Sentence examples for medication entity detection from inspiring English sources

Table  2 shows the performances of the medication entity detection process.

Details of the medication entity detection process can be found in our earlier publications [ 37- 39].

The hybrid algorithm achieved 95.0%/91.6%/93.3% of P/R/F on medication entity detection and 98.7%/99.4%/99.1% of P/R/F on attribute linkage.

Medication entity detection, including medication name and attribute detection, was designed in-house based on the MALLET conditional random field (CRF) package [ 35, 36].

In a double-annotated, gold-standard based evaluation of real-world medication reconciliation data, the proposed algorithm showed good capability in medication entity detection, attribute linkage and medication matching.

The output of the medication entity detector (Bullet A in Figure  2) was the input of the attribute linkage algorithm instead of the gold-standard annotations.

Finally, the patient population investigated in the study usually has long medication lists documented in clinical notes, providing a potentially more suitable foundation for medication entity and discrepancy detection algorithms.

Approximately 68% of the errors (cause 1) were ascribed to the omission of medication entities by the entity detection algorithm due to abbreviations used in the clinical notes (e.g. "abx" for "antibiotics" and "NS" for "normal saline"), misspellings (e.g. misspelled "Affrin" as "Afrin" and "Nutropin" as "Neutropen") and uncommon medication names (e.g. "Pedia Sure").

The recall propagation showed that both the entity detection and medication matching processes contributed to the loss of detection on discrepant medications; therefore, further refinements in these components are required to improve medication discrepancy detection.

The performance was lower on the overview notes, with 82.3% on entity detection and 64.8% after medication matching.

For discrepant medications in the discharge summaries, the proposed algorithm achieved a 100% recall on entity detection and an overall recall of 88.0% after medication matching.