Editing AI for Telemedicine (section)

== <span style="color: #FFFFFF;">Applying</span> ==
'''Remote symptom triage with LLM + structured assessment:'''
<syntaxhighlight lang="python">
from openai import OpenAI
import json

client = OpenAI()

TRIAGE_SYSTEM_PROMPT = """You are a clinical triage AI assistant. 
Given patient symptoms, determine urgency level and likely conditions.

Always respond in JSON with:
{
  "urgency": "emergency|urgent|routine|self_care",
  "time_to_care": "immediate|same_day|48h|1_week|self_care",
  "top_conditions": ["condition1", "condition2", "condition3"],
  "red_flag_symptoms": ["symptom if present"],
  "recommended_action": "string",
  "confidence": "high|medium|low"
}

Emergency triggers (always override to emergency):
- Chest pain + shortness of breath
- Signs of stroke (FAST)
- Severe allergic reaction
- Active suicidal ideation with plan
- Uncontrolled bleeding"""

def triage_patient(symptoms: str, patient_context: dict) -> dict:
    """
    Triage patient based on symptoms and context.
    patient_context: age, sex, chronic conditions, current medications
    """
    prompt = f"""Patient: {patient_context['age']}yo {patient_context['sex']}
Conditions: {', '.join(patient_context.get('conditions', []))}
Medications: {', '.join(patient_context.get('medications', []))}

Presenting symptoms: {symptoms}

Provide triage assessment."""

    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": TRIAGE_SYSTEM_PROMPT},
            {"role": "user", "content": prompt}
        ],
        temperature=0.1,
        response_format={"type": "json_object"}
    )
    result = json.loads(response.choices[0].message.content)
    # Safety override: always flag for immediate review if emergency
    if result['urgency'] == 'emergency':
        result['alert_clinician'] = True
    return result

# Remote patient monitoring deterioration model
def predict_deterioration(rpm_data: list[dict], model) -> dict:
    """Predict 48h deterioration risk from RPM time series."""
    # rpm_data: list of daily readings {weight, bp_sys, bp_dia, spo2, hr, symptoms_score}
    import pandas as pd, numpy as np
    df = pd.DataFrame(rpm_data).sort_values('date')
    # Feature engineering
    df['weight_change_3d'] = df['weight'].diff(3)
    df['bp_trend'] = df['bp_sys'].rolling(3).mean() - df['bp_sys'].rolling(7).mean()
    features = df[['weight', 'weight_change_3d', 'bp_sys', 'bp_dia', 'spo2',
                   'hr', 'symptoms_score', 'bp_trend']].iloc[-1].values
    risk = model.predict_proba([features])[0][1]
    return {'deterioration_risk_48h': float(risk), 'alert': risk > 0.65}
</syntaxhighlight>

; Telemedicine AI tools
: '''Symptom checking''' → Babylon Health, Ada Health, Buoy Health, K Health
: '''Remote dermatology''' → Skin Analytics DERM, DermAI, Miiskin
: '''Ambient documentation''' → Nuance DAX Copilot (Microsoft), Suki, Abridge, DeepScribe
: '''RPM platforms''' → Biofourmis, Vivify Health, Current Health, Dexcom Clarity
: '''Virtual care AI''' → Teladoc with ML, Amazon Clinic, 98point6
</div>

<div style="background-color: #8B4500; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">