Copy-forward notes corrupting model inputs

1. System failure identification

   Notes carried forward between encounters propagate findings that are no longer true. Problem lists accumulate resolved conditions, and models consuming them score against a patient who no longer exists. The record grows longer and less true with every encounter.

2. Data pipeline diagnostics

   Copied content is indistinguishable from newly authored content in the record. Provenance is not retained, so no consumer can weight a finding by whether it was reasserted or merely inherited. The system preserves the text and discards the fact of its reuse.

3. Predictive computational model

   Retain provenance on every field: authored, copied, or unchanged since a prior encounter. Models can then discount inherited findings, and staleness becomes measurable per problem. Provenance is cheap to capture at the moment of copying and impossible to reconstruct later.

4. Workflow integration

   Expose provenance as subtle visual weight in the note view rather than as a warning. The reader needs to know which lines were reasserted today, and does not need an interruption. A warning on every inherited line would become the next thing dismissed.

Back to the case study database