SigBase n., prescription directions
Prescription directions read into fields a pharmacist can approve.
SigBase reads the shorthand directions on a prescription into 18 fields, sends anything it can’t read for certain to a pharmacist, and exports the approved result to FHIR and NCPDP.
- My part
- Design and build
- Built with
- TypeScript
- For
- Pharmacists
The problem
what arrives at the pharmacy
The directions on a prescription arrive as shorthand, such as “take 1 tab po bid prn pain”. Before a pharmacy system can check a dose or print a label, someone has to turn each line into structured fields.
Plate to comeA line and its fields, waiting for review.
Constraints
what it has to work within
A wrong field can hurt someone, so a pharmacist approves every line and each approval is recorded. No patient details may come in. The output has to fit the formats pharmacy systems already read.
Plate to comeAn approval and its audit trail.
What I built
the decisions, in order
Grammar first
A written grammar reads each line. A language model may suggest a reading only for the parts the grammar leaves open, and it never writes to the database.
See Plate III.
It declines on purpose
When the grammar can’t read a part for certain, it marks that part as unread and hands it to a pharmacist.
See Plate III.
Eighteen fields to approve
Each line becomes 18 fields, from dose and route to frequency and indication. The pharmacist checks and approves them, and every approval goes into an audit trail.
See Plate II.
No patient details, by design
Every imported file is checked for patient identifiers, and the whole file is rejected if one turns up.