(8) Predicting high blood glucose levels through voice characteristic analysis in patients with CFRD

Thursday, November 3, 2022

12:00 PM – 12:40 PM ET

Presenting Author(s)

Pichatorn Suppakitjanusant, MD (she/her/hers)

Research Fellow
Emory university
ATLANTA, Georgia, United States

Co-Author(s)

NK

Nittaya Kasemkosin, PhD

Department Head
Faculty of Medicine Ramathibodi Hospital, Mahidol University, United States
BO

Boonsong Ongphiphadhanakul, MD

Professor of Medicine
Faculty of Medicine Ramathibodi Hospital, Mahidol University, United States
SW

Samuel Weinstein

Undergraduate student
Emory University, United States
WH

William R. Hunt, MD

Associate Professor
Emory University
Atlanta, Georgia, United States
VS

Viranuj Sueblinvong, MD

Associate Professor
Emory University, United States
VT

Vin Tangpricha, MD, PhD

Professor of Medicine
Emory University, United States

Background: Cystic fibrosis-related diabetes (CFRD) is usually clinically silent, and patients may remain asymptomatic for years [1]. Prolonged high blood glucose levels in patients with CFRD are worsened lung disease, poorer nutritional status, and increased mortality [2]. It is important to monitor glucose levels and give the appropriate dose of insulin to prevent complications. CFRD patients need finger pricks up to four times a day for tracking their sugar levels. This procedure is painful and unpleasant which reduces the willingness for controlling the disease. Moreover, the blood drawing procedure is hard to be done in a telemedicine clinic during the pandemic.

We are interested in developing a non-invasive tool for monitoring glucose levels by analyzing the characteristics of the voice. We hypothesized that high blood glucose levels cause laryngeal soft tissue swelling and lead to changes in voice characteristics [3]. Studies in patients with diabetes without CF have demonstrated a potential use of this technology [4]. The purpose of this study is to predict high blood glucose levels using voice characteristic analysis in patients with CFRD.

Methods: A prospective cross-sectional study was performed on adults with CF recruited from the CF Clinic at Emory Healthcare from March to December 2021. We recorded 3-second voice samples of a sustained /a/ vowel. Voice characteristics including fundamental frequency (F0), fundamental frequency variation (vF0), jitter, shimmer, noise-to-harmonic ratio (NHR), and relative average perturbation (RAP) were analyzed using the Computerized Speech Lab with the Multi-Dimensional Voice Program. Statistical analysis based on matching voice characteristics with individual’s point-of-care blood glucose testing (POCT) in their visit.

Results: There were 43 patients with CFRD in this study (25 male and 18 female individuals). Patients with POCT glucose levels≥ 200 mg/dL had a similar mean age to patients with POCT glucose levels < 200 mg/dL (32 ± 13 vs 35 ± 14 years old, p=0.494). Patients have been diagnosed with CFRD for 9 ± 7 years. The mean HbA1c level in CFRD patients is 7.2 ± 2.2 %. An acoustic parameter analysis categorized by gender [Table1] shows vF0 was significantly lower in individual who has POCT glucose level ≥ 200 mg/dL compared with those with POCT glucose level < 200 mg/dL in both male (1.14 ± 0.14 % VS 1.58 ± 0.73 %, P < 0.05) and female (0.75 ± 0.07 % VS 1.26 ± 0.56 %, P < 0.05). Multivariate analysis showed that vF0 was significantly associated with the duration of CFRD diagnosis (P = 0.019) and glucose level (P = 0.012) after controlling for age and body mass index [Table 2]. Logistic regression analysis showed that vF0 (P = 0.042) and duration of CFRD (P = 0.028) significantly predict high blood glucose levels when other patient characteristics were considered [Table 3]. Moreover, there is a significantly lower NHR (P=0.049) in the female subjects who were poor glycemic control (HbA1c > 7%) compared to those who were well glycemic control (HbA1c ≤7%).

Conclusions:

There is potential to use the human voice as a non-invasive tool for glucose level measurement in CFRD patients in the future. Fundamental frequency variation was able to predict high blood glucose levels. This study is ongoing and will collect additional data on other ranges of glucose levels.

Acknowledgements:

References: