Error Detection in Patients’ Pharmaceutical Data: Application of Ontology-Based Text Miner

Medication errors in patients’ medical records can influence the healthcare quality and cause risks for them. It is, therefore, crucial to apply appropriate procedures to reduce these errors. This study sought to develop a software for detecting medication errors through qualitative analysis of patients’ medical records.

The software was developed using object-oriented analysis and Java. The text was first pre-analyzed using a framework known as Stanford Core NLP. In the next stage, the text was turned into a semi-structured passage to be connected to Dr Onontology using Apache Jena framework. The name and dosage of available drugs were then extracted in the physician order forms and the patient progress notes. The areas of mismatch were identified through comparing the data obtained from these two forms.

Software assessment was conducted in two stages. In the first stage, the capability of the software in proper recognition of medicine’s name was measured, as100 completed forms containing physician order forms with a total number of 1014 drugs were used for text mining and error detection. After running the analysis in the error detection software, 93% of the drugs were properly recognized. In the next stage, comparisons were made between the physician order forms and the patient progress notes to find possible mismatches. Out of 1000 recorded drugs in the analyzed forms, the software was able to properly detect mismatches in 91.8% of the cases. The medication data available in i2b2 were used for conducting the assessment.

Given that medical records are of paramount importance and their human analysis is a complicated and time-consuming process, deployment of a text miner with the capability of quality analysis could facilitate error detection efficiently and effectively.