Extract Events V1

Your task is to analyze the following question and extract NEW medical insights or events specific to the patient.

CRITICAL DUPLICATE CHECK:

Before returning ANY event, you MUST check the "Previously Extracted Insights" list below.

• If an event with the SAME DATE and SAME TYPE (key) already exists → DO NOT include it
• If the event description is very similar to an existing one (even in different words/language) → DO NOT include it
• Only return genuinely NEW events that are NOT already saved
• When in doubt, DO NOT include the event (prefer false negatives over duplicates)

EXTRACTION RULES:

• Extract patient-specific information (NOT generic medical content from studies/trials)
• Extract life events relevant to health: smoking, drinking, exercise, diet, work, etc.
• Extract BOTH past events AND future appointments/plans
• DO NOT extract events that are simply being mentioned or referenced from context

EVENT TYPES (key field):

• diagnosis - Medical diagnoses, conditions
• symptom - Symptoms reported by patient
• medication - Medications, dosage changes
• treatment - Treatments, therapies, procedures
• test - Lab tests, medical exams, imaging
• appointment - Future or past medical appointments, consultations
• activity - Exercise, lifestyle activities
• reminder - General reminders, tasks to do
• other - Other relevant patient information
• null - Non-medical life events

DATE HANDLING:

• TODAY'S DATE: {dayOfWeek}, {isoDate}
• Past events: Use exact date if mentioned, otherwise 'unknown'
• Future events (appointments): Calculate date from context (e.g., "tomorrow" = tomorrow's date, "next Monday" = calculate)
• Relative dates: "yesterday", "last week", "3 days ago" → calculate ISO date

OUTPUT FORMAT (valid JSON array):

[ ⟨ "insight": "Brief description of the event", "date": "YYYY-MM-DD or 'unknown'", "present": true/false (is this still relevant/active?), "key": "diagnosis|symptom|medication|treatment|test|appointment|activity|reminder|other|null" ⟩ ]

If no NEW events found (or all are duplicates), return: []

---

User's Question: {questionText}

---

Previously Extracted Insights (ALREADY SAVED - DO NOT DUPLICATE): {events}

---

NEW INSIGHTS ONLY (return [] if none are truly new):