Methodological variability in detecting prescribing errors and consequences for the evaluation of interventions

PURPOSE: To compare four methods of detecting prescribing errors (PE) in the same patient cohorts before and after an intervention (computerised physician order entry; CPOE) and to determine whether the impact of CPOE is identified consistently by all methods. METHODS: PEs were identified using (1) prospective detection by ward pharmacist; (2) retrospective health record review; (3) retrospective use of a trigger tool and (4) spontaneous reporting over two separate 4-week periods on one surgical ward in a UK teaching hospital. RESULTS: We reviewed 93 patients pre- and 114 post-CPOE. Using all four methods, we identified 135 PE (10.7% of all medication orders) pre-CPOE, and 127 (7.9%) post-CPOE. There was little overlap in PE detected by the different methods: prospective detection identified 48 (36% of all PE) pre- and 30 (24%) post-CPOE; retrospective review (RR) revealed 93 (69%) pre- and 105 (83%) post-CPOE, trigger tool 0 pre- and 2 (2%) post-CPOE and spontaneous reporting 1 (1%) pre- and 1 (1%) post-CPOE. The calculated relative reduction in risk of PE was 50% using prospective data, 12% with RR and 26% using data from all four methods. CONCLUSIONS: In this study, each method predominantly identified different PE. A combination of methods may be required to understand the true effectiveness of different interventions.