Background
Inappropriate prescribing has been shown to be an independent risk factor for adverse drug events (ADE) and several studies reported that ADEs can cause hospitalization [
1]. Moreover, half of these hospital admissions are preventable [
2,
3]. Different strategies were already described to prevent inappropriate prescribing, i.e. 1) a clinical pharmacist attending physician rounds, 2) computerized physician order entry (CPOE) and 3) CPOE with clinical decision support system (CDSS) [
4].
In order to augment the safety and quality of patients’ therapy, front-office bedside clinical pharmacy services were set up in many European countries, after these were implemented in the UK, Canada and the USA since the ‘70s [
5]. Bedside, embedded clinical pharmacists are typically involved in medication reconciliation and review, medication counselling at discharge and targeted projects improving medication use [
6]. However, due to limited healthcare budget, which is the case in many European countries, bedside clinical pharmacy services are not implemented on a hospital-wide basis but restricted to high-risk patient populations such as geriatric patients characterized by polypharmacy, critically ill patients or patients admitted at the emergency department or on surgical wards with rapid patient turnover [
5,
6].
With the increase in information and implementation of CPOE, medical and treatment data are available in a structured way which led rapidly to embedding basic and advanced CDSS in the CPOE (CPOE/CDSS) in order to support appropriate prescribing [
7‐
9]. During the prescription process, basic CDSS analyse the data in the CPOE based on clinical rules and provide alerts automatically signalling clinical problems such as drug-drug interactions (DDI), basic dosing guidance, drug-allergy checking, duplicate therapy, etc. Nevertheless, basic CDSS is only providing support at the step of prescribing and not during follow-up treatment, without taking into account relevant biochemical parameters [
9]. More advanced CDSS also contain clinical rules combining several sources of information on the characteristics of individual patients such as laboratory values [
9‐
13].
Studies have shown a clear benefit of CPOE/CDSS with a significant decrease in prescription errors and ADEs [
7,
14‐
18]. Additionally, Rommers et al. showed the added value of advanced clinical rules in preventing ADEs when used in combination with basic CDSS [
10]. Conversely, the rise in health information technology has induced new pitfalls such as ‘alert fatigue’, i.e. the ignorance of both relevant and non-relevant alerts by health professionals because of moderate to low clinical relevance of some alerts, especially for basic clinical rules [
18‐
20]. Eppenga et al. demonstrated the improved (but still not optimal) clinical relevance of medication alerts when including more patient-related characteristics [
13]. Furthermore, sophisticated CDSS requires well-established digital communication, an effective integration of data and frequent updates [
19].
In the University Hospitals Leuven, Belgium, medication surveillance is supported by both bedside clinical pharmacists on high-risk wards (geriatrics, pediatrics, emergency department, trauma surgery, abdominal surgery and septic orthopedic surgery) and a basic CPOE/CDSS. Given the limitations of the basic CDSS (i.e. alert fatigue; no integration of biochemical parameters) and the limited implementation of bedside clinical pharmacy services along with the aim for medication surveillance and review on a hospital-wide basis, which is driven by the hospital accreditation standards, University Hospitals Leuven has developed a new back-office clinical service: the “Check of Medication Appropriateness” (CMA).
This service consists of advanced clinical rule alerts aiming exclusively at hospital pharmacists, followed by pharmacists’ performed actions aiming at physicians in case of medication inappropriateness. The objective of this manuscript is to describe the development of CMA and to evaluate the preliminary results of this new clinical service.
Discussion
During the study period, a back-office CMA service, embedding 0.5 FTE hospital pharmacists, yielded 24,943 electronic notes and 637 electronic notes supplemented by phone calls, concerning potentially very harmful ADEs or DRPs. When analysed without the automatic warnings for sequential therapy, 96% of the checks were performed in four specific pharmacotherapeutic categories: drug use in renal insufficiency, drugs with high potential of QTc interval prolongation, drugs with a restricted indication or dosing and overruled very severe DDIs. Acceptance rate of pharmacy notes and pharmacy notes supplemented by phone calls, carried out in the abovementioned categories, was 56% and 83%, respectively.
As expected, geriatric patients (> 75 years) and children (< 18 years) were only rarely included, since bedside clinical pharmacy services are already provided on a daily base at both wards. Patients admitted at emergency and cardiac surgery departments were most frequently involved. These results contribute to a risk based assessment in order to prioritize future investments in bedside clinical pharmacy services at these specific wards. Anticoagulants, mentioned in 6 of the 78 clinical rules, are a major drug class for which advices were formulated. These results suggest that current knowledge on the potential dangers of (the novel oral) anticoagulants, when used in patients with decreased renal function or when used in combination with other anticoagulants, is lacking. These results indicate that it is worthwhile to look for potential ADEs with these agents. Concerning other issues revealed by CMA, action was already undertaken e.g. for the use of meropenem high dose (2 g, q8h) a newsletter was sent to the pediatric hematology-oncology ward.
Our results indicate that the implementation of CMA is a significant addition to the standard services provided by the currently implemented basic CDSS, which is running on a hospital wide basis but only supporting at the step of prescribing without taking into account relevant laboratory values, and also to bedside clinical pharmacy services, which provides support at any time of the treatment but only runs for a limited and highly selected patient population. CMA, which combines automated screening of multiple data sources in the patient’s electronic medical record with the back-office evaluation of the patient’s therapy by a trained hospital pharmacist, has the ability to provide pharmacotherapeutic support on a hospital wide basis. With maximum integration of patient specific characteristics with details on drug treatment, a personalized advice can be given. The organization of our CMA service was well considered before implementation and is based on algorithms, which rely on evidence-based literature and practice-based experience, user-friendly flowcharts and uniform advices which need approval by the P&T committee before implementation. The obtained kappa value (0.79) also proves that there is a homogeneity between the pharmacists in checking the high risk prescriptions.
There are a lot of studies describing the implementation and evaluation of CPOE, basic and advanced CPOE/CDSS, but there are few studies [
10,
23‐
26] describing services like the CMA, consisting of alerts aiming exclusively at pharmacists, followed by pharmacists’ performed actions in case of medication inappropriateness. Our CMA system is very similar to the pharmacy adverse drug event alerting system (ADEAS) developed by Rommers et al. [
10] They formulated a comparable list of clinical rules (
n = 121) divided in risk categories. Like CMA, the generated alerts go to the pharmacist and not directly to the prescribing physician to prevent alert fatigue with the physician. They use ADEAS as a tool for the hospital pharmacist for more clinical ward-based activities, whilst CMA is developed as a back-office clinical service [
10]. In our opinion, a pharmacy decision support system like CMA and ADEAS is an essential added value to support medication surveillance and pharmacotherapy in European countries in which healthcare budget supporting bedside clinical pharmacy is scarce. Since clinical rules generally rely on literature-based evidence and practice-based evidence, we believe that the implementation of a similar service is possible in other European centers.
Our current adoption of the CMA service still demonstrates some limitations. First, when looking into the results without the automated suggestions for sequential therapy, the ratio of actions (
n = 3205) to the total number of clinical rule alerts (
n = 39,481) is low. Only 8% of the alerts were considered clinically relevant by the hospital pharmacists, which implies a large number of false positive alerts and unnecessary checks by the hospital pharmacist. This number of irrelevant alerts seems high, but a comparable percentage has also been observed in some previous studies which investigated one or more specific advanced medication alerts [
12,
13,
23‐
25]. In the study of Rommers et al., a similar percentage (7.8%) of the alerts resulted in advice to prevent possible ADEs [
23]. The reasons why clinical rule alerts were classified as not relevant by the hospital pharmacist were not systematically analysed in our study. However some reasons for non-relevance were frequently identified: 1) the dosage was already adjusted, 2) the drug was (temporarily) stopped, 3) the monitored laboratory value or clinical parameter was already reverted to within the reference limits, 4) repeated alerts which were already evaluated by the pharmacist and 5) double alerts with the same content. The same reasons were already mentioned by de Wit et al. and Rommers et al. [
12,
23] In the near future we plan to increase specificity of the CMA by further reformulating and fine-tuning the alert criteria of the clinical rules (e.g. by identifying and integrating more patient characteristics or parameters like weight, specific dose regimens, clinical symptoms etc.). Unfortunately, the inclusion of more alert criteria is limited because some patient characteristics are not structurally electronically documented in the medical record. So further automation is dependent on the digitalization of patients’ characteristics, which was already mentioned by Eppenga et al. [
13] To prevent repeated alerts, the alert could be suppressed until a relevant change of one of the alert criteria makes the alert reappear. Already reverted laboratory values could be a result of delay in data delivery, which can be resolved by integration of the CMA in HIS resulting in a direct link with the laboratory system. Furthermore, we also have to reconsider the basic CDSS for DDIs. The limited number of actions carried out for overruled DDIs indicates that many of these DDIs have to be overruled by the physician without a high risk for the patient. These unnecessary overrules might definitely lead to frustration and alert fatigue for the prescribing clinicians. For these interactions, it should be considered to omit the overrule by decreasing the severity score and/or by reassessing and reformulating the time interval of the defined interaction in CDSS.
Second, the acceptance rate of the automatically electronic notes suggesting a switch from an IV to oral formulation of a bio-equivalent drug is relatively low: only 33% of the automated notes led to a therapy switch. Therefore, the content of the alert criteria and the validity of the automatic screening algorithm for sequential therapy should be evaluated.
Additionally, for the four selected domains, a total acceptance rate of 69% was observed. This in in line with the ADEAS study, where 128 (63%) out of 204 actions led to a documented modification in therapy [
23].
However, for the alerts for which only an electronic notes was sent for the four selected domains, a moderate acceptance level of 56% was seen. This implies that almost half of the advices was not accepted or perhaps not read by the physician. Along with the reasons mentioned by clinicians to deny the advice, the clinical relevance of clinical rules for which an advice is usually not accepted needs to be assessed. In the near future, a satisfaction survey is planned for physicians to evaluate their general experiences with this service, their overall reasons for agreeing or disagreeing with the pharmacotherapeutic advice and their specific wishes or comments for future expansion.
In contrast, the majority of advices given by a phone call on top of the electronic note was accepted (83%). This was expected since a phone call is only performed for a medication order with a very high risk of ADE or DRP. It also indicates that an additional phone call may have a greater impact on the physician with a higher acceptance rate. Therefore it should be considered whether a phone call should be carried out faster.
Third, the acceptance level may also have been overestimated which is inherent to the retrospective evaluation of the acceptance rate: in some cases the physician could have modified the prescription independently of the pharmacy alert.
In conclusion, a number of benefits but also limitations were identified aiming for the optimization of CMA. Next to further improving the CMA (focusing on increasing the specificity and reconsidering the basic CDSS for DDIs), an expansion of the service is planned to cover a much wider range of drugs or drug classes and to cover more patient groups with the aim of avoiding even more preventable ADEs. First, clinical rules focusing on a specific pharmacotherapeutic domain will be developed, i.e. focusing on antimicrobial stewardship, anticoagulation therapy, post-operative pain management and total parenteral nutrition therapy. Second, as it has been described in literature that critically ill patients are more prone to suffer from ADEs, an ICU-focused implementation of the CMA service is also planned. Lastly, future research (i.e. an interrupted time series analysis) is foreseen to measure the additional effect of the advanced clinical rules on top of basic CPOE/CDSS and bedside clinical pharmacy in preventing PIMs in a controlled design. After this analysis, we need to reconsider if some alerts might be better situated as CDSS warnings to support physicians at the moment of prescribing in case immediately action is necessary.