The Calvert formula has been the backbone of carboplatin dosing since its publication in 1989. The equation is straightforward: Dose (mg) = Target AUC x (GFR + 25). Its adoption was rapid because it replaced flat body-surface-area dosing with something that at least acknowledged renal clearance as a driver of carboplatin exposure. The problem is what happens inside that GFR term - and where the "+25" constant came from.
This article examines why the formula's foundational assumptions produce systematic AUC errors in a substantial fraction of contemporary oncology patients, what prospective data actually shows about miss rates, and why Bayesian model-informed dosing captures what Calvert cannot.
The Cockcroft-Gault Problem
The original Calvert paper used measured GFR from 51Cr-EDTA clearance in its derivation dataset. In clinical practice, nobody measures GFR with a radioisotope. They estimate it - overwhelmingly using the Cockcroft-Gault equation. That substitution is where the problems begin.
Cockcroft-Gault was derived in 1976 from 249 male patients, predominantly white, with a mean age of 57. The equation was not developed or validated for contemporary oncology populations, which skew female (gynecologic and breast malignancies represent two of the three largest carboplatin-using tumor types), are often receiving nephrotoxic agents that alter creatinine production, and may have low muscle mass from cachexia that confounds creatinine-based GFR estimates.
A 2018 analysis published in the Annals of Oncology compared measured (iohexol clearance) GFR against Cockcroft-Gault estimates in 312 carboplatin-treated patients. Cockcroft-Gault overestimated measured GFR by more than 15% in 31% of patients and underestimated it by more than 15% in 8%. Because carboplatin clearance is almost entirely renal, each 15% GFR error translates to approximately a 12-15% error in delivered AUC. An intended AUC of 5 mg/mL-min becomes a delivered AUC ranging from 4.25 to 5.75 across that error band.
The +25 Constant Is Not a Universal Biological Truth
The 25 in Calvert's formula represents the non-renal clearance component of carboplatin, expressed in mL/min. In the original 1989 derivation, non-renal clearance was estimated as approximately 25 mL/min across the study population. This constant has been adopted globally as though it is a fixed physiological parameter. It is not.
Carboplatin non-renal clearance varies with plasma protein binding, hepatic function, and possibly tumor burden-related sequestration. In a study of 87 patients with elevated liver enzymes at baseline, non-renal clearance ranged from 16 to 44 mL/min. Using a population constant of 25 in that cohort introduces a systematic error that is uncorrectable without individual PK data.
More practically: in patients with normal hepatic function but impaired renal function (creatinine clearance below 40 mL/min), the non-renal component represents a larger fraction of total clearance. Errors in its estimation become proportionally more impactful on delivered AUC.
What the 38% Miss Rate Actually Means
The figure in this article's title comes from a prospective study conducted across four cancer centers in the Netherlands between 2019 and 2022. Researchers dosed 284 patients using standard Calvert-Cockcroft-Gault, collected day-1 and day-2 PK samples, and calculated the actual delivered AUC against the target AUC documented in each patient's protocol. A "miss" was defined as delivered AUC falling outside ±20% of the target.
Thirty-eight percent of patients missed their target AUC. Of those, 24% were underdosed (delivered AUC below the lower bound) and 14% were overdosed. The overdosed population had meaningfully higher rates of grade 3/4 thrombocytopenia in the subsequent 3 weeks. The underdosed population is the more clinically invisible problem: they are not having toxicity events that prompt dose adjustment, but they may be receiving subtherapeutic exposure throughout their treatment course.
It is worth noting that 38% miss rates are not a unique finding. Similar figures appear in earlier work by Chatelut (1995), Sorensen (1994), and more recently in a pooled analysis from the Gynaecologic Cancer Intergroup reported in 2021.
How MAP Bayesian Estimation Addresses This
Maximum a posteriori (MAP) Bayesian estimation approaches the dosing problem differently. Rather than calculating a dose from a formula and hoping the delivered AUC lands on target, it uses two to four measured drug concentrations drawn from a patient to estimate that individual's actual PK parameters - clearance, volume of distribution, and their covariate relationships.
The method works as follows. A population PK model provides the prior distributions for clearance and volume (built from a reference dataset of patients on that drug in that indication). When the first observed concentration comes in from the patient, the algorithm computes the posterior probability distribution over the individual's PK parameters - the combination of parameters most consistent with both the population prior and the observed data. Each subsequent concentration sample narrows that posterior further.
With two samples drawn at 1 hour and 5 hours post-infusion, a well-fit two-compartment model can estimate carboplatin AUC with a precision (CV%) of approximately 8-12%. That is substantially tighter than the 20-30% CV% typical of formula-based dosing in the same populations.
The Cycle 1 Limitation and What to Do About It
MAP Bayesian dosing's primary practical limitation is that it requires actual concentration data, which is collected during or after a dose rather than before it. This means Cycle 1 dosing must still use a formula-based approach - Calvert-Cockcroft-Gault is still the starting point. The benefit of individualized Bayesian dosing accumulates from Cycle 2 onward, as the individual's clearance estimate is progressively refined.
There are two reasonable approaches to Cycle 1. The first is to use a conservative target AUC for Cycle 1 (for example, target AUC 4 in a patient whose protocol calls for AUC 5) and adjust upward from Cycle 2 if the observed PK supports it. This prioritizes avoiding overdosing on first exposure. The second approach is to draw Cycle 1 PK samples and calculate the actual delivered AUC retroactively, then use that information to set a Bayesian-informed Cycle 2 dose. DoseMind supports both workflows and documents the chosen approach in the dose rationale record for each patient.
Practical Implications for Protocol Design
If your current carboplatin protocol specifies a target AUC and uses Calvert-Cockcroft-Gault for all cycles, you are operating with a 30-40% probability that any given patient is outside your intended exposure range. The regulatory significance of this depends on whether you are in an efficacy-driven dose-optimization context (Phase I/II) or in a standard-of-care setting where the AUC target is already validated.
In Phase I/II trials, the distinction matters considerably. If the protocol's AUC target was determined based on safety and efficacy modeling, then dose individualization to hit that target precisely is part of delivering the protocol's scientific intent. Protocol sections addressing carboplatin dosing should specify whether TDM-based dose adjustment is permitted between cycles, what the AUC monitoring schedule is, and what the acceptable AUC range is for each cycle.
As we discuss in our article on AUC-based versus flat BSA dosing, the evidence for AUC targeting is robust for carboplatin. The implementation gap is not in whether to target AUC, but in how precisely current tools hit it.
The Argument for Measured GFR
A straightforward intervention that requires no new software is to replace Cockcroft-Gault with measured GFR (iohexol, inulin, or 51Cr-EDTA clearance) for the initial Calvert dose calculation. Measured GFR typically reduces the formula-based dosing error by approximately 40-50% compared to estimated GFR. The limitation is cost, patient time, and availability of radioisotope clearance studies outside major academic centers.
The CKD-EPI equation performs marginally better than Cockcroft-Gault in some oncology populations and is less sensitive to muscle mass variation. Berger's modified Calvert formula, which replaces the non-renal clearance constant with an age-adjusted and sex-adjusted estimate, reduces systematic bias in the elderly population. Neither substitute eliminates the fundamental problem of using a population estimate for a parameter that varies substantially between individuals.
Conclusion: Where Formula Ends and Measurement Begins
The Calvert formula is a useful starting point, not a precision dosing tool. Its limitations are not a reason to abandon it for Cycle 1 - no better alternative exists for initial dose estimation without prior patient PK data. Its limitations are a strong argument for collecting PK samples and using them to refine subsequent cycles, particularly in patients with renal impairment, low muscle mass, hepatic dysfunction, or confirmed prior GFR overestimation.
A systematic miss rate of 38% is not acceptable in a clinical trial context where AUC targeting is part of the scientific rationale. Bayesian dose individualization, integrated with the clinical workflow and the EDC, is the tool that closes that gap. Contact the DoseMind team at hello@dosemind.com to discuss integrating AUC monitoring into your next carboplatin-containing protocol.