Source and Attribution for Patient Assistance Clinical Trial

Data Characteristics in This Category

Patient Assistance Program (PAP) data originates primarily from pharmaceutical companies, charitable foundations, patient service organizations, and medical institutions. This data exists in both structured formats (e.g., patient registration forms, medication records, financial eligibility proofs) and unstructured formats (e.g., medical records, diagnostic reports, genetic test reports, imaging reports). Data updates frequently, with patient status, medication regimens, and financial situations changing dynamically over time. Document structures vary; for instance, medical records may contain multiple free-text descriptions, while diagnostic reports often include standardized diagnostic codes and descriptions. Key fields include patient ID, disease diagnosis, medication history, eligibility criteria compliance, income verification, past treatment history, and adverse event records. Units cover dosage (mg, mL), frequency (times/day, week), time (year, month, day), and financial amounts (RMB, USD).

Constraints Imposed by These Characteristics on "Source and Attribution"

The data characteristics of patient assistance clinical trial pre-screening impose specific requirements on source and attribution. Unstructured text (e.g., medical records) requires advanced natural language processing techniques for information extraction to accurately identify key eligibility criteria fields. High-frequency data updates necessitate knowledge base capabilities for incremental updates and version management to prevent citing outdated information. Diverse document structures require flexible text segmentation strategies to avoid truncating or obscuring critical information. For example, diagnostic codes and descriptions in a diagnostic report must be cited as a whole to ensure complete diagnostic information. Due to the high sensitivity of the data, sources must be strictly limited to the scope authorized by the patient and traceable to the original document and specific passage to meet compliance requirements. Standardized processing of fields and units, such as consistent conversion of dosage units, is crucial for accurate pre-screening logic.

Configuration Settings

Configuration Item	Recommended Value	Rationale
`Chunk size` (Segment Length)	`500-800 characters`	Balances context continuity in long documents with retrieval efficiency in short documents, preventing key information truncation.
`Recall count` (Recall Count)	`Top 8-12 entries`	Ensures coverage of potential relevant information from multiple source documents, improving the comprehensiveness of pre-screening.
`Similarity threshold` (Similarity Threshold)	`0.78-0.85`	Balances precision and recall of retrieval results, reducing the risk of false positives and false negatives.
`Rerank result count` (Rerank Return Count)	`Top 5 entries`	Focuses on the most relevant few document snippets, reducing model processing load and improving response speed.
`maxContext`	`4096-8192 token`	Accommodates the context requirements of longer documents like medical records and diagnostic reports, ensuring information completeness.
`PARSE_FILE_TIMEOUT_SECONDS`	`600 seconds`	Addresses the time required to parse large medical record files, preventing timeouts that lead to file processing failures.

Three Common Mistakes

Symptom: The system displays "Cited Knowledge Base (1 entry)," but more than one relevant document exists. Reason: Inadequate knowledge base segmentation strategy, causing relevant information to be scattered across multiple segments, but only one is retrieved, or Recall count (Recall Count) is set too low.
Symptom: Pre-screening results show incorrect patient medication dosage or frequency information. Reason: The knowledge base failed to accurately identify and standardize different expressions for dosage and frequency units when processing unstructured medical record text.
Symptom: After integrating the model, testing prompts "Request timeout." Reason: PARSE_FILE_TIMEOUT_SECONDS is set too short, failing to adequately process complex or large medical documents, or maxContext is too large, leading to excessive model inference time.

How to Confirm Proper Configuration

Select a set of typical patient data containing various document types (medical records, diagnostic reports, test results). Conduct pre-screening tests and check if each response cites all relevant document snippets.
For specific patient eligibility criteria, adjust the Similarity threshold (Similarity Threshold) and observe changes in the accuracy and recall rate of pre-screening results until business requirements are met.
Simulate uploading documents of different sizes and complexities. Monitor the success rate of file processing within PARSE_FILE_TIMEOUT_SECONDS to ensure all documents are effectively parsed.

Note: 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.