Vancomycin in neonatal sepsis: predictive performance of a Chinese neonatal population pharmacokinetic model and clinical efficacy evaluation

Abstract

Background

In the neonatal population, individual calculation and adjustment of vancomycin (VCM) doses has been recommended based on population pharmacokinetics (PPK) methods.

Objective

Our previous study established a Chinese neonatal VCM PPK model. The main goal of this study was to evaluate the predictive performance of this PPK model for VCM trough concentration.

Methods

The data on neonatal severe infection patients treated with VCM were retrospectively collected. The predictive performance of this PPK model was expressed using mean prediction error (MPE), mean absolute prediction error (MAPE), sensitivity and specificity. Linear regression analysis was used to compare predicted and measured VCM concentrations. We drew the receiver operating characteristic (ROC) curve to evaluate the predictive efficacy of the ratio of area under the concentration-time curve over 24 hours to minimum inhibitory concentration (AUC_0-24/MIC) and trough concentration for clinical efficacy.

Results

A total of 40 neonates with Gram-positive bacterial sepsis were included. After VCM treatment, 32 (80%) neonates were clinically cured. Eight cases were a clinical failure: the trough concentrations and AUC_0-24 were lower than that of the clinical cure patients (8.70±4.30 vs 14.30±4.50 mg/L, p=0.003; 404.30±122.80 vs 515.40±131.70, p=0.037). The measured and predicted trough concentration were 11.16 (5.96, 16.53) mg/L and 10.13 (6.61, 15.73) mg/L, respectively. The MPE and MAPE were 4.62% and 13.26% (5.30%, 25.88%), respectively. The proportion of MAPE <30% in the adjusted regimen was higher than the initial regimen (89.66% vs 65.00%, p=0.039). Predictions of sensitivity and specificity by this PPK model were 88.24% and 94.29%, respectively. The coefficients of determination of linear regression analysis were 0.9171 and 0.9009 for the initial and adjusted regimen, respectively. The AUC_0-24 was correlated with the trough concentration (r=0.587, p<0.001). The ROC curve indicated that the optimal cut-off points for predicting clinical efficacy were AUC_0-24/MIC >425.47 and trough concentration >9.45 mg/L.

Conclusion

This PPK model has good predictive performance in Chinese neonatal patients. Both AUC_0-24/MIC and trough concentration can predict the clinical efficacy of antibacterial treatment.

Introduction

Vancomycin (VCM), a glycopeptide antibiotic, is used as a first-line therapy for neonates with suspected or confirmed methicillin-resistant Staphylococcus aureus (MRSA).¹ More than 80% of VCM is excreted unchanged in the urine. The metabolic processing of VCM can be affected by a variety of factors such as the patient’s weight, age, underlying diseases, liver and kidney function, infection position, and bacterial resistance.² The immune system of neonates, especially premature infants, is immature and susceptible to bacterial infection. Optimising VCM dosing to rapidly achieve adequate drug exposure is imperative in treating neonatal sepsis, particularly when treating invasive MRSA infections.¹ Neonatal weight, renal function, and other organs’ functions are constantly changing, coupled with the influence of severe infection on the haemodynamics of neonates, all of which can affect the pharmacokinetic parameters of VCM in the neonatal population.³ VCM therapeutic drug monitoring (TDM) guidelines recommend that individual calculation and adjustment of VCM doses be based on population pharmacokinetics (PPK) methods.^{4 5} Therefore, it is necessary to analyse the pharmacokinetics of VCM in neonates with severe infection, and to establish a neonatal PPK model for guiding the clinical individualised administration of VCM.

Several practice guidelines of VCM TDM and previous literature recommend that the VCM trough concentration should be maintained at 10–20 mg/L and the ratio of area under the concentration-time curve over 24 hours to minimum inhibitory concentration (AUC_0-24h/MIC) at ≥400 to achieve the best antibiotic exposure level, improve clinical efficacy, and reduce renal toxicity.^4–8 Clinical pharmacokinetic/pharmacodynamic studies have shown that the parameter predicting the clinical and bacteriological efficacy of VCM is the AUC_0-24h/MIC.^{8 9} The VCM TDM guidelines of the Infectious Diseases Society of America recommend that an individualised target of AUC_0-24h/MIC ratio of 400 to 600 (assuming a VCM MIC of 1 mg/L) should be advocated to achieve clinical efficacy while improving patient safety.⁵ Related literature has reported that the PPK model based on AUC_0-24/MIC could be used to optimise the administration of VCM.^{7 8}

In our previous research, we used NONMEM to establish a pharmacokinetic model of VCM in neonatal patients, which can be used in the design of a VCM individualised dosing regimen for neonatal patients.¹⁰ The main goal of this study was to evaluate the predictive performance of this PPK model for VCM trough concentration in Chinese neonatal patients with severe infection. The pharmacokinetic model was used to calculate the neonatal patient’s pharmacokinetics, predict the trough concentration of VCM, and provide reference for the clinical individualised medication of VCM in neonatal patients with severe infection. In addition, our study compared the predictive performance and determined the threshold value of trough concentration and AUC_0-24/MIC for clinical efficacy.

Methods

Patients and data collection

All neonates diagnosed with Gram-positive bacterial blood stream infection and treated with VCM in the neonatal intensive care unit of the Affiliated Suzhou Hospital of Nanjing Medical University (Suzhou, China) from July 2017 to December 2018 were included in our retrospective study. Patients’ data were obtained from the Neonatal Medical Database in March 2019. Inclusion criteria were: (1) postnatal age <28 days; (2) the blood culture shows Gram-positive bacteria and the patient received VCM for anti-infection treatment during hospitalisation; (3) one or more serum VCM trough concentrations were determined. Exclusion criteria were: (1) patients with incomplete or missing clinical and laboratory information, such as having no serum creatinine level during neonatal intensive care unit hospitalisation; (2) trough concentrations of VCM were unsuitable, such as below or exceeding the limit of quantification, no steady-state trough concentration, or incorrect sampling time (not 0.5 hour before the next dose); (3) there were no drug susceptibility results in the aetiology report; (4) the patient had received VCM for <7 days.

The initial dosage regimen of VCM (Lilly Suzhou Pharmaceutical Co, Ltd, H20080356, 500 mg) was 10–15 mg/kg. Those neonates within 7 days of birth were administered VCM once every 12 hours, and the neonates between 7 and 28 days of birth were administered VCM once every 8 hours. Each administration was performed through intravenous infusion over 1 hour.

The main analysis indicators of each neonate were recorded as follows: (1) demographic data—gender, age, weight, corrected gestational age; (2) infection diagnosis; (3) initial and reviewed pathogenic culture and drug susceptibility; (4) the start time and treatment of VCM, including initial and adjusted dosage regimens; (5) steady-state trough concentrations after initial and adjusted regimens; (6) changes of biochemical indexes and inflammatory indexes before and after medication.

Outcome assessments of clinical efficacy

According to the guidelines for clinical research on antimicrobial drugs,¹¹ clinical efficacy was evaluated after 7–14 days of medication. Clinical efficacy was evaluated as clinical cure and clinical failure. Clinical cure was defined as complete resolution or significant improvement of pre-treatment infection signs and symptoms such that no additional antimicrobial therapy was needed. Clinical failure was defined as no apparent response or an incomplete response needing additional antibiotic therapy for infections. Microbiological success was defined as eradication or presumed eradication of Gram-positive bacteria. The primary outcomes were the clinical cure and microbiological success rates at the end of treatment.

Population pharmacokinetic analysis of VCM in Chinese neonates

VCM was intravenously administered over a 1 hour period. The steady-state trough concentration of VCM was measured (1 mL of venous blood was taken as a blood sample) just after 48 hours and 30 min before the next dose.⁴ The target serum VCM trough concentrations ranged from 10–20 mg/L. The serum trough concentrations of VCM were measured by chemiluminescent enzyme immunoassay. The automatic chemiluminescent immunoassay (ARCHITECT I2000SR), VCM kit, and quality control reagent were all products of Abbott Company, Illinois, USA.

The VCM PPK model of our previous study evaluated the PPK of VCM in Chinese neonates by nonlinear mixed effect model. This model collected 154 samples of VCM serum trough concentrations from 91 neonates.¹⁰ We used a one-compartment model with first order elimination to describe the structure pharmacokinetic model, and physiological maturity model was employed to screen covariates. The exponential model described the interindividual variation, and the proportional model described the residual variation.¹⁰ The included covariates were weight (WT), postmenstrual age (PMA), and serum creatinine (SCR). The final model identified WT, PMA and SCR as the main factors affecting VCM clearance in neonates. The parameter estimation results of the final model are shown in Equation 1 and Equation 2 below. Bootstrap and normalised predictive distribution error showed the satisfactory stability and prediction performance of the final model. The final model indicated that 96% of neonates with low SCR (15 μmol/L) were not getting AUC_{0-24 h}/MIC ≥400, according to the current guidelines. The daily dose of VCM can be calculated according to the PPK model of neonatal VCM, shown in Equation 3.¹⁰

$${\displaystyle \begin{array}{ll}CL\left(L\cdot h{r}^{-1}\right)& ={\theta }_{CL}\times {\left(\frac{WT}{70}\right)}^{0.75}\\ & \times \frac{1}{1+{\left(\frac{{\theta }_{MT}}{PMA}\right)}^{{\theta }_{HILL}}}\times {\left({\theta }_{SCR}\right)}^{\frac{30}{SCR}}\end{array}}$$ (1)

$${\displaystyle V\left(L\right)={\theta }_V\times {\left(\frac{WT}{70}\right)}^1}$$ (2)

$$\begin{gathered} {\displaystyle \mathrm{D}\mathrm{O}\mathrm{S}\mathrm{E}\left(\mathrm{m}\mathrm{g}\cdot {\mathrm{d}}^{-1}\right)\ge 400\times \mathrm{M}\mathrm{I}\mathrm{C}\times 4.6 \\ \times {\left(\frac{\mathrm{W}\mathrm{T}}{70}\right)}^{0.75}\times \frac{1}{1+{\left(\frac{37.6}{\mathrm{P}\mathrm{M}\mathrm{A}}\right)}^{5.46}}\times {\left(1.2\right)}^{\frac{30}{\mathrm{S}\mathrm{C}\mathrm{R}}}} \end{gathered}$$ (3)

Prediction of PPK model in Chinese neonatal patients

We embedded this PPK model in the SmartDose software, which can be used in the design of the VCM individualised dosing regimen for neonatal patients.¹² The SmartDose software was developed by Gao et al ¹² as a web platform (https://smartdose03.hfjk.net:28080/%23/tdm/account/login). It is a decision support system for individualisation of VCM dosage. It is developed using the maximum a posterior Bayesian estimation (MAPB) by the open-source language R combined with the PPK characteristics of VCM in Chinese neonatal patients. It provides initial design and adjustment of dose regimens based on the TDM results, as well as a user-defined module to facilitate optimal VCM therapy. SmartDose has high computational reliability, which is validated by NONMEM, the gold standard pharmacokinetic software.¹² SmartDose software has a good ability to predict VCM serum concentration in the individualised medication of VCM in the Chinese adult population.^{13 14} The operation interface of SmartDose software for VCM individualised therapy in neonates is shown in online supplemental figure 1.

Validation of the initial dosage regimen of VCM: In the SmartDose software, we selected the patient population as neonates, and entered the general clinical data of the patients, including gender, corrected gestational age, weight, and SCR value; the patient’s plasma clearance (CL) could then be calculated. We selected the target range of trough concentration and the dosing frequency to predict the required VCM dosage regimen and trough concentration.

Verification of the adjusted dosage regimen of VCM: We selected the patient population as neonates in the SmartDose software, and input the general clinical data of the patients, including gender, corrected gestational age, weight, and SCR value, etc, selected the target range of trough concentration and dosing frequency, and then entered the patient’s initial dosage regimen and TDM results in the adjusted dosage regimen module. Finally, the SmartDose software calculated the blood drug concentration at any time and drew a drug–time curve, which can predict the trough concentration value after adjustment of administration.

Predictive performance of the PPK model

For each neonate in the external validation cohort, VCM concentrations were predicted using the parameters of the PPK model and simulating the actual dosing regimen given to the neonate. Model-predicted VCM concentrations were then compared with the corresponding measured VCM concentrations. Predictive performance of TDM was expressed using bias, precision, sensitivity and specificity. As described by Sheiner and Beal,¹⁵ the mean prediction error (MPE) and mean absolute prediction error (MAPE) for the first and adjusted trough concentrations were used as measures of precision and bias.^{8 15 16} These are calculated by Equations (4) and (5):

$${\displaystyle \mathrm{M}\mathrm{P}\mathrm{E}\left(\mathrm{b}\mathrm{i}\mathrm{a}\mathrm{s}\right)=\frac{1}{\mathrm{N}}\sum _{\mathrm{i}=1}^{\mathrm{N}}\frac{\mathrm{C}\mathrm{p}-\mathrm{C}\mathrm{m}}{\mathrm{C}\mathrm{m}}\times 100}$$ (4)

$${\displaystyle \mathrm{M}\mathrm{A}\mathrm{P}\mathrm{E}\left(\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\right)=\frac{1}{\mathrm{N}}\sum _{\mathrm{i}=1}^{\mathrm{N}}\left|\frac{\mathrm{C}\mathrm{p}-\mathrm{C}\mathrm{m}}{\mathrm{C}\mathrm{m}}\right|\times 100}$$ (5)

where Cp represents VCM concentration predicted by the model and Cm refers to the measured VCM concentration. In addition, the final model was used to calculate the number of patients with MPE within ±20% and ±30%. The final model with low MPE and MAPE values and a high number of prediction errors within ±20% and ±30% was considered acceptable.^{16 17}

Sensitivity and specificity were calculated according to the following Formulas (6) and (7)¹⁸:

$${\displaystyle \mathrm{S}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{i}\mathrm{t}\mathrm{y}=\mathrm{t}\mathrm{r}\mathrm{u}\mathrm{e}\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}÷\left(\mathrm{t}\mathrm{r}\mathrm{u}\mathrm{e}\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}+\mathrm{f}\mathrm{a}\mathrm{l}\mathrm{s}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{g}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}\right)}$$ (6)

$${\displaystyle \mathrm{S}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{t}\mathrm{y}=\mathrm{t}\mathrm{r}\mathrm{u}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{g}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}÷\left(\mathrm{t}\mathrm{r}\mathrm{u}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{g}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}+\mathrm{f}\mathrm{a}\mathrm{l}\mathrm{s}\mathrm{e}\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}\right)}$$ (7)

where true positive means both predicted and measured concentrations were in the therapeutic range; true negative means both predicted and measured concentrations were out of the therapeutic range; false positive means predicted concentration was in, and measured concentration was out of, the therapeutic range; and false negative means predicted concentration was out of, and measured concentration was in, the therapeutic range. The target therapeutic concentrations for intermittent VCM dosing were 10–20 mg/L for trough concentration.

Trough concentration and AUC_0-24 relationship

Following model evaluation, the relationship between trough concentration and AUC_0-24 was examined. Bayesian estimates of CL for each neonate from the PPK model were used to calculate AUC_0-24 at the time that VCM trough concentrations were collected. AUC_0-24 was calculated as the daily dose/CL. For a given trough concentration, the proportion of neonates with that trough concentration who achieved an AUC_0-24 ≥400 was calculated. An AUC_0-24 ≥400 mg· h/L would predict an AUC_0-24/MIC ≥400 for an MIC of ≤1 mg/L.

Statistical analysis

Statistical analysis was performed using SPSS 22.0 software. The counting variables were expressed as case number (n). The χ² test was used to compare the counting variables. The normal distribution variables were expressed as mean±SD (x±s), and the comparison between groups was performed by independent sample t test; the non-normal distribution variables were expressed as median (quartile) (M (QL, QU)), and the method of rank and inspection was used for comparison between groups. We drew the receiver operating characteristic (ROC) curve to obtain the best decisive threshold of AUC_0-24/MIC and trough concentration for clinical efficacy. The comparison of clinical predictive efficacy of AUC_0-24/MIC and trough concentration was done using MedCalc software. A value of p<0.05 indicated that the difference was statistically significant. Linear regression models were used to compare predicted and measured drug serum concentrations, and to evaluate potential relationships between trough concentration and AUC_0-24/MIC.

Results

Patient characteristics

There were a total of 40 patients included in this study, 20 males (50%) and 20 females (50%); 26 (65%) of the 40 cases were premature babies. The gestational age ranged from 26.00–40.29 weeks, with an average of 33.8±4.2 weeks; the corrected gestational age was 30.00–45.57 weeks, with an average of 35.71 (33.14, 40.25) weeks. The corrected weight was 1.16–5.12 kg, with an average of 2.47±1.03 kg.

Outcome assessments of clinical efficacy

All neonates were diagnosed as having neonatal sepsis; the distribution of the infection diagnosis, bacteria in blood, and drug sensitivity are shown in table 1. Forty-three strains of Gram-positive bacteria were cultured in the blood of the 40 neonates. The distribution of pathogenic bacteria and MIC are shown in table 1. One strain of methicillin-resistant Staphylococcus epidermidis (MRSE) had an MIC=4. The mixed other bacteria cultured in blood were six strains of Klebsiella pneumonia (extended spectrum beta-lactamases positive (ESBL+)), one strain of Escherichia coli (ESBL+), one strain of Pseudomonas aeruginosa, and one strain of Candida albicans. Neonates whose cultures had mixed Gram-negative bacteria were all sensitive to meropenem, and were treated with meropenem during VCM therapy. Fluconazole was used to treat Candida albicans. Meropenem was the main antibacterial agent concomitantly used during VCM therapy; other related nephrotoxic agents were vasopressors and furosemide (table 1). Among the 40 neonates, according to clinical evaluation criteria, 32 (80%) were clinically cured after VCM treatment, five (12.5%) were clinical failure, one (2.5%) was transferred to another hospital for further treatment, and two (5%) were discharged automatically. The clinical efficacy rate was 80% (32/40). Among the eight patients with clinical failure: two had cultures with Klebsiella pneumoniae (ESBL+); two had strains of MIC=2 (one of MRSE and one of Staphylococcus cephalis), although the trough concentrations of VCM were in the range of 10–20 mg/L (12.79 mg/L and 18.01 mg/L); and another four patients had trough concentrations of VCM <10 mg/L. We evaluated only the Gram-positive bacteria clearance rate for VCM clinical efficacy to reduce the impact of mixed bacteria. The Gram-positive bacteria of blood culture in 40 neonates were all cleared after VCM treatment, that is, the Gram-positive bacteria clearance rate was 100%.

Table 1.

Summary of the demographic and clinical characteristics

	Neonates (n=40)	Clinical cure (n=32)	Clinical failure (n=8)	t/Z/χ	P value
Gender (n (%))
Female	20 (50.00)	14 (43.75)	6 (75.00)	1.406	0.236
Gestational age (weeks)	33.8±4.2	33.9±4.9	33.7±4.1	0.103	0.919
Gestational age <37 weeks (n (%))	26 (65.00)	22 (68.75)	4 (50.00)	–	0.416
Gestational age at vancomycin treatment (weeks)	35.71 (33.14, 40.25)	35.43 (33.07, 39.72)	38.43 (34.29, 41.29)	−1.177	0.239
Weight at vancomycin treatment (kg)	2.47±1.03	2.50±0.80	2.50±1.10	0.232	0.817
Infection diagnosis (n (%))
Neonatal sepsis	40 (100.00)	32 (100.00)	8 (100.00)	0.000	1.000
Neonatal pneumonia	34 (85.00)	27 (84.38)	7 (87.50)	0.000	1.000
Neonatal purulent meningitis	28 (70.00)	23 (71.88)	5 (62.50)	0.007	0.931
Necrotising enterocolitis	5 (12.50)	3 (9.38)	2 (25.00)	–	0.257
Gram-positive bacteria in blood (n (%))
Staphylococcus aureus	8 (18.60)	8 (25.00)	0 (0.00)	1.182	0.277
MRSA	8 (100.00)	8 (25.00)	0 (0.00)	1.182	0.277
CNS	24 (55.80)	20 (62.50)	4 (50.00)	–	0.690
MRCNS	17 (70.80)	15 (46.88)	2 (25.00)	0.518	0.472
Enterococcus	11 (25.60)	10 (31.25)	1 (12.50)	0.384	0.535
Distribution of MIC (n (%))
MIC ≤1	36 (83.72)	30 (85.71)	6 (75.00)	0.044	0.834
MIC=2	6 (13.95)	4 (11.43)	2 (25.00)	–	0.308
MIC=4	1 (2.33)	1 (2.86)	0 (0.00)	–	1.000
Mixed bacteria (n (%))	9 (22.50)	7 (21.88)	2 (25.00)	0.000	1.000
Klebsiella pneumoniae (ESBL+)	6 (66.70)	4 (57.14)	2 (100.00)	–	0.580
Escherichia coli (ESBL+)	1 (11.10)	1 (14.29)	0 (0.00)	–	1.000
Pseudomonas aeruginosa	1 (11.10)	1 (14.29)	0 (0.00)	–	1.000
Candida albicans	1 (11.10)	1 (14.29)	0 (0.00)	–	1.000
Daily dose of vancomycin (mg/kg)	42.47 (30.37, 45.11)	42.07 (29.97, 44.93)	44.75 (31.15, 56.44)	−1.144	0.253
Course of vancomycin (days)	16.50 (11.75, 21.00)	18.50 (14.00, 21.00)	12.50 (10.00, 17.25)	−1.715	0.095
Trough concentration	11.88±4.84	14.30±4.50	8.70±4.30	−3.163	0.003
AUC0-24	482.19±140.35	515.40±131.70	404.30±122.80	−2.159	0.037
CL (L/h)	0.22±0.15	0.21±0.15	0.22±0.14	0.872	0.386
Concomitant agents
Meropenem (n (%))	16 (40.00)	12 (37.50)	4 (50.00)	–	0.690
Fluconazole (n (%))	5 (12.50)	3 (9.38)	2 (25.00)	–	0.257
Vasopressors (n (%))	29 (72.50)	22 (68.75)	7 (87.50)	0.384	0.535
Furosemide (n (%))	6 (15.00)	4 (12.50)	2 (25.00)	–	0.580

AUC_0-24, area under the concentration-time curve over 24 hours; CL, clearance; CNS, coagulase negative staphylococcus; ESBL, extended spectrum beta lactamases; MIC, minimum inhibitory concentration; MRCNS, methicillin-resistant coagulase negative staphylococcus; MRSA, methicillin-resistant Staphylococcus aureus.

Trough concentration and correlation analysis with clinical efficacy

The VCM dosage regimens were adjusted in 21 of the 40 neonates. Sixty-nine samples of trough concentrations were collected in the end: 40 initial trough concentration samples, and 29 samples of trough concentration after the second and third adjusted dosage regimen. The dosage regimens of VCM were 42.07 (29.97, 44.93) mg/kg/day) and 44.75 (31.15, 56.44) mg/(kg·day) in the initial regimen and adjusted regimen, respectively. The treatment course of VCM was 16.50 (11.75, 21.00) days. The steady-state concentrations of the initial regimen and adjusted regimen were 9.53 (5.54, 16.53) mg/L and 11.92 (6.95, 16.75) mg/L, respectively. There was no statistically significant difference between the initial and adjusted trough concentrations (Z=−0.760, p=0.447), as shown in table 2. The proportions of initial trough concentration and adjusted trough concentration in the range of 10–20 mg/L were 45% and 55.2%, respectively.

Table 2.

PPK model to predict the steady-state trough concentration of VCM initial and adjusted regimen in neonates with severe infection

	Total	Initial regimen	Adjusted regimen	t/Z/χ	P value
Number	69	40	29	–	–
Trough concentration (mg/L, M (QL, QU))
Measured value	11.16 (5.96, 16.53)	9.53 (5.54,16.53)	11.92 (6.95,16.75)	−0.760	0.447
<10 mg/L	31 (44.93)	20 (50.00)	11 (37.90)	0.990	0.320
10–20 mg/L	34 (49.27)	18 (45.00)	16 (55.20)	0.696	0.404
>20 mg/L	4 (5.80)	2 (5.00)	2 (6.90)	–	1.000
Predicted value	10.13 (6.61, 15.73)	9.36 (6.21, 16.32)	10.13 (7.66, 15.05)	−0.333	0.739
MPE (%)	4.62	10.92	3.57	–	–
MAPE (%, M (QL, QU))	13.26 (5.30, 25.88)	13.76 (5.27, 32.38)	12.54 (4.90, 22.91)	−0.505	0.614
MAPE <20% (n (%))	44 (63.77)	24 (60.00)	20 (68.97)	0.585	0.444
MAPE <30% (n (%))	52 (75.36)	26 (65.00)	26 (89.66)	4.256	0.039
Sensitivity (%)	88.24	94.44	81.25	–	0.323
Specificity (%)	94.29	95.45	92.31	–	1.000

M, median; MAPE, mean absolute prediction error; MPE, mean prediction error; PPK, population pharmacokinetics; QL, lower quartile; QU, upper quartile; VCM, vancomycin.

Eight out of 40 cases were clinical failure; the results of correlation analysis between trough concentration and clinical efficacy indicated that the trough concentrations and AUC_0-24 were lower than clinical cure patients (8.70±4.30 vs 14.30±4.50 mg/L, p=0.003; 404.30±122.80 vs 515.40±131.70, p=0.037). The courses of VCM treatment for clinical failure patients were shorter than that of clinical cure patients, but were not statistically different (12.50 (10.00, 17.25) days vs 18.50 (14.00, 21.00) days, p=0.095), as shown in table 1. There were no statistically significant differences in the site of infection diagnosis, distribution of pathogenic bacteria, proportion of MIC, proportion of mixed bacterial infection, concomitant antimicrobial agents, vasopressors, and furosemide drugs.

Predictive performance of PPK model

The measured steady-state trough concentration of 69 cases was 11.16 (5.96, 16.53) mg/L; the trough concentration value predicted by the PPK model was 10.13 (6.61, 15.73) mg/L. The MPE and MAPE between the predicted and measured concentration was 4.62% and 13.26% (5.30%, 25.88%); the MPE of the adjusted regimen and initial regimen were 3.57% and 10.92%, respectively. Among 69 serum samples, the PPK model made a good prediction for 52 (75.36%) concentrations, which means the MAPE was <30%. The proportion of MAPE <30% in the adjusted regimen was higher than in the initial regimen (89.66% vs 65.00%, p=0.039). Predictions of sensitivity and specificity by this PPK model were 88.24% and 94.29%, respectively. The prediction of the initial and adjustment regimen of VCM in neonates with severe infection by the PPK model is shown in table 2. The results showed that there was no statistical difference between the MAPE after the adjustment and the initial regimen (Z=−0.505, p=0.614). The scatter diagrams of correlation between the measured trough concentration and the predicted trough concentration of VCM are shown in figure 1A and B. According to the results of linear regression analysis, this PPK model SmartDose showed good predictive performance expressed by the coefficient of determination (r) of 0.9171 and 0.9009 for initial regimen and adjusted regimen of VCM, respectively.

Figure 1.

Predictive performance of the vancomycin pharmacokinetic models. (A) Predicted versus observed serum concentrations in the initial regimen of vancomycin (r=0.9171). (B) Predicted versus observed serum concentrations in the adjusted regimen of vancomycin (r=0.9009).

AUC_0-24/MIC and trough concentration in predicting clinical efficacy

There were 69 trough concentrations and AUC_0-24 data from 40 neonates. According to the results of linear regression analysis of AUC_0-24 and trough concentration, the AUC_0-24 was correlated with the trough concentration (r=0.587, p<0.001) (figure 2A). For MIC values of 0.5 mg/L, AUC_0–24/MIC ≥400 was achieved for 100% (2/2). When the MIC increased to 1 mg/L, target attainment dropped to 70.2%% (40/57). For MIC values of 2 and 4 mg/L, no infants achieved AUC_0–24/MIC ≥400 (figure 2B). We drew the ROC curve to obtain the best decisive threshold of AUC_0-24/MIC and trough concentration for clinical efficacy. ROC curve results showed that the area under the curve of AUC_0-24/MIC was 0.762, and the area under the curve of trough concentration was 0.790. The results showed that the trough concentration and AUC_0-24/MIC had good predictive value for clinical efficacy (p=0.005, p=0.011). The best predictive value was AUC_0-24/MIC >425.47 (sensitivity was 75.90%, specificity was 72.70%, and the corresponding Youden index was 0.49) and trough concentration >9.45 mg/L (sensitivity was 86.20%, specificity was 63.60%, and the corresponding Youden index was 0.50), as shown in figure 3.

Figure 2.

(A) Correlation between AUC_0-24 and trough concentration of vancomycin (r =0.5866, p<0.001). (B) Scatter diagram of AUC_0-24/MIC and trough concentration. All of the AUC_0-24/MIC <400 when MIC=2 mg/L and 4 mg/L. AUC_0-24, area under the concentration-time curve over 24 hours, MIC, minimum inhibitory concentration.

Figure 3.

ROC curves of AUC_0-24/MIC and trough concentration (AUC_0-24 0.762 and 0.790, respectively). When AUC_0-24/MIC >425.47 (sensitivity 75.90%, specificity 72.70%, Youden index 0.49) or trough concentration >9.45 mg/L (sensitivity 86.20%, specificity 63.60%, Youden index 0.50), the curve has a good predictive value. AUC, area under the curve; AUC_0-24, area under the concentration-time curve over 24 hours; MIC, minimum inhibitory concentration; ROC, receiver operating characteristic.

Discussion

VCM has a good therapeutic effect on severe neonatal infections caused by Gram-positive bacteria,^{19 20} but previous studies in the literature have reported that the rate of serum trough concentrations in the range of 10–20 mg/L after empirical dosage regimen is low (41%). There is a poor correlation between the VCM dose given and the trough concentration achieved.^{19 21} In our study, the rate of trough concentration with the empirical initial regimen in the range of 10–20 mg/L was 45%, and the rate of trough concentration with the empirical adjusted regimen was 55.2%. These results indicated that the rate of trough concentration in the range of 10–20 mg/L could not be significantly improved by empirical dosage adjustment of VCM. VCM showed significant individual differences in the treatment of critical neonatal infections. Besides the monitoring of VCM serum trough concentration, the use of PPK methods to guide the adjustment of the VCM dosage regimen should be recommended.4 5 9 22–24

This study evaluated the predictive performance of the PPK model for the prediction of VCM serum concentration in neonates with severe infection. We used NONMEM to establish a pharmacokinetic model of VCM in neonatal patients. Comparing the measured trough concentration with the predicted trough concentration, the PPK model showed good predictive performance expressed by the coefficients of determination (r) of 0.9171 and 0.9009 for initial regimen and adjusted regimen of VCM, respectively. The sensitivity and specificity of TDM predictive performance by this PPK model were all >85%. The MPE and MAPE of VCM trough concentrations were, respectively, 4.62% and 13.26% (5.30%, 25.88%); the MPE of the adjusted regimen and initial regimen were 3.57% and 10.92%, respectively. There was less error of trough concentration in the adjusted regimen. Among 69 serum samples, the PPK model made a good prediction for 52 (75.36%) concentrations (MAPE <30%), which was similar to the results of validation of the adult population.¹⁷ The proportion of MAPE <30% in the adjusted regimen was higher than in the initial regimen (89.66% vs 65.00%, p=0.039), indicating that this model made a better predictive performance for trough concentration of the adjusted regimen. There were 17 concentrations with APE >30%. We analysed the covariates in the model, and found that the following factors may have an impact on the predictive accuracy of the PPK model on VCM trough concentrations in neonatal severe infection. (1) The SCR value changes greatly before and after VCM treatment, which may affect the prediction error of the VCM concentration in the PPK model, resulting in a large error between the measured trough concentrations. In the clinical practice of VCM individualised medication, adjusting the dosage regimen solely based on the SCR value of VCM pre-administration will still result in a large error between the predicted values and measured values. We recommend that the SCR levels be measured multiple times during the VCM treatment, and the error between the predicted trough concentration value and the measured value can be reduced by adjusting the dosage regimen according to the varied SCR levels during VCM treatment. (2) Neonates are different from adult patients; days after birth and weight vary greatly from day to day, and the pharmacokinetics of VCM are highly variable in this population due to differences in maturation and development. Individual diversity may affect the calculation of VCM pharmacokinetic parameters and the prediction of serum concentration by PPK software, resulting in a large error between predicted and measured values. (3) The patient’s liver and kidney function, albumin level, the rational dosage regimen of VCM, and the interaction of drugs will also affect the prediction of VCM blood trough concentration.^{25 26}

In our study, we included 40 neonates with Gram-positive bacterial sepsis, and 32 (80%) were clinically cured after VCM treatment. Eight cases were clinical failure; the results indicated that the trough concentrations and AUC_0-24 were lower than clinical cure patients (8.70±4.30 vs 14.30±4.50 mg/L, p=0.003; 404.30±122.80 vs 515.40±131.70, p=0.037). The trough concentration and AUC_0-24 were strongly correlated with clinical efficacy. VCM TDM guidelines and searches of the literature have confirmed that VCM trough concentration and AUC_0-24 in the target range can significantly improve the clinical efficacy.^4–8 The trough concentration of VCM should be maintained at 10–20 mg/L, and AUC_0-24h/MIC should be ≥400 to achieve the best antibiotic exposure level, improving clinical efficacy.^4–8

AUC_0-24/MIC is now often used to evaluate the effects of antibacterial drugs.^{4 5 27} When AUC_0-24/MIC is ≥400, bacteria can be cleared more quickly and the patient’s clinical symptoms can improve faster.⁹ In our study, the AUC_0-24 was correlated with the trough concentration of VCM (r=0.587, p<0.001). There were 28 (70%) cases of AUC_0-24/MIC ≥400, which accounted for a relatively high proportion. Swets²⁸ has suggested that AUC_0-24 values under 0.7 have ‘low’ accuracy, values between 0.7 and 0.9 have ‘moderate’ accuracy, and values >0.9 have ‘high’ accuracy.^{29 30} The ROC curve results of our study indicated that the AUC_0-24 for the clinical efficacy of AUC_0-24/MIC and trough concentration was 0.762 and 0.790, respectively. The predictive performance of the trough concentration and AUC_0-24/MIC was of ‘moderate’ accuracy, which suggests that AUC_0-24/MIC and trough concentration were able to reliably predict clinical efficacy (p=0.005, p=0.011). The cut-off of best predictive value was AUC_0-24/MIC >425.47 and trough concentration >9.45 mg/L.

The major limitation of our study was the small number of neonates in our dataset. Only 69 samples from 40 patients were collected, which limits our ability to generalise the results. However, the results of this study may be worthwhile and helpful for VCM individual treatment in neonates. Generalisation of the results should be treated carefully, and further evaluation studies should be conducted when more data are collected.

Conclusion

The results of this study showed that the VCM PPK model had good predictive performance on the trough concentration of VCM in neonates with severe infections, especially for the adjusted regimen. It can be combined with the calculation of the PPK model to provide individualised administration for neonates. It can predict clinical efficacy of antibacterial treatment better when the AUC_0-24/MIC is >425.47 or the trough concentration is >9.45 mg/L.

What does this paper add?

What is already known on this subject

• NONMEN is currently considered the gold standard for the population pharmacokinetics (PPK) model in neonatal population.

• The precision of the PPK model used in individual calculation and adjustment of vancomycin (VCM) dosage is essential for the prediction of clinical efficacy.

What this study adds

• We previously established a Chinese neonatal VCM PPK model. We found the PPK model has good predictive performance for the trough concentration of VCM in Chinese neonatal patients with severe infection, especially for the adjusted regimen.

• The trough concentration was correlated with AUC_0-24; both correlated with clinical efficacy. Both AUC_0-24/MIC and trough concentration can predict the clinical efficacy of antibacterial treatment better when AUC_0-24/MIC is >425.47 or trough concentration is >9.45 mg/L.

Ethics statements

Patient consent for publication

Obtained.

Ethics approval

This study was approved by the Ethics Committee of the Affiliated Suzhou Hospital of Nanjing Medical University (No. K2017024).