Learn the reasoning, not just the number.

Concentration-dependent killing & the post-antibiotic effect

Aminoglycosides are concentration-dependent killers: raise the peak concentration and bacteria die faster and more completely, in contrast to beta-lactams, whose kill rate depends on time spent above the MIC rather than how far above it the drug climbs. A single large dose therefore does more killing than the same total amount split into several smaller doses — the opposite of the intuition that "more frequent" means "more effective."

That single large dose keeps working across a long dosing interval because of the post-antibiotic effect (PAE): bacterial regrowth stays suppressed for roughly 0.5–8 hours after the concentration falls back below the MIC (Craig WA, Ebert SC. Scand J Infect Dis Suppl 1990;74:63–70; Zhanel GG, Hoban DJ, Harding GKM. DICP 1991;25(2):153–63). A second effect points the same way: after first exposure, surviving organisms transiently down-regulate aminoglycoside uptake — adaptive resistance — so re-dosing before concentrations have fallen buys less killing than the arithmetic suggests. What makes the large dose tolerable is separate physiology: uptake into the renal proximal tubule and the cochlear hair cells is carrier-mediated and saturable, so a brief high peak followed by a genuine drug-free window deposits less drug in those tissues than the same daily amount trickled in continuously.

Together these are the rationale for extended-interval dosing (EIAD). Be precise about what the trials showed, though: in humans EIAD is as effective as traditional multiple-daily dosing, not clearly better — the reproducible advantage is less nephrotoxicity, with no excess ototoxicity (Barza M, Ioannidis JPA, Cappelleri JC, Lau J. BMJ 1996;312(7027):338–45, doi:10.1136/bmj.312.7027.338; 21 randomised trials, n = 3,091). The pharmacodynamics predict superiority; the clinical evidence delivers equivalence plus a safety margin (see Toxicity).

The PK/PD indices — Cmax:MIC and AUC₂₄:MIC

The classic efficacy driver for aminoglycosides is the peak-to-MIC ratio, conventionally targeted at Cmax:MIC 8–10. Be honest about where the number comes from. Moore, Lietman, and Smith (J Infect Dis 1987;155(1):93–9, doi:10.1093/infdis/155.1.93) pooled 236 patients with Gram-negative infection across four gentamicin/tobramycin/amikacin trials and found a graded dose–response between rising peak:MIC and clinical response — not a threshold at 8–10. Kashuba and colleagues (Antimicrob Agents Chemother 1999;43(3):623–9, doi:10.1128/AAC.43.3.623) then showed that reaching Cmax:MIC ≥ 10 within the first 48 hours predicted a ~90% probability of temperature and leukocyte-count resolution by day 7 in Gram-negative nosocomial pneumonia — surrogates, not cure — while their own partitioning breakpoints fell lower, at Cmax:MIC 4.5–4.7. So 8–10 is a defensible convention with a soft floor beneath it, not a cliff edge.

Two rules govern how this tool applies the ratio. First, the MIC sets a floor, never a ceiling: a high-MIC isolate raises the target peak, but a low-MIC isolate does not license dropping below the conventional band — that band is the expected peak of a standard weight-based dose, and dosing under it discards the post-antibiotic-effect margin for no gain. The target only ever moves up with the organism. Second, the ratio governs only where concentration-dependent killing of the target pathogen is the goal. It does not govern Gram-positive synergy (the aminoglycoside works with a cell-wall agent, and the deliberately low 3–5 band limits cumulative toxicity over a 4–6 week course) or cystitis (urine concentrations run far above serum, so serum Cmax:MIC is not the driver) — and the calculator will refuse to MIC-anchor either indication.

One more definition, because two different numbers both get called “the peak.” Cmax is the concentration the instant the infusion ends; the distributed peak is the concentration about 30 minutes later, once the drug has finished distributing out of plasma — and it is the number aminoglycoside target bands are defined against, which is why the classical peak draw is 30 minutes post-infusion. The gap between the two is e^ke·0.5 — small in renal failure (~5% at CrCl 30) and not small in normal function (~13% at CrCl 80, ~20% at CrCl 120). The calculator's levels screens compare the distributed peak to the band; the first-dose (empiric) screen still predicts and flags the end-of-infusion Cmax, so a borderline empiric flag just over the band may sit inside it once distributed — read it with that in mind.

A complementary index, AUC₂₄:MIC, captures total exposure across the dosing interval rather than the single peak moment, and Bland, Pai, and Lodise (Pharmacotherapy 2018;38(12):1229–38, doi:10.1002/phar.2193) argue it is the more reliable predictor of killing. Their targets are pre-clinically anchored and scale with how much killing the situation demands: roughly 30–50 — the exposure associated with net bacterial stasis — for non-critically-ill patients with low-burden Gram-negative infection, rising to roughly 80–100 — the exposure associated with a 1-log₁₀ kill — for serious systemic infection. USCAST's 2025 analysis targets a comparable AUC:MIC near 84 (Bhavnani SM et al. Open Forum Infect Dis 2025;12(12):ofaf426, doi:10.1093/ofid/ofaf426). Human outcome data behind all of these remain thin, which is why this tool displays Cmax:MIC as the primary target and treats AUC-based targeting as supporting, opt-in information — it is the index the Bayesian fit (see the method ladder) reports directly, since it falls naturally out of a fitted clearance without needing a separately-timed peak sample.

Toxicity — trough, cumulative dose, ototoxicity

Where the peak drives efficacy, the trough is the practical safety guardrail — the number that tells you whether the drug is clearing between doses, which is why every target table in this tool pairs a peak range with a trough ceiling rather than a peak alone. But be clear what it is and is not: the trough is a marker of accumulation, not the measured driver of harm. In the cohorts that looked, ototoxicity tracked age and cumulative dose — not trough height, and not peak height.

Nephrotoxicity deserves its own paragraph, because it changes management daily. It is proximal tubular injury, typically appearing after 5–7 days of therapy, and it is usually reversible once the drug stops. Risk multiplies with duration, with concurrent nephrotoxins (vancomycin, iodinated contrast, NSAIDs, loop diuretics), with hypovolaemia, and with age. Note the practical limit on the extended-interval advantage: the nephrotoxicity benefit of once-daily dosing is best established for short courses — roughly a week or less — so a long course does not inherit it automatically.

Nephrotoxicity is usually reversible; ototoxicity is the toxicity that dominates prolonged courses and is often permanent. In prospectively monitored mycobacterial cohorts the rate is high: 32 of 87 (37%) among patients randomised to daily versus thrice-weekly streptomycin, kanamycin or amikacin (Peloquin CA, Berning SE, Nitta AT, et al. Clin Infect Dis 2004;38(11):1538–44, doi:10.1086/420742), and 30 of 77 (39%) in a low-dose amikacin NTM cohort (Aznar ML, Marras TK, Elshal AS, Mehrabi M, Brode SK. BMC Pharmacol Toxicol 2019;20:37, doi:10.1186/s40360-019-0302-1). Risk tracks with age and, above all, with cumulative dose: in the Toronto cohort ototoxicity was associated with a median total amikacin exposure of 1.81 g/kg (IQR 0.85–3.13) among those affected, against a median total course of 50.5 g — odds ratio 1.62 per g/kg (95% CI 1.08–2.43). A separate prediction model in streptomycin-treated tuberculosis put the gradient at roughly 7% added risk per additional gram of cumulative aminoglycoside (Adeyemo AA, Adeolu J, Akinyemi JO, Omotade OO, Oluwatosin OM. Front Neurol 2024;15:1461823, doi:10.3389/fneur.2024.1461823) — a different drug in a different disease, so read it as a direction of travel, not a number to dose by.

A common and consequential misconception: thrice-weekly dosing does not reduce ototoxicity risk compared to daily dosing (Peloquin and colleagues, Clinical Infectious Diseases 2004;38(11): 1538–44) — the ear does not "reset" between doses the way the drug-free interval lets the MIC-vs-regrowth balance reset. Toxicity tracks cumulative exposure, not dosing frequency, so switching a patient to TIW dosing to protect their hearing is not supported by the evidence; it is a real dose-schedule decision, not a toxicity-mitigation one.

Because ototoxicity is often insidious and can outlast the course itself, courses expected to run long — NTM therapy above all — warrant a baseline audiogram before starting and periodic audiometry through treatment, not just serum-level monitoring. A stable trough does not rule out a rising cumulative dose; both need tracking.

When to draw levels

The right draw time depends on which method a regimen is being individualized by. The two extended-interval methods below deliberately sample after the first dose and need no steady state; traditional multiple-daily dosing is the one that does.

Hartford / extended-interval nomogram — a single random level drawn 6–14 hours after the start of the first dose's infusion; the level and its exact draw time are plotted against the nomogram's boundary lines to read off the interval (Nicolau DP, Freeman CD, Belliveau PP, Nightingale CH, Ross JW, Quintiliani R. Antimicrob Agents Chemother 1995;39(3):650–5, doi:10.1128/AAC.39.3.650). Three limits travel with it: it is a calibrated instrument, valid only at the dose it was built for — 7 mg/kg gentamicin or tobramycin, and for amikacin only at 15 mg/kg, plotting half the measured level (a gent/tobra dose below 7 can be normalised by multiplying the level by 7/dose; anything above, and any other amikacin dose, is off-calibration and the calculator refuses the nomogram there). It applies to a first dose only. And it was not validated in critical illness, burns, ascites, pregnancy, dialysis, or CrCl <20 — exactly the settings where volume and clearance drift away from the population values its curves assume.

Extended-interval, high-peak individualization — two levels drawn after the first dose, timed to capture the elimination slope directly, again without waiting for steady state; this is the two-level (Sawchuk-Zaske) approach applied at the very start of therapy rather than after several doses have accumulated.

Traditional (multiple-daily-dose) monitoring — a peak drawn 30 minutes after the infusion ends and a trough drawn immediately before the next dose, both at steady state (typically after the third or fourth dose) — the one approach here where waiting matters, because traditional dosing's smaller, more frequent doses take longer to reach a stable peak/trough pattern.

A single pre-dose trough — the least precise input, but a common one. A trough is evidence about accumulation, not about the peak: one pre-dose level cannot locate the peak, so the calculator predicts the peak from the dose on population PK and keeps your measured trough as the trough it is. Do not read the peak it shows alongside as an individualised measurement.

Never draw during the infusion. Every back-extrapolation on this page assumes the elimination phase; inside the infusion the concentration is still rising and the maths runs backwards. The calculator rejects such a level outright rather than returning a number — unguarded, a gentamicin level drawn 15 minutes into a 30-minute infusion once produced a peak estimate ~50% below truth and a ~50% dose increase. Draw at or after the end of infusion; the ideal peak draw is 30 minutes after it.

The method ladder

Before the methods, the mental model they all serve. Two levers, and they are close to independent: the dose sets the peak (Cmax ≈ dose / Vd) and the interval sets the trough (how many half-lives fit between doses). That is why a peak problem is a dose problem and a trough problem is an interval problem — and why typical Vd matters: about 0.25–0.30 L/kg in a well patient, rising toward 0.35–0.5 L/kg in sepsis or third-spacing and falling in cachexia. When Vd is larger than the population assumes, the same mg/kg produces a lower peak — which is exactly why expected-peak tables over-predict in the sickest patients.

The bands this calculator actually uses. Gentamicin / tobramycin: extended-interval peak 18–22 mcg/mL with trough <1 (ideally undetectable); traditional multiple-daily 6–10 with trough <1; Gram-positive synergy 3–5 with trough <1. Amikacin: serious Gram-negative peak 60–80 with trough <1; NTM as in the section below. These are the numbers the screens compare against — worth knowing before you read a recommendation.

Tod, Padoin, and Petitjean's 2001 review (Clin Pharmacokinet 2001;40(11):803–14) frames aminoglycoside individualization as a ladder, not a single method — climb only as far as the patient's data and the clinical stakes require:

1. Linear dose adjustment — the simplest rung: a single level plus the assumption that concentration scales linearly with dose, enough to nudge an already-close regimen. 2. Nomogram (ODA / Hartford) — one random level at a validated draw time maps directly to an interval, no PK math required by the user. 3. Non-Bayesian two-level (Sawchuk-Zaske) — two levels let the elimination rate and volume of distribution be back-calculated directly, individualizing Ke and Vd rather than assuming population values. 4. Bayesian / MAP — population priors (informed by age, weight, renal function) combine with whatever levels exist, however sparse or early, to produce the most defensible individual fit; unlike the two-level method it works from a single level. Neither method requires steady state — but the two-level method requires you to say which situation you are in (see below), because the same pair of levels implies two different volumes of distribution depending on the answer.

Each rung relaxes an assumption the one below it depends on. Use the simplest method that actually fits the patient in front of you — reach for the next rung only when that assumption breaks (an unstable renal function, a level drawn off-nomogram, or too few levels for a reliable two-point slope).

The question the levels can't answer for you

When you hand the two-level method a pair of levels, the calculator asks something that feels like paperwork and is not: were these drawn on the first dose, at steady state, or somewhere in the middle? Here is why it cannot guess. Two levels give you a slope, and the slope gives you Ke — how fast the patient clears the drug. Ke comes out the same no matter where in the course you are, because leftover drug from earlier doses decays at exactly the same rate as the dose you just gave. The slope is blind to the schedule.

Vd is not. To turn a measured level into a volume of distribution, the tool must know how much drug is in the patient — and at steady state, some of it arrived before this dose. The steady-state form divides by an accumulation factor, 1 − e^−Ke·τ. Call a steady-state pair a first dose and the fitted Vd comes out roughly half of truth — and the recommended dose with it. The tool will not silently pick for you, and when the resulting Vd lands outside the plausible 0.20–0.70 L/kg range it says so, because a Vd of 0.15 L/kg for amikacin is not unusual physiology — it is a failed model.

Three answers on the calculator's levels screen: the first dose (nothing on board — the Hartford/EIAD situation), steady state (4–5 half-lives on an unchanged interval, so what goes in each interval comes out), and mid-course (a few doses in, or the spacing changed — neither closed form is exactly right, so the tool asks for the previous dose and quantifies the error, below).

Mid-course: the drug that was already there

A two-level (Sawchuk-Zaske) fit models one dose. If your levels were drawn on the third dose, part of every level is leftover from the second — and the fit credits all of it to the dose you just gave. You can put a number on the error. The previous dose's residual, as a fraction of the current dose's contribution at any shared time, is r = (D_prev / D_current) × e^−Ke·gap — everything else cancels (Vd, the infusion duration, the ramp term), so it needs only the two dose amounts, the hours between them, and Ke. Carry-over is then r / (1 + r) of each measured level: a short half-life and a long gap make it negligible; a long half-life, a short gap, or a dose that changed make it real. The consequence is one-directional — the fit over-states the peak this dose produced, so it under-states Vd by about that fraction, and under-doses. The calculator flags carry-over once it reaches 5% and routes you to the Course tab, which fits the actual dated schedule so no dose is credited to the wrong administration.

“Extend the interval” — to what?

When a measured trough is over target, the answer is not just “extend it.” The calculator answers extend to what: the shortest interval on the clinical ladder (Q8/12/18/24/36/48) whose predicted trough clears the target at the dose that will actually be dispensed — checked after rounding, never shorter than the interval the patient is already on. If no clinical interval clears it, it says so and switches to dose-by-level: give one dose, then re-dose only when a measured level has fallen below target, rather than implying a guarantee no interval can make. And a high trough is fixed by lengthening the interval, not by trimming the dose — a cut lowers the peak you need for killing while barely shortening the time above the toxicity threshold.

NTM amikacin — daily vs thrice-weekly

For nontuberculous mycobacterial (NTM) pulmonary disease, the 2020 ATS/ERS/ESCMID/IDSA guideline (Daley CL et al. Clin Infect Dis 2020;71(4):e1–e36, doi:10.1093/cid/ciaa241) sets IV amikacin levels that depend on the regimen, not just the drug — and this tool deliberately departs from one of them, so the two are separated here rather than blended:

Guideline / primer — 15 mg/kg daily (or 5×/week) → peak 35–45 mcg/mL; 15–25 mg/kg thrice-weekly → peak 65–80 mcg/mL at 25 mg/kg; trough undetectable for both (as restated in Cimino C et al. Open Forum Infect Dis 2024;11(4):ofae128, doi:10.1093/ofid/ofae128).
What this tool targets — Daily 10–15 mg/kg → peak 35–45, trough <5. Thrice-weekly (TIW) 15–25 mg/kg → a graduated band: 45–65 by default, with 65–80 carried only as the guideline “expected level” at 25 mg/kg — not a validated efficacy target (the real-world median achieved TIW peak is ~47 mcg/mL); trough <5, ideally undetectable.

Applying the wrong regimen's target to the wrong schedule — daily's 35–45 to a TIW patient, or the reverse — is a real, clinically consequential mix-up, which is why this tool checks the stated target against the selected regimen and warns on a mismatch rather than silently accepting either number.

Two caveats worth holding onto. First, these targets are pulmonary-derived — the 2020 guideline sets no separate target for extrapulmonary or rapidly-growing-mycobacteria (RGM) disease. Rather than extrapolate the pulmonary rows, the calculator targets 8–10× the isolate MIC there, floored at the practical 35–45 band and capped at the achievable amikacin ceiling of 80 mcg/mL; with no MIC supplied it defaults to 35–45, the common low-MIC M. fortuitum case.

Second — and this is the trap worth naming — the TIW peak of 65–80 is not derived from Cmax:MIC at all. It is simply the concentration a 25 mg/kg dose is expected to produce. The tempting rationalisation is that 65–80 is “Cmax:MIC 8–10 at an amikacin breakpoint MIC of 8”: the arithmetic works and the pharmacology does not. Mycobacterial susceptibility is governed by CLSI M24/M62, not M100, and there has never been an 8 mcg/mL NTM amikacin breakpoint — for M. avium complex the intravenous amikacin breakpoints are S ≤16 / I 32 / R ≥64 mcg/mL (Gill CM, Chamberland R, Abate G. Pathogens 2025;14(6):583, doi:10.3390/pathogens14060583, citing CLSI 2023). At those MICs a genuine Cmax:MIC of 8–10 would demand a peak of 128–160 mcg/mL — unreachable at any tolerable dose, which is exactly why NTM peak targets are descriptive rather than PK/PD-derived. Nor is 8 the Gram-negative breakpoint any more: CLSI M100 (33rd ed, 2023) sets amikacin S ≤4 / R ≥16 and USCAST 2025 sets S ≤2 / S-TDM 4 / R ≥8 (Bhavnani SM et al. Open Forum Infect Dis 2025;12(12):ofaf426, doi:10.1093/ofid/ofaf426), while EUCAST retains ≤8 — three bodies that disagree, so none of them is “the” breakpoint. Where a real MIC is available, Cmax:MIC 8–10 against that isolate is the honest calculation.