Data Characteristics in This Category
CAR-T cell therapy product data comes from various sources. These include clinical trial reports, drug monographs, regulatory approval documents, academic papers, and internal pharmaceutical company R&D documents. Data update frequencies vary. Clinical trial data typically releases periodically during and after trials. Drug monographs and regulatory documents update with approval progress or version iterations. Most documents are unstructured text in PDF format. They contain extensive medical terminology, experimental data tables, charts, and references. Key fields include target name, cell source, manufacturing process, indications, adverse reactions, dosage regimens, clinical efficacy data (e.g., ORR, CR, PFS, OS), and safety indicators (e.g., CRS grade, ICANS grade). Data units involve dosage (e.g., cells/kg), time (e.g., days, months), percentages (e.g., %), and various biological indicators.
Constraints Imposed by These Characteristics on Workflow Orchestration
The multi-source and heterogeneous nature of CAR-T cell therapy data requires workflow orchestration to handle various file types and data formats during preprocessing. The high proportion of unstructured text demands robust information extraction capabilities. For example, workflows must accurately extract key efficacy and safety data from clinical trial reports. Inconsistent data update frequencies mean workflow designs should support a combination of scheduled and manual triggers to adapt to different data source update cycles. Clinical trial data, with its complex structure and specialized terminology, particularly requires high accuracy in model comprehension and information extraction. Additionally, unit information for dosage and time in the data necessitates strict unit standardization within the workflow for numerical calculations or comparisons. This prevents discrepancies caused by inconsistent units. For categorical indicators like CRS grade and ICANS grade, the workflow must accurately identify and process them numerically or as labels.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 4000 characters | Upper limit for the length of a single key paragraph in most clinical trial or regulatory documents, ensuring context completeness. |
Chunk size (Segment Length) | 800 characters | Balances long text processing efficiency with semantic integrity, preventing truncation of critical information. |
Recall count (Recall Count) | 5 items | In a highly specialized context, increasing recall helps cover more potentially relevant but differently expressed information points. |
Similarity threshold (Similarity Threshold) | 0.75 | Slightly higher than general thresholds, considering the precision requirements of medical terminology, reducing interference from irrelevant information. |
Rerank result count (Rerank Return Count) | 3 items | Refines the final returned results, focusing on the most relevant information, improving the accuracy and efficiency of consultation responses. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Addresses potentially long parsing times for large PDF files, preventing parsing failures due to timeouts. |
Three Common Mistakes
- When processing clinical trial reports, numerical efficacy indicators (e.g.,
ORRpercentage) are extracted as empty. This occurs because regular expressions or entity recognition models are not correctly configured to handle the association between percentage signs and numerical values. - The workflow calls an external Python function for data processing, and the function returns error code
500. This happens because the Python environment lacks necessary medical data processing libraries, or function parameters do not match workflow node outputs. - When a form input node is nested as a sub-application within the main workflow, its dialog box remains visible. This occurs because the node's
visibleproperty is not explicitly set tofalsein the sub-application configuration.
How to Verify Configuration
- Run the workflow on CAR-T cell therapy documents from different sources. Check the extraction results for key fields (e.g., target, indications, adverse reactions, efficacy data). Ensure field values are complete and units are correct.
- Simulate user queries to retrieve efficacy or safety information for specific CAR-T products. Verify that the workflow's returned results accurately cite data from the documents. Compare these results with the original documents to cross-check values and descriptions.
- Examine workflow execution logs. Confirm that no parsing failures or timeout errors occur when processing documents containing numerous charts and complex tables. Pay attention to the effect of the
PARSE_FILE_TIMEOUT_SECONDSparameter.
The values provided are common starting points. Measure them 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.