This study shows that it is feasible to reliably extract from the EMR the total duration of antibiotic therapy, broken down per indication and including post-discharge treatment, enabling an automated assessment of the appropriateness of antibiotic treatment duration. General inspection, and if indicated, optimization of the dataset (e.g., excluding erroneous prescriptions) is necessary to enable the use of data for further analysis. The total duration of therapy was accurately extracted from the EMR in 96% of infections, but the registered indication did not match the indication documented in the patient records in 17% of cases, which was mainly due to inaccurate selection of the mandatory indication in the EMR. Only 50% of these indication errors affected the evaluation of appropriateness of LOT, but local validation of the datasets is therefore necessary and the error-rate should be considered when data is used for quality measurement purposes. With regard to our secondary study aim, guideline non-adherence due to excessive treatment duration varied from 10% to 50% per indication, showing possibilities for quality optimization.
Including therapy after hospital discharge in the treatment duration is necessary, as a considerable amount of excess antibiotic use occurs after discharge, which was previously shown for CAP [2]. In our study, especially in ENT infection, complicated UTI and SSTI a high proportion (64%-83%) of antibiotic treatment courses was continued post-discharge, which made out 62-63% of total treatment duration. This emphasizes the need to include post-discharge prescriptions in the assessment of total treatment duration.
In previous studies the total duration of therapy linked to indication was either assessed by manual chart review [7, 8, 13] or by linking the antibiotic therapy to the ICD code [5, 6]. We used an electronic data extraction method, that can be used by all hospitals using EPIC software and that enables to assess a large amount of data more efficiently and more specifically. Determining the inclusion criteria and definitions, for example considering the last registered indication as the definitive treatment indication, and validation of the extracted data require time. However, as opposed to manual data assessment, the majority of time and expertise needs to be invested once, at the start of the project. Thereafter, surveillance can be performed automated as the syntax can be reused. A prerequisite is that the hospital EMR requires indication registration for each prescribed antibiotic. Although we did not show this in our study, the guideline-adherence of treatment duration and further specifications can be presented per hospital department, enabling the fine tuning of the targets for improvement.
Our inclusion criteria ensured a reliable dataset for the most common infections. We were able to confirm that by applying our inclusion criteria, the total duration of therapy was accurately extracted for almost all indications, with prescriptions in the OPAT setting being the most important exception. The error-rate of 4.5% of the electronically extracted LOT was far lower than the 11.5% discrepancy Dyer et al found [6]. As electronic prescriptions are becoming the norm, we expect error-rates due to written prescriptions to drop further in the future. Unfortunately, 17% of prescriptions were linked to an inaccurately selected indication, of which half affected the assessment of guideline-adherent LOT. In our previous study we found similar proportions of inaccurately selected indications in the hospitals where the mandatory indication registration was recently introduced [10]. Based on the Capability-Opportunity-Motivation Behaviour model, designed by Michie and colleagues, behaviour can only occur when an individual has the capability, opportunity and motivation (including habitual process) to perform the behaviour [14]. The physicians of the participating centres did not receive any information yet about why the mandatory indication registration was implemented in the prescribing software, which may have caused lack of motivation. Furthermore, data was extracted one year after implementation of the mandatory indication registration. In our previous study we already saw that habituation decreases the error rate [10]. We therefore believe that the error rate can be decreased when information, education and feedback are given.
Finally, we showed the opportunities of using mandatory indication registration and assessment of the appropriateness of treatment duration to identify clear targets for ASP. The treatment duration for mild-to-moderate severe CAP, for example, was shown to be too long in 40% of patients, considering the guideline recommendation of five days. Each day of antibiotic therapy is associated with 4% increased odds of experiencing an adverse event [4]. For example, three days of ß-lactam therapy for CAP patients was shown to be non-inferior to eight days of therapy [15], while seven days of therapy instead of three is associated with a 1.19-fold increase in experiencing an adverse drug event [4]. This emphasizes the necessity for ASP to monitor treatment duration.
Limitations
Due to our inclusion criteria we did not evaluate all antibiotic courses prescribed during hospital admission. We focused on patients that were initially empirically treated to enable reliable data extraction, as this method was already used and validated in our previous study [10]. In addition, we disregarded difficult-to-treat infections and nosocomial infections, as the local guidelines usually do not apply to these infections. Nevertheless, univocal local guideline recommendations were available for 51% (1745/3390) of the included treatment courses, in which guideline compliance could be assessed. We do believe that the records that were included gave a good representation of antibiotic use for the most common infections. Furthermore, we only evaluated antibiotic courses that were electronically prescribed. Written prescriptions, pre-admission prescriptions or antibiotics that were prescribed after transfer to other hospitals were therefore missed. When assessing the guideline adherence, this should be taken into account. The final point of consideration is the accuracy of the indication selection by prescribers. As human errors are inevitable and the error rate may fluctuate over time, the accuracy of the dataset should be checked regularly. This requires manual chart review, but that can be limited to a relatively small sample of patients, for example 10% of the extracted data.