Background and Importance The digitisation of hospital drug prescriptions has enabled the collection of a huge amount of data. Developing pharmaceutical decision support systems is facilitated thanks to artificial intelligence and the collected data.
Aim and Objectives Build a predictive algorithm that can detect prescriptions requiring pharmaceutical intervention (PI).
Material and Methods The algorithm was developed using machine learning techniques. Data collected during four years were extracted from patients’ records from the prescription assistance software of a selected hospital. Various variables were used, including PIs generated by clinical pharmacists.
Results We used 1,961,176 drug prescriptions, including 312,591 PIs, to develop the matrix of the predictive algorithm in R. The model classifies each drug prescription according to the presence or the absence of a PI. The results after a random forest statistical model are encouraging, yet perfectible, especially the sensibility.
A new approach of model construction is undergoing including a pharmaco-ontology gathering the characteristics of the drugs based on the summaries of product characteristics. This will allow the model to learn the context of the prescription leading to a PI and detect PIs with new data in a similar context. Such pharmaco-ontology exists regrouping only drug-drug interactions.1
Conclusion and Relevance Pharmaceutical decision support systems usually predict PIs thanks to rules designed by pharmacists. Our model aims to detect these high-risk prescriptions thanks to machine learning and previous data validated by clinical pharmacists in their daily practice. The ontology will help associate a context to each PI previously detected and predict PIs on new data. Integrating this model into prescribing assistance software will make it easier for clinical pharmacists to detect PIs.
The predictive algorithm developed in our research project is not a substitute for pharmaceutical analysis of prescriptions. It is an expert system for the identification of risk situations that will be integrated into a team approach to clinical pharmacy practices.
References and/or Acknowledgements 1. Cossin S, Lebrun L, Lobre G, Loustau R, Jouhet V, Griffier R, Mougin F, Diallo G, Thiessard F. Romedi: An Open Data Source About French Drugs on the Semantic Web. Stud Health Technol Inform. 2019 Aug 21; 264:79–82. doi: 10.3233/SHTI190187. PMID: 31437889.
Conflict of Interest No conflict of interest
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.