Data Characteristics
Hematologic oncology clinical trial pre-screening involves highly specialized and diverse data. Key data sources include clinical trial registries (e.g., ClinicalTrials.gov), medical literature databases (e.g., PubMed, Embase), genomics and proteomics data platforms (e.g., TCGA, cBioPortal), and hospital electronic health record (EHR) systems. Data update frequencies vary; clinical trial registration information may update weekly, while genomic data has longer update cycles. Document structures are complex, including unstructured medical text (e.g., clinical diagnoses, treatment plans, pathology reports), semi-structured study protocol descriptions (inclusion/exclusion criteria, study design), and structured biomarker data and patient demographic information. Fields and units adhere to strict medical standards, such as ECOG Performance Status (0-5 points), LDH (U/L), Beta-2 Microglobulin (mg/L), and gene mutation site descriptions (e.g., FLT3-ITD, NPM1 mutation status).
Constraints on Tool Calling and Plugins from These Characteristics
The specialized and diverse nature of hematologic oncology data imposes specific requirements on tool calling and plugin capabilities. Unstructured medical text requires robust Natural Language Processing (NLP) tools for entity recognition and relation extraction to structure information like inclusion/exclusion criteria. Semi-structured study protocol descriptions, especially complex nested logical inclusion/exclusion criteria, demand plugins capable of parsing complex Boolean expressions. Structured biomarker data requires tools that can perform precise numerical comparisons and range evaluations. Inconsistent data update frequencies mean the knowledge base needs regular synchronization with external data sources and incremental update capabilities to prevent pre-screening result deviations due to outdated data. Strict field and unit specifications require tools to perform rigorous data type validation and unit conversion during parameter passing, preventing calculation errors or logical judgment failures caused by unit mismatches. For example, when an inclusion/exclusion criterion involves patient age, the tool must ensure age data from different sources is handled uniformly.
Configuration Settings
| Configuration Item | Suggested Value | Rationale |
|---|---|---|
maxContext | 32000 tokens | Hematologic oncology clinical trial protocol documents are generally long, containing detailed inclusion/exclusion criteria and study designs. |
Chunk size (Segment Length) | 800–1200 characters | Ensures the completeness of medical entities and contextual information within a segment, facilitating subsequent extraction. |
Recall count (Recall Count) | Top 10–15 entries (Top 10–15 items) | Improves recall rate, covering more potentially relevant clinical trial information. |
Similarity threshold (Similarity Threshold) | 0.75–0.85 | Balances recall and precision, avoiding the retrieval of irrelevant or weakly related trials. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Ensures sufficient parsing time when processing large clinical trial protocol PDF or HTML files. |
Knowledge Base Incremental Sync Frequency (Knowledge Base Incremental Sync Frequency) | Daily at midnight | Registries like ClinicalTrials.gov update daily, ensuring data timeliness. |
Three Common Mistakes
- Symptom: The plugin calls an external search tool, but the returned result list is empty. Cause: Search keywords do not accurately match the index fields of the external data source, or the external API request parameters are incorrectly formatted.
- Symptom: The chart tool in the workflow outputs an empty chart or incomplete data. Cause: Parameter passing to the chart tool in the AI conversation is non-standard, lacking necessary numerical or categorical fields, preventing the tool from generating a valid chart.
- Symptom: After calling the knowledge base upload interface, some document content is missing or formatted incorrectly. Cause: The external program did not perform effective structural processing when retrieving HTML content, or the upload interface has limited parsing capabilities for non-standard HTML.
How to Confirm Correct Configuration
- Ask questions using different phrasings for key inclusion/exclusion criteria. Observe if the tool consistently recalls the correct clinical trial information.
- Select multiple clinical trial documents known to contain specific biomarker data. Test if the tool calling process can accurately identify and extract these biomarker fields and their corresponding values.
- Simulate a complete patient pre-screening process, from patient medical record data input to the final output of a list of eligible trials. Check if all tool calling steps in the process return a
200status code and if the results align with the expected logic.
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.