Vancomycin dosing should not be a black box.

The state of vancomycin dosing tools is strange.

What we believe

Six commitments that shape every line of code in the engine.

The math, exposed.

Here’s the actual Colin 2019 clearance equation Vancomyzer evaluates for your patient, verbatim from the engine source. It is the same equation a black-box tool would compute internally. The difference is whether you ever get to see it.

// Adult clearance — Colin 2019, two-compartment, weight + maturation + age decline + SCr
// Source: Clin Pharmacokinet. 2019;58(6):767-780. Eqs 6-13, Table 3.

const CL = THETA_CL
  * Math.pow(weight_kg / 70, 0.75)  // allometric size scaling
  * FMat(PMA_weeks)            // sigmoidal maturation (≈1.0 for adults)
  * FDecline(PMA_years)          // 50% reduction at age 61.6 years
  * Math.exp(-THETA_SCR * (scr_mgdl - SCRstd));

And the Bayesian posterior, in three lines:

// Posterior ∝ likelihood(observed levels | params) × prior(params)
// MAP estimate via multi-start Nelder-Mead in log-space
// Bounded — a single outlier observation cannot override the population prior

That’s the entire trick. There is no proprietary algorithm. There is no closed-source Bayesian magic. Anyone with a graduate-level pharmacometrics course can audit the engine. That is the point.

Literature Reproducibility is a public scoreboard: every published vancomycin case we test our calculator against, with the live delta between our output and the published value. If our engine drifts, the build fails before it ships.

Where the data comes from.

One of the more common arguments against free Bayesian tools is that they rely on too few studies — that priors derived from one study per subpopulation are statistically fragile. The argument is correct as stated. It is also not what Vancomyzer does.

Our default adult prior is Colin 2019, which is itself a pooled population PK analysis across 14 published studies, ~2,554 patients, ~8,303 vancomycin concentrations, spanning neonates through elderly. It was published explicitly to be the “single coherent prior across populations” reference for vancomycin dosing tools — and it’s the same published model the commercial tools build on.

“We use a proprietary Bayesian model trained on data from our 1,000+ hospital partners.”
— every commercial vendor’s marketing page, paraphrased.

What we won’t do.

Promises about what a tool does are easy. Promises about what it refuses to do are where the safety actually lives.

• ✕

  We will not claim FDA clearance we don't have. Vancomyzer is non-device clinical decision support under 21st Century Cures Act §3060, scoped accordingly. The disclaimer on every page is real, not legal noise.

• ✕

  We will not fit your patient on a single outlier level by quietly loosening the prior. The fit is bounded; the residual is shown; the advisory tells you when to draw another level.

• ✕

  We will not pretend the calculator works for pediatrics, dialysis, continuous infusion, or extreme renal failure until those subpopulations have been validated and shipped with their own safety rails.

• ✕

  We will not gate basic AUC-guided dosing behind a hospital contract. The free tier is the same engine as Hospital tier — only EMR integration, BAA, custom branding, and audit logs are paid features.

• ✕

  We will not replace clinician judgment. Every recommendation comes with the math, the residuals, and the assumptions so a pharmacist can override it with full context.