Automation architecture
Automatic generation workflow
System components
- Data capture: Audio transcription and patient contextual data
- Intelligent processing: Cognitive framework with specialized agents
- Structured generation: Reports conforming to predefined JSON schemas
- Automatic integration: Data delivery through callbacks for persistence
Types of automation
Automatic document generation
SofIA automatically generates structured clinical documentation based on:- Consultation transcription: Real-time processed doctor-patient conversation
- Patient contextual data: Information provided in
patientData - Predefined schema: Structure defined in
toolsArgsthat specifies what information to generate - Automatic validation: Clinical coherence verification before report delivery
Automatic form completion
The system can be used to automatically complete existing forms:- Fill mode: Functionality that allows filling specific fields
- Data mapping: Correspondence between generated information and EHR fields
- Field validation: Verification that data meets form requirements
Specialized agents
Clinical documentation agent
Responsible for generating structured clinical notes:- Transcription analysis: Processing of medical content from the consultation
- Contextualization: Integration with previous patient data
- Coherent structure: Information organization according to defined schema
- Medical validation: Clinical coherence verification of content
Medical coding agent
Specialized in assigning standard codes:- Supported terminologies: ICD-10, SNOMED CT, LOINC as configured in the schema
- Automatic analysis: Identification of codifiable diagnoses and procedures
- Code suggestions: Proposals based on clinical content
- Code validation: Verification of suggested code appropriateness
Review and quality agent
Ensures documentation quality and coherence:- Coherence control: Internal consistency verification in the report
- Error detection: Identification of possible clinical inconsistencies
- Schema validation: Confirmation of compliance with required JSON structure
- Quality improvement: Content refinement before final delivery
Automation configuration
Required properties for automation
To enable automation, SofIA SDK requires:toolsArgs: JSON schema that defines what information to generate and how to structure ithandleReport: Callback function that receives the generated reportuserIdandpatientId: Identifiers for traceability and contextpatientData(optional): Patient contextual information to enrich generation
handleReport callback operation
ThehandleReport callback is the main integration point:
- Data reception: Receives structured report according to defined schema
- JSON format: Data arrives in validated JSON format
- Execution timing: Executes when user requests documentation generation
- Persistence responsibility: Callback must manage saving in EHR/HIS system
Implementation best practices
Effective schema design
To optimize automation:- Specific schemas: Define relevant fields for consultation type
- Clear descriptions: Provide detailed descriptions for each field
- Appropriate validations: Include restrictions that reflect clinical reality
- Terminology references: Use
referenceto enable automatic coding
Integration management
For successful integration:- Data validation: Verify structure and content of received reports
- Error handling: Implement failure management in saving process
- Traceability: Maintain records of who, when, and what was generated
- Versioning: Control schema versions for compatibility
Quality considerations
To maintain high automation quality:- Human review: Establish professional validation processes when necessary
- Progressive configuration: Start with partial automation and gradually increase
- Monitoring: Supervise quality and accuracy of generated reports
- Feedback: Use feedback to improve schemas and configurations
Typical use cases
General medicine consultation
- Symptom documentation: Automatic capture of reported symptomatology
- Examination recording: Documentation of physical findings and vital signs
- Therapeutic plan: Automatic generation of discussed treatment plan
- Follow-up: Automatic scheduling of control appointments
Medical specialties
- Specialized consultations: Schema adaptation to each specialty’s needs
- Procedures: Automatic documentation of performed interventions
- Specific follow-ups: Generation of evolution reports according to protocols
Administrative documentation
- Discharge reports: Automatic generation of hospital discharge documents
- Medical certificates: Creation of certifications based on consultation
- Interconsultation reports: Documentation for referrals to other specialists
Limitations and considerations
Technical limitations
- Schema dependency: Automation quality depends on JSON schema design
- Available context: Generation is based on information available during consultation
- Audio processing: Requires sufficient audio quality for accurate transcription
- Connectivity: Needs stable connection for real-time processing
Professional responsibilities
- Medical supervision: Professional maintains responsibility for generated content
- Clinical validation: Review and validate information before final persistence
- Medical decisions: Clinical decisions remain the professional’s responsibility
- Regulatory compliance: Ensure usage complies with applicable regulations
Ethical and legal considerations
- Patient consent: Ensure patient is informed about AI usage
- Data privacy: Maintain confidentiality according to applicable regulations
- Traceability: Record automation usage for future audits
- Quality of care: Do not compromise care quality for efficiency gains
Monitoring and continuous improvement
Important metrics
- Documentation accuracy: Evaluate precision of generated reports
- Processing time: Monitor latency from capture to delivery
- Utilization rate: Measure what percentage of consultations use automation
- User satisfaction: Evaluate healthcare professionals’ perception
Improvement processes
- Quality analysis: Periodic review of generated reports
- Schema optimization: Schema adjustment based on real usage
- Training: Continuous education on usage best practices
- Source updates: Maintenance of system knowledge sources
Next steps
- Clinical chat: Use the conversational assistant to complement automation
- Clinical data schemas: Design effective schemas for your practice