Symptom Entry — Where Triage Begins
A short chat message launches the entire triage pipeline.
- 1 in 5: Avg symptom checkers used (US adults yearly)
- 12 words: Median input length (first complaint message)
- ~70-80%: Free-text NLP accuracy (symptom extraction)
- Headache: Common entry symptom (top reported complaint)
Free-text vs structured intake
Chatbots parse loose language into structured symptom tags.
Entity extraction
NLP model tags body location, symptom type, and modifiers.
Ambiguity handling
Vague phrases trigger a clarifying question immediately.
Follow-Up Questions — Hunting for Red Flags
The AI asks targeted questions to narrow the diagnosis space.
- 5-8: Avg follow-up questions (per triage session)
- 12+: Red flag question types (chest pain, breathing, etc)
- ~65%: Completion rate (finish full questionnaire)
- 3-5 min: Time per session (typical chat duration)
Severity probing
Pain scales and duration questions quantify the complaint.
Red flag checklist
Fixed questions screen for danger signs every time.
Branching logic
Each answer changes which question comes next.
Symptom Pattern Matching
Collected answers are compared to a library of known symptom clusters.
- 800+: Symptom pattern library (condition templates)
- Bayesian: Matching approach (+ ML classifiers)
- ~80%: Top-3 accuracy (correct condition in top 3)
- Critical: False negative concern (missed emergencies)
Pattern libraries
Curated symptom-condition maps built from clinical data.
Probabilistic scoring
Each candidate condition gets a likelihood score.
Safety-first bias
Ties are broken toward the more urgent outcome.
Urgency Classification
Matched patterns are converted into a single urgency score.
- 3: Urgency tiers (home, doctor, emergency)
- 0-100: Scoring scale (composite urgency score)
- High: Red flag weight (dominates final score)
- ~70: Escalation threshold (triggers emergency tier)
Composite scoring
Severity, duration, and red flags combine into one score.
Threshold bands
Score ranges map directly to the three care tiers.
Override rules
Any single red flag can force emergency regardless of score.
Care Recommendation Delivered
The patient receives one clear, actionable triage outcome.
- 3: Outcome options (home / doctor / ER)
- ~60%: User trust in verdict (follow the recommendation)
- Disclaimer: Liability safeguard (not a diagnosis)
- Always: Re-check prompt (worsening symptoms guidance)
Clear verdict card
One prominent card states the recommended action.
Explainability
Key symptoms driving the decision are listed alongside it.
Safety net
Users are told to seek help sooner if things worsen.