters indicative of hypoglycemia | 4 hours | Explainable AI methods will be used to identify voice parameters indicative of hypoglycemia |
| Physiological parameters indicative of hypoglycemia | 4 hours | Explainable AI methods will be used to identify physiological parameters indicative of hypoglycemia |
Other
| Measure | Time frame | Description |
|---|---|---|
| Change in hypoglycemic symptoms across the glycemic trajectory | 4 hours | Hypoglycemic symptoms will be assessed using the Edinburgh Hypoglycemia Scale (higher score means more symptoms). |
| Change in cognitive performance across the glycemic trajectory. | 4 hours | Cognitive performance will be assessed using the Digit Symbol Substitution Test (higher score means better cognitive performance). |
Countries
Switzerland
Outcome results
None listed