Data Characteristics in This Category
Pharmacovigilance data in telemedicine primarily comes from patient reports, Electronic Health Records (EHR), wearable device data, and remote consultation records. This data updates frequently. Patient self-reports can occur in real-time, while EHR data updates according to consultation or examination cycles. Document structures vary, including free-text descriptions of adverse events, structured patient demographics, medication records, and diagnostic results. Fields cover drug names, dosages, administration routes, adverse event types (e.g., AE_TYPE), occurrence times, severity (e.g., SEVERITY_GRADE), and outcomes. Units commonly include milligrams (mg) and milliliters (ml) for dosage, and time is precise to hours, minutes, and seconds.
Constraints Imposed by These Characteristics on "Deployment and Upgrade"
The high update frequency and diversity of telemedicine pharmacovigilance data require FastGPT deployments to have efficient data ingestion capabilities. This prevents information lag. Free-text descriptions of adverse events demand stronger text understanding and entity extraction from the model, necessitating configuration of more powerful language models and RAG capabilities. The coexistence of structured and unstructured data means knowledge base construction must support multimodal data indexing and effective integration during retrieval. Dispersed and sensitive data sources impose strict requirements for data security and privacy protection; the deployment environment must meet compliance standards. Additionally, the nature of telemedicine implies potential network instability, requiring optimization of model response speed and fault tolerance mechanisms.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 2048–4096 | Accommodates complex patient histories and detailed adverse event descriptions, preventing truncation of critical information. |
Chunk size (Segment Length) | 500–800 characters (characters) | Balances context completeness and retrieval efficiency, adapting to long medical texts. |
Recall count (Recall Count) | 8–12 entries (items) | Ensures retrieval of sufficient relevant adverse event cases or drug instructions from the knowledge base, improving answer accuracy. |
Similarity threshold (Similarity Threshold) | 0.75–0.85 | Filters out low-relevance content, reduces noise, and focuses on core pharmacovigilance information. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds (seconds) | Handles parsing of large EHRs or batch reports, preventing data processing interruptions due to timeouts. |
rerank_model | bge-reranker-large | Improves the quality of retrieval result ranking, especially when dealing with semantically similar but distinct medical terms. |
Common Pitfalls
- Slow application response, particularly noticeable delays when processing complex queries. This can result from insufficient machine resources (CPU, memory) to support high concurrent requests or large model inference.
- Inaccurate knowledge base retrieval results, failing to link to relevant drug adverse event information, leading to generic responses. This can result from an unreasonable knowledge base segmentation strategy or improper reranker model configuration, failing to effectively identify deep semantic connections.
- System logs showing "OpenAI API connection error" or similar external interface call failures. This can result from incorrect
OPENAI_API_KEYconfiguration or improper network proxy settings, preventing access to external language model services.
Verification Steps
- Conduct question-answering tests on typical drug adverse event report texts. Verify if the model accurately identifies drugs, adverse event types, and severity, and provides relevant advice or warnings.
- Simulate high-concurrency remote consultation scenarios. Observe FastGPT service CPU, memory, and network I/O usage to ensure system resource consumption is within reasonable limits and response times are stable.
- Upload and parse virtual patient EHRs containing various structures (e.g., tables, free text). Check if the knowledge base fully ingests and correctly indexes all critical information, and if it can be retrieved by subsequent queries.
- Check the connection status of external language models and reranker models via the FastGPT management interface. Ensure all dependent services are operating normally.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.