Junior doctors’ prescribing work after-hours and the impact of computerized decision support

• Junior doctors spent most of their after-hours prescribing time transcribing other doctors’ orders.
• This is a new task brought about by partial implementation of CPOE.
• Unlike doctors on ward rounds, when working alone, junior doctors are receptive to decision support.
• Only if useful and well-designed decision support is implemented will it assist doctors prescribing after-hours.

Background and objectivecomputerized provider order entry (CPOE) systems with integrated decision support (DS) can reduce prescribing errors, but their impact may vary depending on the clinical setting. This study aimed to assess the impact of partial implementation of CPOE on junior doctors’ prescribing work after-hours and to examine differences in junior doctors’ use of DS during transcribing and their own prescribing tasks.MethodsTwelve junior doctors at a 350-bed teaching hospital in Sydney, Australia were shadowed between 16:30 and 22:30 over eight weeks for 65 h. CPOE was available on all wards except for the emergency department (ED). All medication tasks, computerized alerts, prescriber responses, and uses of reference material were recorded.ResultsOf 306 medication orders entered into the CPOE, 78.4% were transcribed from paper ED charts. A total of 113 alerts were triggered, most (78%) were read but only 6 (5%) resulted in prescribers changing an order. Reference material was accessed three times more frequently when junior doctors made their own prescribing decisions than when they transcribed other doctors’ orders, but a similar proportion of alerts was read during decision-making and transcribing tasks.ConclusionJunior doctors spent most of their after-hours prescribing time transcribing other doctors’ orders. This is a new task brought about by partial CPOE implementation. Junior doctors read computerized alerts and used online reference material to support their decision-making. However they rarely made changes to a medication order following alert generation, suggesting the alert information was often not clinically relevant.