From Bench-Top to Bedside: A Prospective In Vitro Antibiotic Combination Testing (iACT) Service to Guide the Selection of Rationally Optimized Antimicrobial Combinations against Extensively Drug Resistant (XDR) Gram Negative Bacteria (GNB)

Introduction Combination therapy is increasingly utilized against extensively-drug resistant (XDR) Gram negative bacteria (GNB). However, choosing a combination can be problematic as effective combinations are often strain-specific. An in vitro antibiotic combination testing (iACT) service, aimed to guide the selection of individualized and rationally optimized combination regimens within 48 hours, was developed. We described the role and feasibility of the iACT service in guiding individualized antibiotic combination selection in patients with XDR-GNB infections. Methods A retrospective case review was performed in two Singapore hospitals from April 2009–June 2014. All patients with XDR-GNB and antibiotic regimen guided by iACT for clinical management were included. The feasibility and role of the prospective iACT service was evaluated. The following patient outcomes were described: (i) 30-day in-hospital all-cause and infection-related mortality, (ii) clinical response, and (iii) microbiological eradication in patients with bloodstream infections. Results From 2009–2014, the iACT service was requested by Infectious Disease physicians for 39 cases (20 P. aeruginosa, 13 A. baumannii and 6 K. pneumoniae). Bloodstream infection was the predominant infection (36%), followed by pneumonia (31%). All iACT recommendations were provided within 48h from request for the service. Prior to iACT-guided therapy, most cases were prescribed combination antibiotics empirically (90%). Changes in the empiric antibiotic regimens were recommended in 21 (54%) cases; in 14 (36%) cases, changes were recommended as the empiric regimens were found to be non-bactericidal in vitro. In 7 (18%) cases, the number of antibiotics used in combination empirically was reduced by the iACT service. Overall, low 30-day infection-related mortality (15%) and high clinical response (82%) were observed. Microbiological eradication was observed in 79% of all bloodstream infections. Conclusions The iACT service can be feasibly employed to guide the timely selection of rationally optimized combination regimens, and played a role in reducing indiscreet antibiotic use.


Introduction
The emergence of extensively-drug resistant (XDR) Gram negative bacteria (GNB) is increasingly recognized as one of the greatest threat to healthcare [1]. Infections caused by these organisms have reported mortality rates of over 50%, and have been associated with increased healthcare costs [2,3]. This is worsened by the fact that the antibiotic pipeline against GNB has dried out. Consequently, physicians are compelled to resort to 'old' antibiotics such as the polymyxins for the treatment of these infections [1,4].
In the recent years, there have been an increasing number of physicians employing combination antibiotic therapy against XDR-GNB, in particularly against Pseudomonas aeruginosa, Acinetobacter baumannii and Klebsiella pneumoniae [5]. The contentions supporting combination antibiotics are aplenty. When two or more antibiotics are combined, there may be potential additive or even synergistic activity. In addition, combination antibiotics may suppress the emergence of resistant subpopulations, which can otherwise be selectively amplified if only one antibiotic was used [6]. On the flipside, irrational use of antibiotic combinations can worsen the already alarming scenario of antimicrobial resistance [6]. This has led experts to conclude that if combination antibiotics are employed, the most appropriate combination scheme should leverage on the latest in vitro and/or animal infection models, and should be individualized depending on organism, susceptibility profile, infection site and the clinical characteristics of each patient [5,7].
In view of the current antimicrobial landscape, a prospective in vitro antibiotic combination testing (iACT) service, with a rapid turn-around time of 48 hours, was developed. The objective of the iACT service is to guide physicians in the selection of an individualized and rationally optimized combination therapy, taking into account the in vitro combination testing results of each strain and the patient's clinical and pharmacokinetic (PK) parameters. In this study, we described the role and feasibility of the iACT service in guiding the selection of individualized and rationally optimized combinations in patients with XDR-GNB infections. In addition, the clinical and microbiological outcomes of patients receiving iACT-guided therapy were described.

Study Design
A retrospective case review was conducted in two Singapore hospitals: Singapore General Hospital, a 1700-bed public hospital and Mount Elizabeth Hospital, a 272-bed private hospital. All adults admitted from 1 st April 2009 to 30 th June 2014 with microbiologic evidence of XDR-GNB or pan-drug resistant (PDR) GNB isolation causing clinical infections and treated with antibiotic regimen guided by iACT were identified from the service's electronic database and retrospectively reviewed. The use of iACT to guide antibiotic combination selection in these patients was elected by their attending ID physicians as part of their clinical management strategy, out of desperate measures. Patients with infections due to an XDR-/PDR-GNB requiring iACT-guided therapy that were classified as cured, and then developed a subsequent infection after a 90-day period due to a different GNB requiring iACT-guided therapy were analyzed as two different cases. This study was approved by the ethics review board of Singapore General Hospital (2012/818/D) and Mount Elizabeth Hospital (PIEC/2012/038) prior to study initiation. The ethics review board of both institutions waived the need for informed consent as the study involved only a retrospective clinical notes review, and all data collected was analyzed anonymously.

Description of the Prospective iACT Service Workflow
Before the implementation of the iACT service, the service was introduced to the Infectious Diseases (ID) physicians in the hospitals, to increase awareness of and provide education about the service. Prior to request of the iACT service for a patient with an infection caused by an XDR-or PDR-GNB isolate, the attending ID physician informed each patient (or immediate family member if the patient was incapacitated) of (i) the poor prognostic outcomes of such resistant infections which most or all antibiotics were not effective against, (ii) the nature of the in vitro testing method as a last-resort measure to guide treatment, (iii) alternative treatment strategies (e.g. use of polymyxin monotherapy or selecting combinations based on anecdotal experience), iv) the clinical aims of the iACT guided treatment, and (v) the potential toxicities of the treatment. Verbal consent was sought and documented in clinical notes by the attending ID physician before iACT was requested.
The process of the iACT service is detailed in Fig 1. In vitro combination testing was carried out immediately upon request of iACT for an individual patient. Simultaneously, a clinical ID pharmacist would review the patient's clinical records and document all relevant clinical information required to develop rationally optimized combination dosage regimens, based on pharmacokinetics (PK) and pharmacodynamics (PD) principles. At 24h, combinations that were at least bacteriostatic (non-cloudy wells) in vitro were identified and fed back as preliminary results to the ID pharmacist. Preliminary recommendations to change antibiotic combinations were made by the ID pharmacist to the attending ID physician, if the initial empiric regimen was not at least bacteriostatic in vitro. At 48h, combinations with bactericidal activity in vitro were elucidated. Based on these in vitro results, the ID pharmacist would select and recommend the most optimal combination(s), with corresponding dosage regimens, to the attending ID physician. Selection was based on the patient's clinical characteristics, site of infection, PK/ PD parameter for efficacy of each antibiotic and the probabilities of PK/PD target attainment at infected sites. Clinical ID pharmacist followed-up and reviewed the patient's case notes within 24h post-recommendation, to ensure that the patient was placed on the recommended antibiotic combination regimens.
In vitro combination testing was carried out in clear 96-well microtiter plates, modified from a method by Aaron et al [13]. Briefly, 100uL of bacterial inoculum was added to a microtiter plate containing 11 antibiotics in single, two-drug and three-drug combinations (100uL of antibiotic solution per well) at clinically relevant unbound concentrations (Table 1) [14][15][16][17][18][19][20][21][22][23][24][25]. This gave approximately 10 5 CFU/mL (1 × 10 5 -5 × 10 5 CFU/mL) of bacteria in each well. The wells were covered and incubated at 35°C for 24h. At 24h, bacteriostatic combinations were identified based on presence/absence of visible growth in each well. The contents of each well were then sampled and measured to ensure <10% loss in volume. The measured contents were centrifuged at 10,000 × g for 15 minutes and the pellet reconstituted to original volume to minimize drug carry-over. Bacterial count was quantified by depositing serial ten-fold dilutions of the sample onto Mueller-Hinton agar plates (BD, Sparks, MD), incubated at 35°C for 24h, and enumerated visually to determine presence of bactericidal activity. The lower limit of detection was 3.9 × 10 2 CFU/ml.

Data Collection and Outcome Measures
All data were extracted from patients' inpatient charts and electronic medical records and documented in a structured form. Baseline characteristics collected included basic demographics, length of stay, presence of co-morbidities, APACHE II score at time of XDR-GNB isolation, history of intensive care unit stay or hospital stay within 3 months of XDR-/PDR-GNB isolation, and history of antibiotic use within 3 months of XDR-/PDR-GNB isolation. Details pertaining to the infection (causative organism, site of infection, date of isolation, in vitro susceptibility data and concurrent infections) and antibiotic use (type and duration of antibiotics used) was also collected. The following roles of the iACT service were documented: (i) confirmation of bactericidal activity in the empirically selected regimen (i.e. no change in the empirical regimen) (ii) change in empiric therapy, based on iACT results, or (iii) reduction of number of antibiotics empirically employed in combination, based on iACT results. If the iACT service recommended changes in an empiric therapy, the reasons for the recommendations were investigated. The time from culture isolation to request for iACT service, as well time from request for iACT service to time of provision of preliminary and final recommendations by the iACT service for each case was also documented.
The following patient outcomes were collected: (i) 30-day in-hospital all-cause and infection-related mortality, (ii) clinical response, and (iii) microbiological eradication (assessed only in patients with bloodstream infections as repeat cultures were not routinely taken for other sites). Adverse drugs events potentially attributable to iACT-guided therapy were documented.

Definitions
XDR and PDR were defined as non-susceptibility to at least one agent in all but two or fewer antibiotic categories and non-susceptibility to all agents in all antimicrobial categories respectively [26]. Bacteriostatic activity was defined as the lack of visible growth in the presence of antibiotics at 24h. Bactericidal activity was defined as 99.9% decrease in the colony count on subculture of an organism in the presence of antibiotics compared to initial inoculum [13]. An empirical regimen was defined as any antimicrobial regimen employed after isolation of XDR-/ PDR-GNB but prior to iACT-guided therapy, selected by the attending ID physician based on individual preference, anecdotal experience or recommendations from published literature. Infections were defined in accordance to definitions provided by the Centers for Disease Control and Prevention (CDC) [27]. Patients were considered immunocompromised if they had acquired immune deficiency syndrome, were neutropenic (absolute neutrophil count < 500/ mm 3 ) or transplant patients on immunosuppressive agents.
Thirty-day in-hospital all-cause and infection-related mortality was defined as all-cause and infection-related death, as determined by the ID physician, that occurred 30 days from isolation of XDR-/PDR-GNB. Clinical response was defined as partial or complete resolution of infection symptoms as determined by the ID physician, irrespective of GNB eradication, while therapeutic failure was defined as worsening at any time or no improvement of clinical conditions by day 14 of therapy as stated by the ID physician. Microbiologic eradication was only assessed for bloodstream infections, and was defined as the clearance of XDR-/PDR-GNB in all follow-up blood cultures. Nephrotoxicity was defined as per RIFLE criteria [28].

Data analysis
Categorical data were presented as numbers and percentages; continuous data were presented as mean ± standard deviation, or median (range), depending on the validity of normality assumption. All data were analyzed using IBM SPSS Statistics 21 (IBM Corp., New York).

Study Population
From April 2009 to June 2014, 40 patients received iACT-guided therapy. Of these, three patients were excluded as medical records were not available. Two patients developed two episodes of XDR-/PDR-GNB infection more than 90 days apart and were each analyzed as two different cases. Hence, a total of 39 cases were included.
The XDR-or PDR-GNB pathogens Table 3 detailed the resistance mechanisms and in vitro susceptibilities of the pathogens. Of the 39 isolates (20 P. aeruginosa, 13 A. baumannii and 6 K. pneumoniae), 31 (79.5%) were XDR and 8 (20.5%) were PDR (four P. aeruginosa, three A. baumannii and one K. pneumoniae). Of the 31 XDR isolates, 19 (61.3%), were sensitive to polymyxin B only. All A. baumannii harbored bla OXA23 and bla OXA51 carbapenemase genes, while 60% (12/20) of the P. aeruginosa isolates harbored genes encoding MBLs (3 bla VIM , 9 bla IMP ). A large variety of mechanisms mediating carbapenem resistance was observed for K. pneumoniae isolates. Amongst the six K. pneumoniae isolates, four (66.7%) K. pneumoniae isolates harbored both genes encoding NDM as well as OXA-181 carbapenemases. Of these, one isolate also demonstrated reduction in porin gene expression, while another had presence of efflux pump as determined by phenotypic methods.

Antibiotic Combinations and Role of the iACT Combination Testing Service
A delay between the time of culture isolation and request for iACT service (median: 4 days; range: 2-47 days) was observed. This delay was attributed to the fact that several ID physicians elected to use empirically selected combinations first upon culture isolation, and only requested for iACT service after the patients failed to respond to the empirically selected combinations. Upon request and isolate receipt, the iACT service provided all preliminary results (based on well turbidity) to the ID physicians within 24h. All final iACT recommendations were provided to the ID physicians within 48h.
Prior to iACT-guided therapy, most cases were empirically prescribed combination therapy (35/59, 89.7%) ( Table 4). More than half (21/39, 53.8%) of the cases were prescribed three or more antibiotics in combination empirically. Against P. aeruginosa, polymyxin B plus a carbapenem plus fluoroquinolones or aminoglycosides were the most common three-antibiotic regimen prescribed empirically, while against A. baumannii, polymyxin B plus a carbapenem plus tigecycline or rifampicin were the most common three-antibiotic regimen prescribed empirically. Empiric therapy was changed by the iACT service in 21 (53.9%) cases (Table 5)-the commonest reasons for change was that empiric antibiotic regimens were not bactericidal in vitro (14/21, 66.7%), or that a more appropriate combination regimen, based on the probabilities of PK/PD target attainment at infected sites, could be used (4/21, 19.0%). In 7 (17.9%) patients, the number of antibiotics used in combination empirically was reduced. The iACT service recommended dose changes in 8 (20.5%) patients; most recommendations involved increasing doses or prolonging infusion time to optimize probability of PK/PD target attainment.
Polymyxin B-based two-or three-drug combinations were the commonest iACT-guided therapy prescribed (33/39, 84.6%). Nebulized colistin was initiated in all patients with   pneumonia who were prescribed intravenous polymyxin B (16/39, 41.0%). The most common combination recommended against A. baumannii infections were polymyxin B plus a carbapenem (7/13, 53.8%). Against the P. aeruginosa isolates, polymyxin B plus carbapenem plus aminoglycosides were most commonly recommended (5/20, 25.0%). Antibiotic combination regimens recommended for K. pneumoniae infections were highly strain-specific; no single regimen was universally effective against all K. pneumoniae strains.

Clinical and Microbiological Outcomes
The outcomes according to infection type and responsible pathogen are detailed in Table 6. Six (15.4%) cases died due to infection; of these, five were critically ill at the time of initiation of iACT-guided therapy (APACHE II score at time of iACT-guided therapy ranged from 17-30 in these patients). For the remaining patient, the initial episode of ventilator-associated pneumonia resolved upon iACT-guided therapy; however, the patient developed a new episode of sepsis within 30 days and died. Clinical response (cure or improvement) was observed in 32 (82.1%) patients. Of note, clinical response was observed for all seven patients in whom the number of antibiotics used in combination was reduced. Out of the 14 patients with bloodstream infections, 11 (78.6%) had documented microbiological eradication. Thirteen (33.3%) patients developed adverse drugs events: 10 (25.6%) developed nephrotoxicity likely secondary to polymyxin B (n = 8) or polymyxin B plus amikacin (n = 2), two (5.1%) developed non-Clostridium difficile-associated-diarrhea, and one (2.6%) developed confusion secondary to levofloxacin. No death related to the adverse events was observed. Adverse events resolved in all patients who survived (7/10, 70.0%).

Discussion
This study explored the role and feasibility of a prospective in vitro combination testing service in our local setting. We found that the iACT service could provide individualized recommendations  Table 4: iACT = in vitro antibiotic combination testing doi:10.1371/journal.pone.0158740.t004 Table 5. Role of Prospective iACT Service in Guiding Antibiotic Combination Therapy. Abbreviations used in Table 4: iACT = in vitro antibiotic combination testing, PK/PD = pharmacokinetic and pharmacodynamic parameters doi:10.1371/journal.pone.0158740.t005  Prospective Combination Testing Service for XDR GNB for the most optimal combination within a rapid turn-around time of 48h and provided an attractive alternative to selecting antibiotic combinations merely based on individual physician's anecdotal experience. In addition, we observed promising rates of clinical response and infectionrelated mortality in patients prescribed iACT-guided combination therapy. The emergence of XDR-GNB worldwide and in Singapore has caused several paradigm shifts in antibiotic therapy [1]. In addition to the revival of old antibiotics, combination therapy has been increasingly accepted as common practice in the treatment of XDR-GNB infections [6]. Unfortunately, choosing an effective combination can be difficult due to several reasons. Traditional single-antibiotic susceptibility results have limited utility in the guiding the selection of combination therapy against XDR-/PDR-GNB. As a result, the vast majority of combinations can only be chosen empirically, based on individual's physician's preference [7]. However, selecting a combination based on anecdotal experience is often counter-productive in our local setting, as resistance is often mediated by more than one mechanism, resulting in strain-specific differences in combinations that are bactericidal, even within a single species. Hence, selecting combinations based merely on anecdotal experience can result in the prescription of an ineffective combination, potentially compromising patient outcomes. In addition, such irrational use of combination antibiotics may perpetuate the cycle of increasing resistance, as well as subject the patients to increased risk of adverse events [7].

Role of iACT
The prospective iACT service was developed to guide physicians in managing patients with XDR-/PDR-GNB infections. The service employed a bench-to-bedside approach to design rationally optimized and individualized antibiotic combination regimens based on PK and PD principles. An in vitro combination testing method, designed to systematically test 80 two-and three-drug combinations within 48h, was first employed to prospectively determine the bactericidal activity of antibiotic combinations in vitro against individual strains [13]. Leveraging on the in vitro results, clinical ID pharmacists then selected the most appropriate antibiotic combination, taking into consideration all relevant PK parameters, such as the patient's comorbidity status, patient's renal and hepatic function and distribution of the antibiotics to the site of infection. Recommendations made by the ID pharmacists do not merely consist of specific antibiotic combinations; instead, individualized dosage regimens (doses, route and infusion time), optimized to achieve PK/PD targets at the site of infection, were recommended to best maximize efficacy and decrease risk of resistance emergence.
In this study, we observed that most physicians empirically prescribed combination therapy prior to iACT-guided therapy, selected based on previously published literature or based on individual physicians' past experience [9,29]. Unfortunately, of these, less than half of these empirically prescribed combinations were found to be bactericidal in vitro, highlighting the inadequacy of selecting combinations based on anecdotal experience in our local setting. Considerable delay was noted between isolation of XDR-GNB and request for iACT service. This is mainly attributable to the fact that the iACT service was not mandatory; as such, several physicians opted to request for iACT service only after patients failed to respond to the initial empiric combination. This can result in the potential benefits conferred by the iACT-guided combination therapy being negated by the delay in time to appropriate therapy, especially in the critically-ill patients. Moving forward, continued education will be required to reinforce the importance of early referral to the iACT service, to ensure the timely prescription of the most appropriate antibiotic combinations.
Our iACT service employed an in vitro combination testing method to guide the selection of antibiotic combinations against each XDR-GNB strain. This can reveal any increased or decreased killing brought about by potential synergy (e.g. mechanistic or subpopulation synergy) or antagonism respectively when two or more antibiotics are combined [30]. In addition to guiding the selection of individualized and rationally optimized combinations in patients with XDR-GNB infections, we found that the iACT service potentially provided other functionalities. In 20% of the patients, the iACT service reduced the number of antibiotics initially used in combination. This suggests that the iACT service plays a role in antimicrobial stewardship, reducing unnecessary or indiscreet antibiotic use.
Overall, we observed an all-cause mortality rate of 21% in our study patients, which compared favorably to mortality rates reported in other studies. In a study by Crusio et al, all-cause in-hospital mortality was 47% in patients infected with carbapenem-resistant GNB and treated with polymyxin B-based combinations [31]. Similar rates of 30-day all-cause mortality (43%) were observed in study by Durante-Mangoni et al [32]. We acknowledged that the observed differences in observed mortality rates may be contributed by a multitude of factors, such as differences in severity of illness. However, it must be highlighted that the combinations employed for patients in these previously published studies were selected based on anecdotal experience or based on results from previously published studies, and not based on strain-specific in vitro combination testing results as per our study. More than 80% of patients receiving iACT guided combination therapy had documented clinical response in our study. Of note, clinical response was observed for all patients, in whom the number of combination antibiotics used empirically was reduced, suggesting that reducing the number of antibiotics in combination based on iACT did not compromise patients' outcomes.
To date, only one study explored the use of prospective combination testing to guide therapy in non-cystic fibrosis patients. In a case series by Nakamura et al, a break-point checkerboard plate technique was employed to guide combination selection against multi-drug resistant P. aeruginosa [33]. The study, however, selected combinations based solely on the results of in vitro testing, and did not take into account patient's PK/PD parameters. Furthermore, the study only reported patients with clinical improvement. In contrast, our study assessed all patients whose antibiotic regimen was prospectively guided by iACT service; in addition to in vitro testing, PK/PD principles were applied as the basis for rational selection of antibiotic combinations and dosing regimens.
Despite our promising findings, this study had several limitations. While we observed promising rates of clinical success and infection-related mortality in our study, we acknowledged that the study was not designed to examine outcomes, as no control group was employed for outcome comparison. Hence, no conclusion could be made with regards to the impact of the iACT service on patient outcomes. Moving forward, a randomized trial will be required to fully evaluate the utility of the service in improving patient outcomes. In addition, given that the iACT service was not mandatory, physicians may be inclined to request iACT service only for certain subset of patients with XDR-GNB infections (e.g. those who did not respond to the initial empiric therapy), which can result in selection bias.

Conclusion
We found that the prospective iACT service can be feasibly implemented to guide the timely selection of rationally optimized combination regimens against XDRGNB, and played an additional role in reducing indiscreet antibiotic use. While further randomized trials will be required to fully evaluate the utility of the service, it is hoped that the prospective iACT service can be continually employed to guide the rational selection of antibiotic combinations, and to reduce indiscreet use of antibiotic combinations.