Premium
Clinical decision support in a hospital electronic prescribing system informed by local data: experience at a tertiary New Zealand centre
Author(s) -
Chin Paul K. L.,
Chuah QianYi,
Crawford Amanda M.,
Clendon Olivia R.,
Drennan Philip G.,
Dalrymple Judith M.,
Barclay Murray L.,
Doogue Matthew P.
Publication year - 2020
Publication title -
internal medicine journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.596
H-Index - 70
eISSN - 1445-5994
pISSN - 1444-0903
DOI - 10.1111/imj.14706
Subject(s) - medicine , medical prescription , clinical decision support system , medical emergency , electronic prescribing , emergency medicine , false positive paradox , decision support system , data mining , nursing , machine learning , computer science
Background An electronic prescribing and administration (ePA) system has been progressively rolled out to Canterbury District Health Board (CDHB, Christchurch, New Zealand) public hospitals since 2014, and is currently used for around 1300 tertiary beds. ePA data can be used to monitor user behaviour, and to evaluate and inform the local customisation of clinical decision support (CDS) tools within the ePA system. Aims To describe retrospectively illustrative vignettes of CDHB ePA analyses that have been used for CDS. Methods Alerts were developed according to a set of common principles agreed upon by the CDHB CDS Working Group. Alerts were informed and evaluated by extracting and parsing data for various time periods during 2016 to 2018 from the CDHB ePA database. Results There was a median of 74 000 prescriptions a month. After examining 525 spironolactone prescriptions, the high dose alert threshold was set at 100 mg with an expected alert burden of 3%. The presence of a ceftriaxone shortage prescribing alert for 1 week was associated with a prescribing rate that was lower than 95% of the preceding 52 weeks. Following review of 367 fentanyl patch alerts, revision of the alert led to false positives falling from 43% to 3% ( P < 0.0001). At the point of firing, 6% of antithrombotic drug interactions alerts led to immediate changes in prescriptions (94% overridden), and a further 22% were changed within 30 min after the alert. Conclusions Local data extracts from ePA systems can inform iterative configuration of the software and monitor user behaviour.