EvidenceComparison with Tucuxi

Comparison with Tucuxi

The Vancomyzer team compared its calculator with Tucuxi, a dosing program developed by the REDS institute at HEIG-VD, Switzerland. On 18 Sep 2026, both programs received the same starting estimates based on Colin’s population equations, similar dosing histories at steady state, and the same two blood levels for 200 simulated patients. The median absolute difference in clearance estimates was 0.84%. The 95th percentile was 4.81%, and the largest difference was 9.22%. Both programs completed all calculations.

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.

How to interpret this comparison. The numerical limits set before the comparison were met. However, the programs handled uncertainty in blood levels differently, so their calculations were not identical. This comparison does not reproduce the full Colin Bayesian model. Tucuxi used 60 repeated doses to approximate steady state. At the sampling times, this differed from the long-term steady-state calculation by no more than 0.014%, using Tucuxi’s saved estimates. A difference in how clearance is calculated from the article and its supplement remains unresolved; neither interpretation has been confirmed by the model author.

Comparison dated 18 Sep 2026, calculator version 2026-09-17.1. The test data, model settings, results and instructions are available for independent review. Repeating the Vancomyzer calculations with version 2026-09-25.1gave the same estimates for these patients. This check covered two blood levels at steady state; it did not cover every calculator workflow.

Check before using blood levels

The Vancomyzer team supplied Tucuxi with the equations and starting assumptions used in this comparison. These were based on Colin’s population equations, with separate assumptions about variability. They were not the complete published Bayesian model or a model approved by its author. Before adding blood levels, Tucuxi had to match independently calculated values for a reference patient (35 years, 70 kg, serum creatinine 0.83 mg/dL; 1000 mg every 12 hours infused over 1.75 hours). A mismatch would have stopped the comparison.

QuantityTucuxiReferenceRelative differenceResult
prior CL (L/h)4.10244.1024< 1 in a millionpass
prior V1 (L)42.900042.9000exactpass
prior Q (L/h)3.22003.2200exactpass
prior V2 (L)41.700041.7000exactpass
SS peak (mg/L, end of infusion)31.920431.9205< 1 in a millionpass
SS trough (mg/L, end of interval)12.927912.9279< 1 in a millionpass
SS AUC per 12 h interval (mg·h/L)243.7608243.7608< 1 in a millionpass

The independent reference calculations are described on the Literature Reproducibility page. This check included the starting estimates before blood levels were added.

Agreement between the two programs · n = 200

Difference between the two programs’ individual (posterior) estimates, Vancomyzer relative to Tucuxi, per parameter. Q and V₂ are reported without a criterion: two levels barely inform them, so both programs stay near the shared prior.

ParameterMedian absolute differenceMean signed difference90th percentile95th percentileLargest difference
Clearance (CL)0.84%+0.50%2.53%4.81%9.22%
Central volume (V₁)0.87%+0.33%2.10%2.57%7.59%
Inter-comp. clearance (Q)0.48%+0.15%1.74%2.66%7.26%
Peripheral volume (V₂)0.03%-0.02%0.08%0.13%0.33%

Acceptance criteria (fixed 17 Sep 2026, before the run)

  • Clearance Median absolute difference ≤ 2%0.84% met
  • Clearance 95th percentile |Δ| ≤ 10%4.81% met
  • Central volume Median absolute difference ≤ 3%0.87% met
  • Failed or excluded calculations ≤ 2% of patients0 of 200 met
  • Every clearance difference > 10% explained0 such cases met

Steady-state exposure computed from each program’s own estimates: AUC₂₄ Median absolute difference 0.85% (95th percentile 4.59%, maximum 8.43%), peak 0.57%, trough 1.54% (maximum 12.52%; the trough is the quantity most sensitive to clearance). Meeting these criteria means the two programs agree on this synthetic cohort; it does not mean the programs are interchangeable in practice.

Accuracy in simulated patients

Because the patients are simulated, their “true” parameters are known. Median absolute percentage error of each program’s estimates against that truth, with the unfitted prior for reference (lower is better).

ParameterBefore blood levelsVancomyzerTucuxi
Clearance (CL)31.4%10.0%10.1%
Central volume (V₁)28.8%25.9%26.2%
Inter-comp. clearance (Q)52.4%52.8%52.2%
Peripheral volume (V₂)25.4%25.0%25.1%

In this comparison, only clearance improves materially with a two-level fit (31.4% with the prior alone, 10.0% for Vancomyzer, 10.1% for Tucuxi). The simulated patient values come from a simulation adapted by the Vancomyzer team from Goti’s work, rather than the complete published Goti model, so the results describe performance in a simulation, not accuracy in patients.

Largest disagreements — and why

The programs give different weight to blood levels that are higher than predicted. Vancomyzer allows more uncertainty around those levels, which can change the clearance estimate. To investigate, the team repeated the calculations for all 14 simulated patients with a clearance difference of 3% or more. A separately written calculation used each program’s approach to blood-level uncertainty in turn.

PatientVancomyzer CLTucuxi CLDifferenceRepeat calculation vs VancomyzerRepeat calculation vs TucuxiEffect of uncertainty assumptions
p1010.8420.771+9.22%0.00%+0.01%+9.22%
p973.6833.417+7.80%0.00%0.00%+7.80%
p1672.7462.561+7.19%0.00%0.00%+7.20%
p1035.4715.122+6.82%0.00%0.00%+6.82%
p1285.8255.461+6.68%0.00%0.00%+6.68%
p11810.77810.125+6.45%0.00%0.00%+6.45%
p852.7562.595+6.18%+0.01%0.00%+6.19%
p1121.6731.577+6.03%0.00%0.00%+6.03%
p1232.5102.386+5.21%+0.01%0.00%+5.22%
p1701.5941.521+4.81%+0.01%0.00%+4.82%
p193.2713.144+4.03%-0.01%0.00%+4.03%
p9611.72211.288+3.84%0.00%0.00%+3.84%
p1602.8202.719+3.74%0.00%0.00%+3.74%
p31.3821.338+3.27%0.00%0.00%+3.27%

The separate calculation matched each program’s clearance estimate to within 0.01% when it used that program’s uncertainty assumptions. This supports those assumptions as the explanation for the differences in these 14 cases. The largest difference was9.22% in case p101. This finding applies to the selected clearance estimates. It does not establish which approach is more accurate in real patients.

Method

  1. Create 200 simulated ICU patients using a model adapted from Goti’s work, then simulate two blood levels with measurement error. The data and full settings are available in the linked methods.
  2. Run Vancomyzer’s calculator (calculator version 2026-09-17.1) on each patient: Colin 2019 starting estimates from patient characteristics, then update them using the two blood levels.
  3. Use the Colin 2019 equations in Tucuxi with prior variability (log-scale SD: CL 0.35, V₁ 0.25, Q 0.5, V₂ 0.5) and Tucuxi’s mixed residual error (1.0 mg/L additive, 15% proportional). Check its prior-only prediction against the reference patient (table above). Then give Tucuxi a 60-dose history approximating steady state and the same two levels per patient and run its Bayesian fit.
  4. Compare the estimates against limits set on 17 Sep 2026, before the analysis. Repeat calculations for clearance differences of 3% or more using each program’s uncertainty assumptions.

Comparator: Tucuxi-core (tucucli). The exact program version, test data, model definitions and analysis scripts are available in the reproducibility materials.

Scope & limitations

  • What was compared

    Both programs used starting estimates and variability assumptions supplied by Vancomyzer, based on Colin’s population equations. Their similar results apply to these simulated patients and shared assumptions. They do not confirm the full published Colin model or establish accuracy in patients.

  • Blood-level uncertainty

    The programs handle blood-level uncertainty differently. Repeating the calculations supported this as the explanation for the 14 clearance differences of 3% or more. Differences reached about 9% in simulated patients with levels well above prediction.

  • Simulated patients only

    The patients and their levels are computer-generated. Real-patient performance has not been evaluated; the Predictive Performance analysis is also synthetic.

  • Two-level sampling

    A post-infusion level and a trough at steady state mainly inform clearance. Agreement on Q and V₂ largely reflects both programs staying near the shared prior.

  • Open-source comparator only

    Commercial Bayesian dosing products were not run; no results for them are shown or implied.

The results met the numerical limits set for this group of 200 simulated patients. The separate check of the largest clearance differences supported blood-level uncertainty assumptions as their explanation. Clinical accuracy and the choice of uncertainty assumptions still need evaluation with patient data.