395 General Discussion The predictive value of CDSS-generated STOPP/START signals may be increased by identifying patients at the highest risk for a ‘true’ inappropriate prescription. In Chapter 4.1, we investigated several potential determinants that might positively or negatively affect the acceptance of STOPP and START signals. Unfortunately, the investigated patient-related potential determinants were poor predictors of acceptance [23]. Thus, it would be interesting for future research to explore whether the PPV could be increased by selecting patients by, for instance, co-morbidity (e.g. excluding an alert for benzodiazepine use in patients with psychiatric illness) or by settings (e.g. primary care, in-hospital, or long-term care facility). 1.3. Improving applicability by increasing the availability of structured electronic patient data The use of software to reduce prescribing errors has already been suggested to add as an additional step to the original six-step WHOmodel for appropriate prescribing [36,37]. Most evidence for the benefits of software assistance in healthcare has been demonstrated for computerised physician/prescriber order entry (CPOE) combined with clinical decision support, which has been shown to reduce prescribing errors by about 50% [38]. Although electronic prescribing of drugs is a common practice, in most European countries, a more advanced approach (i.e. not exclusively based on a patient’s medication list) is necessary to alert for potentially inappropriate prescribing. The STOPP/START considers various patient data (e.g. clinical conditions, problems, diseases, laboratory values and measurements), requiring the structurally coded documentation of all relevant data. In OPERAM, we used a stand-alone CDSS to manually document a patient’s medical history, while clinical conditions are often registered in free-text in EHRs. Developments in software techniques (e.g. natural language processing) will likely facilitate converting unstructured clinical context into coded information shortly, as a source for clinical decision support [39]. Recent research has demonstrated that contextualised drug-drug interaction management had a greater clinical utility than basic drug-drug interaction management in hospitalised patients by suppressing irrelevant alerts based on clinical context [40]. Thus, further improvements in such techniques may allow for the automated linkage of drugs to clinical conditions, thereby facilitating the efficient software detection of potentially unindicated drug use; the first step of the pharmacotherapy analysis of the medication review process [41]. This feature would be critical in screening for potentially inappropriate prescribing, given that the most frequently generated signal and recommended action in OPERAMwas to discontinue a drug without a clear indication (STOPP A1), which is in fact an implicit criterion. 5
RkJQdWJsaXNoZXIy MTk4NDMw