Optimalisation of pharmacotherapy and patient safety by development of advanced clinical decision support systems in hospital and first line setting

In 1999, the US Institute of Medicine released a report titled “To Err is Human: building a Safer Health System”, which addressed the potential impact of medical errors on health care and patient morbidity and mortality. According to this report, in 1993 alone, about 7000 deaths in the US could be directly attributed to medication errors. As a result of these unsettling findings, during the last two decades, much effort has been put into reducing medication errors (ME’s) and associated (potential) Adverse Drug Events (pADE’s). However, more recent investigation implies that ME’s remain relatively common, with most of these errors occurring during the prescription process. Depending on the type of study, ME’s might occur in 0.3% to 39.1% of all prescriptions. Although rare, ME’s might give rise to ADE’s which in turn could lead to patient hospitalization. For example, in the Dutch HARM study, Leenderstse et al. found that up to 5.6% of unplanned admissions were due to ADE’s. Furthermore, almost half of these admissions (46.5%) were potentially preventable. Additionally, the occurrence of ADE’s in hospitalized patients is associated with an increased cost of care and negative patient outcomes such as increased length of stay and increased patient morbidity and mortality. A recent meta-analysis by Laatikainen et al. estimated that ADE’s might affect up to 19% of hospitalized patients, with about one third of them (32.3%) being preventable. Health information technology constitutes a major component of 21st century health care and has been widely proposed as a strategy to reduce and prevent ME’s and ADE’s. Electronic health records allowed for the implementation of computerized physician order entry (CPOE), through which physicians can enter clinical and treatment data in a structured way. This, in turn, facilitated the rapid adoption of clinical decision support systems (CDSS) which can provide guidance and decision-making advice to clinicians in order to support the appropriate prescribing of medication. Sometimes, CDSS are further categorised into “basic” or “advanced” systems. Advanced systems can be described as CDSS that incorporate context-or patient specific characteristics in its decision making algorithms (e.g. laboratory values such as electrolytes or renal function, patient disease, etc.). Several systematic reviews have shown that the implementation of CPOE and/or CDSS can lead to a reduction of up to 50% in ME’s and potential ADE’s. Medication errors occurring during the prescription-stage of the medication process might even be reduced by 70%. Evidence for effects on patient outcomes such as mortality or length of stay are harder to evaluate, mostly due to the significant heterogeneity of analysed studies. While the implementation of CPOE and CDSS can lead to major improvements regarding certain forms of medication errors, new types of problems can also arise. Concerning CDSS, issues regarding the generation of alerts with low specificity and positive predictive value (PPV) have been frequently reported. This problem can lead to “alert fatigue”, in which alerts, including those with a potential to cause severe ADE’s, are systematically ignored by clinicians due to an overload of clinically irrelevant notifications. Depending on the type of CDSS feature (e.g. drug-interaction checks, drug dosing assistance, drug allergy checker, etc.), clinician alert override rates up to 90-95% have been described in the literature. Furthermore, frequent updates and adjustments of CPOE and CDSS are required in order retain system operability. The development of “advanced” CDSS systems (cf. supra) has been proposed as a method of reducing alert fatigue and improving system specificity. Indeed, studies of CDSS incorporating patient-specific factors such as laboratory values have demonstrated a significant increase in the proportion of clinically relevant alerts and PPV. However, in most cases, even with these more advanced systems, PPV and alert specificity remained low, implying additional room for improvement. In the University Hospitals Leuven, a tertiary 1995-bed hospital in Belgium, a home-grown electronic patient file (with CPOE and CDSS) called “Clinical Workstation” (KWS) has been developed and is currently utilized by caretakers on a hospital-wide basis (excluding intensive care units, which utilize a different CPOE without coupled CDSS). Since 2011, KWS has also been implemented in other hospitals under a collaboration known as the NexUZhealth partners. The CDSS incorporated in KWS provides a number of features to assist clinicians in the medication prescription process, such as dosing guidance, drug duplication information, drug-drug, drug-pregnancy and drug-food interactions and drug-allergy or ADE reporting. Most of these CDSS features can be regarded as basic, incorporating little to no patient specific factors in their algorithms. Recently, an additional clinical pharmacy service called Check of Medication Appropriateness (CMA) has been implemented in the University Hospitals Leuven. This service encompasses the back-office screening for potentially inappropriate medications (PIM’s) on automatically validated queries (combining clinical, laboratory and medication-related data). A trained hospital pharmacist assesses the generated queries and, when deemed necessary, carries out an intervention in the form of an electronic note for the treating physician in the patient’s medical record or, in the case of severe PIM’s, through telephonic contact of the treating physician. One of the main findings during the evaluation of CMA as a clinical service was the limited number of interventions carried out for the overruled drug-drug interaction (DDI’s) alerts labeled as “very severe” by the KWS CDSS system. Indeed, only 939 (5%) of the 18.902 “very severe” DDI overrules that were registered during the 18-month study period led to an action by the hospital pharmacist. Additionally, preliminary data registrations concerning the overruling of interactions show that up to 85% of generated “very severe” interaction alerts are overruled by treating physicians. This low number of interventions combined with a high overrule rate indicates a low specificity of the generated alerts with a potential risk for alert fatigue, thus suggesting room for further improvement. Following these results, an operational mandate was approved with the aim of further optimising the available features of the CDSS integrated in KWS. A team of two hospital pharmacists (each 0.2-0.3 FTE) will fulfil this mandate for a starting period of three years in cooperation with several partners (Pharmacy & Therapeutics committee, involved specialists, IT-services, etc.). However, the optimisation of these features should be based on best available evidence and outcome changes following implementation of new or advanced features in the CDSS should be registered and analysed in order to further optimise possible future interventions (cf. Plan, Do, Check, Act cycle according to Deming). Therefore, this trial will mainly focus on the incorporation of evidence based data to contribute to the optimization of the available CDSS on a hospital-wide basis through multiple centres associated with the NexUZhealth collaboration and analysing the impact of the implementation of these new ‘standard of care’ interventions on specific outcomes (e.g. number of alerts generated). This with the aim of further improving patient safety and caregiver satisfaction.