← Transparent Dosing·Literature Reproducibility

Literature Reproducibility

Published vancomycin values run through Vancomyzer's engine. Cases that use the same model as the engine (Colin 2019) are pass/fail checks of the implementation. Cases from other published models or patient cohorts are shown for context only. These cases run in the automated test suite (npm test); a drift outside tolerance fails the suite.

Vancomyzer has not yet been validated in real patients. Its equations are checked against published values and synthetic test cases; external validation with patient data is planned.

Summary
2 / 2 same-model reproductions within tolerance
Not pass/fail: 3 cross-model references and 1 reference band, excluded from these statistics.
Largest |AUC₂₄ difference|
0.06%
Largest |CL difference|
0.06%

Same-model reproductions (Colin 2019) · pass/fail

The published value comes from Colin 2019, the model the engine uses, so the engine should reproduce it within the stated tolerance.

Colin 2019, abstract — typical adult (35 y, 70 kg, SCr 0.83 mg/dL)

✓ Within tolerance

What it tests: The engine's Colin 2019 prior for the published typical adult (35 y, 70 kg, SCr 0.83 mg/dL) should match the published CL (4.10 L/h), V1 (42.9 L) and the AUC₂₄ that CL implies at 1000 mg every 12 h, each within 1%.

Same model as the engine. The published values describe a model-typical individual, not a real patient, so this checks that the equations are implemented as published; it does not test clinical accuracy.

Patient: 35 y M · 70 kg · SCr 0.83 mg/dL·Regimen: 1000 mg every 12 h over 2 h
Inputs as published. Model-typical individual from the Colin 2019 abstract, not a real patient. Colin 2019 has no sex covariate; sex is entered only because the calculator asks for it.
PublishedEngine (Colin 2019)DifferenceTolerance
AUC₂₄ (mg·h/L)487.8487.5−0.06% ✓±1%
CL (L/h)4.104.10+0.06% ✓±1%
V₁ (L)42.942.9+0.00% ✓±1%
How the published value was obtained

Extraction: CL and V1 read from the published abstract. AUC₂₄ derived from the published CL: 2000 mg/day / 4.10 L/h = 487.8 mg·h/L.

Verification: V1 42.9 L, V2 41.7 L, CL 4.10 L/h and Q 3.22 L/h for a 35-year-old, 70-kg patient with SCr 0.83 mg/dL are stated in the abstract (PMID 30656565). AUC₂₄ is derived as daily dose divided by the published CL.

Tolerance: ±1% on CL, V1 and AUC₂₄. The published values come from the same equations the engine implements, so the only expected difference is rounding of the published values (CL 4.10 L/h is rounded to about ±0.12%). A larger difference means the implementation changed and must be investigated.

Colin 2019, Discussion (p. 8) — 60 y, 65 kg, SCr 0.97 mg/dL

✓ Within tolerance

What it tests: The engine's Colin 2019 prior for a 60-year-old, 65-kg patient with SCr 0.97 mg/dL should match the published CL (2.55 L/h) and the AUC₂₄ it implies at 750 mg every 12 h, each within 1%. This exercises the age-decline and serum creatinine terms together.

Same model as the engine. Colin 2019 states CL = 2.55 L/h for this patient; the engine's equations give about 2.551 L/h. This checks that the age and renal terms are implemented as published; it does not test clinical accuracy.

Patient: 60 y M · 65 kg · SCr 0.97 mg/dL·Regimen: 750 mg every 12 h over 1.5 h
Inputs as published. Model-typical patient described in the Colin 2019 Discussion, not a real patient. Colin 2019 has no sex covariate; sex is entered only because the calculator asks for it. Height is not specified.
PublishedEngine (Colin 2019)DifferenceTolerance
AUC₂₄ (mg·h/L)588.2587.9−0.05% ✓±1%
CL (L/h)2.552.55+0.05% ✓±1%
How the published value was obtained

Extraction: CL (2.55 L/h) read from page 8. AUC₂₄ derived from the published CL: 1500 mg/day / 2.55 L/h = 588.2 mg·h/L.

Verification: Checked against the open Colin 2019 PDF (Discussion, page 8). Quote: "vancomycin clearance in a 60-year-old, 65-kg patient with a SCR of 0.97 mg dL⁻¹ (85.7 μmol L⁻¹) is 2.55 L h⁻¹ (0.039 L h⁻¹ kg⁻¹)."

Tolerance: ±1% on CL and AUC₂₄. The published CL comes from the same equations the engine implements; it is rounded to three significant figures (about ±0.2%). A larger difference means the age-decline or serum creatinine term changed and must be investigated.

Cross-model references · context only

The published value comes from a different model or summarizes a patient cohort. The engine's Colin 2019 value is shown next to it; these cards never pass or fail.

Smit 2020 final-model clearance at 130 kg (different model)

Different model · context only

Different model — difference shown for context, not a pass/fail test.

What it shows: Smit 2020's typical clearance at 130 kg (5.72 × (130/70)^0.535 = 7.97 L/h) and the AUC₂₄ it implies at 2250 mg every 12 h, next to the engine's Colin 2019 values for the same inputs.

Vancomyzer does not use Smit 2020; it is shown because it is a published obesity model. Smit 2020 was developed in adults undergoing bariatric surgery with normal renal function and in non-obese volunteers, with no ICU patients. The difference shows how far two published models can disagree for the same weight; it does not show which is more accurate for a given patient.

Patient: 35 y F · 130 kg · SCr 0.8 mg/dL · 165 cm·Regimen: 2250 mg every 12 h over 4 h
Illustrative inputs (not from the source). Only body weight enters the Smit 2020 clearance equation. Age, sex, height (BMI 47.8) and SCr were chosen by Vancomyzer to describe a younger adult with normal renal function; they change only the Colin 2019 value. Smit 2020 population: 20 morbidly obese adults undergoing bariatric surgery (110.6-234.6 kg, age 23-54, eGFR 60 or more) and 8 non-obese volunteers; single dose; no ICU patients.
Published (different model or cohort)Engine (Colin 2019)Difference (context only)
AUC₂₄ (mg·h/L)564.9676.2+19.7%
CL (L/h)7.976.65−16.5%
How the published value was obtained

Extraction: CL evaluated from the Smit 2020 final-model equation at TBW 130 kg: 5.72 × (130/70)^0.535 = 7.97 L/h. AUC₂₄ derived as daily dose / CL = 4500 / 7.97 = 565 mg·h/L.

Verification: Final-model clearance equation CL = 5.72 × (TBW/70)^0.535 checked against the PMC7015748 full text. The value for this card is evaluated from that equation at TBW 130 kg.

Tolerance: Not applicable: cross-model reference (different model), no pass/fail.

Adane 2015 abstract — cohort median AUC₂₄ and population clearance, BMI ≥ 40 (different model)

Different model · context only

Different model — difference shown for context, not a pass/fail test.

What it shows: The published cohort median AUC₂₄ (583 mg·h/L, IQR 514–726) and population clearance (6.54 L/h, one-compartment model) in 31 adults with BMI ≥ 40, next to the engine's Colin 2019 values for one approximated cohort-median patient at 2000 mg every 12 h.

Adane 2015 measured three steady-state concentrations per patient in 31 adults with BMI ≥ 40 at one hospital and fitted a one-compartment model. The published AUC and CL summarize that cohort, while the engine value is for one approximated patient, so they are not expected to match and the difference is not a test of accuracy.

Patient: 50 y F · 147.9 kg · SCr 0.9 mg/dL · 173 cm·Regimen: 2000 mg every 12 h over 3.5 h
Inputs partly approximated (not verified). Weight (147.9 kg) is the published cohort median; height 173 cm gives BMI 49.4 (published median 49.5). Age 50 years, SCr 0.9 mg/dL and female sex are approximations chosen by Vancomyzer and have not been verified against the paper.
Published (different model or cohort)Engine (Colin 2019)Difference (context only)
AUC₂₄ (mg·h/L)582.9 (513.8–726.2)698.8+19.9%
CL (L/h)6.545.72−12.5%

AUC₂₄ values in parentheses: interquartile range.

How the published value was obtained

Extraction: Median AUC₂₄ 582.9 mg·h/L (IQR 513.8–726.2) at a median dose of 4000 mg/day, and population mean CL 6.54 L/h from a one-compartment model, read from the abstract. These summarize the cohort; they are not values for this card's inputs.

Verification: Cohort values checked against the PubMed abstract (PMID 25644478): n = 31, median weight 147.9 kg, BMI 49.5 kg/m², Cockcroft-Gault ClCr 124.8 mL/min/1.73 m², median dose 4000 mg/day, median AUC₂₄ 582.9 mg·h/L (IQR 513.8–726.2), population mean V 0.51 L/kg and CL 6.54 L/h. The abstract also states that 24-hour urine creatinine clearance was collected. Patient-level values (Table 1) were not checked.

Tolerance: Not applicable: cross-model reference (cohort statistic, different model), no pass/fail.

Carreno 2017 abstract — full-data AUC₂₄ range across four population models, obese adults (different models)

Different model · context only

Different model — difference shown for context, not a pass/fail test.

What it shows: The engine's Colin 2019 Bayesian AUC₂₄ from two illustrative levels, next to the range of full-data AUC₂₄ estimates (437–489 mg·h/L) that Carreno 2017 reported across four other population models in 12 obese adults.

Carreno 2017 found that Bayesian estimates from a peak and a trough approximated the full-data AUC better than trough-only or midpoint-and-trough estimates. The published range is a cohort result for real patients and other models, while this card's patient values and levels are illustrative, so the difference is context only.

Patient: 61 y M · 130 kg · SCr 1 mg/dL · 170 cm·Regimen: 1250 mg every 12 h over 2.5 h·Levels (hours after the start of the dose): 25 mg/L at 3.5 h, 10 mg/L at 11.5 h
Illustrative inputs (not from the source). Illustrative values chosen by Vancomyzer, not from the paper. Age 61 years and BMI 45 match the published cohort medians; weight, height, sex and SCr do not come from the paper. Cockcroft-Gault with total body weight gives about 143 mL/min for these inputs, not the published median of 86 mL/min.
Published (different model or cohort)Engine (Colin 2019)Difference (context only)
AUC₂₄ (mg·h/L)463.0 (437–489)395.7−14.5%

AUC₂₄ values in parentheses: range of full-data estimates across four population models; 463 is the midpoint, derived for display.

How the published value was obtained

Extraction: Range of full-data AUC₂₄ estimates (437–489 mg·h/L) read from the abstract; midpoint 463 derived by Vancomyzer. The levels on this card (25 mg/L 3.5 h after the start of the dose, 1 h after the end of a 2.5 h infusion; 10 mg/L at 11.5 h) are illustrative and were chosen by Vancomyzer.

Verification: Checked against the PubMed abstract (PMID 28289024): n = 12, median age 61 years, median creatinine clearance 86 mL/min (the abstract does not state the method), median BMI 45 kg/m², full-data AUC₂₄ estimates 437–489 mg·h/L across four models. Per-patient data were not available.

Tolerance: Not applicable: cross-model reference (cohort result from other models, illustrative inputs), no pass/fail.

Published reference band · no engine run

Published results from a multi-model comparison, shown for context. Vancomyzer was not run on these patients.

Patanwala 2022 (abstract, Table 1, section 3.2) — three population models in 188 ICU adults

◇ Reference band · not pass/fail

What it shows: Published context: three population models run in one Bayesian program (Tucuxi) gave cohort-mean AUC₂₄ values from 469 to 562 mg·h/L for the same 188 ICU adults, and agreed on the AUC dosing category in 48% of estimations. Vancomyzer was not run on this cohort.

Same 188 ICU patients and 466 AUC₂₄ estimations, three population models. Cohort-mean AUC₂₄ ranged from 469 (Goti) to 562 (Colin), about 20% apart, and the three models agreed on the AUC dosing category in 48% of estimations (pairwise 59–68%). The choice of model can change dosing decisions.

Cohort: 188 adult ICU patients · 466 AUC₂₄ estimations · mean age 58 ± 17 y · 63% male · APACHE III 62 ± 22 · 39% mechanically ventilated · 35% on vasopressors · Royal Prince Alfred Hospital, Sydney, 2019-2020
Published cohort-mean AUC₂₄ (mg·h/L) per population model
Goti 2018 (via Tucuxi)
469 ± 148
Colin 2019 (via Tucuxi)
Model Vancomyzer uses for dosing
562 ± 172
Thomson 2009 (via Tucuxi)
517 ± 164
200500800
How this relates to Vancomyzer: Vancomyzer uses Colin 2019 for dosing. This card shows published values only. Vancomyzer was not run on this cohort, so how its estimates would compare with these results is not known.

Limitations

  • These cases check the engine against published values; they are not clinical validation. Vancomyzer has not yet been validated in real patients. Its equations are checked against published values and synthetic test cases; external validation with patient data is planned.
  • Only the same-model reproductions (Colin 2019) are pass/fail tests. They use model-typical individuals rather than real patients, so they confirm that the equations are implemented as published, not that doses are accurate for patients.
  • Cross-model references compare the engine (Colin 2019) with a different published model or a cohort statistic. Different model — difference shown for context, not a pass/fail test. A difference does not show which model is more accurate for a given patient. Some of these cards use approximated or illustrative inputs; each card says which.
  • No pediatric, dialysis or post-transplant cases. These are outside Vancomyzer's scope (adults not on renal replacement therapy).
  • Sources considered but not added as cases: Rybak/ASHP 2020 (narrative recommendations only, no worked patient example to reproduce); Pai 2014 (aggregate results across the cohort, no per-patient demographics with AUC); Turner 2018 (aggregate medians and interquartile ranges per program across 19 ICU patients; not yet added as a reference band); Neely 2014 cohort trough (the comparison was circular: the fitted level was the value being compared); Shingde 2020 single-sample Bayesian (the candidate "published AUC" was derived rather than read from the paper, so the comparison would have been circular). Patanwala 2022 is shown above as a reference band.
  • A "within tolerance" result does not mean a recommendation is correct for any individual patient. Every clinical decision remains the responsibility of the treating clinician.
  • These cases run in the automated test suite (npm test); a drift outside tolerance fails the suite. A case outside tolerance is still shown on this page, with an amber badge.