Data Characteristics
Smart triage systems primarily use medical knowledge bases, clinical guidelines, drug inserts, disease diagnosis and treatment protocols, and historical medical records as data sources. This data combines structured formats (e.g., disease codes, symptom dictionaries) and unstructured formats (e.g., physician notes, medical literature). Clinical guidelines and drug inserts typically have fixed revision cycles, updating quarterly or annually. In contrast, epidemiological data or new drug information may update more frequently, in real-time or near real-time.
Medical knowledge bases often use hierarchical classifications, including standard fields like disease definitions, etiologies, symptoms, diagnoses, and treatment plans. Drug inserts have fixed sections, such as indications, dosage and administration, and adverse reactions. Fields and units involve numerous medical terms, units of measurement (e.g., mg, ml, mmol/L), and time units (e.g., hours, days, weeks).
Constraints on Citation and Traceability
The data characteristics of smart triage quality documentation impose specific requirements on citation and traceability. The hierarchical structure and high density of specialized medical terminology mean that a single text block might not provide complete context in RAG (Retrieval Augmented Generation). This requires tracing back to higher-level sections or related concepts.
Using multiple data sources requires the system to distinguish and label information from different origins, such as differentiating official guidelines from clinical experience summaries. Frequently updated medical information means document version management is crucial to ensure citations always refer to the latest, authoritative versions.
Strict compliance requirements, such as medical device registration certificates and drug approval numbers, must be clearly displayed in citations to meet regulatory demands. Identifying and precisely matching units of measurement and specialized fields prevents incorrect citations due to unit confusion or misinterpretation of fields.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Chunk size (Chunk Size) | 800–1200 characters | Medical texts have strong contextual relevance; this ensures each chunk contains sufficient information. |
Recall count (Retrieval Count) | Top 5–8 items | Improves recall rate for relevant medical knowledge, covering multi-dimensional information. |
Similarity threshold (Similarity Threshold) | Calibrate based on actual measurements | Adjust based on the vectorization effectiveness of specific medical terms to avoid over-generalization or missed retrievals. |
Rerank result count (Reranked Return Count) | Top 3 items | Optimizes sorting for multi-source knowledge bases, prioritizing authoritative and highly relevant content. |
maxContext | 4000 tokens | Ensures the large language model can process complex medical questions involving multiple cited passages. |
ENABLE_DOC_VERSIONING | true | Mandates document version management to ensure the timeliness and accuracy of citation sources. |
Common Pitfalls
- Missing or incomplete citation sources in generated content. This might occur if the
Similarity threshold(Similarity Threshold) is set too high, preventing slightly less relevant but valid passages from being retrieved. - Citations pointing to outdated or deprecated medical guidelines. This typically happens if
ENABLE_DOC_VERSIONINGis not enabled or if the document version management mechanism is misconfigured. - The
Knowledge base search(Knowledge Base Search) node in the workflow fails to correctly parse knowledge bases referenced by variables, leading to an incorrect search scope or empty data. This might be due to the knowledge base variable format passed during the API call not matching the system's expectations.
Validation Steps
- Query with typical disease cases. Verify that all cited document snippets in the generated answer can be traced back to their original documents. Check that document titles, sections, and version numbers are correct.
- Test with questions containing specific medical units of measurement. Confirm that numerical values and units in the citation sources are accurate and unambiguous.
- Simulate a medical knowledge base update. Check if the smart triage system prioritizes citing the latest version of relevant documents and can correctly identify and label differences between new and old versions.
- Verify that when the knowledge base contains multiple similar medical documents from different sources, the system prioritizes citing documents from official or authoritative sources.
The values provided are common starting points and should be measured against specific 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.