Automated calculation and reporting of vancomycin area under the concentration–time curve: a simplified single-trough concentration-based equation approach

ABSTRACT

Vancomycin, a crucial antibiotic for Gram-positive bacterial infections, requires therapeutic drug monitoring (TDM). Contemporary guidelines advocate for AUC-based monitoring; however, using Bayesian programs for AUC estimation poses challenges. We aimed to develop and evaluate a simplified AUC estimation equation using a steady-state trough concentration (C_trough) value. Utilizing 1,034 TDM records from 580 general hospitalized patients at a university-affiliated hospital in Ulsan, we created an equation named SSTA that calculates the AUC by applying C_trough, body weight, and single dose as input variables. External validation included 326 records from 163 patients at a university-affiliated hospital in Seoul (EWUSH) and literature data from 20 patients at a university-affiliated hospital in Bangkok (MUSI). It was compared with other AUC estimation models based on the C_trough, including a linear regression model (LR), a sophisticated model based on the first-order equation (VancoPK), and a Bayesian model (BSCt). Evaluation metrics, such as median absolute percentage error (MdAPE) and the percentage of observations within ±20% error (P20), were calculated. External validation using the EWUSH data set showed that SSTA, LR, VancoPK, and BSCt had MdAPE values of 6.4, 10.1, 6.6, and 7.5% and P20 values of 87.1, 82.5, 87.7, and 83.4%, respectively. External validation using the MUSI data set showed that SSTA, LR, and VancoPK had MdAPEs of 5.2, 9.4, and 7.2%, and P20 of 95, 90, and 95%, respectively. Owing to its decent AUC prediction performance, simplicity, and convenience for automated calculation and reporting, SSTA could be used as an adjunctive tool for the AUC-based TDM.

INTRODUCTION

Vancomycin is a glycopeptide antibiotic effective against Gram-positive bacteria and is considered the first-line therapy for methicillin-resistant Staphylococcus aureus (MRSA) infections (1). Characterized by a narrow therapeutic index and significant interindividual pharmacokinetic variability, vancomycin requires therapeutic drug monitoring (TDM) (2, 3). Although the area under the concentration–time curve (AUC) divided by the minimum inhibitory concentration (MIC) is known as the best predictor of the activity of vancomycin against MRSA (4, 5), the steady-state trough concentration (C_trough) has traditionally been used as a practical surrogate marker for TDM (6). However, recent guidelines in the United States and Japan recommend using the AUC as the preferred indicator for vancomycin TDM (7, 8). This shift toward AUC-based TDM is motivated by its potential to reduce vancomycin-associated acute kidney injury and the limited correlation between AUC and C_trough values (9, 10).

The US and Japanese guidelines advocate the Bayesian method for AUC estimation. However, the Bayesian method requires specialized software and skilled personnel, leading to difficulties in resource allocation and timely reporting (11, 12). Alternatively, US guidelines propose using a first-order pharmacokinetic equation with two concentration values. However, it requires peak concentration (C_peak) measurement, in addition to C_trough, and demands precise time records for drug administration and blood sampling, which burdens patients and medical personnel performing related tasks (13). These resource- and time-related challenges hinder the practical application of AUC-based TDM (14).

Beyond the guideline-recommended methods, AUC also can be estimated using a first-order pharmacokinetic equation with a single C_trough value. This requires estimating the volume of distribution (Vd) or elimination rate constant (K_e) using population pharmacokinetic models (15, 16). Fewel proposed an AUC estimation model (VancoPK) utilizing a first-order pharmacokinetic equation and a newly developed Vd estimation model, and the method was implemented on a website (17). VancoPK has shown good accuracy and precision in multiple external validation studies (18–20). However, using such web tools or Bayesian programs for AUC calculation requires additional work for entering patient demographic information, dosing history, and vancomycin concentration data. For an estimation model to be well-utilized in clinical practice, ease of use is an important factor (21), and reducing such input workload is key to enabling effective translation to clinical practice (11).

Acknowledging these challenges, the authors recognize the need for a new AUC estimation method to promote AUC-based TDM. This study aimed to develop a novel AUC estimation model that utilizes a single C_trough value under intermittent dosing. The input variables and calculation method of the model should be simple enough to be easily integrated into the institution’s information system, while showing comparable AUC estimation accuracy to existing methods.

MATERIALS AND METHODS

Study design and data sets

Three retrospective data sets were used for this study: one for model development and two for validation. For the development data set, we collected medical records from 1,780 TDM consultation accompanied by one–two concentration measurements (TDM occasions) for 812 patients at Ulsan University Hospital (UUH) from March 2020 to July 2023. Adult patients receiving intermittent intravenous vancomycin administration without renal replacement therapy were enrolled, and TDM occasions with appropriate steady-state trough concentration measurements were selected. Briefly, criteria for inclusion were as follows: age ≥18 years; no renal replacement therapy within 1 week before TDM; steady-state conditions with regular dosing (administering the same dose three or more times before TDM); acceptable sampling times for C_trough (concentration sampled within ±1 hour of the scheduled time for the next dose). Additionally, to avoid bias, up to three treatment periods (periods of consecutive vancomycin administration) per patient and three TDM occasions per period were included. Treatment periods in which administration was discontinued more than four times the dosing interval were considered separate treatment periods. Reference AUC24 values for model fitting were calculated using the Bayesian program MwPharm++ (version 2.3.1.136; Mediware a. s., Praha, Czech Republic) with the population model previously developed using the patient data from UUH (22, 23). The performance of the developed model was assessed through internal validation (subject-wise tenfold cross-validation) and further externally validated using two independent data sets: TDM consultation data from Ewha Womans University Seoul Hospital (EWUSH) and published literature from the Mahidol University Faculty of Medicine Siriraj Hospital (MUSI). The same selection criteria as those for the model development data set were applied for the EWUSH data set, except for the criteria for limiting the number of TDM records included per patient. Data extraction and research were approved by the Institutional Review Board of UUH (IRB file No. UUH 2023–07-008) and EWUSH (IRB file No. SEUMC 2023–09-023), and the requirement for patient consent was waived. The detailed data collection and case selection flow is described in Appendix A.

Model development

Assuming drug elimination by first-order kinetics, regular bolus intravenous administration, and steady state, the AUC24 can be expressed as an equation for a single drug dose (dose), Vd, and C_trough (16).

$${\displaystyle \begin{array}{c}AUC24=\frac{Dose\times 24}{Vd\times ln\left(1+\frac{Dose}{\left(Vd\times C_{trough}\right)}\right)}\end{array}\left(1\right).}$$

To determine the Vd used in this formula, we created a Vd estimation model with the actual body weight (BW) as the input variable: $\mathrm{V}\mathrm{d} = {\mathrm{v}}_0+{\mathrm{v}}_1\times \mathrm{B}\mathrm{W}$. The coefficients of the Vd model were obtained using a robust nonlinear regression method. The developed model was named simple single-trough equation for AUC (SSTA). The development process is detailed in Appendix B. We implemented the SSTA within UUH’s hospital information system (HIS) on a pilot basis to demonstrate its ease of application.

Comparative models

We evaluated the three models using a single C_trough value for comparison with the developed model. First, a linear regression (LR) model using the C_trough value as the only input variable was generated from the UUH data. Second, the VancoPK model was utilized, which is based on a first-order pharmacokinetic equation, estimating Vd using age and BW ($\mathrm{V}\mathrm{d}=0.29\times \mathrm{a}\mathrm{g}\mathrm{e}+0.33\times \mathrm{B}\mathrm{W}+11$) and numerically deriving K_e. Third, a Bayesian model using only a single C_trough value (BSCt) was utilized using MwPharm++.

Performance metrics

The percentage error (PE) of AUC prediction was defined as

$${\displaystyle P{E}_i\left(\mathrm{\%}\right)=\frac{\left(y_{i,predicted}-y_{i,reference}\right)}{y_{i,reference}}\times 100,}$$

where y_{i, reference} represents the reference AUC24 value for each evaluation data and y_{i, predicted} represents the predicted AUC24 values from each model. Reference AUC24 values for EWUSH data were determined using a first-order pharmacokinetic equation with peak and trough concentrations. For MUSI data, they were calculated using the linear trapezoidal method with seven concentrations. The performance evaluation metrics included the median PE (MdPE) for bias, interquartile range of PE (IQRPE) for precision, median of absolute PE (MdAPE), and percentage of observations with a PE within ±20% (P20) for accuracy. For the EWUSH data set, agreement based on clinical categories (subtherapeutic, <400 mg∙h/L; therapeutic, 400–600 mg∙h/L; and toxic, >600 mg∙h/L; assuming a MIC of 1 mg/L) was evaluated, and additional scrutiny was performed through stratification analysis to review the factors affecting performance (age, BMI, eGFR [estimated glomerular filtration rate by 2009 Chronic Kidney Disease Epidemiology Collaboration equation], dosing interval, and number of regular dosing before TDM). MdPE exceeding ±10% was considered marked bias. Cases exhibiting outlying errors were also examined. Outlying errors were defined as PE < (Q1–1.5 × IQR) or > (Q3 +1.5 × IQR).

Statistics

Data processing, statistical analysis, and visualization were conducted using the R version 4.3.0 (R Foundation for Statistical Computing, Vienna, Austria). Robust regression parameters of SSTA and LR were estimated using the MM estimation procedure with the robustbase package (version 0.99–1) in R. For each performance metric, 95% confidence intervals (CIs) were estimated using the bootstrap percentile method with 5,000 bootstraps. Additionally, the 95% CIs for the difference in the P20 metric between the SSTA and comparative models were calculated. McNemar’s exact test was used for pairwise comparisons of P20, with the significance level at 0.05. The overall performance of SSTA was considered acceptable if the McNemar’s exact test result is not significant when compared with that of VancoPK and BSCt. The agreement of clinical classification was measured using the overall percentage agreement and linear weighted kappa value. Locally estimated scatterplot smoothing (LOESS) curves were used to aid in the visual exploration of trends in the scatter plots.

RESULTS

Characteristics of enrolled patients and TDM data

The data used for model development (UUH data set) were selected following the criteria outlined after collecting 1,780 TDM occasions from 812 patients. Among them, 1,170 TDM occasions from 580 patients met the target patient population and TDM data suitability criteria. Additionally, two TDM occasions from two patients and 134 occasions from 58 patients were excluded due to exceeding the limit of three treatment episodes and three occasions per episode, respectively. Ultimately, the UUH data set included 1,109 vancomycin concentrations on 1,034 TDM occasions from 580 patients (Fig. 1). For the external validation, the EWUSH data set was collected from 405 patients, resulting in 963 TDM occasions. Finally, 326 TDM occasions from 163 Korean patients were included based on the criteria (Fig. S3). The infusion duration of EWUSH data was consistently recorded as 1 hour for all TDM occasions. The patient characteristics, dosing history, and TDM data are shown in Table 1. The indications for vancomycin and MRSA identification are listed in Table S1.

Fig 1.

Flow chart of patient and TDM data enrolment for the UUH data set. Abbreviations: LoQ, limit of quantification; RRTx, renal replacement therapy.

TABLE 1.

Characteristics of patient demographics, dosing history, and TDM data for model development (UUH data set) and external validation (EWUSH data set)^d

Characteristics	UUHa	EWUSHa
Number of subjects	580	163
Male sex, N (%)	362 (62.4)	95 (58.3)
Age, years	68 (57, 77)	72 (60, 80)
Body weight, kg	60.0 (52.1, 67.7)	57.7 (49.6, 67.7)
Height, cm	163 (156, 170)	163 (155, 170)
Body mass index, kg/m2	22.6 (20.4, 25.3)	22.2 (19.5, 24.7)
Serum creatinine, mg/dL	0.67 (0.52, 0.86)	0.60 (0.46, 0.82)
eGFRb, mL/min/1.73 m2	96.3 (81.9, 107.8)	97.9 (79.7, 115.0)
Number of treatment episodes	647	172
Number of TDM occasions	1,034	326
Single dose, mg	1,000 (800, 1000)	750 (550, 1,000)
Vancomycin dosing interval, N (%)
3 hours	-	2 (0.6)
4 hours	-	14 (4.3)
6 hours	2 (0.2)	58 (17.8)
8 hours	27 (2.6)	51 (15.6)
12 hours	945 (91.4)	179 (54.9)
18 hours	16 (1.5)	-
24 hours	43 (4.2)	22 (6.7)
48 hours	1 (0.1)	-
Infusion duration, N (%)
1 hour	997 (96.4)	326 (100)
1.3 hour	1 (0.1)	-
1.5 hour	6 (0.6)	-
2 hour	30 (2.9)	-
Daily dose per weight, mg/kg/day	30.6 (25.9, 35.9)	31.1 (22.5, 42.9)
AUC24 by the Bayesian method, mg∙hr/L	484.5 (381.9, 591.7)	560.2 (441.8, 684.8)
Number of trough concentrations	1,034	326
Number of non-trough concentrationsc	75	326
Measured trough concentration, mg/L	12.6 (8.8, 17.1)	16.7 (11.5, 21.1)

^a Number (%) or median (Q1; Q3).

^b Calculated by the 2009 CKD-EPI equation.

^c Used concentration data with the sampling time outside of the trough definition (sampled within ±1 h of the scheduled time for the next dose). In the EWUSH dataset, they were all post-distributional peak concentrations (1–3 hours post-infusion).

^d AUC24, area under the concentration–time curve normalized to 24 hours; eGFR, estimated glomerular filtration rate; MRSA, methicillin-resistant Staphylococcus aureus; TDM, therapeutic drug monitoring.

The additional external validation data, MUSI data, were collected from literature sources, including 20 TDM occasions from 20 Thai patients. Their mean age was 63.8 years, mean weight was 58.9 kg, and creatinine clearance was 71.1 mL/min. The average weight-adjusted daily dose was an average of 16.4 mg/kg, with a median dosing interval of 12 hours (range: 8–48 hours). Using seven concentration values to calculate the AUC24 values via the linear trapezoidal method, and the average AUC24 was 643.5 mg∙h/L.

Model development and internal validation

The Vd estimation equation for the SSTA was determined through a robust nonlinear regression analysis as $Vd = 21.9 + 0.53\times BW$. Substituting this into Equation (1) yields the SSTA equation:

$${\displaystyle \begin{array}{c}AUC24=\frac{Dose\times 24}{\left(21.9+0.53\times BW\right)\times ln\left(1+\frac{Dose}{\left(\left(21.9+0.53\times BW\right)\times C_{trough}\right)}\right)}\end{array}\left(2\right).}$$

The LR model, obtained through robust linear regression analysis, was represented as $AUC24=168.3+24.68\times C_{trough}$. The corresponding C_trough values for AUC24 classification thresholds of 400 and 600 mg∙hr/L were back-calculated as 9.4 and 17.5 mg/L by LR, respectively. The internal validation results for SSTA and LR showed a cross-validation MdPE of 0.0 and 0.1%, IQRPE of 5.3 and 10.7%, MdAPE of 2.6 and 5.6%, and P20 of 99.3 and 96.2%, respectively.

External validation—overall performance

External validation was conducted using the EWUSH and MUSI data sets. BSCt could not be evaluated in the MUSI data because information on sex, height, and number of doses per day, which are required for BSCt, was not disclosed in the referenced literature. The performance metrics are listed in Table 2. In the EWUSH data, SSTA showed significantly higher P20 values than LR (4.6% [95% CI: 0.3–8.6%]; P = 0.036). No significant differences were observed with VancoPK (−0.6% [95% CI: –1.8% to 0.6%], P = 0.625) and BSCt (3.7% [95% CI: –0.3% to 7.4%], P = 0.058). In the MUSI data set, SSTA showed a 5% higher P20 than LR; however, the difference was not statistically significant (95% CI: 0%–15%, P = 1.000). There was no difference in P20 with VancoPK. Figure 2 and 3 illustrate the relationship between the reference and predicted AUC24s in the EWUSH and MUSI data sets, respectively.

TABLE 2.

External validation results showing AUC prediction performance of various models^b

Data set (No. of TDM occasions)	Performance metric	Prediction model
		SSTA	LR	VancoPK	BSCt
EWUSH (N = 326)	MdPE, % (95% CI)	−1.5 (-3.2,–0.5)	4.3 (1.9, 6.4)	−2.2 (-3.5,–1.4)	1.5 (-0.2, 2.8)
	IQRPE, % (95% CI)	13.3 (11.2, 15.0)	19.0 (16.7, 21.8)	12.4 (10.6, 14.8)	15.1 (12.8, 17.1)
	MdAPE, % (95% CI)	6.4 (5.7, 7.5)	10.1 (9.2, 11.2)	6.6 (5.8, 7.2)	7.5 (6.6, 8.5)
	P20, % (95% CI)	87.1 (83.4, 90.5)	82.5 (78.2, 86.8)	87.7 (84.0, 91.1)	83.4 (79.1, 87.4)
MUSI (N = 20)a	MdPE, % (95% CI)	0.1 (-5.2, 2.3)	−0.8 (-9.4, 2.7)	−7.2 (-10.6,–2.0)	.
	IQRPE, % (95% CI)	10.4 (4.7, 18.2)	14.2 (7.0, 26.2)	11.1 (5.6, 16.7)	.
	MdAPE, % (95% CI)	5.2 (2.3, 8.7)	9.4 (3.5, 13.6)	7.2 (4.4, 10.6)	.
	P20, % (95% CI)	95 (85, 100)	90 (75, 100)	95 (85, 100)	.

^a Regarding the infusion time, it is described as “The infusion time was 1 and 2 h for doses of 500 and 750–1500 mg, respectively.” in the referenced literature. The time was not specified for the 675 mg dose; however, since it was between the two dose ranges, the infusion time of the case was roughly considered to be 1.5 hours.

^b BSCt, Bayesian approach using a single trough concentration; CI, confidence interval; EWUSH, Ewha Womans University Seoul Hospital (Seoul, Korea); IQRPE, interquartile range of percentage error; LR, linear regression model; MdAPE, median absolute percentage error; MdPE, median percentage error; P20, percentage of estimated AUC24 within 20% of reference AUC24; SSTA, simple single-trough equation for AUC; TDM, therapeutic drug monitoring.

Fig 2.

Relation of reference and predicted AUC24 in the EWUSH data set. Scatter plots comparing reference AUC24 values calculated using the 2-point first-order equation and AUC24 values predicted by various models in the EWUSH data set: (A) SSTA, (B) LR, (C) VancoPK, and (D) BSCt. The points on the graphs indicating the individual TDM occasions are colored according to the dose per BW. The solid line represents the line of identity. Dashed lines indicate clinical cut-off values for AUC24. The gray lines represent LOESS curves to assist in the visual exploration of trends.

Fig 3.

Relation of reference and predicted AUC24 in the MUSI data set. Scatter plots show the relationship between reference AUC24 values and AUC24 values predicted by various models in the MUSI data set: (A) SSTA, (B) LR, and (C) VancoPK. Dashed lines indicate clinical cut-off values for AUC24. The gray lines represent LOESS curves to aid in the visual exploration of trends.

Table 3 presents a contingency table and linear weighted kappa values revealing the concordance between the clinical classification based on the reference AUC24 values obtained using the two-point first-order pharmacokinetic equation and classifications derived from various models using the EWUSH data. The overall percentage agreements were 83.1% for SSTA, 74.5% for LR, 82.2% for VancoPK, and 81.3% for BSCt. The overall percentage agreement for the classification according to the target concentration ranges of C_trough (15–20 mg/L) with the target AUC24 range (400–600 mg∙h/L) was 64.7%.

TABLE 3.

Contingency table showing the agreement between AUC24s by a two-point first-order equation and various models in the clinical categories from the EWUSH data set^a

Clinical category (Ctrough, mg/L; AUC24, mg∙hr/L)	AUC24 by first-order equation			Weighted kappa (95% CI)
	<400	400–600	>600
Ctrough
<15	54	71	5	0.587 (0.526, 0.648)
15–20	0	65	32
>20	0	7	92
AUC24 by SSTA
<400	46	14	0	0.778 (0.722, 0.833)
400–600	8	119	23
>600	0	10	106
AUC24 by LR (corresponding Ctrough)
<400 (<9.4)	40	13	0	0.668 (0.603, 0.733)
400–600 (9.4–17.4)	14	91	17
>600 (>17.4)	0	39	112
AUC24 by VancoPK
<400	46	18	0	0.768 (0.712, 0.823)
400–600	8	116	23
>600	0	9	106
AUC24 by BSCt
<400	45	13	1	0.754 (0.696, 0.812)
400–600	9	105	13
>600	0	25	115

^a AUC24, area under the concentration–time curve normalized to 24 hours; BSCt, Bayesian approach using a single trough concentration; CI, confidence interval; C_trough, trough concentration; EWUSH, Ewha Womans University Seoul Hospital (Seoul, Korea); LR, linear regression model; SSTA, simple single-trough equation for the AUC.

External validation—stratification analysis and inspection for cases with outlying errors in the EWUSH data set

None of the models exhibited marked bias or inaccuracy across the case groups stratified by age, BMI, eGFR, and number of regular doses before TDM (Fig. 4; Table S2). However, SSTA presented a relatively large negative bias with an MdPE of −9.6% in the patients with obesity group (BMI ≥30 kg/m², N = 23). In cases with a dosing interval of 3 hours (N = 2), the SSTA and VancoPK illustrated a marked negative bias and inaccuracy. In cases with a dosing interval of 4 hours (N = 14), the BSCt exhibited a marked positive bias and inaccuracy.

Fig 4.

Relation of various factors and percentage error of prediction models in the EWUSH data set. Error distribution in patient groups divided according to the factors affecting vancomycin pharmacokinetics is presented as boxplots: (A) Age, (B) BMI, (C) eGFR, (D) number of regular dosing before TDM, and (E) dosing interval.

The PE distribution of SSTA had a skewness of −0.477 and a kurtosis of 3.73. There were 22 cases (6.7%) with outlying errors (defined as PE <−28.6% or >24.4%), including 15 cases with negative errors and seven cases with positive errors. There were 34 cases with PE <-20% and eight cases with PE >20%. Cases exhibiting outlying errors in the SSTA also demonstrated a PE <-20% or >20% in VancoPK (Table S3).

Implementation of SSTA on HIS

Figure 5 shows a sample screenshot of the HIS at UUH, showing the SSTA application in calculating the AUC and providing the recommended doses. As the input values, including vancomycin concentration, single dose, and BW, are entered, the system calculates the AUC24 value according to SSTA and automatically computes the recommended dose targeting the AUC24 of 500 mg∙h/L, which is based on a proportional relationship between the AUC and dose. The recommended dose is a single dose when maintaining the current dosing interval. If one wishes to change the dosing interval, they can convert it to a daily dose, then divide it by 24, and then multiply by the desired dosing interval value to determine the new single dose.

Fig 5.

Annotated screenshot of the test result view of the EMR at UUH as an example of SSTA application in calculating AUC24 and providing recommended doses. When input values including measured vancomycin concentration, body weight (entered at the time of prescription through OCS or retrieved from the EMR’s most recent value), current dose (entered at the time of prescription through OCS), and target AUC24 (entered at the time of prescription through OCS or default value) are entered, output values would be calculated and reported.

DISCUSSION

Herein, we developed the SSTA, which streamlines the AUC24 calculation by incorporating only a single steady-state C_trough, BW, and dose as input variables. We evaluated its performance by comparing it with other C_trough-based AUC estimation models. In the external validation results using the EWUSH data set, the SSTA demonstrated superior performance to LR in all metrics, with a statistically significant increase of 4.6% in P20. SSTA showed a 0.6% decrease compared to VancoPK and a 3.7% increase compared to BSCt in P20; however, these differences were not statistically significant. In the MUSI data set, SSTA outperformed LR and VancoPK across all metrics. Overall, SSTA exhibited greater performance than LR and showed comparable performance to those of other sophisticated models, such as VancoPK and BSCt, which require more input variables. Moreover, by implementing in HIS, the calculation process of the SSTA can occur seamlessly without significant human effort. The implementation difficulty of SSTA would be similar to eGFR reporting, along with serum creatinine concentration, which is already widely implemented (24). Just as equations for the eGFR turn creatinine test results into a more clinically valuable index of renal function, SSTA could allow clinical laboratories to report vancomycin concentration values as more meaningful AUC values.

The criteria for determining the clinically acceptable level of AUC estimation performance are not entirely clear. However, theoretically, to achieve the target range of 400–600 mg∙h/L when administering a dose targeting 500 mg∙h/L, the AUC estimation error should be within 20%. This margin is sometimes considered an acceptable threshold for precision or accuracy (7, 25, 26). The P20 of the SSTA in the two external validation cohorts was 87.1% and 95%, indicating that a significant proportion of TDM occasions fell within acceptable error margins. However, caution is warranted due to the fat-tailed distribution of PE. Extreme negative errors could lead to an underestimation of the AUC and potential overdosing. Since the SSTA model does not consider infusion time, bias may occur depending on the dosing interval and infusion time. Additionally, the Vd model with actual BW as the only variable did not correct for the fat mass impact. Consequently, in the stratification analysis, the SSTA showed relatively large errors in the obesity group (BMI ≥30 kg/m²) and in patients with a dosing interval of 3 hours. Moreover, patient factors such as severe burns, trauma, septic shock, and critical illness, which can affect vancomycin pharmacokinetics (27, 28), may also impair the performance of the SSTA. For these extreme patients, two-point TDM should ultimately be performed. However, for general patients, the SSTA, with its focus on convenience, can be expected to play a complementary role.

In our study, the target range for C_trough corresponding to the AUC24 target range (400–600 mg∙hr/L) by LR was 9.4–17.4 mg/L. This was lower and broader than the existing target range of 15–20 mg/L in the 2009 US vancomycin TDM guidelines; it is consistent with previous studies, revealing a relationship between AUC24 and C_trough values (29, 30). The SSTA exhibited superior performance in all quantitative metrics compared to LR. Regarding concordance based on the clinical classifications of AUC24, SSTA showed higher concordance with the two-point first-order equation than LR or the traditional C_trough target range of 15–20 mg/L. Moreover, when implemented in the institution’s information system, the calculation process of SSTA can occur seamlessly without requiring significant human effort, providing immediacy and convenience equivalent to conventional TDM. There is no reason to not use SSTA instead of conventional trough-based TDM or a recalibrated trough target for dosing appropriate for the AUC target.

VancoPK exhibited better IQRPE and P20 values than SSTA on the EWUSH data set. However, considering the marginal difference in metrics and the difficulty of obtaining accurate time information in a clinical setting (31), the advantages of VancoPK may diminish. Furthermore, the calculation process for VancoPK requires a numerical solution to obtain K_e. There is a trade-off between computational cost and application difficulty versus prediction accuracy. However, while the performance difference between SSTA and VancoPK appears insignificant, SSTA has a clear benefit regarding computational cost and application difficulty.

The Bayesian approach is considered the preferred method for pharmacokinetic analysis with limited sampling (32). However, BSCt showed less favorable performance metrics than SSTA in our study. Considering that the evaluation using the EWUSH data was based on the reference value, using only two vancomycin concentrations and a one-compartment model, the alignment with the two-compartment model used in BSCt may be compromised. The suitability of a population pharmacokinetic model could also be a concern. However, previous studies have reported clinical classification concordance of AUC24 using Bayesian methods with a single C_trough in the range of 71.4%–76.8% (33, 34). Thus, the 81.3% concordance observed in the BSCt model was not sufficiently low to question its validity. Additionally, the reported IQRPEs range from 9% to 24% depended on the Bayesian programs when using a single C_trough (35). The IQRPEs for BSCt was 15.1% in the EWUSH data set, indicating moderate performance.

One major limitation was that the reference AUC for model development was generated using Bayesian methods based on one or two vancomycin concentration values. There is a potential for bias in the reference AUC for model fitting. Another limitation arose from the diverse dosing conditions and patient groups that were incompletely represented in the external validation data. Therefore, caution is required when applying the SSTA model in cases where the dosing interval is <6 hours or >24 hours, resulting in extremely lower or greater intervals compared to the standard 12 hours, or in cases that have not been thoroughly evaluated, such as in a patient with severe obesity or pediatric patients, and with a sampling time outside 1 hour from the next scheduled dose time.

In conclusion, SSTA performance was superior or comparable to that of other C_trough-based AUC estimation models, and its application to an institution’s information systems was convenient, allowing easy automated calculation. Therefore, SSTA could be a replacement for conventional trough-based TDM, and it could complement other AUC estimation methods. It could serve as a practical solution to the implementation challenge of AUC-guided dosing, especially in institutions facing resource constraints.

DATA AVAILABILITY

The data for the SSTA development are openly available in figshare at https://doi.org/10.6084/m9.figshare.26494405.v1.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/aac.00699-24.

Supplemental material. aac.00699-24-s0001.docx.

Appendix A, data collection and case selection protocol; Appendix B, simple single-trough equation for AUC (SSTA) derivation and equation characteristics; Appendix C, additional figures and tables.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.