Extract patient-reported medication use for review¶
A pharmacist reviewing a call needs to see what the caller actually said. This example sends a synthetic transcript to Claude and extracts medication mentions, reported adherence, barriers and explicit follow-up requests. The pharmacist reviews the evidence and decides what action to take.
Source transcript¶
Pharmacist: What would you like to review today?
Caller: I missed two doses of Medicine A because the refill cost too much. Could the pharmacist call me about the refill?
Caller: I also take Medicine B.
Review the extracted evidence¶
| Medication | Reported status | Evidence |
|---|---|---|
| Medicine A | Missed doses | “I missed two doses of Medicine A” |
| Medicine B | Unclear | “I also take Medicine B.” establishes the mention. The caller gives no adherence status. |
The cost barrier has its own quote, “the refill cost too much”. The follow-up request is the caller’s question, “Could the pharmacist call me about the refill?” A model validator checks every quote against the transcript. It also requires a supporting quote for each stated adherence status.
Run the example¶
From a checkout of Instructor, set ANTHROPIC_API_KEY and ANTHROPIC_MODEL for the Claude deployment you want to use. The script prints the structured result for the synthetic call above.
The example passes the transcript in Instructor’s validation context so a fabricated quote fails validation and can be retried. Verbatim matching establishes that the words occur in the transcript. The reviewer checks that they support the extracted status. Runnable example · Validation context