Editing Metadata Effectively in GenAI Validation

Metadata commands define how metadata should be updated within the Metadata Editor. Writing clear and structured commands helps ensure that updates are interpreted correctly and that regenerated tables reflect the intended logic.

Using the Command Interface

Enter commands directly in the editor. Type / to open the command menu and insert available elements such as datasets, variables, and statistical methods. The menu supports structure, but clear logic and intent must be defined by the user.

Best Practices for Writing Commands

1. Structure your commands

• Number each instruction (1, 2, 3...)
• Write one instruction per line
• Keep each instruction focused on a single action

Examples:

1. Use the ADSL dataset to remove subjects with FASFL != 'Y'

2. Calculate >1 row by filtering REGIMENN > 1

3. Add percentage to the "Number of Participants Evaluable for DLTs" row

2. Be specific about variables and datasets

• Use exact variable names (e.g., TRT01P instead of TRT01A)
• Specify datasets when relevant (e.g., ADSL, ADAE)

Examples:

Use TRT01P instead of TRT01A

Use AEBDSYS1 instead of AEBODSYS

3. Define filtering criteria explicitly

• Use clear conditions with operators (=, !=, >, <)
• Specify both variable and value

Examples:

Remove subjects with FASFL != 'Y'

Calculate >1 row by filtering REGIMENN > 1

4. Describe calculation logic clearly

• Specify aggregation methods (count, summation)
• Define grouping when relevant

Examples:

Count distinct subjects grouped by AEBDSYS1 and COHORT

The "Total" column should be the summation of DL 1, DL- 1, DL- 2 columns

5. Specify output formatting

• Define exact structure and spacing
• Include examples when possible

Examples:

Use the format: median (min, max)

95% CI should be without bracket

6. Use exact text matches

• Quote exact row/column names
• Preserve spacing and special characters

Examples:

Filter by SMQ01NAM = 'Peripheral neuropathy (SMQ)'

7. Add context for specific cases

• Specify when logic applies to certain rows or classifications

Examples:

Add percentage to the "Number of Participants Evaluable for DLTs" row

Apply filtering only for SMQ classification

8. Specify N and n calculations when required

• Define whether it applies to big N or small n
• Specify dataset and calculation logic

Examples:

1. Use TRTA from ADAE dataset to calculate n (small n, table body)

2. Use the summation of all TRTxxA from ADSL to calculate big N

3. For crossover studies: use TRT01P from ADSL for big N calculation

Summary

• Structured
• Specific
• Clear in logic and formatting
• Precise in references

This ensures accurate metadata updates and reliable table regeneration.