Late-onset sepsis (LOS) or nosocomial sepsis is a severe infectious neonatal condition characterized by bacteremia and a systemic inflammatory response that clinically manifests after the first 72 h of life and typically originates from indwelling central lines or catheters during hospital admission (Shane et al., 2017). Premature and/or very low birth weight (VLBW) neonates are at an increased risk for LOS due to an immature immune system and frequent exposure to invasive procedures in neonatal intensive care units (NICUs) (Stoll et al., 2004). Coagulase-negative staphylococci (CoNS) are the predominant pathogen in LOS (Dong and Speer, 2015). Approximately 21% of VLBW infants experience at least one episode of culture-proven LOS (Stoll et al., 2002). LOS has a high morbidity and mortality rate in premature and/or small for gestational age infants (Stoll et al., 2004). In cases of (suspected) LOS, the current standard of care is initiating treatment with broad-spectrum antibiotics (an aminoglycoside and a penicillin derivative) until the pathogen is identified by blood culture, which can take up to 48 h. Thereafter, narrow-spectrum antibiotic treatment targeting the isolated pathogen is initiated.

Intravenous (i.v.) vancomycin is given for 7 days in neonates with CoNS-positive LOS. Standardized loading doses of vancomycin are initiated, and subsequent doses may be corrected and individualized by means of therapeutic drug monitoring (TDM). However, the bactericidal efficacy of vancomycin is currently not assessed by readily accessible or reliable pharmacodynamic (PD) markers. Current vancomycin dosing regimens rely on indexed pharmacokinetic (PK) efficacy targets, based on the ratio of area under the curve (AUC_0–24h) and minimal-inhibitory concentration (MIC) (Pacifici and Allegaert, 2012; Abdulla et al., 2021; Lee et al., 2021). As these PK/PD indices rely heavily on the MIC, these may be inadequate as MICs are highly variable between patients, bacterial strains, and over time (Rathi et al., 2016).

PD markers that accurately describe the early and gradual bactericidal action of vancomycin in the case of CoNS-positive LOS are needed. Bacterial DNA load (BDL), determined through real-time quantitative PCR (RT-qPCR), is a marker used for this purpose. The method has a short turnover time, is sensitive, biologically accurate, and most of all, requires a small volume of blood (Van den Brand et al., 2014; van den Brand et al., 2018). The objective of this study was to assess changes in BDL to characterize the early response to vancomycin treatment in CoNS-positive LOS. A population-PK/PD model was developed for quantifying the effect of vancomycin on BDL in CoNS-positive LOS. In addition, potential covariates that influence the measure of the bactericidal effect were explored. Using this model, we aimed to gain insight into the vancomycin bactericidal effect over time and evaluate PD and clinical responses within the first hours of vancomycin treatment. A quantified vancomycin effect on BDL could support the accuracy of TDM.

In this study, bacterial DNA loads were used as a PD marker to evaluate the bactericidal effect of vancomycin in premature and/or VLBW neonates with CoNS-positive late-onset sepsis. In patients demonstrating a bactericidal response, the time profile of the BDL could be quantified with an empirical PK/PD model. Using this model, the time course and measure of the bactericidal effect of vancomycin could be described (Figure 4).

To our knowledge, the present study is the first to quantify the bactericidal effect of vancomycin in premature and VLBW neonates with CoNS-positive LOS by assessing time profiles of BDL using RT-qPCR. Bacterial PCR has been investigated in the past as a potential diagnostic tool and for antimicrobial susceptibility screening in the context of LOS, and varying results have been reported, with some studies suggesting that BDL measured through PCR methods could serve as a surrogate for blood culture in diagnosing some infectious diseases (Van den Brand et al., 2014; Chahbouni et al., 2015), while others demonstrated that PCR-based diagnostics were inferior to conventional blood culture (Tröger et al., 2016; Morrissey et al., 2017). All studies indicate that PCR-based diagnostics are feasible, however.

After assessing previously published population PK models for vancomycin, the model by Marqués-Miñana et al. (2010) was found to best describe the observed vancomycin levels of our study population (Seay et al., 1994; Grimsley and Thomson, 1999; Capparelli et al., 2001; Kimura et al., 2004; Anderson et al., 2007, and; Oudin et al., 2011). The selected model was developed using data from 70 neonates admitted to an NICU and treated with vancomycin, with a large proportion of patients comparable to our study population (Marqués-Miñana et al., 2010). The model was internally and externally validated by the original authors.

For patients responding to vancomycin treatment, time profiles of BDL were described using a turnover model (Figure 2). There was a positive relation between slope S and PMA when PMA was included as a power function with an estimated exponent of 8.23, indicating that older patients demonstrated a larger bactericidal effect than younger patients at equal vancomycin concentrations. However, it is questionable whether this perceived age dependency in the bactericidal activity of vancomycin is truly the underlying process for the increased BDL decline for patients of higher PMA. Innate and adaptive immunity is immature for premature neonates, and it is possible that increased bactericidal action at higher PMA is a product of higher immune activity, rather than an increased effect of vancomycin (Collins et al., 2018). However, the data did not support the estimation of IIV on k _death , nor could PMA be estimated as a covariate on k _death . Nonetheless, including PMA in the model as an exponential function on slope S significantly improved model fit and explained 5.3% of the IIV in slope S. Still, the remaining IIV in slope S in the final model was 58.8% and could be considered very high for classical PKPD models. This high variability is likely the result of oversimplification of the model. A true mechanistic antimicrobial PKPD model incorporates a complicated system of true bacterial growth, decay, and sigmoidal E _max effects, as well as resistance mechanisms, ideally with multiple levels of IIV. Oversimplification of these underlying mechanisms results in a model where all these types of variability have been combined in a single, large IIV for slope. As for high IIV in BDL₀, this is strongly supported by the data since the included patients demonstrated enormous variability in BDL at the first vancomycin dose.

A number of mechanistic PKPD models for antimicrobials have been published; however, these models have been exclusively applied in in vitro and animal studies. No such models could be found using clinical PD data. An overview of published antimicrobial PKPD models has been provided in a database by Minichmayr et al. (2022). A single similar publication was found that modeled the effect of vancomycin on CoNS colonization in central line-associated LOS using an in vitro hollow fiber infection model and a rabbit model (Ramos-Martín et al., 2016). The authors found that based on a translational model using their preclinical data, currently accepted dosing guidelines of AUC/MIC ≥400 were potentially too low for neonates ≤29 weeks GA and argued that efforts should be made for developing more efficacious dosing regimens in central line-associated LOS, optimizing bactericidal efficacy, minimizing toxicity, and preventing drug resistance.

Since the analyzed population was small, bootstrapping methods were deemed unsuitable for evaluating parameter uncertainty and model stability (Dosne et al., 2017). Thus, SIR was used for this purpose. SIR converged after five iterations (Supplementary Figure S2), and reliable RSE and 95% confidence intervals were obtained (Table 2). Based on the SIR results, the model was deemed accurate.

Simulation-based model evaluation using pc-VPC (N = 1,000) indicated that the final model could adequately predict the observed BDLs in the study population. Therefore, the final constructed model was deemed suitable for this study and provided sufficient insight that the effect of vancomycin on BDLs as determined through RT-qPCR can be described in a population-PKPD model.

Six patients demonstrated an increasing or constant BDL, as demonstrated in Figure 1B. These “persisting septic” BDL responses were evaluated by a neonatologist in our research team using the recorded clinical response of the patient. Indeed, five of these patients had clinically persistent CoNS bacteremia during vancomycin therapy, where the primary source of LOS was CoNS colonization of the central venous line (CVL) in situ, and removal of the CVL resulted in clinical improvements and negative blood cultures. Although CVL infections are a frequent source of nosocomial sepsis, there seems to be little consensus on whether CVL removal is beneficial in CoNS bacteremia (Cairns et al., 1995; Benjamin et al., 2000; Benjamin et al., 2001). It has been found that in over 70% of CoNS CVL infections, line retention does not interfere with antimicrobial efficacy (Cairns et al., 1995). However, Benjamin et al. (2001) stated that CVL removal should be considered for patients with persistent sepsis, identified as four consecutive blood cultures positive for CoNS. Two of our patients presented erratic BDL profile measurements, showing no specific pattern in BDL change over time (Figure 1C). In one of these two patients, CoNS bacteremia was secondary to an infected thrombus. Presence of an infected thrombus has been implied as a risk factor for persistent or recurrent staphylococcal sepsis in multiple case reports (Hubbard et al., 2016; Mani and Chandrasekharan, 2022). Incremental degradation of the colonized thrombus due to shearing stress and releasing CoNS-infected debris into the bloodstream at irregular time intervals could explain the observed erratic BDL profile. The other patient suffered from an infected peripheral venous line, complicated by the presence of a pustule at the ankle. Likewise, irregular mechanical stress at the primary site of infection could release high loads of infected material into the bloodstream at random intervals. For some patients, most measured BDLs were below or near the quantification limit following vancomycin treatment (Figure 1D). In these cases, the isolated CoNS culture was susceptible to amikacin. During this study, empirical amikacin and benzylpenicillin treatment was a clinical routine until diagnostic blood culture results. Therefore, susceptible CoNS exposed to 48 h of amikacin is expected to demonstrate substantial bacterial killing, explaining the absence of CoNS BDL during subsequent vancomycin therapy.

There were some limitations to this study. First, since RT-qPCR quantifies the total bacterial DNA in the study sample, it could not distinguish between DNA from living or dead bacteria. While it is known that circulating free DNA has a half-life of 1–2 h and is cleared through macrophage-mediated phagocytosis and enzymatic degradation in the spleen and liver, to our best knowledge, the rate at which dead bacteria are cleared from the neonatal bloodstream is unknown (Kustanovich et al., 2019). To account for the time delay between vancomycin dosing and BDL decrease in the model, we attempted to estimate the lag time (T_lag). However, including a T_lag in the model did not increase model fit and overcomplicated the model. Using total BDL allowed for a comprehensive approximation of bactericidal activity. Another limitation of the study was the risk of sample contamination. CoNS are not only the most frequent pathogen in LOS in developed countries but also the predominant contaminating micro-organism in blood samples (Huebner and Goldmann, 1999). Therefore, efforts should be made to minimalize the contamination risk. For instance, assessment of bacterial density could be considered or multiple sample sources could be used (Kassis et al., 2009). This was not performed during this study, as this would result in an unacceptable burden due to increased blood sampling. Regardless of this, culture-based assessments should be combined with careful clinical examination of patients to minimize the risk of unnecessary treatment due to sample contamination. The final limitation of the study is that no BDL profiles in the absence of antibiotic treatment were available. Therefore, it was difficult to distinguish between natural bacterial growth and treatment and thus estimate k _growth , k _death , and slope S as separate parameters. Unfortunately, no values of k _growth or k _death of CoNS, either as initial estimates or fixed parameters, could be found in the literature. A single study was identified that investigated the mechanistic PKPD relations between vancomycin and CoNS based on in vitro and animal data, though no parameter estimates were published in the model (Ramos-Martín et al., 2016). In our final model, it was assumed that k _growth was a zero-order constant dependent on BDL at T = 0 and k _death . As cellular multiplication relies on cell doubling, it is most certainly a first-order process. Moreover, relating k _growth to BDL at T = 0 results in a function in which the BDL cannot exceed this value at the cost of model accuracy. By estimating the BDL₀ with IIV, the model could more accurately predict BDLs above the last BDL before the first vancomycin dose.

There were a number of strengths to this study and its implications. First, an empirical model was developed to describe the bactericidal action of vancomycin in CoNS-positive LOS. CoNS are the predominant infective pathogen in LOS, accounting for approximately 53.2%–77.9% of all culture-proven LOS cases in developed countries (Dong and Speer, 2015). Vancomycin is the first-in-line antibiotic in CoNS-positive LOS, and dosing guidelines in neonatology are currently based on an AUC_24h/MIC index, in which a target of at least 400 is generally associated with efficacy (Pacifici and Allegaert, 2012). However, PK/PD indices heavily rely on the MIC, which is associated with considerable variability between bacterial strains, patients, and occasions (Rathi et al., 2016). Moreover, these indices treat bactericidal action as a binary “all-or-nothing” response. This implies that bacterial killing is only active at concentrations above the MIC and inactive at levels below the MIC, whereas in reality, bacterial killing changes dynamically with concentration. The method proposed here incorporates gradual bacterial killing as a function of vancomycin concentration and provides a more nuanced insight into bactericidal dynamics, independent of the MIC. This could be of particular benefit in the context of TDM, as improved concentration targets could be identified. Second, a relatively large number of drug concentrations and BDLs were available for each enrolled patient. Blood sampling in neonatology comes with considerable risk, and the number of samples collected per patient is hampered in the research context (Howie, 2011). Therefore, studies with a large amount of samples per patient are infrequent and valuable in this population. The relatively large amount of measurements per patient in our study allowed for more accurate depictions of the underlying PK and PD principles. A third strength of this study was that bacterial blood colonization was determined through multiplex RT-qPCR. The method used was validated and evaluated in the clinical setting and allowed for quantification of CFU eq/mL by adjusting measured DNA load for sample volume and CoNS genome load (Van den Brand et al., 2014; van den Brand et al., 2018). Thus, a surrogate marker for blood colonization that could be quantified within 8 h provided information on the bactericidal action of vancomycin in this study. Last, BDL profiles indicating treatment non-response (Figures 1BD) were compared with the corresponding clinical records by a neonatologist to assess whether these patients did not respond clinically. By doing so, the assumption to create the PKPD model based only on data of patients demonstrating a decrease in BDL was confirmed.

This study demonstrates that a decrease in BDL in CoNS-positive LOS can be quantified and predicted as a function of vancomycin concentration over time for patients who respond to vancomycin therapy. If developed further by combining preclinical data with clinical data, this would allow for more nuanced and precise dosing regimens, as compared to the currently used “all-or-nothing” dosing guidelines based on MIC targets. Moreover, it is expected that the applicability and accuracy of TDM could significantly improve if more evidence-based targets are identified.