Comparative performance of two drug interaction screening programmes analysing a cross-sectional prescription dataset of 84,625 psychiatric inpatients.

BACKGROUND: Clinical decision support software (CDSS) solutions can automatically identify drug interactions and thereby aim to improve drug safety. However, data on the comparative performance of different CDSS to detect and appropriately classify interactions in real-life prescription datasets is limited.
OBJECTIVE: The aim of this study was to compare the results from two different CDSS analysing the pharmacotherapy of a large population of psychiatric inpatients for drug interactions.
METHODS: We performed mass analyses of cross-sectional patient-level prescriptions from 84,625 psychiatric inpatients using two CDSS - MediQ and ID PHARMA CHECK(®). Interactions with the highest risk ratings and the most frequent ratings were reclassified according to the Zurich Interaction System (ZHIAS), a multidimensional classification that incorporates the OpeRational ClassificAtion of Drug Interactions (ORCA) and served as a reference standard.
RESULTS: MediQ reported 6,133 unique interacting combinations responsible for 270,617 alerts affecting 63,454 patients. ID PHARMA CHECK(®) issued 5,400 interactions and 157,489 alerts in 48,302 patients. Only 2,154 unique interactions were identified by both programmes, but overlap increased with higher risk rating. MediQ reported high-risk interactions in 2.5 % of all patients, compared with 5 % according to ID PHARMA CHECK(®). The positive predictive value for unique major alerts to be (provisionally) contraindicated according to ORCA was higher for MediQ (0.63) than for either of the two ID PHARMA CHECK(®) components (0.42 for hospINDEX and 0.30 for ID MACS). MediQ reported more interactions, and ID PHARMA CHECK(®) tended to classify interactions into a higher risk class, but overall both programmes identified a similar number of (provisionally) contraindicated interactions according to ORCA criteria. Both programmes identified arrhythmia as the most frequent specific risk associated with interactions in psychiatric patients.
CONCLUSIONS: CDSS can be used for mass-analysis of prescription data and thereby support quality management. However, in clinical practice CDSS impose an overwhelming alert burden on the prescriber, and prediction of clinical relevance remains a major challenge. Only a small subset of yet to be determined alerts appears suitable for automated display in clinical routine.