Skip to content

Risk Analysis (Expanded Draft)

Note: Structured, ID-based quality risk tracking is being maintained in Docs/Quality/RiskAnalysis.md with linkage to requirements/design/tests through Docs/Quality/TraceabilityMatrix.md. This document remains the broader risk narrative.

This document captures technical, clinical, operational, and regulatory risks derived from the 23 Jan 2026 protocol and the current project direction (OmniPod DASH + Dexcom G7 + BB algorithm integration). It will evolve with implementation details.

Clinical Safety Risks

  • Hypoglycemia in early pregnancy due to increased insulin sensitivity.
  • Postprandial hypoglycemia from increased meal bolus fraction (90%).
  • Hyperglycemia from under-dosing during late-pregnancy insulin resistance.
  • DKA/HHS risk if pump delivery fails or CGM data is missing.
  • Nausea/vomiting and unpredictable intake causing dosing mismatch.
  • Overly aggressive targets (90/100 mg/dL) increasing time below range.

Algorithm and Control Risks

  • Adaptation rate too slow for rapid insulin resistance changes.
  • Adaptation rate too aggressive leading to instability or oscillation.
  • Meal announcement ambiguity (less/usual/more) yields inconsistent dosing.
  • Failure to revert meal bolus fraction when postprandial lows occur.
  • Incorrect basal behavior during CGM downtime (weight-based fallback).
  • Offline basal fallback may over- or under-deliver insulin if it diverges from current needs, especially after prolonged disconnection.

Device and Integration Risks

  • CGM data gaps, sensor failure, or calibration issues cause unsafe dosing.
  • Bluetooth interruptions (> 2 hours) delay controller updates.
  • Bluetooth permission denial or restricted access prevents CGM onboarding.
  • Pump command failures or occlusions cause missed insulin delivery.
  • OmniPod DASH delivery constraints conflict with algorithm assumptions.
  • OmniPod DASH onboarding settings (basal schedule, max bolus, reminders) conflict with algorithm-driven dosing if left user-configurable.
  • Water exposure or disconnection limits (removed for showering) create gaps.
  • iOS background execution limits can delay algorithm runs if BLE events are absent.

User Experience and Adherence Risks

  • Users misunderstand alarms and do not perform confirmatory fingersticks.
  • Users fail to perform ketone checks when required.
  • Inconsistent meal announcements (missing meals or misclassified sizes).
  • Alarm fatigue leading to ignored low/high alerts.
  • Incorrect infusion set replacement timing leads to delivery failure.

Data Quality and Monitoring Risks

  • Missing CGM data degrades time-in-range metrics and analyses.
  • Incorrect calculation of pregnancy-specific metrics (63-140 mg/dL).
  • Loss or corruption of device logs impairs safety review and auditability.
  • Incomplete documentation of adverse events and device issues.

Operational and Study Risks

  • Enrollment constraints (eligibility, exclusions) reduce target sample size.
  • Requirement for Fiasp and medication washout complicates onboarding.
  • Inadequate training increases misuse of the system.
  • Follow-up cadence not maintained, reducing safety oversight.
  • Long-acting insulin supplementation (high dose scenario) misapplied.

Regulatory and Compliance Risks

  • Pregnancy configuration is off-label; may require IDE approval.
  • Adverse event reporting timelines not met (72-hour severe events).
  • DSMB oversight requires reliable, timely data summaries.
  • Data de-identification failures could breach confidentiality.

Security and Privacy Risks

  • CGM/pump data transmission or storage breaches.
  • Improper handling of identifiable data in logs or exports.

Mitigations (Initial)

  • Conservative low-glucose safeguards and escalation pathways.
  • Explicit workflows for CGM downtime and pump failure fallbacks.
  • Clear Bluetooth permission prompts and error states for CGM setup.
  • Clear, guided alarm response flows with fingerstick validation.
  • Robust ketone action plan prompts and documentation.
  • Audit trails for dosing, targets, algorithm state, and device faults.
  • Lock or streamline pump onboarding settings to align with algorithm control.
  • Data redundancy for CGM streams and periodic integrity checks.
  • Clinician-controlled target adjustments; limit user self-adjustment.
  • Use multiple wake strategies (BLE events, BGTaskScheduler, HealthKit background delivery, optional silent push) and implement catch-up logic on resume.
  • Make offline mode obvious, log the transition, and cap offline basal duration.

Open Questions

  • Final safety thresholds for alarms and escalation in the app context.
  • How to implement or integrate DSMB-ready reporting and summaries.
  • OmniPod DASH API constraints on dosing precision and failure states.
  • Expected minimum data completeness for analysis in this app context.

Traceability (Initial)

  • R-CLIN-TARGETS: Pregnancy targets (63-140 mg/dL, fasting 70-95).
  • Risk: Hypoglycemia if algorithm or targets too aggressive.
  • R-ALG-LOW-TARGETS: Add 90/100 mg/dL targets.
  • Risk: Increased time < 54 mg/dL.
  • R-ALG-MEAL90: Meal bolus 90% of learned requirement.
  • Risk: Postprandial hypoglycemia or overshoot.
  • R-ALERTS: Low/high alert logic and response.
  • Risk: Alarm fatigue, missed intervention.
  • R-KETONE: Ketone checks and escalation.
  • Risk: DKA if checks skipped.
  • R-CGM-DOWN: CGM downtime behavior.
  • Risk: Unsafe basal/bolus during missing CGM.
  • R-DATA-METRICS: Time-in-range computation and logs.
  • Risk: Incorrect outcome reporting.

Additional Risk Controls (Proposed)

  • Lock pregnancy targets to clinician control; log all changes.
  • Require acknowledgment for missed meal announcements or long CGM gaps.
  • Safety guardrails for rapid target reductions (stepwise, time-based).
  • Automated prompts for ketone checks at defined thresholds.
  • Data completeness checks with user notification when < 70% CGM coverage.

Residual Risks and Acceptance

  • Some hypoglycemia risk is inherent to tighter targets in pregnancy.
  • CGM and pump failure risks cannot be eliminated; require robust fallback.
  • User adherence remains a variable; training and alerts reduce but do not remove.