A multifaceted strategy to optimize pharmacokinetics of antimicrobial therapy in patients with hospital-acquired infections—a monocentre quality improvement project

Abstract

Objective

We assessed the efficacy of a quality improvement programme to optimize the delivery of antimicrobial therapy in critically ill patients with hospital-acquired infections (HAI).

Patients and methods

Before–after trial in a university hospital in France. Consecutive adults receiving systemic antimicrobial therapy for HAI were included. Patients received standard care during the pre-intervention period (June 2017 to November 2017). The quality improvement programme was implemented in December 2017. During the intervention period (January 2018 to June 2019), clinicians were trained to dose adjustment based on therapeutic drug monitoring and continuous infusion of β-lactam antibiotics. The primary endpoint was the mortality rate at day 90.

Results

A total of 198 patients were included (58 pre-intervention, 140 intervention). The compliance with the therapeutic drug monitoring-dose adaptation increased from 20.3% to 59.3% after the intervention (P < 0.0001). The 90-day mortality rate was 27.6% in the pre-intervention period and 17.3% in the intervention group (adjusted relative risk 0.53, 95%CI 0.27–1.07, P = 0.08). Treatment failures were observed in 22 (37.9%) patients before and 36 (25.7%) patients after the intervention (P = 0.07).

Conclusions

Recommendations for therapeutic drug monitoring-dose adaptation and continuous infusion of β-lactam antibiotics were not associated with a reduction in the 90-day mortality rate in patients with HAI.

Introduction

Despite developing international guidelines for the management of sepsis, treatment failure and death rates remain high in critically ill patients.^1,2 Optimizing antimicrobial therapy, such as early sepsis recognition and prevention of delayed effective treatment, is a major approach to reducing sepsis morbidity and mortality.^3–5 In addition to these approaches, there is a growing interest in the concept of personalized optimization of the pharmacokinetics of antimicrobial therapy in septic patients,^6–8 notably because the antibiotic concentrations reached in the blood and at the site of infection are highly variable among them.⁹ Several studies have notably demonstrated that administering β-lactam antibiotics as a continuous infusion reduced the risk of treatment failure.^10–12 A strategy of dose adaptation based on daily therapeutic drug monitoring was also shown to increase the attainment of pharmacokinetic and pharmacodynamic targets,¹³ but data on clinical outcomes remain scarce.^14,15

Among the cause of sepsis, hospital-acquired infections (HAI) are a significant health concern worldwide and a public health priority in Europe. In 2017, the ECDC estimated that there are more than 2.5 million cases of HAI per year in Europe, accounting for a significant proportion of disability (infected patients lose an average of 7.7 quality-adjusted life years) and premature death.¹⁶ Up to 50% of patients hospitalized in ICUs for 3 days or more develop HAI, most frequently pneumonia.¹⁷

We hypothesized that the early and individual optimization of the pharmacokinetics of antimicrobial therapy could improve the outcomes of critically ill patients with HAI. We have thus implemented a local quality improvement programme aiming to promote the continuous infusion of β-lactam antibiotics with dose adaptation of empiric antimicrobial therapy to blood drug monitoring. We thus designed a before–after study to test in a monocentre study the effects of a quality improvement programme for optimizing the pharmacokinetics of antimicrobial treatment on the mortality rate of patients with HAI.

Patients and methods

Ethics approval and consent to participate

The protocol was approved by the Société Française d’Anesthésie-Réanimation institutional review board (CERAR, IRB 00010254—2021-127). Consent was waived according to French law because the trial was an institutional quality improvement initiative applied to all patients.¹⁸ Written information about the use of personal medical information was nonetheless delivered to patients and their next of kin.

Setting and patients

The study was conducted in one ICU of the university hospital of Nantes, France. Patients over 18 years old were eligible in case of antimicrobial therapy for HAI. Patients were not suitable for inclusion in case of pregnancy or if a decision to withdraw life-sustaining therapies was already taken at the time of HAI diagnosis.

Study design

We conducted a before-and-after study to investigate the effects of the implementation of a quality improvement programme.

The control phase (before the intervention) consisted of all consecutive patients treated for HAI before the medical training began (June to November 2017). The quality improvement programme was introduced over one month (December 2017), during which no patient data were collected. The intervention phase consisted of all consecutive patients treated for HAI for 18 months (January 2018 to June 2019). Patients were followed up until day 90.

During the control phase, a local protocol for the choice of the spectrum and standardization of daily doses of empiric antimicrobial therapy was already available (Supplemental Tables S1 and S2 (available as Supplementary data at JAC Online))⁴. Still, the type of infusion (bolus or continuous) and the decision to monitor the antibiotic blood concentration were left at the discretion of the attending physician.

During the quality improvement programme, physicians, residents and nurses received formal education and clinical training on continuous infusion of β-lactam antibiotics and therapeutic drug monitoring for dose adaptation for β-lactam antibiotics, aminoglycosides and quinolones. Briefly, for β-lactam antibiotics, therapeutic drug monitoring was recommended between the 24th and 48th hour of empiric antimicrobial therapy and to adapt the daily dose to reach 100% of the time over 4-fold the minimal inhibitory concentration (MIC).¹⁹ For aminoglycosides, the dose was adapted to the blood concentration reached 30 minutes after a 30-min prolonged infusion, with a minimum target of 8-fold MIC.²⁰ Finally, for quinolones, repeated dosing made it possible to obtain a curve whose area under the curve (above the MIC threshold) allowed us to establish whether the target was reached.²¹ In case of under- or overdosing, indicative dose adjustments were available (Supplemental Table S2), but the final decision was taken by clinicians on the basis of the difference to target and organ failure evolution. It was recommended to repeat the therapeutic drug monitoring after dose adaptation.

During the intervention phase, the training was repeated every 6 months, and the clinicians had to adhere to the quality improvement programme. During these 18 months, antimicrobial therapy had to conform to the pharmacokinetic optimization bundle, and patients were followed up until day 90.

Outcomes

The primary endpoint was a 90-day mortality rate. The secondary endpoints included compliance with the protocol, the duration of invasive mechanical ventilation and ICU hospitalization and the rate of treatment failures.

Follow-up

For the infection documentation, the type of bacteria was collected (Gram-negative bacilli, Gram-positive cocci or a combination of both). After bacterial identification, when the MICs were unavailable, we considered the EUCAST breakpoints tables. Detailed information explaining the instructions for data collection and definitions for outcomes was developed before data collection. For quality assurance purposes, data were electronically checked for uniformity and completeness.

Definitions

Adherence to the quality improvement programme was considered when: (i) therapeutic drug monitoring-dose adaptation was considered when treatment blood concentration was measured at least once, and (ii) when the β-lactam antibiotic was initiated as a continuous infusion.

Underdosage was considered when the residual or steady state blood concentration was below 4xMIC for β-lactam antibiotics or 8× MIC for aminoglycosides. Overdose was considered when the concentration was above 20× MIC.

Treatment failure was defined as no apparent response or incomplete response, and persistence or progression of most or all the pre-therapy signs and symptoms. Relapse was defined as an HAI with one or more bacteria common with the first episode separated by an asymptomatic period, and recurrence when no common pathogen was found between the two episodes of HAI separated by an asymptomatic period.

For dominant bacteria in culture, antimicrobial resistance was defined according to the EUCAST breakpoints.

Statistical analysis

Numeric variables were expressed as mean (±SD) and discrete outcomes as absolute and relative (%) frequencies. Group comparability was assessed by comparing baseline demographic data and follow-up duration between groups. The comparison of the primary outcome between the two phases was performed with a multivariate Cox regression adjusted on age, SOFA score and cause of admission (medical versus trauma versus other). Data were checked for multicollinearity with the Belsley–Kuh–Welsch technique, and proportional hazards were studied according to Schoenfeld residuals.

Normality and heteroskedasticity of continuous data were evaluated with Shapiro–Wilk and Levene’s test, respectively. Continuous outcomes were compared with ANOVA, Welch ANOVA or Kruskal–Wallis tests according to data distribution. Discrete outcomes were compared with chi-square or Fisher’s exact test accordingly. The duration of mechanical ventilation and ICU hospitalization were compared between the two periods using a log-rank non-parametric test.

As exploratory post hoc analyses, the rates of the primary outcome were compared in the following subgroups: continuous infusion (yes versus no), drug therapeutic monitoring (yes versus no), continuous infusion and drug therapeutic monitoring (yes versus no), and HAP versus non-respiratory HAI.

The alpha risk was set to 5%, and two-tailed tests were used. Statistical analysis was performed with EasyMedStat (v.3.17; www.easymedstat.com).

Results

Population

One hundred and ninety-eight patients were included in our study, 58 during the control period and 140 during the intervention period. Demographic data are presented in Table 1. The median SOFA score at admission was 7 ± 4 before and 6 ± 4 after the intervention (P = 0.55). The most frequent cause of ICU admission was severe trauma (55%) during the pre-intervention period and medical (46%) in the intervention period (P = 0.003).

The characteristics of HAI are described in Table 2. The repartition of hospital-acquired infection sites did not appear to be an imbalance between the two periods (P = 0.50), with pneumonia being the most frequent cause (60% and 51%). Hospital-acquired infection onsets were 4 ± 3 days versus 4 ± 4 days, respectively (P = 0.39). The primary causative pathogens were Gram-negative bacilli (55% pre-intervention and 46% intervention, P = 0.61), and amoxicillin/clavulanate ac. was the most frequently prescribed antimicrobial therapy (48.3% versus 48.2%, P = 0.99).

Compliance with the multifaceted strategy

Compliance with the local protocol is shown in Table 3. The continuous infusion of β-lactam antibiotics was performed in 21.4% of patients during the intervention period (not recorded in the control phase). Therapeutic drug monitoring-dose adaptation was applied in 12 (20.3%) patients before the intervention and in 82 (58.6%) patients during the intervention (P < 0.0001). When therapeutic drug monitoring-dose adaptation was applied during the intervention period, only 48.2% of measurements reached the therapeutical targets, with 16.9% under dosage and 33.7% overdoses (see Supplemental Table S3 for molecule-specific rates). Dose adaptation was performed in three (5.2%) patients before versus 17 (12.2%) after the intervention (P = 0.13). Therapeutic drug monitoring-dose adaptation consisted in increasing daily dose in five (3.6%) patients, and in decreasing it in 12 (8.6%) patients (Table 3).

Primary outcome

Out of 58 patients, 16 (27.6%) were dead at day 90 in the control period, as compared with 25 (17.9%) patients in the intervention period (relative risk 0.71, 95%CI 0.46–1.16, P = 0.16, Figure 1). We then performed a multivariate analysis to consider the imbalance of baseline data that could have altered our estimation of the intervention effect (see Supplemental Table S4). After adjustment, the hazard ratio for mortality at day 90 in the intervention period was 0.53 (95%CI 0.27–1.07, P = 0.08).

Subgroup analyses

The hazard ratio for the primary outcome was 0.82 (95%CI 0.35–1.95) in the subgroup of patients with HAP, and 0.61 (95%CI 0.24–1.56) in patients with non-respiratory infections (Table 4). The intervention effects in subgroups based on compliance to guidelines are described in Supplemental Table S5. In an a posteriori defined subgroup, 90-day mortality occurred in four out of 34 (12.2%) patients treated with continuous infusion and drug therapeutic monitoring and 164 out of 66 (24.2%) patients with incomplete adherence to the quality improvement programme (hazard ratio 0.41 (95%CI 0.14–1.25), P = 0.12).

Secondary outcomes

Secondary outcomes are presented in Table 4. The median length of stay in the ICU was 32 ± 35 days in the pre-intervention period and 25 ± 22 days in the intervention period (P = 0.39). There was a non-statistically significant trend towards a lower risk of treatment failure (37.9% before the intervention versus 25.7% during the intervention, P = 0.07).

Discussion

This quality improvement programme aimed at optimizing the pharmacokinetics of antimicrobial therapy but did not significantly enhance the outcomes of critically ill patients hospitalized with HAI. However, the rates of compliance to the protocol increased after its implementation, and the dose and type of infusion of antimicrobial therapy were significantly improved after the intervention.

When therapeutic drug monitoring-dose adaptation was applied during the intervention period, nearly half of the dosages were out of the targeted ranges. Several physiological modifications observed in critically ill patients can explain this observation. Notably, the cases of underdosages are related to the capillary leakage observed during severe inflammation that increases the volume of distribution of hydrophilic antibiotics,²² hypoalbuminaemia that increases the free fraction of drugs^23,24 and cardiac hyper-flow induced by systemic inflammation that increases the renal clearance.²⁵ Overdoses can be explained by the high rate of acute kidney injury in critically ill patients with HAI. Our results are consistent with other studies that have observed failures to meet pharmacokinetic targets in up to 75% of critically ill patients^13,15,24 and support the rationale to measure the blood concentrations of antibiotics to adapt the daily dose in critically ill patients. Importantly, we found that when drug monitoring is recommended for the dose adaptation of empiric antimicrobial therapy, it results in therapeutic intervention in 12.2% of patients. This low percentage, which can decrease the medico-economic interest in such a practice, could be explained by modifying the spectrum of antimicrobial therapy after bacterial identification. This result suggests restricting the drug monitoring to patients at high risk of not meeting the pharmacokinetics targets or measuring the blood trug level of antibiotics only after bacterial identification.

The definition of blood concentration targets is critical when developing a protocol for pharmacokinetic optimization. The sampling time frame needs to be adapted to each antibiotic family. Most studies investigating the achievement of pharmacokinetic targets for aminoglycosides were observational cohorts, and concentrations above 8× MIC at the end of a bolus infusion were associated with favourable outcomes in critically ill patients.^26,27 For β-lactam antibiotics, unfavourable outcomes were more associated with blood concentrations not reaching 100% of the time spent above 4× MIC.²⁸ We aimed to measure the area under the curve for quinolones because their efficiency is both time and dose-dependent. Still, this approach is complex in clinical practice because it requires several blood measures. For this, a limited-sample strategy based on Bayesian estimation could easily be envisaged. The necessary adaptation of the targets to each treatment increases the difficulty of implementing protocols of optimization of pharmacokinetics in routine care and could explain that even if the rates of therapeutic drug monitoring-dose adaptation increased with our intervention, it was performed only in 58.6% of the patients at the end of the study.

Therapeutic drug monitoring may not be sufficient to improve patient outcomes because proper dose adaptations remain challenging. While standardized dose adaptation based on therapeutic drug monitoring failed to reach the targets in close to 10% of patients,^13,29 data are still missing to implement individual adaptation to measured concentrations. In most of the protocols of dose adaptation based on drug monitoring, the concentration targets are based on MIC of the causing pathogens. In our study, the MIC were inconstantly available because we recommended performing the first dosage during empiric antimicrobial therapy. We have made this choice because we reasoned that it could reduce the time to reach efficient treatment by rapidly diagnosing under dosage in the most severe patients. In most cases, we received the results of the blood concentration dosage at the same time as bacterial identification, and we were thus able to adapt the antibiotic dose to the MIC and the observed blood concentration. In the cases of late determination of the MIC concentration, such as in the cases of slow-growing bacteria or polymicrobial infections, we used the EUCAST breakpoint to define bacterial susceptibility up to getting the MIC measure. It thus remains possible that the predefined targets were not adequate for a few days in the case of resistant pathogens. In addition, the biological parameters that influence the pharmacokinetics of antibiotics (renal clearance, the volume of distribution) can vary over time with modifications of the medical status and the medical support. Regular monitoring could thus be required to enable an improvement of patient outcomes. Finally, the dosages were performed in the blood because it is easily accessible. Still, the diffusion to the site of infection varies from one antibiotic to another and from one organ to another.

There is no consensual outcome to assess the efficacy of antimicrobial stewardship in critically ill patients. The COMBACTE-STAT network has proposed to evaluate the rates of infection relapse within 48 hours of the end of the treatment,³⁰ while the ranking of the most desirable composite outcomes could also be an attractive alternative.^31–33 In this study, the primary outcome was 90-day mortality is meaningful and clinically relevant, but it is also highly dependent on medical history and cause of admission.³⁴ This primary outcome could have reduced the statistical power to demonstrate the efficacy of this quality improvement programme, and the use of patient-centred outcomes or biomarker-based evaluation could be tested in the future.^35,36

The adherence to the improvement programme did not overpass 50% for drug therapeutic monitoring, and patient care complied with both interventions only in 20% of cases. Similarly, low adherence to quality improvement programmes has been reported in studies evaluating the treatment of acute respiratory distress syndrome³⁷ or sepsis.² Several considerations are important to interpret these observations. First, we recommended the drug therapeutic monitoring to adapt the probabilistic therapy, which is challenging since these dosages are frequently not accessible on night and weekend duties. Second, the baseline application rates are a significant determinant. In our study, no patient received a continuous infusion of antimicrobial therapy before the intervention; the quality improvement programme resulted in a small but significant 20% increase in this procedure. Third, adherence to new guidelines is a slow process. For instance, only 65% of patients treated for myocardial infarction received optimal treatment 10 years after guidelines were released.³⁸ The compliance to the surviving sepsis campaign did not surpass 36% 8 years after their first publication.² However, our programme implementation may have been suboptimal since we did not apply specific tools to counteract barriers, such as the Consolidated Framework for Implementation Research.³⁹ Even if low adherence is a frequent limit to before–after studies, this study design remains useful because, contrary to randomized control trials, it tests the intervention applicability in clinical practice, helping to discriminate achievable and non-achievable targets. Finally, we have chosen this design because of the high risk of cross-contamination between study arms in case of randomization to a bundle or standard care in the same ICU, and because the estimation of the intervention effects by before–after studies is fair as compared to randomized clinical trials.^40–42

Our study has several limitations. First, it is a single-centre, before–after design, which did not make it possible to demonstrate causality between the quality improvement project and outcomes. Second, nearly half of the HAIs were treated with amoxicillin/clavulanic acid. This result is explained by a local probabilistic antimicrobial policy considering the risk of resistant bacteria in HAP as low up to 10 days after trauma.⁴ However, the continuous infusion of β-lactams has mainly been described with piperacillin/tazobactam,^10–12 and the frequent use of amoxicillin/clavulanic acid could have reduced the compliance to this mode of delivery. Even if the estimation of the quality improvement programme effect was close to the one reported in a recent controlled study evaluating therapeutic drug monitoring of piperacillin/tazobactam, this observation limits the extrapolation of our results in ICUs using higher rates of broad-spectrum β-lactams. Third, we did not assess the evolution of the resistance profile in our ICU because we did not assess the colonization with resistant bacteria on ICU discharge in the pre-intervention period. We also did not compare the rates of spectrum escalation with and without the intervention. Fourth, we aimed to optimize the pharmacokinetics of several classes of antibiotics. So, we cannot exclude that the intervention could be more efficient in β-lactams antibiotics on which most previous studies have been focused.^6,43 Finally, the quality improvement compliance rate did not surpass 60% during the intervention period. Even if this rate of guideline application is consistent with recent observations made in ICUs,^2,17,44,45 it has probably reduced the statistical power of the study.

In conclusion, implementing a quality improvement programme to optimize antimicrobial therapy based on therapeutic drug monitoring and continuous infusion of β-lactams was not associated with a reduction in mortality at day 90 in critically ill patients with HAI. The impact on dose optimization was substantial, with nearly half of the patients being in overdose or under dosage. However, the intervention application rates were low and confidence intervals for the findings were wide. The study may have had limited power to detect a clinically significant difference and a possible individual benefit.

Acknowledgements

We thank the CIC004 immunity and infection for technical support.

Funding

This study was supported by internal funding.

Ethics approval and consent to participate

The protocol was approved by the Société Française d’Anesthésie-Réanimation institutional review board (CERAR, IRB 00010254—2021-127). Consent was waived according to French law because the trial was an institutional quality improvement initiative applied to all patients.

Transparency declarations

A.R. reports receiving grants and consulting fees from Merck and bioMerieux. Other authors declare that they have no conflicts of interest involving the work under consideration for publication. No compensation was received for this study.

Authors’ contributions

C.L., M.B. and R.L.F. contributed to the acquisition and interpretation of data; revised the manuscript critically for important intellectual content. Y.H., N.G., A.B., P.J.M., D.D.D.L., K.A. and R.C. made substantial contributions to the interpretation of data and revised the work critically for important intellectual content. M.G., E.D. and R.B. performed the drug monitoring and revised the work critically for important intellectual content. R.C. contributed to the analysis and interpretation of the data, and revised the work critically for important intellectual content. A.R. contributed to the conception of the work and interpretation of data, and drafted the work. All the authors approved the version to be published; and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Supplementary data

Tables S1 to S5 are available as Supplementary data at JAC Online.

References