Effectiveness of Electronic Guidelines (GERH®) to Improve the Clinical Use of Antibiotics in An Intensive Care Unit

The objective of the study was to evaluate the capacity of GERH®-derived local resistance maps (LRMs) to predict antibiotic susceptibility profiles and recommend the appropriate empirical treatment for ICU patients with nosocomial infection. Data gathered between 2007 and 2016 were retrospectively studied to compare susceptibility information from antibiograms of microorganisms isolated in blood cultures, lower respiratory tract samples, and urine samples from all ICU patients meeting clinical criteria for infection with the susceptibility mapped by LRMs for these bacterial species. Susceptibility described by LRMs was concordant with in vitro study results in 73.9% of cases. The LRM-predicted outcome agreed with the antibiogram result in >90% of cases infected with the bacteria for which GERH® offers data on susceptibility to daptomycin, vancomycin, teicoplanin, linezolid, and rifampicin. Full adherence to LRM recommendations would have improved the percentage adequacy of empirical prescriptions by 2.2% for lower respiratory tract infections (p = 0.018), 3.1% for bacteremia (p = 0.07), and 5.3% for urinary tract infections (p = 0.142). LRMs may moderately improve the adequacy of empirical antibiotic therapy, especially for lower respiratory tract infections. LRMs recommend appropriate prescriptions in approximately 50% of cases but are less useful in patients with bacteremia or urinary tract infection.


Introduction
Healthcare-associated infections are those developed by patients as result of care received in hospital (known as nosocomial infections) or any other healthcare setting [1]. Nosocomial infections are those that appear at least 72 h after hospital admission and were not previously present or incubating [2,3]. They indicate the care quality delivered by hospitals and are related to increased morbidity and mortality and a longer hospital stay, representing an important public health problem and increasing healthcare costs [4]. Nosocomial infections are estimated to affect 5-10% of patients admitted to hospital [5], although their prevalence varies among departments and hospitals, being threeto five-fold higher in intensive care units (ICUs) than in other hospital areas [6,7]. They are considered to be the sixth cause of death in Europe and the USA, but around one-third of them could be prevented by infection control programs and adequate hygiene measures [8]. The magnitude of the attributable mortality is controversial and depends on the type of infection, the severity of the patient (APACHE II scale), and the length of hospital stay, among many other factors [9,10]. However, the earliest possible prescription of an appropriate empirical antibiotic treatment of nosocomial infections is known to be a key factor to improve the survival of ICU patients [11].
In ICU patients, nosocomial infections are often produced by multi-resistant microorganisms [12], complicating prescribing decisions. The scant development of new active ingredients has prompted novel strategies to extend the usefulness of existing antibiotics against severe infections [13], including the local study of bacterial resistance phenotypes. This allows a more precise selection of antibiotics based on local knowledge of the microorganisms most frequently responsible for infection in each hospital area [14].
Since 2012, Spanish hospitals have implemented programs to optimize the use of antibiotics (Programas de Optimización del Uso de Antimicrobianos (PROA)). The main objectives are to improve the clinical outcomes of patients with infections; minimize antimicrobial-related adverse effects, including the development and spread of resistance; and promote more effective treatments with lower health costs. PROA recommendations include accessibility to microbiological data, knowledge of bacterial resistance percentages (especially in the hospital area), and the implementation of measures to assist prescribing decisions. Computerized clinical decision support systems have proven useful to meet these objectives [13].
PROA implementation in our hospital (Torrecardenas Hospital Complex, Almeria, Spain) led to the development in 2014 of a computerized clinical decision support program for the prescription of antimicrobials, named in Spanish the Guía Electrónica de Resistencias Hospitalarias (GERH ® ). Using this system, updated identification and susceptibility data from microbiological studies in all hospitalized patients can be rapidly communicated to physicians using a secure hospital intranet system. Two GERH ® -based applications were launched: local resistance maps (LRMs) and preliminary microbiological reports with therapeutic recommendations (PMRTRs). To date, these applications have only been available to ICU physicians.
The objective of this study was to evaluate the capacity of LRMs to predict antibiotic susceptibility and resistance profiles and improve the appropriateness of empirical treatments for ICU patients with nosocomial infections, including bacteremia and/or lower respiratory or urinary tract infections.

Methods
GERH ® is based on the Microsoft.NET Framework with Visual C# and SQL and Open Database Connectivity (ODBC) to the laboratory information system (Sistema de Información de Laboratorio (SIL)) of the hospital microbiology laboratory. It is installed on a central server and has been used by ICU physicians since 2014 to consult all susceptibility results stored in the SIL since 2006. The data are organized according to the hospital department, date/date interval, sample type, microorganism/s isolated, and antimicrobials tested, and graphs are created for the ready visualization and interpretation of these data [15].

Local Resistance Maps (LRMs)
LRMs ( Figure 1) graphically depict information on the frequency of isolated microorganisms ( Figure 1A), bacterial susceptibility ( Figure 1B-D), and antibiotic activity ( Figure 1E). The physicians select and access the graphs via touch screens connected to the hospital intranet. Data are automatically updated every 24 h to include new records from the central GERH ® server [15].
Antibiotics 2020, 9, x; doi: FOR PEER REVIEW www.mdpi.com/journal/antibiotics  The LRMs were derived from analyses of the outcomes obtained in all in vitro susceptibility assays for bacteria isolated in ICU patients with bacteremia, lower respiratory tract infection, or urinary tract infection within a defined time interval, commonly the 12-month period before the consultation. The graphs depict accumulated information for the selected time period on the antibiotic susceptibility profile of isolated bacteria, indicating the likelihood that the infection in question is caused by specific bacteria as well as the expected activity of antibiotics against them. This allows the ICU physician to make informed decisions about the treatment of cases based on the local bacterial epidemiology and on the predicted susceptibility profile of the bacteria isolated.
This instrument followed the recommendation of the Spanish Society of Infectious Diseases and Clinical Microbiology (Spanish abbreviation: SEIMC) for accumulated reports on antimicrobial susceptibility to include solely microorganisms obtained from human clinical samples with susceptibility results verified by clinical microbiologists. When the same microorganisms are isolated multiple times in the same patient, those with a change in their resistance phenotype to one or more antibiotics are considered [16].

Study Design
This retrospective study compared the concordance obtained each year from 2007 to 2016 between susceptibility data from the antibiograms of microorganisms isolated in blood cultures, respiratory samples (bronchial aspirate, bronchial brushing, sputum, bronchoalveolar lavage, and/or tracheal secretion), and/or urine samples from all ICU patients meeting clinical criteria for infection and the susceptibility data depicted by LRMs for the same bacterial species, based on accumulated data for the whole year. In the LRMs, bacteria were defined as susceptible to an antibiotic when at least 75% of clinical isolates of this bacterial species were susceptible according to in vitro tests [15]. Concordance was defined as agreement between the susceptibility evaluated by LRMs and the susceptibility obtained in the in vitro study in the microbiology laboratory, i.e. when bacteria were considered as susceptible or resistant by both methods, including "intermediate resistance" within the "resistant" category. There was no concordance when the bacteria were considered susceptible by one approach and resistant by the other. Duplicate bacteria with the same identification and antibiogram were excluded, whether obtained from the same sample or isolated in multiple samples from the same patient.
The adequacy of empirical antibiotic treatments was also retrospectively studied, considering them appropriate when active against the bacterial pathogen causing the infection [13]. Accordingly, we determined whether each antibiotic prescribed was active against the bacteria isolated in the different clinical samples from each patient, based on the antibiogram data. The percentage adequacy of empirical prescriptions if they had been based on LRM recommendations was calculated as the number of times that LRM results for the susceptibility of a bacterium to each antibiotic agreed with the result of the susceptibility study as a percentage of the total number of isolates tested. The adequacy of the actual empirical treatment prescribed by physicians was evaluated with reference to the activity of the antibiotic(s) against each bacterium isolated in the different clinical samples according to the corresponding antibiograms. Data on the antibiotics prescribed in ICU patients were retrospectively obtained from the Spanish national nosocomial infection surveillance program (Spanish abbreviation: ENVIN).
Although the instrument is available in the ICU, therapeutic decisions do not have to be based on the data it provides. For this reason, we did not consider or gather data on the compliance of prescribed treatments with LRM recommendations.

Study Variables
Data gathered from antibiogram results, LRMs, and the ENVIN platform were: type of infection, date of sample gathering, type of sample (blood, respiratory, or urine), bacteria identified in each sample, antibiogram of the microorganism(s) isolated, concordance between antibiogram and LRM data, empirical antibiotic treatment prescribed, and concordance with the antibiogram result.

Statistical analysis
In the statistical analysis, the chi-square test was used to compare the adequacy of the actual empirical treatment with the adequacy of the LRM-recommended treatment, considering p < 0.05 to be significant. Table 1 compares the susceptibility data for each bacterium and antibiotic according to in vitro studies with those provided by LRMs after analyzing the information accumulated during the previous year. During the study period (2007-2016), the results of 22,520 in vitro trials were compared to the LRM data, obtaining an average concordance of 73.9%. In other words, the susceptibility or resistance described by LRMs for each bacterium-antibiotic association agreed with the in vitro study results in 73.9% of cases. C, the percentage of concordance in the assessment of antibiotic susceptibility of each bacterium between the information provided by LRMs and that obtained in the in vitro susceptibility study; N, for each year, the number of times in which each antibiotic was tested against bacteria of this group (number of trials with bacterial susceptibility against this antibiotic and comparison with the information provided by LRMs).

Concordance Between LRMs and Susceptibility In Vitro
For enterobacteria, the concordance ranged from a mean of 95.9% for amikacin over the study period (range 89.7-100%) to mean of 63.6% (range 47.6-86.7%) for ciprofloxacin/levofloxacin. In other words, when an enterobacterium was isolated, the expected susceptibility outcome was the same according to both the LRM and antibiogram in 95.9% of cases for amikacin and in 63.6% of cases for ciprofloxacin/levofloxacin. An intermediate degree of concordance was obtained for the other antibiotics studied.
For non-fermenting Gram-negative bacilli, the concordance widely varied among different antibiotics, obtaining the highest percentage agreement for colistin (88.3%; range 87.5-97.3%) and tobramycin (86.2%; range 58.2-100%) and lower degrees of concordance for amikacin For the remaining bacteria under study (S. pneumoniae, enterococci and Haemophilus spp.), there were few comparative data and the mean percentage agreement was generally high but showed a very wide range.
Considering all bacteria and antibiotics included in the 22,520 comparisons conducted during the study period, the highest percentage concordance between LRMs and antibiograms was observed for daptomycin (99.7%; range 99.0-100%), vancomycin (98.7%; range 95.5-100%), teicoplanin (92.5%; range 83.0-100%), linezolid (92.3%; range 79.3-100%), and rifampicin (90.9%; range 53.6-99.0%). In summary, the susceptibility data offered by LRMs for bacteria on which the GERH ® has this information agrees with the antibiogram result in >90% of cases. Table 2 compares the in vitro and LRM susceptibility data for each bacterium and antibiotic by year and by infection type. The mean percentage concordance was 73.5% in lower respiratory tract infections (range 66.5-80.0%), 69.3% in urinary tract infections (range 54.7-81.6%), and 76.1% in bacteremia (range 68.9-80.8%). These findings indicate that the susceptibility information provided by LRMs for these infections is in agreement with the actual susceptibility of the isolated bacteria in 73.5%, 69.3%, and 76.1% of cases, respectively.   Table 3 displays the percentage adequacy of LRM-recommended empirical prescriptions for bacteria in relation to the actual susceptibility observed for them. Antibiotics with a percentage adequacy >80% in the empirical antibiotic prescription were amikacin, colistin, daptomycin, linezolid, teicoplanin, and vancomycin. Thus, in relation to amikacin, out of 981 isolates of enterobacteria or non-fermenting Gram-negative bacilli isolated in samples, 904 were susceptible to this antibiotic and 77 were resistant, while its use was recommended by LRMs in 854 of cases, giving a percentage adequacy of 87.1%. Accordingly, if amikacin had been used as empirical treatment when recommended by LRMs, this treatment would have been appropriate in 87.1% of cases in which enterobacteria or non-fermenting Gram-negative bacilli were isolated. For daptomycin, linezolid, teicoplanin, and vancomycin the LRM-recommended treatment would have been appropriate in 99.7%, 92.0%, 92.2%, and 98.7% of cases in which a Gram-positive coccus (staphylococcus, enterococcus, or pneumococcus) was isolated.

Discussion
A major factor in the emergence of bacterial resistances is the inappropriate prescription of antibiotics [17], estimated to represent 30-50% of all antibiotic prescriptions [18]. For this reason, analysis of the antibiotic susceptibility of microorganisms is not only of major epidemiological and clinical importance but provides invaluable support for prescribing decisions. The use of computerized systems based on laboratory susceptibility results assists physicians in the selection of treatments without replacing their own clinical judgement. Various studies have demonstrated that these systems can improve healthcare, reduce inappropriate prescriptions and pharmaceutical costs, monitor antibiotic resistances, and diminish the morbidity and mortality of patients [15,[19][20][21].
GERH ® is integrated within the routine clinical workflow of our ICU, offering a predictive model that provides timely recommendations [13] and is designed to increase the percentage of patients who receive appropriate empirical antibiotic therapy, as recommended in previous studies [22,23]. According to the present findings, LRM-recommended prescriptions would have been appropriate in terms of the susceptibility of isolated bacteria in 57.6% of lower respiratory tract infection cases, 41.4% of bacteremia cases, and 54.9% of urinary tract infection cases. Higher percentages were published for these infections in the ENVIN study (2018), ranging between 63% and 72% [24]. Nevertheless, the use of LRMs in our ICU would have significantly improved the adequacy of empirical treatment prescriptions in lower respiratory tract infections by 2.2% (p = 0.018), although no significant improvement would have been achieved in the cases of bacteremia (3.1%; p = 0.070) or urinary tract infection (5.3%; p = 0.142). These improvements are modest but similar to previous reports [25][26][27], contributing to evidence that these systems can assist clinical decision-making and improve the adequacy of empirical antibiotic treatments, as previously affirmed [28].
Therapeutic recommendations are provided by LRMs before the responsible microorganism has been defined, and their percentage adequacy is less than when recommendations are made after identifying the etiological agent but before testing its susceptibility [29]. This is the case with PMRTRs, another GERH ® instrument, whose prescription recommendations were reported to be appropriate in >82% of cases and to achieve an improvement of 40% in the adequacy of prescriptions for each clinical situation, as we noted in a previous publication [15].
The main study limitation was that it did not consider whether or not physicians had consulted LRMs (available since 2014) before prescribing antibiotics, preventing assessment of the impact of LRM consultations over time on the adequacy of empirical antibiotic therapies. LRMs were designed to inform clinicians about the local epidemiology related to nosocomial infections, allowing them to base empirical antibiotic prescriptions on the likelihood of infection with a specific microorganism and on the accumulated activity of different antibiotics against bacteria isolated in a given focus. The aim was not to replace the judgment of clinicians, which may or may not coincide with LRM recommendations. For this reason, clinicians were not asked to state whether or not their prescription followed these recommendations, thereby preserving their prescribing autonomy. As a novel instrument, an adaptation period can be expected before it is accepted and implemented by physicians, who are also influenced by the perception of resulting improvements in antibiotic prescribing and outcomes.
As currently designed, LRMs do not yield information in relation to other PROA objectives such as the improvement in clinical outcomes and the reduction in antibiotic resistance rates, adverse effects, pharmacological interactions, antibiotic consumption, or pharmaceutical costs. These instruments could be improved by the incorporation of new functionalities that monitor and respond to these objectives and tailor empirical therapy recommendations to the clinical situation of each patient. For instance, prescription decisions could be further supported by integrating data for each patient on clinical observations, laboratory results (biochemistry and microbiology), radiology findings, and/or the concentrations of antibiotics in each tissue sample [30], along with the antibiotic susceptibility data.

Conclusions
Although GERH ® -derived LRMs proved to have a high capacity to predict antibiotic susceptibility and resistance profiles, they produce only a moderate improvement in the adequacy of empirical antibiotic therapy, which is significantly greater in cases of lower respiratory tract infections. According to these findings, LRMs are useful to recommend appropriate prescriptions in approximately 50% of cases but less so in patients with bacteremia or urinary tract infections.