Data Characteristics
Cardiovascular interventional drug safety data primarily comes from national adverse drug reaction monitoring systems, hospital electronic medical records, follow-up records, and medical literature. Data updates are frequent; adverse event reports may be submitted weekly or monthly, and medical literature is continuously published. The data document structure is complex, typically including patient basic information, diagnoses, medication details (including device use), adverse event descriptions, event assessment results, and treatment measures. Adverse event descriptions are often unstructured text, containing medical terminology and abbreviations. Device-related information includes model, batch, and implantation date. Some fields involve specialized scale scores, such as bleeding risk scores and ischemic event scores.
Constraints Imposed by Data Characteristics on Deployment and Upgrade
High-frequency data sources require deployment solutions with incremental update capabilities. This avoids resource waste and service interruptions from full knowledge base rebuilds. Unstructured text and medical terminology demand more sophisticated text segmentation strategies and embedding model choices in FastGPT to ensure accurate semantic understanding. Integrating data from multiple sources requires handling different data formats and field mappings to build a unified knowledge structure. Key identifiers like device models and batches need effective extraction and indexing during knowledge base construction to support quick queries based on specific devices. Specialized scale scores require specific data preprocessing, potentially involving regular expressions or rule matching, to convert numerical information into retrievable text descriptions.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
UPLOAD_FILE_MAX_SIZE | 500 MB | Original medical records and follow-up records may contain many images or scanned documents, leading to large file sizes. |
maxContext | 3000 Tokens | Cardiovascular interventional adverse event descriptions may involve complex medical histories and multiple medications, requiring a longer context window. |
Chunk size | 800 characters | Ensures a single segment can contain a complete adverse event description while avoiding excessive length that leads to information redundancy. |
Similarity threshold | 0.75 | The medical field demands high recall precision; increasing the threshold helps filter out irrelevant results. |
Recall count | Top 10 entries | Increases the number of recalled items to capture more potentially relevant medical literature or historical cases. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Processing large PDFs or documents with complex structures can take a long time, preventing timeout interruptions. |
Common Pitfalls
- Knowledge base index construction stalls. The management interface shows "indexing" status for an extended period. This often happens when unexpected special characters or format errors in the data preprocessing stage prevent the text parser from processing correctly.
- Model responses contain much irrelevant or repetitive information. Answers appear verbose and lack specificity. This is due to an improper text segmentation strategy that fails to effectively identify the core content of adverse event reports, leading to context confusion.
- New data cannot be retrieved. User queries do not yield answers from the latest adverse event reports. This usually means the incremental update mechanism is misconfigured, data synchronization intervals are too long, or update tasks fail without timely alerts.
Verification Steps
- Upload a typical adverse event report document containing complex medical terminology and device models. Check if the knowledge base correctly identifies and segments key information.
- Ask questions about specific device models and adverse event types. Verify if the model can recall and integrate relevant reports. Compare the accuracy and completeness of the answers.
- Simulate a new adverse event report data import. Check if the knowledge base's incremental update task triggers as expected and completes index updates within the specified time.
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.