Impact of the MIC of piperacillin/tazobactam on the outcome for patients with bacteraemia due to Enterobacteriaceae: the Bacteraemia-MIC project

Abstract

Introduction

Clinical breakpoints for the interpretation of antimicrobial susceptibility testing of microorganisms in the laboratory are set up by international expert committees to help guide decisions for antimicrobial therapy.¹ The EUCAST² uses several criteria for establishing breakpoints, including MIC distributions for large collections of microorganisms, phenotypic and genotypic resistance markers, the pharmacokinetic (PK) and pharmacodynamic (PD) properties of the antimicrobials and, most importantly, if available, the observed clinical outcome of patients treated with licensed doses of those drugs relative to MIC.^1,3 Unfortunately, clinical data supporting the establishment of clinical breakpoints are usually lacking, particularly for antimicrobials not recently approved and so in vitro data, animal experiments and PK-PD stochastic models are frequently the only data available for use.⁴

Piperacillin/tazobactam is a broad-spectrum β-lactam/β-lactamase inhibitor combination with activity against Gram-negative and Gram-positive bacteria. As a β-lactam, the best PK-PD predictor for efficacy is the length of time at which free drug concentrations remain above the MIC for the microorganism (ƒT_>MIC);⁵ the target is to obtain free drug concentrations above the MIC for at least 40%–50% of the dosing interval.^6,7 EUCAST recommends considering Enterobacteriaceae with MICs of ≤8 mg/L as susceptible, MICs of 16 mg/L are intermediate and those with >16 mg/L as resistant;² patients with isolates showing intermediate MIC can be treated with higher doses. A fixed concentration of 4 mg/L tazobactam is considered for susceptibility testing of the piperacillin/tazobactam combination.

The objective of this study was to evaluate the impact of piperacillin/tazobactam MIC on the clinical outcomes of patients with bacteraemia caused by Enterobacteriaceae treated with piperacillin/tazobactam when the isolate was fully or borderline susceptible.

Patients and methods

This study is part of the Bacteraemia-MIC project, aimed at investigating the impact of the MICs of different antimicrobials on the outcome of bacteraemia due to Enterobacteriaceae. A prospective observational multicentre cohort study was conducted between January 2011 and December 2013 at 13 Spanish university hospitals associated with the Spanish Network for Research in Infectious Diseases (REIPI) (www.reipi.org). Patients aged >17 years with monomicrobial bacteraemia due to Enterobacteriaceae and initially treated with β-lactam or a fluoroquinolone monotherapy were identified by daily communication with the microbiology laboratory, and were included in the Bacteraemia-MIC database if the first antibiotic dose was administered within the first 12 h after blood cultures were obtained and was administered for at least 48 h. Patients with any of the following criteria were excluded: transient bacteraemia (occurring after an invasive procedure such as a digestive tract endoscopy that cleared without treatment and had no complications); polymicrobial bacteraemia; combination antibiotic treatment with the drug of interest; non-hospitalized patients; patients with a do-not-resuscitate order; neutropenic patients (total neutrophil count <500/mm³); and survival less than 24 h after blood cultures were obtained. The study was approved by the Ethics Committee of the Hospital Universitario Virgen Macarena (reference number 1578), which waived the need to obtain consent because of the observational nature of study. In this analysis only patients initially treated with piperacillin/tazobactam in monotherapy were included.

Variables and definitions

Outcome variables were: clinical response at day 21, irrespective of whether piperacillin/tazobactam continued for the full course of treatment or was changed to another antibiotic; clinical response at the end of treatment with piperacillin/tazobactam; and all-cause mortality until day 30. Mortality at days 7 and 14 were also analysed. Clinical response at day 21 was classified as: ‘clinical cure’ if all symptoms and signs of infection had been completely resolved and all antibiotic therapy for the episode had ceased; ‘improvement’ if all symptoms and signs of infection had been completely resolved, but antibiotic therapy for the episode was still needed; and ‘failure’ if there were any persistent, recurrent or new symptoms or signs related to infection present, or death occurred. Clinical response at the end of therapy with piperacillin/tazobactam was classified as: ‘clinical cure’ if all symptoms and signs of infection had been completely resolved and no further antibiotics were needed; ‘improvement’ if all symptoms and signs of infection were completely or partially resolved, but antibiotic therapy for the episode continued with another drug because of de-escalation, switch to an oral drug or a more convenient parenteral drug for outpatient therapy; and ‘failure’ if a switch to or addition of another antibiotic was needed because infection-related symptoms or signs worsened or persisted, or death occurred. Clinical response was decided by a trained clinical investigator at each site; however, because of the subjectivity of this endpoint, an investigator (J. R.-B.) blinded with respect to exposure (piperacillin/tazobactam MIC) assessed the clinical response data of all cases according to vital signs, evolution of Pitt^8,9 and SOFA¹⁰ scores, and laboratory data. When the assessment of the blind investigator for the clinical response was different from that indicated by the local investigator, a query with the indication to review the case was sent to the local team, and an agreement was reached according to the available data.

All data were collected by local trained clinical investigators. Training was provided by a teleconference-assisted presentation based on the study protocol procedures; one investigator (J. R.-B.) was accessible for solving doubts during the study. Queries for missing and inconsistent data were sent when appropriate. The data included: demographics; height; weight; site of acquisition, classified as community, healthcare-associated or nosocomial according to Friedman's criteria;¹¹ chronic underlying diseases; severity of underlying condition according to the Charlson index¹² and McCabe score;¹³ severity of acute condition according to Pitt and SOFA scores; invasive procedures; source of bacteraemia according to clinical and microbiological criteria;¹⁴ severity of the systemic inflammatory response syndrome at presentation; basal creatinine level; aetiology; treatment; and outcome. Additionally, vital signs were recorded daily, if available, until day 7, and thereafter weekly, Pitt and SOFA scores were recorded at end of therapy with piperacillin/tazobactam and at days 14 and 21, and all key laboratory data were also recorded as available. Patients were followed for 1 month after the onset of bacteraemia to assess clinical outcome, including mortality.

Estimating exposure to piperacillin/tazobactam

The software used to predict serum piperacillin concentrations in individual patients was ID–ODS^TM (Individually Designed Optimum Dosing Strategies) (http://www.optimum-dosing-strategies.org/). ID–ODS^TM is a dose optimization tool powered by R^® software [version 2.15.3; Institute for Statistics and Mathematics (http://www.r-project.org/)] with an extensive model library built from population PK models published in peer-reviewed literature. ID–ODS^TM is based on patient demographics and laboratory information that is readily available at the bedside, and incorporates Monte Carlo simulation and Bayesian feedback into the design of personalized dosing regimens via a graphical user interface. Piperacillin concentration–time profiles were computed, based on the first-order two-compartment intravenous infusion model of Felton et al.,¹⁵ written in R^® language using published mean population PK parameter values for drug clearance, volume of distribution and transfer rate constants. Individualized patient concentrations were predicted using actual body weight and calculated creatinine clearance, which were covariates in the original model. Change in calculated PK parameters was allowed to ensure that changing physiological variables were incorporated during the time course of therapy. Protein binding of 30% was assumed for calculating the free piperacillin concentrations¹⁶ since the model used described only total piperacillin concentrations. The model-based predictions of free piperacillin concentrations at 0.6 min time intervals were then used to calculate the %fT_>MIC in the last recorded dosing interval and the concentration/MIC ratios, estimated at half the dosing interval (MID concentration/MIC) or at the end of the dosing interval (END concentration/MIC).

Microbiological studies

The antibiotic susceptibility of the first blood culture isolate obtained from each patient was determined at each study hospital by disc diffusion or automated broth microdilution methods. All isolates included in the study were frozen at −80°C and sent to the reference laboratory (Hospital Universitario Virgen Macarena) where identification was confirmed and the piperacillin/tazobactam MIC was determined for each isolate by manual broth microdilution, following the EUCAST methodology.¹⁷ Briefly, piperacillin (Sigma-Aldrich, Madrid, Spain) MICs in the presence of sodium salt tazobactam (Sigma-Aldrich) were determined by standard ISO broth microdilution (BBL-Mueller-Hinton-II cation-adjusted broth; Becton-Dickinson, Madrid, Spain). The panels were prepared freshly with concentrations of piperacillin from 0.06 to 128 mg/L, and a fixed concentration of 4 mg/L of tazobactam.

ESBL production was determined according to EUCAST recommendations by phenotypic method.¹⁸ PCR assays were used to characterize the β-lactamase produced. PCR conditions and previously described group-specific primers covering the most frequent ESBLs in our area (SHV,¹⁹ CTX-M-9,²⁰ CTX-M-1²¹ and TEM¹⁹) were used to amplify the bla genes, and the amplicons were sequenced.

Statistical analysis

A classification and regression tree was created (CART software 6.0; Salford Systems) to identify a potential piperacillin/tazobactam breakpoint for the studied outcomes, both in crude analysis and including confounders. Because a breakpoint could not be found, the isolates were classified according to MIC as low MIC (≤4 mg/L), borderline MIC (8–16 mg/L) and high MIC (≥32 mg/L). Percentages were compared using the χ² or Fisher's exact test, as appropriate. For univariate analyses, quantitative variables were appropriately dichotomized. Contingency tables were used to calculate relative risk (RR) with 95% CI. Multivariate analysis was performed by logistic regression. Variables with univariate P < 0.1 were introduced into the models and selected using the backward stepwise approach. The MIC of piperacillin/tazobactam was forced into the final models. Interactions of interest were tested. The prediction ability of the models was studied by calculating the area under the receiver operating characteristic (ROC) curve. All tests were performed using SPSS 18.0.

Results

During the study period, 1058 episodes of bacteraemia due to Enterobacteriaceae were included in the Bacteraemia-MIC cohort. Of these, 287 were initially treated with piperacillin/tazobactam. Demographic, epidemiological, clinical and microbiological data and the outcomes of the episodes are shown in Table 1. Escherichia coli and Klebsiella pneumoniae were the most frequent pathogens. ESBL-producing Enterobacteriaceae were identified in 13 (4.7%) episodes: 10 were E. coli, 2 K. pneumoniae and 1 Enterobacter cloacae (5.7%, 4.3% and 5.9% of the total specific species, respectively). The ESBL genes were characterized in all of them: bla_CTX-M-15 (six isolates); bla_CTX-M-14 (three); bla_CTX-M-1 (two); bla_CTX-M-9 (one); bla_CTX-M-27 (one); bla_CTX-M-28 (one); and one isolate produced two ESBLs. The most common source of bacteraemia was the biliary tract.

Outcomes according to MIC are shown in Table 2. Since CART analysis could not find an MIC breakpoint that predicted outcome, either in crude analysis or when confounders were added (see below), we compared the outcomes of patients with isolates showing low MICs (≤4 mg/L; n = 248, 86.4%) and borderline MICs (8–16 mg/L; n = 27, 9.4%). In 12 patients (4.2%), the isolate was resistant (>16 mg/L) and these were not included for further analysis. No significant differences between the low MIC group and the borderline MIC group were detected, except for type of acquisition. Nosocomial acquisition was more frequent among borderline MIC cases (Table 1). With respect to the piperacillin/tazobactam dose, there was no difference between the low MIC group and the borderline MIC group. Estimated exposures to piperacillin/tazobactam, measured as %fT_>MIC, were, as expected, higher among low MIC isolates (Tables 1 and 3).

The median duration of therapy with piperacillin/tazobactam was 3 days for patients in whom it was changed to another drug and 10 in those who were treated with this antibiotic for the whole course; 216 patients (78.5%) were changed from piperacillin/tazobactam to another antibiotic during the course of treatment (80.2% of patients with low MIC and 63.0% with borderline MIC, P = 0.04). The most common reasons for the switch were: de-escalation in 167 patients (64.1% of patients with low MIC and 29.6% with borderline MIC, P < 0.0001) or an unsatisfactory clinical evolution in 23 patients (8.9% of patients with low MIC and 3.7% with borderline MIC, P = 0.71). The most frequently used antibiotics after piperacillin/tazobactam were quinolones (29.2%), cephalosporins (26.9%), amoxicillin/clavulanic acid (23.6%) and carbapenems (17.1%), with no significant differences between the low MIC group and the borderline MIC group (data not shown).

A crude analysis of clinical outcome at day 21 that compares low MIC and borderline MIC cases according to patient characteristics and clinical and microbiological variables is shown in Table 4. Charlson index, Pitt score, severe sepsis or septic shock, need for mechanical ventilation and respiratory tract source were associated with an increased risk of failure at the end of antimicrobial treatment, while biliary tract source and change of treatment were protective.

A multivariate analysis was then performed (Table 5). The presence of a low or borderline MIC showed no influence on cure rates. Interactions between MIC and severe sepsis/shock, and between MIC and source of bacteraemia, were tested, but showed no significant modification effect. The area under the ROC curve for this model was 0.82. When multivariate analyses were performed, considering the MIC of piperacillin/tazobactam as a continuous variable, the model showed similar results; the OR (95% CI) for MIC was 0.97 (0.8–1.17), P = 0.72. Estimated exposure to piperacillin/tazobactam, calculated as %fT_>MIC, had no impact on the results, either when added to the model or used instead of MIC [OR (95% CI) was 1.008 (0.94–1.08), P = 0.82]. The inclusion of MID concentration/MIC and END concentration/MIC had no impact on the results (data not shown).

We also analysed the risk factors associated with failure at end of treatment with piperacillin/tazobactam (Table S1, available as Supplementary data at JAC Online). The following clinical variables showed a significant association with failure under univariate analysis: mechanical ventilation; Charlson index; Pitt score; nosocomial acquisition; severe sepsis or septic shock; respiratory or vascular catheter source, whereas a biliary source was protective. The variables associated with increased risk of failure in the final multivariate model (Table 5) were Pitt score and biliary tract source (protective). Again, borderline MIC showed no association; severe sepsis/shock or source of bacteraemia did not show significant modification effects. The area under the ROC curve for the model was 0.79. When the MIC of piperacillin/tazobactam was used as a continuous variable, the model showed similar results; the OR (95% CI) for MIC was 0.95 (0.81–1.11), P = 0.49. Again, %fT_>MIC had no impact [OR (95% CI) was 0.99 (0.96–1.04), P = 0.85]. MID concentration/MIC and END concentration/MIC had no impact (data not shown).

Finally, the univariate analyses of variables associated with 30 day mortality are shown in Table 6. The following variables were significantly associated with 30 day mortality, while treatment change and biliary tract source were shown to be protective factors (P < 0.01): age >65 years; mechanical ventilation; Charlson index; Pitt score; present severe sepsis or septic shock; and abdominal or respiratory source of bacteraemia.

The final multivariate model is shown in Table 5. According to this model, a biliary tract source showed an independent protective effect against mortality, whereas severe sepsis or septic shock, Pitt score and Charlson index were associated with increased mortality. Borderline MIC of piperacillin/tazobactam showed no association. Again, severe sepsis/shock or source of bacteraemia did not show significant modification effects. In this model, the area under the ROC curve was 0.85. Again, the MIC of piperacillin/tazobactam as a continuous variable caused no change; the MIC showed an RR of 1.05 (95% CI = 0.9–1.24), P = 0.5. %fT_>MIC did not improve the model [OR (95% CI) was 0.99 (0.94–1.04), P = 0.59] and MID concentration/MIC and END concentration/MIC showed no impact on this model (data not shown). We additionally performed analyses using 7 and 14 day mortality as outcomes; the adjusted OR (95% CI) for piperacillin/tazobactam borderline MIC was 1.98 (0.15–25.31) and 1.93 (0.32–11.48), respectively.

The outcomes according to MIC were analysed in the subgroup of patients with a source of bacteraemia different from biliary or urinary tract (71 patients); only seven of them had isolates with a borderline MIC. No trends towards higher failure or mortality were shown; in fact, both failure and mortality were 14.3% (one patient) among those with borderline MIC and 35.9% and 26.6% among those with a low MIC, respectively.

The results of the multivariate analysis and CART classification tree (Figure S1, available as Supplementary data at JAC Online) indicated that Charlson index, biliary source, age and severe sepsis or septic shock were the important prognostic factors for this kind of infection. The MIC had no influence. The sensitivity and specificity of the analysis were 69.9% and 89.7% respectively; the AUC (0.84) showed an excellent classification and diagnostic performance of the model.

Discussion

In this prospective cohort study and using carefully designed and executed assessments of outcome, we did not find that a borderline piperacillin/tazobactam MIC was associated with a worse outcome than a lower MIC among patients with bacteraemia due to Enterobacteriaceae treated with this antibiotic. The influence of other variables was as expected: severity of chronic underlying illness as measured by the Charlson index; severity of acute condition according to Pitt score; presentation with severe sepsis or septic shock were independent factors associated with a worse prognosis; while a biliary tract source had a protective effect. An overall interpretation of these results would be that the specific MIC of piperacillin/tazobactam, if susceptible, has no significant impact on outcome, which would therefore reinforce established breakpoints. Specifically, we think that our data are not enough to suggest that EUCAST susceptibility breakpoint should be changed to 16 mg/L. In fact, to interpret the data presented accurately, several important considerations should be mentioned.

Clinical studies are needed to help establish susceptibility breakpoints, but, because such studies are scarce,^22–25 decisions about breakpoints are mainly based on in vitro and PK-PD studies. Designing and developing an observational study with this objective (finding an MIC cut-off to predict clinical outcomes) presents a challenge, since it requires a large number of patients infected with organisms with different MICs treated with a specific antibiotic, carefully chosen inclusion criteria that avoid exposure bias, accurate outcome definitions and assessment, and appropriate control for confounders. One systematic review of previous studies performed in Enterobacteriaceae that provided information on MIC-related outcomes disclosed that they were not designed for that purpose and so lacked essential data such as dosing, determination of MIC using reference methods or control for confounders, among other limitations.²⁵ Consequently, meta-analyses of data from those studies provided limited results. We tried to overcome these limitations in our study.

None the less, some limitations must be taken into account, including low statistical power to detect differences because of a smaller number of patients with a borderline MIC and the fact that we mainly evaluated the impact of empirical piperacillin/tazobactam, because the therapy was changed in a substantial proportion of patients; however, in bacteraemic patients, empirical therapy is an important determinant of outcome.²⁶

In this study, we compared the clinical outcomes of patients with low (≤4 mg/L) and borderline susceptible (8–16 mg/L) MICs; we included isolates with MIC = 16 mg/L (considered intermediate according to EUCAST breakpoints) because some PK studies using Monte Carlo simulation suggested that satisfactory target attainment rates can be achieved for pathogens with MICs up to 16 mg/L when piperacillin/tazobactam is administered at 4.5 g every 6 or 8 h in continuous infusion (94% and 82%, respectively), but are much lower for higher MICs or other doses.¹⁵ In our study, most patients received 4.5 g every 8 h, using either a bolus or continuous infusion (infusion duration ≥3 h). Unfortunately, due to the low numbers, we could not make a comparison between these two methods of administering the drug. Whatever the infusion time, this dose seems to be appropriate according to our data, and is the usual dose in most European countries.⁷ We cannot provide data for 3.375 g every 8 h, which is frequently used in the USA and other countries, but we would expect lower probability for target attainment.

Interestingly, even though we could not measure piperacillin/tazobactam levels due to funding limitations, our estimates of exposure suggested that most patients with susceptible isolates reached at least 50% of the time above the MIC for the isolate with the dose used. Piperacillin/tazobactam concentrations have been shown to vary widely in critically ill patients,^27–29 although it should be noted that the majority of our patients were not critically ill (only 3.5% were under mechanical ventilation at the onset of bacteraemia), although a substantial proportion had severe sepsis or severe shock. Preliminary data of piperacillin/tazobactam concentrations measured in a very similar population also supported the notion that the PK-PD target was attained in a sample of similar patients with bacteraemia due to Enterobacteriaceae.³⁰ Our results would therefore mainly apply to patients who are not critically ill.

Another critical variable is the infection source. A previous small series of bacteraemic infections caused by ESBL-producing E. coli showed that the piperacillin/tazobactam MIC had no influence on urinary tract infections, but had a possible impact on other sources.³¹ The most common source in our cohort was the biliary tract, even more than the urinary tract, reflecting the fact that piperacillin/tazobactam is frequently used in empirical therapy for healthcare-associated biliary tract infections in Spain. This a relatively benign source when compared with nosocomial pneumonia or other complicated intra-abdominal infections;²⁶ in our analysis, it was a protective factor. Even though the source effect was controlled for in the multivariate analysis, we do think that more studies investigating the impact of piperacillin/tazobactam MIC within the susceptible range should be carried out for other types of infection.

One aspect of specific interest is the influence of piperacillin/tazobactam MICs in the treatment of patients with infection caused by ESBL-producing Enterobacteriaceae. Previous data suggested that piperacillin/tazobactam at the doses used in this study would be efficacious for the treatment of bacteraemia due to susceptible ESBL-producing E. coli, particularly if the sources are the urinary or biliary tracts,^31,32 but some doubts were raised for isolates with borderline MICs from other sources.³³ The ESBL producers with borderline MICs were not associated with worse outcomes in univariate or multivariate analyses in our study; however, only two of these cases had a source different than urinary or biliary tract, precluding any further analysis.

Other limitations of our study should also be considered when interpreting the results. First, the sample size was small, particularly in the borderline MIC group, which limited its power to detect the potential influence of MIC. None the less, this is, to the best of our knowledge, the largest study to date specifically to investigate this topic. Secondly, piperacillin/tazobactam was not always maintained throughout the full course of treatment of the patient. Since we had anticipated that this would be the case, we collected the reasons why therapy was changed, included clinical response at the end of piperacillin/tazobactam therapy as one of the outcome variables, and considered any change as failure, unless performed for de-escalation or dosage convenience. Finally, piperacillin/tazobactam exposure was estimated using a PK model because we could not measure plasma concentrations.

In conclusion, we did not find that higher piperacillin/tazobactam MIC within the susceptible or intermediate susceptibility range had a significant influence on the outcome of patients with bacteraemia due to Enterobacteriaceae. These results applied mainly to non-critically ill patients receiving 4.5 g every 8 h with biliary and urinary tract infections. More studies of other populations are needed.

Other investigators from the REIPI/GEIH-SEIMC BACTERAEMIA-MIC group

Marina de Cueto (Unidad Clínica Intercentros de Enfermedades Infecciosas, Microbiología y Medicina Preventiva, Hospitales Universitarios Virgen Macarena y Virgen del Rocío, Seville, Spain), Ana Maria Planes Reig (Departamento de Microbiología, Hospital Universitari Valld'Hebron, Barcelona, Spain), Fe Tubau Quintano (Servicio de Microbiología, Hospital Universitario de Bellvitge-IDIBELL, Barcelona, Spain), Carmen Peña (Servicio de Enfermedades Infecciosas, Hospital Universitario de Bellvitge-IDIBELL, Barcelona, Spain), M. Elvira Galán Otalora (Hospital de la Santa Creu i Sant Pau, Barcelona, Spain), Carlos Ruíz de Alegría (Servicio de Microbiología, Hospital Universitario Marqués de Valdecilla, Santander, Spain), M. Isabel Morosini (Servicio de Microbiología, Hospital Universitario Ramón y Cajal and Instituto Ramón y Cajal de Investigación Sanitaria (IRYCIS), Madrid, Spain), José Antonio Lepe (Unidad Clínica Intercentros de Enfermedades Infecciosas, Microbiología y Medicina Preventiva, Hospitales Virgen Macarena y Virgen del Rocío, Seville, Spain), José Miguel Cisneros (Unidad Clínica Intercentros de Enfermedades Infecciosas, Microbiología y Medicina Preventiva, Hospitales Universitarios Virgen Macarena y Virgen del Rocío, Seville, Spain), Shawna Morey (Vassar Brothers Medical Center, Poughkeepsie, NY, USA, and Optimum Dosing Strategies, Bloomingdale, NJ, USA) and Mohd-Hafiz Abdul-Aziz (University of Queensland, Brisbane, Australia).

Funding

The study was funded by the Instituto de Salud Carlos III, Ministry of Economy and Competitiveness, Spain (FIS; PI10/02021) co-financed by European Development Regional Fund ‘A way to achieve Europe’ ERDF, Spanish Network for Research in Infectious Diseases (REIPI RD12/0015). J. A. R. is funded, in part, by an Australian National Health and Medical Research Council Research Fellowship (APP1048652).

Transparency declarations

B. A. has been a scientific advisor for AstraZeneca, Merck, Pfizer, Novartis, Astellas and Gilead, and has been a speaker for AstraZeneca, Merck, Pfizer, Astellas, Gilead and Novartis. R. C. is currently Chairman of EUCAST and has participated in educational programmes organized by AstraZeneza and MSD. J. A. R. has been a scientific advisor for InfectoPharm and Merck. J. R.-B. has been a scientific advisor for AstraZeneca, Merck, Roche, Achaogen, InfectoPharm and Basilea, and has been a speaker for Astellas, AstraZeneca, Merck, Pfizer and Novartis. All other authors: none to declare.

Acknowledgements

References

Author notes