Data Characteristics in This Category
Smart triage in pharmacovigilance primarily uses data from drug inserts, adverse event reports (e.g., CIOMS I forms), medical literature, clinical trial data, and post-market drug surveillance databases. This data updates frequently, especially for ongoing monitoring of adverse events after new drugs launch. Document structures typically include standardized fields such as patient demographics, drug names, dosages, administration routes, adverse event descriptions (MedDRA codes), event timing, and outcomes. Fields often involve medical terminology and specialized abbreviations. Units cover dosage units (mg, g, IU), time units (days, hours), and frequency units (times/day).
Constraints Imposed by These Characteristics on "Multi-turn Conversation and Prompts"
The specialized and time-sensitive nature of pharmacovigilance data demands high accuracy in multi-turn conversations. Precise matching of medical terminology and contextual understanding are critical. Incorrect or ambiguous understanding can lead to serious medical risks. Frequent data updates require models to quickly absorb new knowledge and avoid providing outdated information. Complex document structures and diverse field types necessitate clear guidance in prompt design to direct the model to extract information from specific fields and perform logical reasoning. Additionally, free-text descriptions in adverse event reports challenge the semantic understanding and summarization capabilities of multi-turn conversations. Models must identify and extract key symptoms, signs, and medication history.
Configuration Settings
| Configuration Item | Recommended Value | Rationale for Value |
|---|---|---|
maxContext | 8 | Retains sufficient context to understand complex medical logic and patient condition evolution in multi-turn conversations. |
Chunk size (Segment Length) | 500 characters (characters) | Accommodates longer descriptive text in medical literature and reports, ensuring semantic completeness. |
Recall count (Recall Count) | 10 entries (items) | Increases coverage when retrieving relevant adverse event information and medication guidelines from the knowledge base. |
Similarity threshold (Similarity Threshold) | 0.75 | Ensures retrieved knowledge snippets are highly relevant to the patient's query, avoiding irrelevant information. |
Rerank result count (Reranked Return Count) | 3 entries (items) | Optimizes the final knowledge snippets presented to the user, focusing on the most core diagnostic or recommendation basis. |
System Prompt | Guides the model to focus on adverse event identification, risk assessment, and to respond in rigorous medical language. | Ensures conversations consistently align with the core goals of pharmacovigilance and maintain professionalism and safety. |
Common Pitfalls
- Symptom: During a conversation, a specific drug is mentioned, but the model fails to correctly identify its known adverse reactions. Reason: Drug information in the knowledge base is not updated promptly, or the prompt does not effectively guide the model to retrieve the latest drug inserts.
- Symptom: A user describes symptoms, but the model's response includes medical terms or diagnostic suggestions unrelated to the symptoms. Reason: The
Similarity threshold(Similarity Threshold) is set too low, leading to the retrieval of imprecise knowledge snippets, or the model's understanding of the context is insufficient. - Symptom: A conversation includes LaTeX-formatted medical formulas or specialized symbols, but they display as raw code in the frontend. Reason: The application frontend does not integrate a LaTeX rendering library, preventing correct parsing and display of complex mathematical or chemical expressions.
How to Verify Correct Configuration
- Input a query containing adverse reactions for a newly launched drug. Check if the model's response cites the latest drug vigilance information.
- Simulate a multi-turn conversation, progressively delving into a complex case. Check if the model can continuously understand the context and provide coherent medical advice.
- Test queries containing MedDRA codes or ATC classifications. Check if the model can accurately identify and link them to relevant drugs or adverse events.
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.