Improving hospital drug safety - identification of medication errors and subsequent development, implementation and outcome evaluation of alert algorithms for their targeted prevention

Any drug prescription requires careful weighting of risks vs. benefits. Failure to do so or to ignore known contraindications, recommended dose-adjustments and other precautions represents a medication error that may result in adverse drug events, i.e. harm to the patient. Clinical decision support systems can routinely detect potential medication errors and issue automated alerts for their prevention. However, current systems typically focus on high sensitivity at the price of low specificity regarding relevance of their alerts. In clinical practice this results in an excessive number of alerts to prescribers with subsequent alert fatigue and indiscriminate alert overriding, i.e. even important warnings are ignored. So medication errors continue to be a theoretically avoidable yet persistent burden for healthcare systems and patients. Therefore this thesis pursued the following objectives: I) Systematic quantification of potential medication errors in a real-life hospital setting. II) Validation of the clinical relevance of selected potential medication errors and associated adverse events. III) Development, implementation and outcome assessment not only of highly sensitive but also of highly specific alert algorithms for the prevention of clinically relevant medication errors. I) Two studies were performed in order to systematically quantify potential medication errors: A local pharmacoepidemiological database including 6.6 million drug administrations during approximately 82000 hospitalizations was successfully developed based on raw data extracted from the electronic medical records system of a tertiary care hospital. After its validation, highly efficient algorithms were developed that identified potential medication errors. They allowed the retrospective assessment of a considerable number of contraindicated and/or critical prescriptions. Sensitivity and specificity regarding clinical relevance of these potential medication errors was enhanced by the use of additional patient-specific laboratory data and repeated clinical validation procedures. With the help of a newly developed interface with ID PHARMA CHECK® - a commercially available clinical decision support system - several ten thousand potential drug interactions, contraindications and dosing errors were identified and assigned to formal severity categories. 48 distinct contraindicated drug interactions were considered as clinically relevant and suitable for display of highly specific alerts within a clinical information systems; 32 alert algorithms required retrieval and implementation of current patient-specific information such as laboratory results in order to reach high specificity. The resulting algorithms were subsequently programmed for routine use with the clinical decision support system. II) Three sub-studies were conducted within the pharmacoepidemiological database and addressed specific safety concerns of pharmacotherapy in clinical practice: The first of these studies identified 1136 hospitalizations with exposure to second-generation antipsychotics. Blood pressure, blood glucose, lipids and body mass index should be routinely monitored in those patients, however they were found to be documented in 97.7, 75.7, 24.6 and 77.4 % of hospitalizations, respectively. 63.4, 70.8 and 37.1% of the patients with hyperglycemia, dyslipidemia and hypertension, respectively, had no pharmacotherapy for these conditions. Among patients exposed to second-generation antipsychotics and concomitant use of drugs featuring a high risk for potentially severe adverse drug events, one case with associated neutropenia and four cases with abnormal QTc-interval were detected. Specific monitoring for such adverse drug events was not performed in 89.8% of patients with related high-risk drug combinations. The second sub-study analyzed the use of benzodiazepines (including “Z-drugs”) that were found to be administered to 48.3% of 53081 patients hospitalized in 2011 and 2012. Validated algorithms identified 3372 patient-days (2.9%) with comedication that significantly inhibits the respective benzodiazepines’ metabolism. Validation revealed 205 cases with clinically relevant medication errors. Among those, 23 cases with associated adverse drug events such as severe CNS-depression, falls with subsequent injuries and severe dyspnea were detected. The third sub-study analyzed 3444 hospitalizations with administrations of selected macrolide and quinolone antibiotics and identified concomitant use of additional QT-prolonging drugs in 1332 (38.7%). Among those we identified 7 events of related QTc-prolongation, but 50.4 % had no ECG-monitoring. Of all patients exposed to the studied antibiotics 547 (15.9%) featured episodes of hypokalemia, an important additional risk factor for potentially lethal Torsade de Pointes arrhythmia. Clinically relevant QT-prolongation was detected in 7 patients. Another 31 patients were exposed to contraindicated comedication with simvastatin, atorvastatin or tizanidine where the risk of pharmacokinetic drug-drug interactions clearly outweighed benefits, 3 thereof with associated adverse events. III) These two studies then aimed to develop new solutions for the prevention of such avoidable potential medication errors and associated adverse drug events: According to our pharmacoepidemiological database overdosing of paracetamol occurred in 988 hospitalizations per year, but in only 11 (0.4 %) this was judged as clinically relevant (≥ 5 g on ≥ 3 consecutive days). A new alert algorithm was developed as part of this study, and in 2014 it was implemented into the hospital-wide electronic drug prescribing system. It automatically detects cases of paracetamol overdosing, and after manual assessment alerts were issued in 23 cases, with subsequent changes to prescriptions in 21 (91.3 %) thereof. While the occurrence of mild and therefore clinically irrelevant acetaminophen overdosing changed only marginally in 2014 (n = 914), no clinically relevant overdosing occurred anymore. The second automated alert concerned metformin overdosing in renal impairment. It has been used in routine clinical practice for three years and generated 2145 automated alerts (about 2 per day). Validated expert recommendations regarding metformin therapy, i.e. dose reduction or stop, were issued for 381 patients (about 3 per week). Follow-up was available for 240 cases, and prescribers’ compliance with recommendations was 79 %. Furthermore, during 3 years we identified 8 local cases of lactic acidosis associated with metformin therapy in renal impairment that could not be prevented, e.g. because metformin overdosing had occurred before hospitalization. Besides these principal studies, spontaneous reports from international pharmacovigilance databases on liver disease associated with the new oral anticoagulant rivaroxaban and allergy-like reactions to herbal medicines were analyzed in two additional studies. In conclusion, local pharmacoepidemiological databases can be created using already available electronic medical information systems. They are an innovative and promising new approach for the identification and management of medication errors and resulting adverse drug events. Such databases are able to quantify clinically relevant medication errors and thereby provide data for a rational selection of targets for new and highly specific preventive safety measures. Any such measures must be integrated into a comprehensive hospital safety concept where local drug safety experts play an important role for the evaluation and communication of medication errors. Within such a system, automated alert algorithms that use also patient-specific information are a cornerstone for the proactive prevention of medication errors and resulting adverse drug events