What data for this category looks like
Data for pre-existing condition determination primarily comes from past medical records, medical insurance settlement details, and application notification documents submitted by claim applicants. The data is updated synchronously with each individual claim application. Document structures include fields such as visit time, diagnosis conclusion, standardized disease codes, medication records, and treating medical institution information. Some texts contain multiple segments of visit details. Field units mostly use standard date formats, general currency units, or unified medical industry disease coding rules.
Constraints for model integration and configuration
Data sources for this category are scattered and have diverse formats, including unstructured medical record texts and structured settlement data. The model integration link must support multi-format file parsing and mixed data alignment. Long-text medical histories may exceed the single-round model context window. Segmented recall or long-text processing logic must be configured. Standardized disease code field verification is required. Rule matching parameters for field extraction must be configured during model integration to ensure accurate identification of key diagnosis information. Data submitted in real time with each case requires the model call link to support low-latency responses, to avoid lengthening the claim initial review process.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–16000 characters | Pre-existing condition determination requires covering complete multi-segment visit medical record texts, to avoid truncation of key diagnosis information |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing long medical records and multi-page PDF files takes a long time; this avoids early termination of the parsing process |
Recall Count | Top 3–5 entries | Prioritize matching past medical history records with the highest relevance to the current claim application, balancing information completeness and model load |
Similarity Threshold | 0.75–0.85 | Filter irrelevant medical records, retain matching results suspected of pre-existing conditions, and reduce invalid model inference |
TOOL_CALL_MAX_STEPS | 2 times | Pre-existing condition determination only requires calling two types of tools: medical record parsing and disease code matching, no multi-round iteration is needed |
UPLOAD_FILE_MAX_SIZE | 500 MB | Adapt to common medical record file sizes in claim applications, avoid invalid large file upload requests |
The parameter values provided on this page are conventional recommendations used as a starting point for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require individual analysis. It is recommended to test on your own samples before finalizing settings.
Three common errors
- A locally deployed FastGPT 4.8.22 cannot call a third-party model, returning an
Invalid API Keyerror. This occurs when the third-party model's API key and access domain name are not correctly configured, and the platform's required request header format is not followed. - Parsing a long medical record may result in loss of key diagnosis code information. This occurs when the
Similarity Thresholdis set too high, filtering low-match but critical pre-existing condition records. - The claim initial review process may time out. This occurs when
PARSE_FILE_TIMEOUT_SECONDSis set too short, terminating the process before long medical records finish parsing.
How to confirm the configuration is correct
- Upload a single standard pre-existing medical record file, verify the integrity of the parsed text, and confirm the parsed result matches the source file content.
- Initiate a simulated claim initial review request, check the model call logs, and confirm the tool call count and return format meet configuration requirements.
- Use the built-in API testing tool to initiate a single call to the third-party model, verify interface connectivity and valid key configuration.
- Adjust the value range of core configuration parameters, compare model outputs across different configurations, and confirm alignment with business judgment logic.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.