A plain-and-precise account of every formula, coefficient, and clinical rule the vancomycin dosing tool uses, laid out so you can check each one against its source.
This tool runs fixed equations only — no AI, no machine learning, ever. The first-dose engine uses five published population PK models (VancoPK 2021, Buelga 2005, Adane 2015, Roberts 2011, Masich 2020). The level-based methods use standard textbook equations. All 9 methods were checked against 33 worked scenarios. It has not been tested in a clinical trial. Read Section 7 (Known limitations) in full.
'Notation: TBW = total body weight; IBW = ideal body weight; AdjBW = adjusted body weight; LBW = lean body weight; BSA = body surface area; SCr = serum creatinine (mg/dL); CrCl = creatinine clearance (mL/min); Vd = volume of distribution (L); CL = clearance (L/hr); Ke = elimination rate constant (hr⁻¹); AUC24 = 24-hour area under the curve (mg·hr/L); τ = dosing interval (hr); tinf = infusion duration (hr).
Devine 1974 in metric units (50 or 45.5 kg + 2.3 kg/in over 60 in; 2.3÷2.54 = 0.91 kg/cm). Devine BJ. Drug Intell Clin Pharm 1974
Standard 0.4 correction. Used as the dosing/CrCl weight when obese (TBW/IBW > 1.2). Bauer, Applied Clinical Pharmacokinetics
Normalizes CrCl to 1.73 m² in the Adane and Roberts models. Mosteller RD. N Engl J Med 1987;317:1098
Feeds LBW and the model-eligibility labels (Adane BMI > 40, Masich BMI > 30).
Each empiric model applies CG with its own weight rule, its own SCr, and its own cap — not interchangeable, transcribed exactly from each published model:
| Model | Weight in CG | SCr used | Cap / normalization |
|---|---|---|---|
| VancoPK | AdjBW if obese, else TBW | raw SCr | min(CrCl, 200) |
| Buelga | AdjBW if obese; TBW if TBW<IBW; else IBW | IDMS-adj* | capped 120 in CL step |
| Adane | TBW | raw SCr | BSA-normalized to 1.73 m² |
| Roberts | AdjBW if obese, else TBW | raw SCr | min(CrCl,120), then /BSA×1.73 |
| Masich | AdjBW if obese, else TBW | raw SCr | min(CrCl, 120) |
*IDMS-adjusted SCr (Buelga only): SCr_adj = SCr × 1.065 + 0.067. SCr is floored to 0.6 mg/dL before CG to stop clearance being over-read in frail/elderly patients — see Section 6.
When no drug level has been drawn, the tool estimates exposure from population averages. Five published models; each has its own Vd and CL equation. For every model:
Concentrations use a one-compartment, first-order, steady-state intermittent-infusion model (Section 3). The five model spans are shown side-by-side so divergence is visible.
Every method below produces a Vd and a Ke, then feeds them into this same core to get Cmax, Cmin and AUC24.
Volume fixed from the population regression; elimination rate is the single unknown, solved by bisection so the model reproduces the one measured steady-state trough. log-linear / 2020 ASHP–IDSA
This is the reference-standard method — both Ke and Vd are truly measured from the patient's own two levels, with no population assumption. Sawchuk RJ, Zaske DE. J Pharmacokinet Biopharm 1976;4:183
For a single level drawn before steady state. The standard “single dose” method (goal-seek Ke against the regression volume). Winter, Basic Clinical PK
For a patient whose renal function is still changing, so steady state can't be assumed. The standard “two doses, two levels” method. Sawchuk–Zaske / Winter (superposition)
A forward calculator: given a Vd and Ke, predict the steady-state peak, trough and AUC for a chosen regimen. Winter, Basic Clinical PK
A utility that converts two clock times plus the number of midnights crossed into an exact elapsed interval (Δt), to feed the two-level method. No PK; arithmetic only.
Builds a dosing map (500–2000 mg post-HD) with the predicted pre-dialysis trough for each, given a dialyzer removal fraction (25 / 35 / 45%) and an HD population Vd (≈1.07 L/kg). Winter / Bauer, HD vancomycin dosing
The efficacy target is the ratio AUC24/MIC = 400–600, per the 2020 consensus guideline:
When MIC is unknown it defaults to 1 mg/L (broth microdilution). At MIC = 1 the ratio equals the raw AUC, so the display is a no-op vs raw AUC — the common empiric case.
At MIC ≥ 2 mg/L, hitting AUC/MIC ≥ 400 would require a raw AUC ≥ 800 (toxic), so the tool surfaces a “consider alternative therapy” message rather than pushing a toxic dose. 2020 consensus (MIC≥2 caveat)
| Method family | What is shown | Cited typical error |
|---|---|---|
| Population / empiric (no levels) | point estimate + “expect wide deviation”; the 5-model span is shown | eGFR ±30% for an individual |
| Two-level (Sawchuk–Zaske) | point estimate — reference standard; deterministic algebra so no CI is meaningful | assay CV ~5–15% + draw-time error |
| One-level | point estimate; noted as less reliable than a two-level fit | assay CV ~5–15% + draw-time error |
| Model estimate (Bayesian-style MAP companion) | point estimate beside the one-level result | within ~10–20% of a 2-level ref; bias ~5–21% |
The “model estimate” MAP companion is an illustrative one-compartment prior fit, not a validated Bayesian engine — see Section 7.
The companion “model estimate” beside the one-level result is an illustrative one-compartment MAP prior fit over a coarse grid — not a validated Bayesian engine (no covariate model, no published priors, no posterior sampling). Treat it as a sanity-check companion.
Reproduces the workbook's demo exactly, but its root-find can fail on inconsistent inputs and hit non-physical roots. It is guarded (AUC withheld, “verify/flag for review”), but the guard is a bounds check, not a clinical judgment — please confirm the guard thresholds and the back-extrapolation are acceptable for your own practice.
The dose recommender optimizes toward a raw AUC ≈ 500 (midpoint of 400–600). MIC scaling drives the efficacy verdict and the MIC≥2 message, but the recommended dose itself is chosen against raw AUC. Confirm this matches your intent for elevated-MIC organisms.
At MIC ≥ 2 the tool surfaces “consider alternative therapy” rather than pushing a toxic dose — but it still displays the raw-AUC computation. Confirm the wording and threshold match your own guidance.
Re-checked 2026-07-27 and still open. The abstract confirms the structure (one compartment, creatinine clearance the only covariate on clearance, volume about 0.8 L/kg against the 0.78 L/kg encoded), but 3.23 and 0.69 remain behind a paywall and unverified. Also newly recorded: the model was derived in 16 patients. Full citation: Masich AM et al. Pharmacotherapy. 2020 Mar;40(3):211–220 (PMID 31957057).
The sex-branch fix to lean body weight corrects a real bug in the original formula, but LBW feeds no dosing model, so it changes only the displayed value. Confirm you are comfortable with LBW being informational.
Five methods are machine-exact to their published formulas and all nine were checked across 33 scenarios — but that only proves the math was copied correctly, not that it is clinically right. There has been no independent clinical validation and no clinical trial.
All PK uses a one-compartment first-order model assuming steady state (except the explicitly pre-steady-state single- and 2-dose methods). It does not model two-compartment distribution, changing renal function within an interval, augmented renal clearance beyond the population models, or CRRT. Use clinical judgment where these apply.
This is a helper, not an answer. Check every number against your own hospital’s protocol before it reaches a patient. The math here is from published formulas — it has not been tested in a clinical trial.