Home
Medication errors are a big issue in the health system, causing inefficiency and patient harm. Regarding patient safety, a global campaign to prevent medication errors was launched by the World Health Organization to highlight the importance of this subject for health care quality. Along these lines, we propose an unsupervised method, called Density-Distance-Centrality (DDC), to detect potential outlier prescriptions.
To address this issue, our work considers past data from electronic records to detect outlier prescriptions in order to avoid wrong prescriptions. To achieve this, we propose an unsupervised algorithm based on graph models to detect prescription outliers. Our approach uses previous prescriptions to automatically learn the threshold between normal and abnormal doses for each medication in electronic medication orders, highlighting potential misuses.
A dataset with 563 thousand prescribed medications was used to assess our proposed approach against different state-of-the-art techniques for outlier detection. In the experiments, our approach achieves better results in the task of overdose and underdose detection in medical prescriptions compared to other methods adopted to deal with this problem. Additionally, most of the false positive instances detected by our algorithm were potential prescription errors.
There is a great advantage for the algorithm to learn the distribution of the institution’s prescriptions. These characteristics allow the use of this method anywhere in the world, with drug-specific standardization, self-adapting to the specifics of each hospital and being able to be automatically updated.
All the content of the work (algorithm, sample dataset, and experiments) is available at the project’s GitHub Page (https://github.com/nlp-pucrs/prescription-outliers) in order to be easily replicated. The sample dataset has no patient data; it only features a subset of the medication dataset, with information on dose, frequency, overdose, and underdose.