Automate Patient Follow-Ups and Post-Discharge Instructions

Most small clinics rely on their EMR's built-in communication tools, like those in athenahealth or eClinicalWorks. These systems are good for generic appointment reminders but cannot interpret the clinical content within a discharge summary. They can send a template message, but they cannot read a note that says “begin physical therapy exercises in 3 days” and schedule that specific reminder.

Consider a 5-provider clinic where a medical assistant (MA) handles post-discharge calls. The MA manually reads each patient's PDF discharge summary, sets calendar reminders, and makes phone calls. For a patient told to “call if temperature exceeds 101°F,” the MA has no way to automate that check-in. This manual process takes 10-15 minutes per patient and is prone to error. With 10 discharges a day, this consumes over 2 hours of staff time and critical instructions are frequently missed.

Even dedicated patient messaging platforms like Klara fall short because they are architected for communication, not clinical interpretation. They can segment patients by diagnosis code but cannot parse the unstructured text where a physician details a specific wound care regimen or medication tapering schedule. The core problem is that these tools operate on structured data, while the most critical follow-up instructions are buried in free-form text.

The result is that MAs spend hours on low-value administrative work instead of direct patient care. Patients receive inconsistent follow-up, which increases the risk of complications, hospital readmissions, and poor outcomes. The clinic cannot scale its operations without hiring more administrative staff to handle the manual workload.

The engagement would begin with a discovery process and a signed Business Associate Agreement (BAA). Syntora would analyze 50-100 anonymized discharge summaries from your EMR to map the language, formats, and types of instructions your clinicians use. This audit determines the specific data points to extract and the logic for the follow-up sequences. You receive a technical scope document detailing this plan for your approval before any code is written.

The system's core would be a HIPAA-compliant AWS Lambda function that triggers when a new discharge document is created. This function uses the Claude API to parse the unstructured text, extracting entities like medication schedules, follow-up appointments, and specific warning signs. We have used this same pattern for complex financial document processing. Pydantic models validate the extracted data before it is stored in a Supabase database, creating a complete audit trail. A FastAPI service then reads this structured data and sends personalized SMS messages via Twilio at the correct intervals.

The final deliverable is an automated system that runs in your own AWS account. Your clinical staff would have access to a simple Vercel-hosted dashboard to monitor upcoming messages, review patient responses, and handle any exceptions the AI flags for human review. The system is designed to operate with a processing time of under 2 seconds per document and would typically have a monthly hosting cost under $50. You receive all the source code and documentation.