AI SaMD Playbook
    Talk to us →
    Back to AI SaMD PlaybookAmendment tool · for PCCP revisions & FDA AI rounds

    PCCP diff checker

    Paste the cleared PCCP on the left and the amended version on the right. Get a side-by-side diff with a section-level summary so you can quickly answer the reviewer's first question · "what actually changed?".

    Changes+7 added−5 removed
    Section-level summary
    • PCCP · DiabetesScreen v1.0+0 / −1
    • PCCP · DiabetesScreen v1.1+1 / −0
    • Description of Modifications+2 / −1
    • Modification Protocol+3 / −2
    • Impact Assessment+1 / −1
    OriginalAmended
    # PCCP · DiabetesScreen v1.0
    # PCCP · DiabetesScreen v1.1
    ## Description of Modifications
    ## Description of Modifications
    - Periodic retraining every 6 months on new institutional data.
    - Periodic retraining every 3 months on new institutional data.
    - Threshold for positive prediction held at 0.62.
    - Threshold for positive prediction held at 0.62.
    - Added support for ingesting de-identified Epic FHIR observations.
    ## Modification Protocol
    ## Modification Protocol
    - Acceptance: sensitivity ≥ 0.85, specificity ≥ 0.90 on held-out test set.
    - Bias monitored across 5 self-reported race groups.
    - Acceptance: sensitivity ≥ 0.87, specificity ≥ 0.90 on held-out test set.
    - Bias monitored across 5 self-reported race groups and 4 age bands.
    - Drift threshold: subgroup AUROC drop > 0.03 triggers investigation.
    ## Impact Assessment
    ## Impact Assessment
    - No change to intended use.
    - No change to intended use.
    - No change to clinician workflow.
    - Workflow: adds inline confidence band in EHR side panel.

    Runs entirely in your browser · no PCCP content leaves this page.