Hypothesis

We hypothesize that a closed‑loop algorithm integrating real‑time CGM, HRV, and sleep stage data can reduce postprandial glucose excursions more effectively than standard self‑guided decisions based on CGM alone.

Mechanistic Rationale

• Autonomic influence on hepatic glucose output: Low HRV reflects heightened sympathetic tone, which stimulates glycogenolysis and gluconeogenesis, raising hepatic glucose production independently of dietary intake (HRV study).
• Sleep‑dependent insulin sensitivity: Slow‑wave sleep enhances peripheral insulin action, whereas fragmented sleep attenuates it, shifting the glucose‑insulin dose‑response curve.
• Combined signal: When HRV indicates sympathetic dominance and sleep metrics show reduced slow‑wave proportion, the liver is primed to release glucose; anticipating this state allows pre‑emptive reduction of carbohydrate intake or light activity to blunt the upcoming rise.
• Feedback advantage: By acting on the multivariate risk score before glucose rises, the algorithm addresses the feed‑forward component that CGM‑only reactive strategies miss, thereby lowering overall glycemic variability.

Testable Prediction

In a randomized crossover n‑of‑1 trial, participants using the multimodal feedback loop will achieve a 15 % lower mean amplitude of glycemic excursions (MAGE) and a 10 % reduction in time spent >140 mg/dL compared with periods when they follow CGM‑only guidance or usual self‑care.

Experimental Design

1. Participants: 30 non‑diabetic adults aged 20‑45 who already wear a CGM and a HRV‑capable wearable for at least two weeks baseline.
2. Intervention phases (each 14 days, randomized order, 7‑day washout):
   • Multimodal loop: Device computes a risk score = weighted sum of normalized HRV (RMSSD deficit), sleep slow‑wave % deficit, and current glucose trend; if score exceeds threshold, a gentle notification suggests a 10‑minute walk or postponement of high‑glycemic snack.
   • CGM‑only: Standard alerts based on glucose >140 mg/dL or rapid rise >2 mg/dL/min.
   • Control: No notifications, participants make decisions as usual.
3. Outcomes: Continuous glucose metrics (MAGE, time in range 70‑140 mg/dL, mean glucose), sleep architecture (from wearable), HRV (nightly RMSSD), and self‑reported burden (Likert scale).
4. Analysis: Within‑subject comparison using paired t‑tests or Wilcoxon signed‑rank; significance set at p<0.05. Effect size calculated as Cohen’s d.

Falsifiability

If the multimodal loop does not produce a statistically significant improvement in MAGE or time in range relative to both CGM‑only and control conditions, the hypothesis is falsified. Likewise, if increased notifications lead to higher anxiety scores without metabolic benefit, the mechanistic premise of beneficial feed‑forward control would be challenged.

Expected Impact

Demonstrating that autonomic and sleep states add predictive value beyond glucose alone would justify integrating multimodal wearable data into decision‑support tools, moving quantified‑self from correlation to causation and reducing reliance on population‑based cutoffs.