# New machine learning methods for extracting features from digital health data with applications to sleep apnea

> **NIH NIH R01** · UNIVERSITY OF MICHIGAN AT ANN ARBOR · 2024 · $367,519

## Abstract

ABSTRACT
While obstructive sleep apnea (OSA) is linked to metabolic syndrome, the results from
randomized clinical trials with positive airway pressure (PAP) therapy, a treatment for OSA, are
inconclusive regarding therapy effects on glucose metabolism and cardiovascular disease.
Continuous glucose monitors (CGM) and ambulatory blood pressure monitors (ABPM) are
increasingly used to elucidate the effects of OSA on glucose metabolism and blood pressure,
however, there is a large gap between the complexity of data from wearable devices and
commonly used crude summary measures (e.g., glucose time in range, 24h blood pressure
mean). Without advancements in algorithms that provide better characterization of temporal
features in CGM and ABPM data, the adverse effects of OSA on glucose metabolism and blood
pressure will likely remain underappreciated, and elucidating the heterogeneity of treatment
effects with PAP therapy will remain difficult. Our proposal is motivated by data from a randomized
clinical trial on the effects of PAP therapy on glycemic measures and blood pressure of patients
with concurrent type 2 diabetes and OSA. Conventional CGM and ABPM summaries lack
sensitivity to differentiate control and treatment groups and detect heterogeneity in treatment
effects. The overall objective of this proposal is to develop novel statistical and machine learning
methods to fully exploit CGM and ABPM data for precision phenotyping of glycemic and
cardiovascular measures. To achieve our objective, we propose the following: (1) To address the
limited statistical power of existing methods for characterizing features of the glycemic state, we
will develop a distributional data analysis framework for CGM-based glycemic measures that will
incorporate global and local temporal characteristics. (2) To address the limitations of existing
methods for ABPM data due to the separate analysis of each blood pressure modality (systolic,
diastolic) and oversimplified division of time into the night (0-6 h) and day, we will develop a tensor
data analysis framework for ABPM data that will integrate all concurrent measurements (systolic,
diastolic, mean arterial pressure, heart rate) aligned by the full 24h time period. We will develop
software to enable the broad application of proposed algorithms to other studies that collect CGM
and ABPM data. Completion of these aims will provide the necessary algorithmic and
computational tools to test whether behavioral interventions and new pharmacotherapeutic
agents improve the glycemic status and blood pressure in population subsets, including those
with prediabetes and hypertension. Reducing the metabolic and cardiovascular risk burden has
unquestionable relevance for the prevention of cardiovascular morbidity and mortality.

## Key facts

- **NIH application ID:** 10851193
- **Project number:** 1R01HL172785-01
- **Recipient organization:** UNIVERSITY OF MICHIGAN AT ANN ARBOR
- **Principal Investigator:** Irina Gaynanova
- **Activity code:** R01 (R01, R21, SBIR, etc.)
- **Funding institute:** NIH
- **Fiscal year:** 2024
- **Award amount:** $367,519
- **Award type:** 1
- **Project period:** 2024-06-01 → 2029-02-28

## Primary source

NIH RePORTER: https://reporter.nih.gov/project-details/10851193

## Citation

> US National Institutes of Health, RePORTER application 10851193, New machine learning methods for extracting features from digital health data with applications to sleep apnea (1R01HL172785-01). Retrieved via AI Analytics 2026-05-25 from https://api.ai-analytics.org/grant/nih/10851193. Licensed CC0.

---

*[NIH grants dataset](/datasets/nih-grants) · CC0 1.0*