Your Patient Isn't Average

Every pharmacokinetic number the calculator shows you — clearance, volume of distribution, the predicted trough — is a population-typical value: the best single guess for a patient with these demographics and this renal function. But no real patient is the typical patient. Two people with identical age, weight, and creatinine can clear cefepime at meaningfully different rates. One runs hot, one runs cold, and the model's point estimate lands somewhere in between.

That scatter isn't noise to wave away — it's measurable, and population PK studies quantify it. Which invites a better question than “what concentration will my patient have?” Instead ask: given everything we know about how patients like this one vary, what fraction of them would actually hit the pharmacodynamic target on this regimen?

Monte Carlo simulation answers exactly that. Generate thousands of virtual patients whose PK parameters scatter realistically around the typical values, dose every one of them with the same regimen, and count how many achieve the target. That fraction is the Probability of Target Attainment (PTA) — and it's the number behind the calculator's Monte Carlo panel. The rest of this page builds it up one piece at a time.

One Dose, Many Exposures

Here is the steady-state free (unbound) concentration profile for a typical patient receiving cefepime 2 g IV every 8 hours as a 0.5-hour infusion (bold cyan line), shown over 24 hours. The dashed orange line marks 4×MIC for an MIC of 4 mg/L — that is, 16 mg/L, the exposure the calculator's efficacy target asks the free trough to stay above. Now press the button: each thin line is a virtual patient whose clearance and volume were sampled from realistic population variability, all receiving the exact same regimen.

Same drug, same dose, same interval — and the curves fan out dramatically. High-clearance patients dive below the target line hours before the next dose; low-clearance patients ride comfortably above it. Same regimen ≠ same exposure.

The core idea: a dosing regimen doesn't produce a concentration profile — it produces a distribution of profiles across the population of patients who could receive it. Everything that follows is about putting numbers on that distribution.

Between-Subject Variability and Log-Normal Sampling

Where do those virtual patients come from? Population PK studies report not just the typical clearance (TVCL) but how much patients scatter around it — the between-subject variability (BSV), expressed as a standard deviation ω (omega). Each virtual patient's clearance is sampled as:

CLi = TVCL × eη,    η ~ N(0, ω²)

Why the exponential? Because clearances can't go negative, and biology is right-skewed: a patient can plausibly clear drug at 3× the typical rate, but never at −1×. Multiplying by eη (a log-normal distribution) guarantees positive values and produces exactly that long right tail. For cefepime, the Barreto 2023 population model reports ωCL = 0.24. Drag the slider and watch 2,000 sampled clearances (typical CL = 7.84 L/hr) reshape.

Between-Subject Variability on Clearance
ω (omega, SD of η) 0.24
Typical CL
7.84 L/hr
CV% = √(eω²−1) × 100
Middle 90% of sampled CL

The same 2,000 random draws are re-scaled as you move the slider (a fixed “z-pool”), so the histogram morphs smoothly instead of reshuffling — the same trick the simulation uses to keep results stable while you explore.

Reading ω: ω = 0.24 corresponds to roughly 24% CV — about one patient in three sits more than ~24% away from the typical clearance. Push ω to 0.5–0.6 (common in critically ill populations) and the right tail stretches far out: a few patients clear drug two to three times faster than average. Those are the patients standard dosing fails.

Scoring One Virtual Patient

A Monte Carlo simulation is just this, repeated: draw one virtual patient (a sampled CL and V pair), compute their steady-state curve, and apply a pass/fail rule. Here the rule is the calculator's efficacy target for critically ill patients: 100% fT>4×MIC — the free concentration must stay above 4×MIC for the entire dosing interval. At steady state the curve's lowest point is the trough, so the test reduces to: is the free trough ≥ 4×MIC (16 mg/L at MIC 4)?

Sampled CL (L/hr)
Sampled V (L)
Free trough (mg/L)
Trough / MIC
Patients drawn
0
Passed
0
Running PTA
Notice something? With an MIC of 4 mg/L, even the typical patient on 2 g q8h (0.5-h infusion) fails this stringent target — their free trough sits well below 16 mg/L. Only the low-clearance tail of the population passes. Keep clicking and watch the running percentage settle: with a handful of draws it jumps around wildly, which is exactly why the real simulation uses thousands.

Run Thousands: the PTA

Drawing patients one at a time builds intuition, but the estimate only stabilizes with volume. Run 1,000 virtual patients and watch the distribution of free troughs pile up — each trough is plotted in ×MIC units, so the pass line is simply 4 on the x-axis. The PTA is the fraction of the population to the right of it.

Patients simulated: 0 / 1,000  ·  PTA:

Free trough distribution across 1,000 simulated patients (MIC 4 mg/L, target trough ≥ 4×MIC). The distribution is right-skewed — the log-normal fingerprint from the previous section, propagated through the PK model.

The PTA you just computed is for one MIC. Repeat the whole exercise across a range of doubling MICs and you get the classic PTA-versus-MIC curve — computed below from 2,000 simulated patients per point. The PK/PD breakpoint is the highest MIC at which PTA still reaches ≥90%: the most resistant organism this regimen can be trusted to cover.

PK/PD breakpoint: computing…

Efficacy Is Only Half the Question

Everything so far asked whether patients get enough drug. But the same variability that leaves fast clearers under-exposed leaves slow clearers over-exposed — and for cefepime, neurotoxicity risk climbs as troughs rise. The calculator therefore scores each virtual patient against two thresholds, splitting the trough distribution into three zones:

< 4×MIC — sub-target (efficacy at risk) 4–8×MIC — on target ≥ 8×MIC — over-exposure (toxicity risk)
Sub-target (<4×MIC)
On target (4–8×MIC)
Over-exposed (≥8×MIC)

The same trough histogram as above, recolored by zone (this chart updates to match your last 1,000-patient run in the previous section). A dose change slides the whole distribution: raise the dose and the red zone shrinks while the orange zone grows.

The two-sided question: dose optimization is not “maximize PTA.” It's enough drug to kill the organism in most patients, without pushing an unacceptable fraction into the toxic range. A regimen with 95% PTA but 30% of patients over-exposed may be a worse choice than one with 90% PTA and 5% over-exposure — that trade-off is a clinical judgment the simulation informs but cannot make.

Reading the Calculator's Panel

The calculator's Monte Carlo panel (on the cefepime tab) runs exactly this pipeline — with a more sophisticated two-compartment cefepime model under the hood. Every element of the panel maps to a concept from this page:

Panel element Concept on this page
MIC input Sets the target line (Section 2) — every threshold is a multiple of it
Efficacy target (100% fT>4×MIC) The PASS rule applied to each virtual patient (Section 4)
Toxicity threshold (trough ≥8×MIC) The orange over-exposure zone (Section 6)
N patients Sample size — more patients, less jitter in the estimate (Sections 4–5)
PTA % Fraction of virtual patients passing the efficacy rule (Section 5)
Over-exposure % Fraction landing in the orange zone (Section 6)

Open the calculator's Monte Carlo panel →

Limitations & References

An honest PTA is only as good as the model behind it. Keep these caveats in view:

Teaching model, not the clinical model. This page uses a simplified one-compartment model so the math stays visible. The calculator itself uses a two-compartment cefepime population model (Barreto 2023), which captures the distribution phase this page ignores.
Simplified variability. BSV is applied to CL and V only — no correlation between them, no residual (assay/model) error, and no covariate effects beyond the fixed typical values used here. Real population models include more structure.
Steady-state assumption. All curves here are at steady state. Early in therapy — before accumulation is complete — exposures are lower, which is precisely why loading doses matter.
PTA is model-dependent. Change the population model, the variability estimates, or the target definition and the PTA changes with them. Garbage parameters in, garbage PTA out. A PTA is a statement about a model of a population — not a guarantee about the patient in front of you.

References & Further Reading

  1. Barreto EF, et al. Population pharmacokinetic model of cefepime for critically ill adults: a comparative assessment of eGFR equations. 2023. [PubMed: 37882514] — source of the cefepime typical clearance and between-subject variability used on this page.
  2. Population Pharmacokinetics course, Module 8: Probability of Target Attainment — interactive PTA-vs-MIC exploration across multiple drugs and regimens.
  3. PD Targets reference — the pharmacodynamic targets (including 100% fT>4×MIC for critically ill patients) and their primary literature.