AI-Guided Remission for Health ChatBots

A new 2026 protocol paper introduces an AI-guided remission chatbot for diabetic foot patients — a closed-loop system that coaches them through the dangerous months after an ulcer heals. Lead author Dr. Lucian Feraru is a limb-salvage surgeon and co-founder of WoundScribe.ai, joined by Armstrong on limb preservation, Klonoff on diabetes technology, Najafi on wearable gait analytics, and Antoszewska on dermatology. The stakes justify the ambition. Up to 65% of healed diabetic foot ulcers recur within five years, and the first three months are the most dangerous because newly healed skin is mechanically fragile. The authors borrow the cancer word — remission — to reframe a healed foot as vulnerable rather than cured, with the goal of maximising ulcer-free days. This is the natural next chapter for diabetic foot care: extending specialist oversight into the home, where recurrence usually happens. ### Walking, dosed like a drug The conceptual heart of the paper is a dosing metaphor wrapped in a control loop. Start near the patient's own step-count baseline, then titrate up or down based on how the tissue responds. Every day the system runs four moves: 1. Capture the last 24 hours of data 2. Classify into a green, amber, or red risk tier 3. Recommend tomorrow's step target 4. Escalate to the clinical team if something looks wrong Three sensor streams feed the loop: - Step counts from accelerometers - Foot skin temperature from infrared thermometry at 0.3°C accuracy - Shoe-wear time from in-shoe sensors The single most important threshold is a left-versus-right temperature difference of 2.2°C — one day means caution, two or more days triggers escalation, because sustained asymmetry predicts ulceration. Week 1 establishes a personal baseline, then the dose climbs about 500 steps per week — but only if there are no hotspots. Progression is gated on physiology, not the calendar. ### Transparent logic, human in the loop The brain is deliberately rule-based, not a black box. The system fails safe, not silent — data are timestamped and range-checked, missing data trigger reminders rather than guesses, and a clinician is always accountable for the call. Each failure mode — missed emergencies, over-trust in automation, alert fatigue, privacy, equity — gets a concrete design response. The value radiates well beyond podiatry. For wound care teams thinking about how proactive patient engagement and education connects to long-term outcomes, this protocol is a clean template — a worked example of dosing a behaviour, watching the tissue respond, and escalating early during the months when recurrence is most likely.