Comparing Methods for Extracting Daily Dosage from Prescriptions
Author Information
Author(s): Haaker Theodorus S, Choi Joshua S, Nanjo Claude J, Warner Phillip B, Abu-Hanna Ameen, Kawamoto Kensaku
Primary Institution: University of Utah
Hypothesis
Which method is best for extracting daily dosage information from free-text prescription signatures?
Conclusion
Both the LLM and RxSig models excel in daily dose extraction from free-text sigs, with RxSig being the more scalable approach.
Supporting Evidence
- The LLM achieved an F1-score of 0.98, while RxSig scored 0.95.
- RxSig had a positive predictive value (PPV) of 0.99, indicating high accuracy.
- The BiLSTM model scored the lowest with an F1-score of 0.71.
- RxSig completed its tasks in 4 minutes, while the LLM took 210 minutes.
- Both the LLM and RxSig showed high sensitivity, consistently above 0.90.
Takeaway
This study looked at different ways to read prescriptions and found that some methods are really good at figuring out how much medicine someone should take each day.
Methodology
Five methods for extracting daily dosage were compared: Parsigs, RxSig, Sig2db, a large language model (LLM), and a BiLSTM model, using a dataset of free-text prescription signatures.
Potential Biases
Potential bias due to the annotation of the dataset being completed before algorithm adaptations.
Limitations
The study relied on a single health system and condition, which may limit the generalizability of the findings.
Participant Demographics
The dataset included 29,835 free-text sigs from 17,295 patients.
Statistical Information
P-Value
P=1.000
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website