Vancomycin equations and pharmacokinetic background

1. What the equations represent

The calculator offers conventional first-order PK estimation and five population models with optional Bayesian maximum a posteriori (MAP) fitting. Both estimate vancomycin clearance in L/h, then evaluate an unchanged repeating maintenance regimen. The principal reported AUC24 is steady-state exposure normalized to 24 hours. A first-dose result is separately labeled.

The selected AUC24 target is 400, 450, 500, 550, or 600 mg*h/L. The clinical 400-600 range primarily concerns serious invasive MRSA with broth-microdilution MIC 1 mg/L. MIC changes the displayed AUC/MIC ratio; it does not automatically change the exposure target. Raising exposure above 600 solely to reach a ratio of 400 is not the calculator's dosing rule. [1]

A population estimate has no fitted patient concentration. One Bayesian level updates clearance while retaining the model's distribution parameters. Two levels can update clearance and central volume, except that Tsai retains fixed central volume. Conventional single-level estimation fixes Vd; conventional two-level estimation derives Vd and elimination from the sampled decline.

The public form reconstructs 1-30 identical regularly spaced doses and up to two concentrations, with times measured from the latest infusion start. It does not represent arbitrary missed, delayed, partially infused, or unequal doses. A first-dose workflow assumes no residual vancomycin before that dose. Changing renal function, changing support, pregnancy, pediatric use, intermittent HD, SLED, peritoneal dialysis, ECMO, and temporary extracorporeal VAD remain outside the implemented pathways. Selecting a steady-state workflow does not establish that steady state occurred.

The equations below describe the implemented mathematics. Population applicability, original-source transcription, computational agreement, and clinical validation are separate questions. The model-specific evidence gaps remain stated in section 10.

2. Symbols and units

The last renal-unit conversion only changes units. It does not make CrCl equal to vancomycin CL. Numeric conversion and calculation use the accepted full input precision. A decimal comma such as 1,5 is rejected; use 1.5.

3. Body size and renal estimates

Conventional body-size calculations

Let H be height in inches. The height corrections below 60 inches and amputation factors are retained calculator conventions. The named Devine framework supplies the 50/45.5 kg base and 2.3 kg/inch increment. [3]

Percent of IBW expresses the weight threshold in familiar terms; the calculation compares weights directly. The 0.4 adjusted-weight factor has published renal-estimation precedent, but this does not validate every threshold in this calculator. [4]

The conventional factor modifies IBW before its renal weight-selection rules. It does not multiply actual weight for Vd or loading-dose calculations, and it does not modify Bayesian body-size calculations.

Bayesian IBW is max(base, base + 2.3 * (H - 60)), with base 50 kg for male and 45.5 kg for female. It therefore retains the base below 60 inches. Bayesian AdjBW uses that IBW. The conventional 120%-of-IBW rule and Bayesian BMI rule are intentionally described separately.

Cockcroft-Gault and model-specific limits

Age is years. The first result is mL/min; the indexed result is mL/min/1.73 m². The original female correction is a 15% reduction. The PDF's denominator 85 is an approximation; the program retains the explicit 0.85 multiplier. [2]

These limits describe the software's retained preprocessing. They are not universal recommendations to cap renal function. No automatic creatinine floor or creatinine rounding is added. Measured conventional workflows derive elimination from levels; their CrCl still informs interval preference. The Bayesian preprocessing also computes an adjusted-weight comparison estimate limited to 1-250 mL/min, but that value is not substituted for the named covariates above.

CKD-EPI 2009 covariate retained for Zhang

This is the retained model covariate, not a replacement for a laboratory's current reported eGFR equation. The original CKD-EPI expression includes the female multiplier. The application does not apply a race factor. The Zhang supplementary table differs on the female factor; exact investigator covariate reproduction remains unresolved. [5] [12]

4. Conventional PK background

A one-compartment model treats distribution as an apparent volume Vd and elimination as a first-order decline once the sampling assumptions are appropriate. Apparent volume is a proportionality parameter, not an anatomical fluid volume. Under constant linear PK, a concentration falls by the same fraction over equal time intervals.

Empiric and current-regimen estimates without levels

The Vd factor is selectable from 0.2 to 0.95 L/kg in 0.05 increments. It remains an assumption in no-level and conventional single-level workflows. The elimination regression is the Matzke relationship: Figure 2 of the original 1984 article reports the same slope and intercept. That study's separate vancomycin-clearance regression and its historical dosing nomogram are not substituted into this pathway. [6]

The supplied background page describes several historical Vd and clearance approaches. Values such as 0.62, 0.7, or 0.9 L/kg reflect different source populations and assumptions. They do not all become simultaneous corrections to one patient's Vd. The program keeps one explicitly selected conventional Vd factor, or uses the measured-level Vd calculation when appropriate. [7]

Finite infusion

A dose takes time to enter the body. Using an instantaneous-bolus equation in place of a finite infusion can change estimated k, Vd, peak, and trough. For a single infusion starting at t = 0 with no residual concentration:

At repeated-dose steady state, residual concentration from earlier doses contributes to the next interval:

T must be shorter than tau. The denominator 1 - exp(-k * tau) accounts for accumulation from previous identical doses. Cmax,ss is the mathematical end-of-infusion concentration, which must not be confused with a later postdistributional measured sample. These finite-infusion relationships are also present in the supplied teaching PDF. [9]

5. One concentration at steady state

The conventional single-level workflow holds Vd at TBW times the selected factor and solves for a positive k such that the concentration predicted by E9 equals the measured concentration at its entered time. It uses the actual entered infusion duration and time from infusion start, including the during-infusion expression when that timing is accepted. Distribution and sample-quality warnings still apply.

The numerical search scans k from 0.001 to 1.5 h^-1 on 400 logarithmic steps, finds a sign change, and bisects up to 100 iterations. Failure to reconcile the concentration is reported. This is a model equation solve with an assumed volume, not independent identification of both clearance and volume.

The legacy Ambrose-Winter expression k = ln(1 + D/(Vd * Ctrough)) / tau follows from an estimated bolus peak of Ctrough + D/Vd. The supplied Word sheet uses entered elapsed time in place of tau. The advanced calculator already uses E9-E10 instead of that approximation. The distinction matters for finite infusions and samples that are not true troughs. [8]

6. Two concentrations at steady state

C1 and C2 must describe a declining postinfusion pair from the same interval without an intervening dose. Both times are measured from the infusion start. Require C1 > C2, T < t1 < t2, and t2 no later than tau. The software requires at least 15 minutes between samples as a duplicate-sample safeguard; that is not proof of adequate clinical separation or completed distribution.

Back-extrapolation to the earlier end-of-infusion time increases the concentration; forward extrapolation toward the next dose decreases it. This sign convention is important when manually checking a result.

At steady state the trough before the analyzed dose equals the trough extrapolated before the next identical dose. This equality gives the second denominator. The program uses the entered T, preserving actual administration history rather than imposing the Word sheet's one-hour assumption.

Linear infusion and logarithmic elimination cross-check

The logarithmic mean approaches the concentration itself when the endpoints coincide. The infusion segment is approximated as a straight line, so this cross-check need not equal exact D24/CL. Agreement does not establish accurate timing, steady state, or stable physiology.

7. Two concentrations after a first dose

This conventional workflow requires zero residual vancomycin before one infusion and no intervening additional dose. It uses the same two-level slope and end-of-infusion back-extrapolation in E11, but sets the pre-dose residual to zero when calculating volume.

The accepted infusion-duration range ends before hour 24. E15 describes that single dose only; it does not add future doses or a loading-plus-maintenance course. A separate first-dose cross-check replaces AUC0-T with the triangular area Cmax * T / 2 and retains the exact postinfusion exponential segment through hour 24.

The first-dose AUC0-24, single-dose AUC0-infinity, and repeated-regimen AUCss24 are different quantities. This is the retained one-compartment first-dose method. It does not implement the later published alpha-distribution-phase compensation method. [16]

8. AUC and average concentration

For any retained model with linear constant clearance, mass balance gives the following steady-state relationships. They apply across one, two, and three compartments.

Caverage,ss is an interval-average concentration in mg/L. It is neither the trough nor the end-of-infusion peak. For intervals longer than 24 hours, D24 and AUCss24 are normalized averages for the repeating regimen; exposure in one particular calendar day can differ. The PDF's average-concentration equation is thus directly useful and equivalent to the implemented AUC calculation. [9]

For a fixed dose, interval, and CL, changing infusion duration changes peak and trough but leaves AUCss24 and Caverage,ss unchanged. If actual historical T is changed before refitting measured levels, the fitted CL may change and therefore AUC can change. That is a different operation from changing a future infusion policy.

The proportional-dose identity is background algebra. The application evaluates explicit candidate or custom doses and reports their recalculated exposure; it does not treat a selected target as proof of the final regimen's actual AUC. Display rounding does not feed back into dose calculation.

9. Infusion duration policies

Choose the policy before calculating generated maintenance options and the mathematical exact-target regimen. The choice is recorded with the input snapshot and calculation report. Enter the actual administered duration for historical doses; a proposed default does not rewrite history. A clinician-entered custom duration remains explicit.

Default rate policy

This policy uses at least 60 minutes and no faster than 10 mg/min, then rounds up to a 15-minute increment. For example, 1000 mg produces T = 1.75 h. Its basis is the reviewed Hospira IV powder label. Product labeling and patient tolerance still apply. [17]

Supplied dose-based schedule

The continuous boundaries also cover unrounded mathematical doses: 1250 mg remains in the one-hour band, while any dose above 1250 mg moves to the 1.5-hour band. The top-band choice resolves the attachment's 2.5-3-hour range without guessing a single duration for every use.

The supplied schedule can be faster than the reviewed label's rate or shorter than its 60-minute minimum. For example, 1250 mg over 1 h is about 20.83 mg/min. The application therefore retains a separate administration review against E18 even when the alternative schedule is selected. The schedule is a supplied option for review against the applicable product and institutional policy; it is not labeled as the default label's dosing instruction.

Administration review compares the entered or generated T with max(1, D/600), before quarter-hour rounding. A duration below this minimum generates a review warning. Such a warning does not silently extend the infusion or change the represented concentration history.

10. Five population priors

These are the population equations before any MAP update. CL and Q are in L/h; volumes are in L; TBW is kg. CrCl and eGFR mean the exact pathway-specific covariates in section 3. Automatic selection uses CRRT first, then BMI at least 30, then critical illness, then the general adult prior. Belabbas requires clinician selection and documented hematologic malignancy.

Goti 2018

The application retains the non-dialysis pathway. The original study included HD, but this does not make this implementation an HD calculator. The detailed final-model table and the body of the 2019 erratum were not obtained in the prior audit; full coefficient and residual-semantics re-verification remains incomplete. [10]

Roberts 2011

The original study used continuous infusion and measured BSA-indexed 24-hour urinary CrCl in septic ICU adults. The application substitutes indexed Cockcroft-Gault and predicts intermittent infusions. It also uses the residual likelihood described in section 12 instead of claiming exact reproduction of the source's exponential-plus-additive error equation. Structural coefficients and reported CV/SD values were checked against primary Table 2. [11]

Zhang 2023

The calculator requires BMI at least 30 although the source includes overweight adults. Table 2 centers weight at 90 kg; the narrative's 100 kg description differs. The code follows the table and retains the female CKD-EPI factor discussed above. The weight-factor lower bound is a software safeguard. RRT and the paper's separate two-loading-dose strategy are not implemented by this pathway. [12]

Tsai 2024 CRRT

The source models time-varying CRRT and temporary mechanical support. The application retains fixed CRRT and the no-tMCS Q branch. Vc remains fixed even with two levels. Effluent, urine output, modality, downtime, and settings are reviewed but are not numerical terms in E23. The residual 1.39 value is interpreted as SD using the source software's convention; an author model file was not obtained. [13] [15]

Belabbas 2023

Documented hematologic malignancy is required. The primary abstract confirms structure, named covariates, and typical CL/Vd; the detailed exponent, modifiers, preprocessing, and variability table remain incompletely verified against full primary text. [14]

Eligibility checks and mathematical execution do not establish local external validity. Full author control streams, all original covariance structures, prospective patient validation, and uncertainty calibration have not been reproduced for this exact implementation.

11. Compartment and infusion predictions

Each Bayesian prior retains its one-, two-, or three-compartment structure during fitting and prediction. The conventional result uses one compartment. For compartments with drug amounts Ai, central concentration is A1/Vc. Elimination occurs from the central compartment; peripheral compartments exchange drug with it.

R(t) is the sum of active infusion rates. With one compartment the sum terms are absent. With two compartments there is one peripheral volume Vp and one Q; with three, the peripheral pairs are V2/Q2 and V3/Q3.

Disposition terms used by the engine

The code transforms the compartment matrix to a symmetric matrix S, then obtains its eigenvalues and normalized eigenvectors. In the following, q1j is the central component of eigenvector j, muj is its negative eigenvalue, and V1 is Vc.

For one compartment, lambda = CL/Vd and a = 1/Vd. For two compartments this is the familiar sum of two exponentials; three compartments produce three terms. These coefficients refer to a unit-dose response, so a general bolus expression with coefficients already multiplied by dose must not be substituted without adjusting units.

E27 explains the two-exponential background in the PDF; the engine evaluates E26 for all structures. The terminal half-life is ln(2) / min(lambda). It is not generally ln(2)/(CL/Vc) when peripheral distribution exists.

Finite infusion and dose-history superposition

For an infusion event starting at time s, let u = t - s. Its concentration contribution is zero at or before its start. Otherwise, integrate the unit-dose disposition over the active infusion. The index j below labels disposition terms within that event:

This event sum allows Bayesian fitting before steady state, provided every represented dose and elapsed sample time is accurate. The public regular-dose form sets starts at 0, tau, 2*tau, through (n-1)*tau. A level entered t hours after the latest start is placed at (n-1)*tau + t. A chart evaluates the same event function; it is not the source of the analytic AUC calculation.

Integrating the full unit-dose response gives sum(a/lambda) = 1/CL. Consequently a single complete IV dose has AUC0-infinity = D/CL even in multiple compartments. This is a mass-balance identity, not an instruction to administer an IV bolus.

12. Bayesian MAP and fit diagnostics

MAP estimates the supported log-scale parameter effects conditional on the selected population model and the observed concentrations. The engine fits CL with one level and CL/Vc with two levels, except that Tsai has no fitted Vc effect. Peripheral parameters stay fixed.

Constants independent of the fitted effects are omitted. The logarithmic variance term matters when residual variance depends on the predicted concentration. The two fitted prior effects are treated as independent; no prior CL/Vc covariance term is included. The local fitted covariance, when available, is calculated separately.

On narrow screens, scroll the variability table horizontally to read all columns.

Omega entries are variances, not standard deviations. Zhang uses the patient additive-error branch. Roberts' likelihood is an implementation approximation; Tsai's SD interpretation and Goti/Belabbas full-table gaps are described above. [10] [11] [12] [13] [14]

Search bounds and review gates

The bounded Nelder-Mead search uses etaCL from -2.3 to 2.3 and, when fitted, etaVc from -1.8 to 1.8. Valid fitted parameters require 0.03 < CL < 40 L/h and 2 < Vc < 800 L. One-effect fitting uses starts 0, -1, and 1; two-effect fitting uses all nine pairs on the -1, 0, 1 grid. Each search permits up to 600 iterations. The best valid objective found is retained with deterministic handling of near ties. These operating limits and search heuristics do not prove a global optimum or a clinically plausible fit.

Nonconvergence, an invalid objective, a parameter boundary, or a nonfinite or absolute standardized residual above 3 withholds proposed primary, exact, custom, and candidate regimens. Entered-history exposure and fitted parameters may remain visible for diagnostic review. The limits are software review gates, not validated toxicity or physiology thresholds. Measured conventional estimates reuse the CL/Vd envelope without clamping the derived result.

Separated converged solutions within 2 J units and more than 0.1 apart in log-parameter distance are disclosed as competing fits. They withhold a local uncertainty range because a single local curvature estimate cannot describe separated solutions. The selected point is not automatically withheld solely for that ambiguity.

13. Approximate ranges and model comparison

When supported, the uncertainty calculation approximates the local curvature of E31 in log-parameter space. It is conditional on the selected model, dose history, measured values, and fixed distribution parameters.

Centered finite differences use steps 0.01 and 0.005. The calculation checks positive curvature, a Hessian condition number no greater than 1000000, and no more than 5% relative SD disagreement between steps. A separate gradient uses [J(x+0.001*e_i) - J(x-0.001*e_i)] / 0.002; its absolute product with the parameter's log SD must be no greater than 0.01. Fit diagnostics, boundaries, and competing minima are also checked. A failed range is reported as unavailable rather than converted into a narrow interval.

Without concentrations, the separate population-prior spread uses sqrt(omega) in E34. That spread is not individualized posterior uncertainty. No statistical range is calculated for conventional estimates. Neither type of Bayesian range is a calibrated 95% confidence or credible interval for this implementation.

AUC ranges use one unchanged entered current regimen, or a clinician custom regimen in an empiric case. They are not attached to an automatically retargeted exact dose: retargeting D24 = target * CL would algebraically hide clearance uncertainty. No fixed regimen means no fixed-regimen AUC range.

Each eligible comparison uses the same patient, administration history, concentrations, and fixed regimen. It refits that prior when concentrations exist and displays failures or exclusions. The program neither averages the priors nor selects a new model from the most favorable AUC. Disagreement identifies a reason to review model suitability and data.

14. Candidate, exact, custom, and loading doses

Candidate grid used by the advanced calculator

Standard options use doses in 250 mg increments and intervals of 8, 12, 24, 36, or 48 h. The grid is screened using these retained software conventions:

The covariate is conventional CrCl for conventional results; the selected prior's specified CrCl or indexed eGFR for Bayesian results. These preferences are program heuristics, not a universal dosing nomogram.

The advanced module first ranks candidates within 400-600 mg*h/L ahead of candidates outside that range, then sorts by score and returns up to six. Floating-point tolerance at the range boundaries is 0.00000001 mg*h/L. A selected AUC target therefore does not guarantee that a generated dose lands exactly on it. If the grid contains no in-range option, that is explicitly reported.

Exact and custom regimens

The mathematical exact-target dose uses E17 with the leading candidate's interval, or the retained 12-hour fallback interval when no candidate is available. Its infusion time follows the selected policy. It remains an unrounded mathematical dose subject to evaluation limits, rather than an administration order.

A custom dose of 100-4000 mg, interval of 4-72 h, and T of 0.5-8 h, with T shorter than tau, is evaluated using the fitted or population PK estimate. Entering a future custom regimen does not refit the patient to that future regimen. E16 and E29 give its exposure and concentrations. Both its AUC and administration status must be read.

Separate loading reference

The retained reference uses 20-25 mg/kg for CRRT, obesity, and general cases when loading is indicated; 20-35 mg/kg applies to critically ill cases without obesity. It caps at 3000 mg and remains separate from maintenance AUC calculation. The CRRT range is tied to the retained model simulation and local review; it does not implement dynamic CRRT clearance. [1] [13]

15. Worked arithmetic examples

These synthetic examples demonstrate the equations and timing conventions. They are not patient-specific prescribing recommendations.

One repeating regimen with known parameters

Let CL = 4 L/h and Vd = 50 L. Then k = 0.08 h^-1 and half-life is about 8.66434 h. For D = 1000 mg every 12 h infused over 2 h:

Using the unrounded C1 and C2 recovers k = 0.08, Vd = 50, and CL = 4 with E11-E12. The first sample is one hour after the two-hour infusion ends. The slight cross-check difference arises from its linear infusion approximation.

The same single dose without prior drug

With zero prior concentration, the single 1000 mg dose over 2 h has Cmax,1 about 18.482026 mg/L, AUC0-infinity = 250 mg*h/L, and exact AUC0-24 about 210.253279 mg*h/L. Those values describe a single dose. They differ from the 500 mg*h/L repeated-maintenance AUCss24 above.

Uncertainty must keep the dose fixed

For the same D24 = 2000 mg/day, a CL range of 3-5 L/h maps to an AUC range of 400 to about 666.67 mg*h/L. The inversion reverses the endpoints. Choosing a different exact dose at each endpoint would force both AUCs back to the selected target and would conceal the exposure uncertainty.

16. How the supplied references are used

The supplied references support the conventional background and provide useful equivalent forms of implemented equations. Their historical settings do not all match this advanced version. The following crosswalk makes those differences explicit.

On narrow screens, scroll the comparison table horizontally to read both columns.

Useful historical algebra

The first two are rearrangements of E9. The third is an algebraic inverse, with no claim to measured renal function. They are documented so a manual comparison is understandable, while the active generation rules remain those in section 14. The general PDF also has layout and shorthand limitations; its "during infusion t = T" annotation describes the end of infusion, while the general during-infusion interval is 0 < t <= T.

Reference access: University of Florida equation PDF; GlobalRPh pharmacokinetic background; GlobalRPh pharmacokinetic overview; GlobalRPh single-level background.

17. Precision and numerical evaluation

Visible values remove unnecessary trailing decimal zeros after display rounding: 65.00 becomes 65, 1.5000 becomes 1.5, and 1750.000 becomes 1750. Significant nonzero digits remain subject to the display precision of that quantity. Full internal precision is retained for subsequent arithmetic, fitting, and comparisons.

The conventional two-level implementation evaluates a near-unity concentration ratio with log1p, and small 1 - exp(-x) expressions with -expm1(-x). These are equivalent arithmetic forms that reduce subtraction error. The first-dose infusion area is evaluated without subtracting two large nearly equal total areas.

The compartment engine's eigenvalue calculations and the MAP optimizer are numerical algorithms. The source recovery package contains the executable implementation and regression evidence. Read a result's fit status and withheld-regimen status before interpreting precision as certainty.

18. Linked references

The numbered references support their named methods and clinical context. They do not endorse every implementation setting. The Word equation sheet and infusion image were supplied for this revision; their crosswalk is in section 16.

1. Rybak MJ, Le J, Lodise TP, et al. Revised consensus guideline for vancomycin monitoring in serious MRSA infection. Am J Health Syst Pharm. 2020;77:835-864. DOI 10.1093/ajhp/zxaa036. IDSA guideline page; ASHP full guideline PDF.
2. Cockcroft DW, Gault MH. Prediction of creatinine clearance from serum creatinine. Nephron. 1976;16:31-41. DOI 10.1159/000180580; PubMed.
3. McCarron MM, Devine BJ. Clinical pharmacy case studies, case number 25, gentamicin therapy. Drug Intell Clin Pharm. 1974;8:650-655. DOI 10.1177/106002807400801104. Historical IBW attribution; below-60-inch and amputation conventions are identified separately.
4. Winter MA, Guhr KN, Berg GM. Impact of various body weights and serum creatinine concentrations on Cockcroft-Gault bias and accuracy. Pharmacotherapy. 2012;32:604-612. DOI 10.1002/j.1875-9114.2012.01098.x; PubMed.
5. Levey AS, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150:604-612. PubMed; Full primary text.
6. Matzke GR, McGory RW, Halstenson CE, Keane WF. Pharmacokinetics of vancomycin in patients with various degrees of renal function. Antimicrob Agents Chemother. 1984;25:433-437. DOI 10.1128/AAC.25.4.433; Original PDF. Figure 2 on page 436 provides the retained elimination regression.
7. GlobalRPh. Pharmacokinetics background and Pharmacokinetic overview. Historical textbook-derived background and comparison equations. Different drug-specific regressions and older treatment targets are not automatically adopted.
8. GlobalRPh. Vancomycin single-level analysis advanced. Ambrose-Winter background; see Ambrose PJ, Winter ME, Vancomycin, in Basic Clinical Pharmacokinetics, 4th ed., 2004:451-476. The exact finite-infusion solve in this application is identified separately.
9. University of Florida College of Pharmacy. Useful pharmacokinetic equations, five-page teaching reference. Supplied copy reviewed visually; the document includes equations for other drugs and routes.
10. Goti V, et al. Hospitalized patients with and without hemodialysis have markedly different vancomycin pharmacokinetics. Ther Drug Monit. 2018;40:212-221. Original DOI; PubMed. 2019 erratum; erratum record. Full-table and erratum-body gaps remain disclosed.
11. Roberts JA, et al. Vancomycin dosing in critically ill patients: robust methods for improved continuous-infusion regimens. Antimicrob Agents Chemother. 2011;55:2704-2709. DOI; Full primary text.
12. Zhang T, et al. How to dose vancomycin in overweight and obese patients with varying renal (dys)function in the novel era of AUC 400-600 mg*h/L-targeted dosing. Published online 2023; Clin Pharmacokinet. 2024;63:79-91. DOI; Full primary text; Supplement.
13. Tsai MT, et al. Population pharmacokinetics of vancomycin in intensive care patients with time-varying temporary mechanical circulatory support or CRRT. Infect Dis Ther. 2024. DOI; Full primary text.
14. Belabbas T, et al. Population pharmacokinetic model and dosing optimization of vancomycin in hematologic malignancies with neutropenia and augmented renal clearance. J Infect Chemother. 2023;29:391-400. DOI; Primary abstract. Detailed full-table verification remains incomplete.
15. Certara. Phoenix Model Object and residual-error conventions. The documented additive standard-deviation convention supports the stated Tsai interpretation as an inference; it does not replace an author model file.
16. Sujjavorakul K, et al. Predicting AUC for first-dose vancomycin using first-order pharmacokinetic equations. Antibiotics. 2023;12:630. DOI; PubMed. The distribution-compensated method is distinct from the retained first-dose implementation.
17. Hospira, Inc. Vancomycin hydrochloride powder for IV solution. DailyMed product label, updated April 1, 2026. Basis of the conservative default and its separate administration comparison.