AI use register
The summary is generated; the reading is yours. This page goes with Chapter 17 and Chapter 18. It records which registry tasks an AI tool may do, which need a named person, what data each tool sees, and how its output is checked. The first rows of the first table are examples; change them to fit your registry.
|
Entry |
| Registry |
|
| Condition |
|
| Completed by (role) |
|
| Date |
|
| Version |
|
Tasks and permissions
| Task |
AI may do it (yes, draft only, no) |
Tool and version |
What data the tool sees |
Named person who reviews |
How the output is checked |
Approved (date) |
| Draft a participant newsletter |
Draft only |
|
No participant data |
Coordinator |
Read in full before it is sent |
|
| Profile missing data by field and site |
Yes |
|
De-identified extract |
Data manager |
One number traced back to its rows |
|
| Suggest vocabulary mappings |
Draft only |
|
Field values, no identifiers |
Data manager |
Each mapping checked; vocabulary release recorded |
|
| Summarize the cohort for a sponsor |
Draft only |
|
Aggregate tables |
Principal investigator |
Every number traced; definitions of confirmed and evaluable checked |
|
| Decide whether a diagnosis is confirmed |
No |
|
|
Qualified reviewer |
Not delegated |
|
| Contact a participant |
No |
|
|
Coordinator |
Not delegated |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Output checks
| Date |
Output |
Number traced |
Rows it came from |
Definitions used (and version) |
Matches? |
Checked by (role) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Vocabulary mapping record
| Source value |
Standard term |
Identifier |
Vocabulary and release |
Mapped by (role) |
Checked by (role) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|