Deep learning application to automatic classification of pharmacist interventions.

BackgroundPharmacist Interventions (PIs) are actions proposed by pharmacists during the prescription review process to address non-optimal drug use. PIs must be triggered by drug-related problems (DRP) but can also be recommendations for better prescribing and administration practices. PIs are produced daily text documents and messages forwarded to prescribers. Although they could be used retrospectively to build on safeguards for preventing DRP, the reuse of the PIs data is under-exploited. ObjectiveThe objective of this work is to train a deep learning algorithm able to automatically categorize PIs to value this large amount of data. Materials and MethodsThe study was conducted at the University Hospital of Strasbourg. PIs data was collected over the year 2017. Data from the first six months of 2017 was labelled by two pharmacists, who manually assigned one of the 29 possible classes from the French Society of Clinical Pharmacy classification. A deep neural network classifier was trained to learn to automatically predict the class of PIs from the processed text data. Results27,699 labelled PIs were used to train and evaluate a classifier. The accuracy of the prediction calculated on the validation dataset was 78.0%. We predicted classes for the PIs collected in the second half of 2017. Of the 4,460 predictions checked manually, 67 required corrections. These verified data was concatenated with the original dataset to create an extended dataset to re-train the neural network. The accuracy achieved was 81.0 %, showing that the prediction process can be further improved as the amount of data increases. ConclusionsPIs classification is beneficial for assessing and improving pharmaceutical care practice. Here we report a high-performance automatic classification of PIs based on deep learning that could find a place in highlighting the clinical relevance of the drug prescription review performed daily by hospital pharmacists.