Label Time Machine

Blueprint

Every tool in the lab is documented the same way, from the operational problem to the MLR questions a reviewer will ask.

1. 1

   Operational problem

   Medical information and medical affairs teams are asked when an indication was added, when a boxed warning changed or what exactly the last label update touched. DailyMed keeps every version, but comparing them means downloading files one by one and reading them side by side.

   The result is slow, error-prone answers and label-change monitoring that depends on someone noticing.

2. 2

   Hypothesis

   If every published version of a label can be scrubbed like a timeline, with key changes flagged and edits highlighted word by word, teams will answer label-history questions in minutes and spot safety and indication changes as soon as they are published.

   Seeing fifteen years of a label at once also makes it a teaching tool for how a product's evidence and risk profile evolve.

3. 3

   Solution architecture

   • Relay (labels.php, PHP 7.4+): validated parameters, one fixed upstream (DailyMed web services), no visitor headers or logs. Unzips each version and parses the SPL XML into sections.
   • Cache of parsed public versions on the server, because a published version never changes.
   • Engine (engine.js): sections grouped by LOINC code, paragraph and word diffs (Myers), milestone rules for new indications, boxed warnings, contraindications and warnings.
   • Interface: canvas timeline of section sizes with a playhead, version player, word-level diff viewer, milestones list and CSV change log.
   • Tests: the relay end to end against a mock DailyMed built from real SPL files; diff correctness on random edits.
   • Installable PWA; a fictional demo label works offline.
4. 4

   HCP and patient impact

   HCPs: faster, accurate answers to questions about what the current label says and when it changed.

   Patients: indirect. Safety updates such as boxed warnings and new contraindications become visible to medical teams as soon as they are published.

   No patient-level data is processed at any point.

5. 5

   Regulatory and MLR considerations

   • Public label data only; no personal or HCP data.
   • For internal monitoring, medical information research and teaching. It is not a regulatory record: check the official label on DailyMed before relying on any detail.
   • No generative AI: every flagged change is computed from the label text and can be traced to it.
   • Data source attribution: DailyMed, U.S. National Library of Medicine. Label Time Machine is independent and not endorsed by NLM or FDA.