logologo

Popular Boards

1 Mentions
Mark Warren, Richard M. Bergenstal, Matthew Hager | Diabetes Technology & Therapeutics | (2024)
Primary Link Pubmed DOI Openalex

Abstract

<b><i>Background:</i></b> Despite new pharmacotherapy, most patients with long-term type 2 diabetes are still hyperglycemic. This could have been solved by insulin with its unlimited potential efficacy, but its dynamic physiology demands frequent titrations which are overdemanding. This report provides a real-life account for a scalable transformation of diabetes care in a community-based endocrinology center by harnessing artificial intelligence-based autonomous insulin titration. <b><i>Methods:</i></b> The center embedded the d-Nav<sup>®</sup> technology and its dedicated clinical support. Reported outcomes include treatment efficacy/safety in the first 600 patients and use of cardiorenal-risk reduction pharmacotherapy. <b><i>Findings:</i></b> Patients used d-Nav for 8.2 ± 3.0 months with 82% retention. Age was 67.1 ± 11.5 years and duration of diabetes was 19.8 ± 11.0 years. During the last 3 years before d-Nav, glycated hemoglobin (HbA1c) had been overall higher than 8% and at the beginning of the program it was as high as 8.6% ± 2.1% with 29.3% of the patients with HbA1c >9%. With d-Nav, HbA1c decreased to 7.3% ± 1.2% with 5.7% of patients with HbA1c >9%. During the first 3 months, d-Nav reduced total daily dose of insulin in one of every five patients due to relatively low glucose levels to minimize the risk of hypoglycemia. Glucagon like peptide 1 (GLP-1) receptor agonists or dual GLP-1 and Glucose-dependent insulinotropic polypeptide (GIP) receptor agonists were prescribed in about a half of the patients and sodium glucose cotransporter 2 inhibitor in a third. The frequency of hypoglycemia (<54 mg/dL) was 0.4 ± 0.6/month and severe hypoglycemia 1.7/100-patient-years. <b><i>Interpretation:</i></b> The use of d-Nav allowed for improvement in overall diabetes management with appropriate use of both insulin and noninsulin pharmacologic agents in a scalable way.

Tags

Sample Definition And Size

The study retrospectively analyzed the first 600 patients with type 2 diabetes using insulin (via pens or syringes) who were prescribed the AI-driven d‑Nav insulin titration program at a community-based endocrinology center in Greenville, North Carolina, USA. The cohort had a mean age of 67.1 ± 11.5 years, mean diabetes duration of 19.8 ± 11.0 years, and mean duration on insulin of 11.9 ± 10.0 years. Retention was 82%, with withdrawals occurring at a mean of 4.8 ± 3.1 months. The study also included a comparison group of 1,103 insulin-using patients not on d‑Nav for baseline demographic comparison.

Study Type

This was a retrospective, real‑world observational implementation study (clinical implementation report) of an AI‑driven autonomous insulin titration program, using de‑identified data collected during routine clinical care between October 2022 and September 2023.

Conflicts Of Interest

I.H. is cofounder and medical director of Hygieia PC.; E.B. is cofounder and CEO of Hygieia, Inc.; M.W. and M.H. are part of an institutional partnership with Hygieia PC.; R.M.B. has received research support, consulting fees, or advisory board roles from multiple diabetes‑related companies including Hygieia. The study was funded by Hygieia, Inc.

Results Summary

Patients used d‑Nav for 8.2 ± 3.0 months. Baseline HbA1c was 8.6% ± 2.1%, with 29.3% having HbA1c > 9%. After 6 months, HbA1c decreased to 7.3% ± 1.2% (P < 0.0001), with only 5.7% > 9%. During the first 3 months, total daily insulin dose decreased in 21% of patients to reduce hypoglycemia risk; overall insulin dose increased by 60.6% from 69.3 to 111.3 units, particularly in those with baseline weekly mean glucose >150 mg/dL. Hypoglycemia (<54 mg/dL) occurred at 0.4 ± 0.6 events/month; severe hypoglycemia occurred at 1.7 events per 100 patient‑years. In a CGM subgroup (n=80), time‑in‑range improved from 47.7% ± 25.5% pre‑d‑Nav to 65.4% ± 17.1% on d‑Nav (P = 0.003). Patient satisfaction was high (mean score 3.8/4).

Referenced In

Mercedes C.
a month ago
AI Papers and Discussion

🤖 AI Wearables for Managing Diabetes ~

In this paper - Fraser et al., 2025 - Researchers reviewed 60 studies on AI-powered wearables - continuous glucose monitors (CGMs), for Type 2 diabetes care. The tech is advancing fast—but major gaps remain before it reaches your doctor's office.

What's New:

Smarter Glucose Prediction: Deep learning models,particularly LSTMs and transformers, can predict blood sugar swings 30-120 minutes ahead of time by spotting patterns humans miss. One model achieved R² = 0.989 accuracy, essentially forecasting your glucose curve in real-time.

AI That Acts, Not Just Predicts: We're moving from forecasting to actual intervention. Reinforcement learning models, like Warren et al., 2024 now can suggest insulin doses autonomously, with results showing HbA1c drops by 1.3%.

Non-Invasive Hacks: Researchers are ditching finger pricks entirely—using smartphone photoplethysmography (PPG), ECG signals, even tear fluid infrared sensors to estimate glucose. One PPG-based deep learning model reached 90.6% accuracy for diabetes detection.

The AI Toolkit Deep-Dive:

>> LSTMs & GRUs (45% of studies): These recurrent neural networks are the workhorses—they process CGM time-series data sequentially, remembering past glucose values to predict future trends.

>> Transformers: They use "attention mechanisms" to weigh which past data points matter most — handling longer prediction windows across diverse populations better than LSTMs.

>> Temporal Fusion Transformers: Combine static data (age, BMI) with time-varying data (glucose, heart rate) in one model. Emerging but promising for personalized predictions.

>> Multi-Agent Reinforcement Learning (MARL): Multiple AI agents compete to identify which patient features (lab values, meds, demographics) most predict adverse glycemic events in hospitals. One study achieved 92.8% precision for hypoglycemia detection.

>> XGBoost & Random Forests: Traditional machine learning still dominates when doctors need interpretability. XGBoost ranked features by importance (SHAP values), showing clinicians why the AI flagged a patient as high-risk.

>> Deep-Ensemble Learning: Stacks multiple neural networks together—CNNs for pattern recognition, BiLSTMs for sequence memory, meta-learners for combining outputs.

Despite the great advances made, there are still some hard truths to tackle:

  • 60% use "black box" models that doctors can't explain - the pattern is unknown.

  • Only 7% report race/ethnicity data.

  • Most studies are tiny (median 150 people).

  • No long-term follow-ups exist - What would we see in 1 year's time?

The authors close off the review --- The tech works in pilots but we need diverse data, transparent models, and real-world validation before this becomes standard care.

Read Full Post
0