Workflow Orchestration for CAR-T Cell Therapy Regulations

Data Characteristics

CAR-T cell therapy regulations and SOP data originate from regulatory documents published by drug administration authorities, internal hospital operating procedures, and pharmaceutical company drug inserts and clinical trial protocols. This data typically exists as PDF and Word documents, with a small portion in structured tables.

Update frequency varies: national regulations update quarterly or semi-annually, hospital SOPs may be revised monthly or quarterly based on clinical practice and regulatory requirements, and drug inserts update with new batches or versions.

Document structures often include chapters, clauses, and annexes for regulations, while SOPs usually feature fixed sections like purpose, scope, responsibilities, procedural steps, and precautions.

Fields and units are strict and diverse, involving dosage (e.g., 10^6 cells/kg), time (days, hours), temperature (℃), and various biological indicators (CD3+ cell percentage).

Constraints Imposed on Workflow Orchestration

The characteristics of CAR-T cell therapy regulatory data impose specific requirements on workflow orchestration.

The hierarchical structure of regulatory documents demands deep semantic understanding and relational querying from the knowledge base. This prevents keyword-only matching that ignores context.

The procedural nature of SOPs requires workflows to simulate decision trees, guiding users step-by-step to information.

Frequent updates necessitate efficient version management and incremental update capabilities in the workflow to ensure query accuracy. For example, if a treatment plan's indication scope changes, the workflow must quickly identify and update relevant knowledge snippets.

For fields involving precise measurements and units, the workflow must strictly maintain numerical and unit consistency during information extraction and answer generation. This prevents safety risks from unit confusion. Complex biological indicators may require workflow integration with external computation or validation modules to verify numerical reasonableness.

Configuration Guidelines

Configuration Item	Recommended Value	Rationale
`Chunk size` (Segment Length)	`800–1200 characters`	Ensures semantic integrity of regulatory clauses or SOP steps, preventing critical information from being cut off.
`Chunk Overlap Length` (Segment Overlap Length)	`100–200 characters`	Ensures contextual continuity, handling cross-paragraph references and logical dependencies.
`Recall count` (Recall Count)	`8–12 items`	Balances recall precision with computational overhead, covering multiple potentially relevant clauses or steps in the regulations.
`Similarity threshold` (Similarity Threshold)	`0.78–0.85` (cosine similarity)	Improves relevance recall, filtering out low-relevance generic content, and focusing on regulatory details.
`Rerank result count` (Reranked Return Count)	`3–5 items`	Filters out the most core and directly relevant regulatory or SOP sections, reducing noise for the model.
`LLM_MAX_TOKENS`	`4096 tokens` or higher	Accommodates the complexity and length of regulatory and SOP documents, ensuring enough context for reasoning.

Common Pitfalls

Answers include outdated regulations or SOP content because the knowledge base synchronization mechanism failed to process source document version updates in a timely manner.
The workflow confuses dosage or time units, for example, misinterpreting mg/kg as g/kg. This occurs when text segmentation severs the association between values and units.
When users ask about specific procedural steps, the workflow provides incomplete operational guidance, returning only partial information. This typically happens when the workflow's logical branch design does not adequately cover all possible paths or conditional judgments.

Validation Steps

Select recently updated CAR-T treatment plan indication change documents. Test relevant questions and verify if answers cite the latest version of the content.
For questions involving critical parameters like drug dosage and administration time, check the accuracy of values and units in the answers. Ensure strict consistency with fields like μg/kg or hours in the original SOP documents.
Simulate a complex multi-step operation process question. Observe if the workflow guides the user through the query step-by-step according to SOP logic, providing complete operational steps, such as the entire process from cell collection to reinfusion.

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.