AI System for Preparing Administrative Summaries of Patient Call Notes

The problem

In daily operations, several requests from one call may be lost within a single note. When the call transcript, patient record number, task type and unit directory are spread across different screens, documents or team records, the issue may not be visible in time.

This fragmented data delays preparation of the call summary and task draft and forces staff to search again for supporting information. Before deciding, the call centre officer must complete missing fields and verify conflicts one by one.

A common exception is that an administrative sentence may contain a symptom requiring clinical assessment. If this context is not checked, an incomplete or incorrect workflow could close clinical content as a routine task, so its output cannot be applied directly.

What we set out to improve

• Number of tasks added or removed by the officer
• Rate of outputs corrected or rejected by the call centre officer
• Time between the source record and the review output

How the system works

1. 01 Define the review scope and the authorised call centre officer role
2. 02 Receive the call transcript, patient record number, task type and unit directory
3. 03 Check source identity, date, version and data freshness
4. 04 Extract administrative requests
5. 05 Match tasks with candidate units
6. 06 Check the exception that an administrative sentence may contain a symptom requiring clinical assessment
7. 07 Place low-confidence or conflicting records in a separate human review queue
8. 08 Write the source identity, rule, output and time to the audit trail
9. 09 Have the call centre officer approve, correct, defer or reject the output

Methods we used

• Administrative-intent classification with clinical expressions placed in a separate queue
• Source, format, version and mandatory-field checks for call transcripts, patient record numbers, task types and the unit directory
• A missing-data warning instead of inference when inputs are absent or conflicting
• Separate assessment of exceptions and the main rule
• Joint retention of the source identity, rule result, call summary and task draft, and human decision

Where people stay involved

Data and security

Call transcripts, patient record numbers, task types and the unit directory are accessed with the narrowest permissions required for the role. Identity and sensitive fields are masked where possible, and only the necessary section rather than the entire raw file is sent to the model. Access, output and the call centre officer's decision are logged; production use does not begin until the organisation approves data retention and location.

Who this suits

Frequently asked questions