Every cancer center has its own grading conventions. This week we built the mechanism for institutions to encode their own preferences into the AI's grading reasoning, expanded EHR data coverage, and made severity the default ordering for adverse event review.
Giving institutions a real say in how the platform grades
Every cancer center has its own grading conventions. A Grade 2 nausea call at one institution may follow different criteria than the same call at another, depending on patient population, trial protocols, and clinical practice patterns. An AI grading system that does not let institutions encode that local knowledge is, by definition, less accurate than one that does.
This week we built the mechanism. Organizations can configure rules specific to their institution that the AI weighs during grading suggestions. We also expanded EHR data coverage materially, and made severity the default ordering for adverse event review so coordinators see the events that matter first.
Here is what shipped, why it matters, and what is next.
Feature Highlight 1: Clinical Preferences for institutional grading rules
Organizations can now configure rules that guide AI suggestions across all four workflow categories:
- CTCAE grading: institution-specific criteria for grade determination
- Drug attribution: custom rules for multi-drug causality analysis
- Drug interaction analysis: organization-specific interaction protocols
- Encounter title generation: preferred formatting and terminology
These preferences are passed as additional instructions into every AI workflow, ensuring suggestions align with how the team actually practices. Preferences can be toggled on and off without deletion, so the team can experiment with different rule sets and roll back cleanly.
The feature ships with a comprehensive end-to-end test suite covering create, read, update, delete operations, organization scoping, authentication enforcement, and toggle behavior. This is the level of test rigor we apply to clinical-grade software.
Feature Highlight 2: Expanded EHR integration coverage
Our Oracle Health integration previously supported about forty FHIR resource scopes. This week, we expanded coverage to over a hundred resource permissions using the updated SMART/FHIR specification.
What this means in practice:
- Full create-read-update-delete operations across clinical domains (Encounters, Observations, Procedures, Medications, CarePlans)
- DiagnosticReports for lab results that inform adverse event grading
- Conditions for comorbidity cross-referencing during CTCAE analysis
- MedicationAdministration for drug tracking in multi-drug attribution workflows
The expanded scope enables automatic extraction of about fifty additional FHIR resource types critical for oncology trials, improving data completeness for AI grading models.
We also improved OAuth error handling across all EHR connections (Epic, Oracle Health, AthenaHealth). Failed authentication now routes users back to the appropriate app context with the integrations panel open, instead of dropping them on a generic error page.
For new users, a "Connect to EHR" dropdown now appears directly on the clinical note input screen, reducing the steps to link an EHR from five clicks to one.
Feature Highlight 3: Severity-ordered adverse event review
Adverse events are now sorted by grade in descending order. Grade 5 at top, Grade 1 at bottom. High-severity events (Grade 3-5) expand by default. Low-severity events collapse to reduce visual noise.
This may sound like a small change. For coordinators reviewing patients on combination therapies with ten or more adverse events, it eliminates the scan-the-whole-list step. The clinically significant events surface immediately. Grade descriptor tooltips on every badge provide instant verification without opening a separate definitions dialog.
Data export and documentation improvements
- CSV export: drug attribution and interaction history can now be exported as CSV files alongside the existing JSON export, making it easier to import into spreadsheets and statistical tools for analysis
- Soft-delete architecture: clinical documentation deletion now uses soft-delete with archive timestamps across eleven tables. This preserves data integrity for audit trails while allowing users to clean up their workspace. The implementation also eliminated over twenty unsafe type assertions throughout the deletion flow.
- Improved invitation flow: when a user receives an organization invitation but is signed into the wrong account, the platform now signs them out automatically and redirects to the sign-in page with the invite link preserved. Previously this scenario produced a confusing error.
Looking Ahead
Next: template rules for common oncology protocols in the Clinical Preferences system, making it faster for new organizations to start with institution-specific grading.
More Friday updates at burna.ai/blog.
