Cardiovascular Interventional Pharmacovigilance: Source and

Data Characteristics

Cardiovascular interventional pharmacovigilance data primarily comes from Medical Device Reports (MDRs), clinical trial reports, real-world evidence (RWE) data, and academic literature. Regulatory bodies, such as the FDA's MAUDE database, collect MDR data. Updates occur monthly or quarterly. Clinical trial reports and RWE data release periodically, based on study cycles and publication strategies.

MDR reports often include structured fields for device information, event descriptions, patient information, and outcomes. They also contain extensive free-text descriptions. Academic literature, primarily in PDF format, follows standard sections like abstract, introduction, methods, results, and discussion. Specific units, such as millimeters (mm), kilograms (kg), and milliliters (mL), describe device specifications and patient physiological parameters. Standardized medical terminology, including ICD-10 codes, is also common.

Constraints on Source and Traceability

The diverse nature of cardiovascular interventional data imposes specific constraints on source and traceability. MDR reports contain free-text and semi-structured data. The RAG system must effectively process non-standardized event descriptions during chunking and embedding to ensure relevant recall. Irregular update frequencies require the knowledge base to support incremental updates and version management, reflecting the latest adverse event information.

Academic literature in PDF format, especially with complex layouts and figures, demands advanced document parsing capabilities. This can lead to incomplete text extraction or formatting loss, affecting citation accuracy. The system must accurately identify and process numerical information with units due to precise device parameters and physiological indicators. This prevents ambiguity or errors in citations. Correct recognition and association of medical codes like ICD-10 are crucial for tracing back to original medical terminology.

Configuration Settings

Configuration Item	Suggested Value	Rationale
`Chunk Size`	`500–800 characters`	Balances the completeness of MDR event descriptions with the semantic coherence of academic literature paragraphs, preventing truncation of critical information.
`Recall Count`	`Top 8`	Increases the number of recalled items to cover more potentially relevant information, considering the complexity of cardiovascular interventional adverse event reports.
`Similarity Threshold`	`0.75`	Ensures high relevance between recalled content and queries, reducing false positives, especially when dealing with specific device models and symptoms.
`Reranked Return Count`	`Top 3`	Streamlines the final displayed citations while maintaining relevance, improving user reading efficiency.
`PARSE_FILE_TIMEOUT_SECONDS`	`180 seconds`	Addresses parsing of large PDF documents, particularly clinical trial reports with multiple pages of figures and complex layouts.
`MAX_PREVIEW_SIZE`	`5 MB`	Allows users to preview larger original report files, facilitating context verification for cited sources.

Common Pitfalls

Cited source URLs in Q&A results are inaccessible or incorrect. This often occurs when the knowledge base's original link is not properly configured with a proxy or CDN path after deployment. This prevents the frontend from correctly resolving the backend's file download address.
Cited content lacks critical numerical values or units, resulting in incomplete information. This happens when the document parsing stage fails to effectively identify and extract numerical-unit combinations from free text, or when chunking separates numbers from their units.
RAG Q&A results return no cited sources, even when relevant content exists in the knowledge base. This may be due to a maxContext parameter that is too small, preventing all retrieved content from being sent to the large language model's context, or a similarity threshold that is too high, filtering out relevant documents.

Verification Steps

For typical cardiovascular interventional adverse event queries, check if the Q&A results return at least 3 valid citations. Click and verify that all citation links are accessible.
Randomly select 5 original documents containing numerical values and units (e.g., MDR reports). Query the system to verify if it accurately cites and presents these values and units, comparing them against the original text for accuracy.
Upload a PDF document with complex tables and figures. Observe if parsing time falls within the PARSE_FILE_TIMEOUT_SECONDS configuration. Check if the parsed text content retains all critical information.
In the knowledge base management interface, randomly select several cardiovascular interventional documents. Verify that their Chunk Size meets expectations and that the chunked content maintains semantic integrity.

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.