Introduction
Pyelonephritis is a bacterial infection of the renal pelvis and kidney. It is a life-threatening infection that can lead to renal scarring and impairment of kidney function [
1]. However, with adequate treatment, the infection can be cured without complications. The incidence of pyelonephritis varies depending on sex and age [
1]. Estimates of outpatient pyelonephritis rates in females are 12–13 cases per 10,000 population annually [
1]. The predominant aetiological agent of pyelonephritis is
Escherichia coli in up to 84% of cases [
1]. International guidelines (IDSA, ESCMID) recommend outpatient management of pyelonephritis with oral ciprofloxacin, levofloxacin, or oral trimethoprim-sulfamethoxazole [
2]. However, antibiotic resistance to these antibiotics within populations of
E. coli is increasing and complicates treatment for pyelonephritis [
2]. Unfortunately, despite a wide range of alternative oral antibiotics having activity against
E. coli, the use of orally available cephalexin, fosfomycin, mecillinam, nitrofurantoin, and trimethoprim at standard doses is excluded by EUCAST
E. coli breakpoints [
3]. There is therefore a limited range of recommended antibiotics for the oral treatment of pyelonephritis. If this range of antibiotics could be increased, it would be of benefit to patients, who could avoid the need for intravenous therapy, and associated costs and hospitalisations would be reduced. An initial step in considering if the range of antibiotics could be increased is an analysis of published antibiotic pharmacological data, including a pharmacokinetic analysis.
Understanding variability in antibiotic pharmacokinetics has relevant practical applications [
4]. Given a schema of administration of antibiotic therapy, patient characteristics which impact on antibiotic pharmacokinetics, and bacterial susceptibility to an antibiotic, it is possible to make predictions about an antibiotic’s clinical efficacy. Alternatively, using a pharmacodynamic clinical efficacy target, patient characteristics which impact on pharmacokinetics and bacterial antibiotic susceptibility, it is possible to individualise antibiotic dosing regimens. In addition, if geographically restricted antibiotic susceptibility is used, it is possible to make predictions about antibiotic efficacy in specific geographical locations. Population pharmacokinetic/pharmacodynamic (PK/PD) modelling and simulation are recommended as a supportive approach to individualise therapy as it allows patient characteristics to be included as covariates of the PK/PD parameters of developed models [
4].
Discussion
There is clinical evidence that suggests oral antibiotics can be effective for the treatment of patients with pyelonephritis. This evidence principally relates to quinolones in the adult population, but also includes oral beta-lactams [
2,
34]. In other populations, where intravenous antibiotics and quinolones are less suitable, i.e. paediatric populations, oral antibiotics including cefixime, ceftibuten, or amoxicillin-clavulanic acid are recommended in preference to intravenous antibiotics [
35]. In the adult population however, there is a limited range of oral antibiotics recommended for the treatment of pyelonephritis reflecting a lack of both clinical data and PK/PD predictions to support recommendations [
2]. On a background of increasing antibiotic resistance, there is a need to increase the oral antibiotic options for the treatment of pyelonephritis and PK/PD analyses can provide a basis for the identification of these alternatives.
The first, and an important finding of the study, was that there were limitations in the quality and relevance of PK models available for use in simulations. Most of the antibiotics under study have been in clinical use for many years and the PK studies do not fulfil current modelling standards. Also, some of the selected antibiotics, e.g. mecillinam, nitrofurantoin, norfloxacin, trimethoprim, are only currently recommended for use in patients with lower urinary tract infection and so, a focus may have been put on site of action data (urine) instead of blood exposure. The limitations in the PK data identified in this study are a finding of major importance. This study demonstrates the need for new pharmacokinetic studies of oral antibiotics in populations that are relevant, e.g. patients in the community with pyelonephritis, as opposed to healthy volunteers or intensive care unit patients.
The PK/PD simulation data suggest there is potential for the use of novel antibiotic dosing regimens for the oral treatment of pyelonephritis. Regarding amoxicillin, a dose of 2500 mg was predicted to be appropriate for the 43% of Leeds isolates reported as sensitive by EUCAST breakpoints. This dose is within BNF-recommended doses of amoxicillin, with a dose of 3000 mg being recommended for prophylaxis. Likewise, according to the results reported here, amoxicillin-clavulanic acid at 2250 mg of amoxicillin may have been an effective treatment for the 62% of Leeds isolates susceptible according to EUCAST. On the contrary, if the MIC
90 or the whole of our studied bacterial population from Leeds are considered, doses higher than 10 g are needed, highlighting the need for local antibiotic susceptibility data to inform treatment recommendations. Although, the predicted effective dose of cephalexin was 4000 mg for EUCAST-susceptible isolates (above BNF-recommended dosing levels), 22% of isolates had a cephalexin MIC of ≤ 4 mg/L, an MIC associated with a high PTA at the BNF-recommended dose of 1500 mg eight hourly. These data are therefore supportive of the recommendation within the IDSA guideline to assess the role of oral broad-spectrum cephalosporins for outpatient treatment of pyelonephritis [
2]. On the other hand, ciprofloxacin, an antibiotic recommended for the treatment of pyelonephritis, was predicted to be effective at standard doses; i.e., our PK/PD simulations predicted ciprofloxacin at standard doses achieved a CFR of > 0.9 at the MIC
90. But interestingly, it was predicted that ciprofloxacin could be prescribed at lower doses whilst maintaining a PTA > 0.9, with doses as low as 100 mg being effective for most patients (71% of population). This prediction has not given consideration to mutant selection thresholds, and the selection of antibiotic resistant mutants would need to be considered in clinical trials [
36]. Fosfomycin was also predicted to be clinically effective at near to standard doses (3500 mg). But this assessment is based on using a PDT of an AUC/MIC ratio of 43. This target was chosen based on published in vivo data, but in vitro data have suggested AUC/MIC targets of around 2000, which, if clinically relevant, would preclude the oral use of fosfomycin for pyelonephritis [
27]. It is also known that fosfomycin develops resistance on treatment, which may again limit the drugs usefulness, though these resistant populations may have paid a biological price and be less fit, and so less clinically relevant [
37].
The PTA and CFR values determined for the oral antibiotics inform the overall study aim of deciding about the potential of the investigated antibiotics to be further investigated for clinical use. The data suggests that multiple antibiotics should be further investigated. What will be needed are pharmacokinetic studies to be able to conduct simulations with more confidence. But given the conducted studies used PK data from healthy volunteers and intensive care unit patients, both groups known to have higher levels of renal excretion than other groups, it is possible that studies in relevant populations may be more favourable to assessments of antibiotic efficacies [
38]. We considered performing sensitivity analysis on the models to examine this further; however, as the results of this study are hypothesis generating, we decided against this approach. Nevertheless, further work based on updated PK data should perform sensitivity analyses to explore data applicability in order to make treatment recommendations. The PK/PD simulations completed were based on a set of bacteraemia isolates from patients with pyelonephritis in Leeds. This has the advantage of being a data set that was based on a defined population, but will have inherent biases. The data are geographically and temporarily restricted, and may represent a bias to more resistant isolates being based on bacteraemia isolates from hospitalized adults as opposed to urine cultures from community patients. Indeed, compared with community
E. coli isolates identified in urine samples from Leeds in 2010–2012, resistance rates used in this study had almost twice the rate of antibiotic resistance reported in community isolates [
39]. Again, re-analysis of predicted antibiotic efficacies completed using
E. coli isolates derived from community-based cases of pyelonephritis may be more favourable to assessments of antibiotic efficacies. Also, the simulations were conducted over the first 24 h of treatment. If simulations were extended beyond 24 h, they may also be more favourable to predictions of antibiotic efficacies.
The safety and tolerability of non-standard doses which are predicted to be clinically effective in this study simulation are not known, and so it must be clear that these non-standard doses are not treatment recommendations; i.e., they should be considered as a suggestion to reconsider standard doses until further research is performed. Oral antibiotics (ciprofloxacin and trimethoprim-sulfamethoxazole) recommended for the treatment of pyelonephritis have been associated with high rates of adverse events of 24–33% respectively [
40]. It is plausible that higher than standard doses of antibiotics may result in treatments with comparable side effect profiles and tolerability.
There are a number of limitations to the presented analyses that we highlight as follows. The analyses do not account for the duration of antibiotic treatment, which might have an impact on the total exposure to the drug and therefore on the response. Moreover, the use of data obtained from critically ill patients introduces a limitation in that the model population may be very different to those receiving the hypothetical oral doses in the community. For example, the variability of pharmacokinetic parameters is likely to be greater in critically ill patients due to factors such as creatinine clearance differences in these patients, presenting a limitation that may affect the final PTA. As we decided to fix the creatinine clearance in an attempt to represent a healthy patient population, it is likely that in doing so, the overall simulated variability of the model is less than the true value. This is in addition to the limitations in selecting absorption parameters from older studies, where formulations used may be different to those currently used, as we were unable to include variance for these parameters. PK/PD targets were obtained from EUCAST rationale documents when available and their reliability impacts on the PTA results. Ideally, the targets were clinically derived; however, EUCAST use multiple sources for targets. We acknowledge therefore that the targets will vary in the accuracy of their estimates. Whilst the data presented are restricted to one patient cohort and so may limit generalisability, they allow a detailed consideration of the approaches being considered, namely the potential for novel and/or individualised dosing regimens for the oral treatment of pyelonephritis based on the susceptibility of the patient isolate, and based on geographically restricted susceptibility data. In the future, geographically restricted (local) antibiotic guidelines could be developed by inputting a specific patient population’s antibiotic susceptibilities into antibiotic simulations utilising known pharmacokinetic data.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.