Converting Creatinine-Based Estimates Into Model Covariates

Why the Scale of a Renal Covariate Matters

Creatinine-based renal estimates can enter a population pharmacokinetic model in different forms. Cockcroft–Gault gives an estimate in mL/min, whereas eGFR equations commonly report a value indexed to a standard body surface area (BSA), in mL/min/1.73 m². These values do not have the same units or necessarily the same numerical scale. If a model uses one in place of the other, its coefficient also has a different interpretation.

This tutorial works through two standard calculations, a unit conversion, and BSA de-indexing. It then shows why a coefficient cannot be copied unchanged when the renal covariate is rescaled. As emphasized in the earlier material on choosing renal function metrics, the calculation is only one part of the decision: the metric must also suit the intended population, drug, and modeling question.

Definition: Cockcroft–Gault creatinine clearance is an estimate of creatinine clearance, conventionally reported in mL/min. Indexed eGFR is an estimated glomerular filtration rate expressed relative to a standard BSA, conventionally in mL/min/1.73 m². De-indexed eGFR is obtained by multiplying indexed eGFR by the individual's BSA divided by 1.73 m²; its units are mL/min.

Calculating Cockcroft–Gault Creatinine Clearance

A commonly used adult Cockcroft–Gault equation is

\mathrm{CrCl}_{\mathrm{CG}}= \frac{(140-\mathrm{age})\,W}{72\,S_{\mathrm{cr}}} \times \begin{cases} 0.85, & \text{female}\\ 1, & \text{male} \end{cases}

Here, age is in years, \(W\) is the selected body weight in kg, and serum creatinine \(S_{\mathrm{cr}}\) is in mg/dL. The result is in mL/min. The conventional female multiplier is part of this equation; it is not a general correction applicable to every renal estimate. The weight convention also needs to be specified: actual, ideal, or adjusted weight choices may differ, particularly at body-size extremes. The equation alone does not settle that choice.

If serum creatinine is measured in μmol/L, convert it to mg/dL using \(S_{\mathrm{cr}}(\mathrm{mg/dL})=S_{\mathrm{cr}}(\mathrm{\mu mol/L})/88.4\). Equivalently, replace the denominator \(72S_{\mathrm{cr}}\) with \((72/88.4)S_{\mathrm{cr}}(\mathrm{\mu mol/L})\). Keep the conversion and rounding consistent; otherwise the two forms can appear to disagree.

Worked Example: Cockcroft–Gault With a Creatinine Unit Conversion

Consider a hypothetical 65-year-old woman with weight 57.6 kg and serum creatinine 0.800 mg/dL. Calculate Cockcroft–Gault creatinine clearance, then verify the calculation using μmol/L.

Using mg/dL, substitute the stated values and the conventional female multiplier:

\mathrm{CrCl}_{\mathrm{CG}} = \frac{(140-65)(57.6)}{72(0.800)}\times0.85 = \frac{4320}{57.6}\times0.85 = 75.00\times0.85 = 63.75\ \mathrm{mL/min}.

The creatinine conversion is \(0.800\times88.4=70.72\ \mathrm{\mu mol/L}\). Using the equivalent denominator coefficient \(72/88.4=0.8144796\),

\mathrm{CrCl}_{\mathrm{CG}} = \frac{75(57.6)}{(0.8144796)(70.72)}\times0.85 = \frac{4320}{57.6}\times0.85 = 63.75\ \mathrm{mL/min}.

Both calculations agree because the conversion coefficient has not been rounded prematurely. This is an estimated creatinine clearance from the stated equation, not a measured clearance.

Calculating and De-Indexing eGFR

For a concrete eGFR calculation, this tutorial uses the 2021 CKD-EPI creatinine equation. It is a race-free equation for adults with serum creatinine in mg/dL. For age in years,

\mathrm{eGFR} = 142 \times \min(S_{\mathrm{cr}}/\kappa,1)^{\alpha} \times \max(S_{\mathrm{cr}}/\kappa,1)^{-1.200} \times 0.9938^{\mathrm{age}} \times 1.012\ \text{if female},

where \(\kappa=0.7\) and \(\alpha=-0.241\) for females, and \(\kappa=0.9\) and \(\alpha=-0.302\) for males. Omit the factor 1.012 for males. The output is indexed eGFR in mL/min/1.73 m². The equation requires the specified creatinine units and should not be applied as though it were the Cockcroft–Gault equation.

To convert indexed eGFR to an absolute estimate for an individual's BSA, use

\mathrm{eGFR}_{\mathrm{absolute}} = \mathrm{eGFR}_{\mathrm{indexed}} \times \frac{\mathrm{BSA}}{1.73\ \mathrm{m^2}}.

A common BSA calculation is the Mosteller formula, with height in cm and weight in kg:

\mathrm{BSA}\ (\mathrm{m^2}) = \sqrt{\frac{\mathrm{height}\ (\mathrm{cm})\times\mathrm{weight}\ (\mathrm{kg})}{3600}}.

De-indexing changes the scale and units; it does not change the underlying eGFR equation or turn the estimate into a measured GFR.

Worked Example: CKD-EPI eGFR and BSA De-Indexing

Use the same hypothetical woman: age 65 years, serum creatinine 0.800 mg/dL, weight 57.6 kg, and height 160 cm. Calculate indexed eGFR and then de-index it.

For a female, \(\kappa=0.7\), \(\alpha=-0.241\), and the sex factor is 1.012. The ratio \(S_{\mathrm{cr}}/\kappa=0.800/0.7=1.142857\), so the minimum term is 1 and the maximum term is \(1.142857^{-1.200}\), approximately 0.8519. Also, \(0.9938^{65}\approx0.6675\). Thus,

\mathrm{eGFR} \approx 142(1)(0.8519)(0.6675)(1.012) \approx 81.72\ \mathrm{mL/min/1.73\ m^2}.

The BSA is

\mathrm{BSA} = \sqrt{\frac{160(57.6)}{3600}} = \sqrt{2.56} = 1.60\ \mathrm{m^2}.

Therefore, the de-indexed estimate is

\mathrm{eGFR}_{\mathrm{absolute}} = 81.72\times\frac{1.60}{1.73} \approx 75.58\ \mathrm{mL/min}.

The indexed and de-indexed results describe different scales: 81.72 mL/min/1.73 m² and approximately 75.58 mL/min, respectively. The hypothetical Cockcroft–Gault result above is 63.75 mL/min. Their difference is not an arithmetic contradiction: the equations estimate different quantities and use different inputs and constructions.

Why Scaling Changes a Covariate Coefficient

A model coefficient is interpreted per unit of the covariate actually entered. If a renal covariate is multiplied by a constant, its coefficient must be divided by that constant to preserve the same predictions. This is a change of parameterization, not evidence that the biological relationship has changed.

R_{\mathrm{absolute}} = cR_{\mathrm{indexed}}, \qquad c=\frac{\mathrm{BSA}}{1.73}.

Suppose a model for typical clearance has the linear form

\mathrm{CL}_{\mathrm{typ}} = \theta_0+\theta_{\mathrm{indexed}}R_{\mathrm{indexed}}.

At a fixed BSA, substituting \(R_{\mathrm{indexed}}=R_{\mathrm{absolute}}/c\) gives

\mathrm{CL}_{\mathrm{typ}} = \theta_0+ \left(\frac{\theta_{\mathrm{indexed}}}{c}\right) R_{\mathrm{absolute}}.

Thus the coefficient on the absolute measure is \(\theta_{\mathrm{absolute}}=\theta_{\mathrm{indexed}}/c\). Its units also change. This equivalence holds for a fixed conversion factor. If BSA varies across individuals, the factor \(c\) varies too; indexed and absolute eGFR are then not related by one common constant across the dataset.

Worked Example: Rescaling a Renal Covariate Without Changing Predictions

Suppose a hypothetical model uses indexed eGFR and has \(\mathrm{CL}_{\mathrm{typ}}=10+0.200R_{\mathrm{indexed}}\), where clearance is in L/h and indexed eGFR is in mL/min/1.73 m². For an individual with BSA 1.60 m² and indexed eGFR 75.0 mL/min/1.73 m², compare the indexed and absolute parameterizations.

First de-index the covariate:

R_{\mathrm{absolute}} = 75.0\times\frac{1.60}{1.73} \approx 69.36\ \mathrm{mL/min}.

The original model predicts

\mathrm{CL}_{\mathrm{typ}} = 10+0.200(75.0) = 25.00\ \mathrm{L/h}.

Here \(c=1.60/1.73\), so the equivalent coefficient on absolute eGFR is

\theta_{\mathrm{absolute}} = \frac{0.200}{1.60/1.73} = 0.21625\ \frac{\mathrm{L/h}}{\mathrm{mL/min}}.

Using the unrounded conversion, \(10+0.21625(75.0\times1.60/1.73)=25.00\ \mathrm{L/h}\), as required. If someone instead keeps the coefficient 0.200 while substituting absolute eGFR, the result is \(10+0.200(69.36)=23.87\ \mathrm{L/h}\). That is a different model, not merely a different display of the same model.

Assumptions, Diagnostics, and Alternatives

These equations rely on serum creatinine being interpretable as a relatively stable marker of renal function in the population where the equation is used. A changing creatinine concentration during acute kidney injury may lag behind changes in filtration, so a steady-state estimate can be misleading. Creatinine production also varies with muscle mass, diet, age, and other factors; unusual muscle mass can make creatinine-based estimates systematically unrepresentative of filtration. Use the stated equation within its intended population and follow the relevant laboratory and clinical conventions.

Before using a calculated covariate, check the input units, equation version, age, sex-specific terms, weight convention where applicable, and BSA units. Then inspect the resulting values and model inputs. Implausible values, sharp discrepancies after a unit conversion, or covariate ranges concentrated at implausible limits can reveal transcription errors, incorrect units, or inappropriate extrapolation. Record whether eGFR is indexed or de-indexed in both the dataset and model documentation.

The alternatives are not interchangeable labels. Cockcroft–Gault estimates creatinine clearance and includes body weight directly. CKD-EPI produces indexed eGFR, which can be de-indexed when an absolute scale is needed. Which measure is appropriate depends on the modeling objective, data availability, and intended interpretation; a difference between estimates alone does not establish that one is erroneous.

Measure entered	Typical units	Scale and coefficient interpretation
Cockcroft–Gault creatinine clearance	mL/min	Per 1 mL/min of the calculated clearance; depends on the selected weight convention.
CKD-EPI indexed eGFR	mL/min/1.73 m²	Per 1 mL/min/1.73 m²; standardized to reference BSA.
De-indexed eGFR	mL/min	Per 1 mL/min of the BSA-adjusted estimate; conversion from indexed values varies with BSA.

A further limitation is that a creatinine-based estimate is a covariate proxy, not a direct measurement of drug elimination. A fitted association between the estimate and clearance does not by itself show that changing renal function causes the modeled change. Model form, parameter estimation, and the biological meaning of the covariate require separate consideration.

Derivation and a Common Pitfall

The coefficient conversion follows directly by substitution. Start with \(R_{\mathrm{absolute}}=cR_{\mathrm{indexed}}\), so \(R_{\mathrm{indexed}}=R_{\mathrm{absolute}}/c\), for fixed \(c>0\). In the indexed model,

\mathrm{CL}_{\mathrm{typ}} = \theta_0+\theta_{\mathrm{indexed}}R_{\mathrm{indexed}} = \theta_0+\theta_{\mathrm{indexed}} \frac{R_{\mathrm{absolute}}}{c} = \theta_0+ \left(\frac{\theta_{\mathrm{indexed}}}{c}\right)R_{\mathrm{absolute}}.

The intercept is unchanged and the slope is divided by \(c\). The units confirm the result: a slope expressed per indexed unit cannot be read as a slope per mL/min without conversion.

Common pitfall: “Both renal estimates are reported as roughly 70, so I can substitute one for the other and keep the model coefficient.” The reasoning fails because the estimates may have different definitions, units, and scales. Indexed eGFR also requires BSA-specific de-indexing to obtain an absolute estimate. The correction is to identify the exact covariate used, document its equation and units, recalculate if needed, and transform or refit the coefficient rather than copying it unchanged. If BSA varies, account for that varying conversion rather than assuming one universal slope transformation.

Key takeaway: State the renal equation, its inputs, and its output units before interpreting a covariate effect. Cockcroft–Gault, indexed eGFR, and de-indexed eGFR are distinct quantities; consistent scaling is essential for meaningful model coefficients and predictions.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions.

A hypothetical 70-year-old man weighs 70 kg and has serum creatinine 1.00 mg/dL. Using the stated Cockcroft–Gault equation, what is his estimated creatinine clearance?
A hypothetical indexed eGFR is 90.0 mL/min/1.73 m² and BSA is 1.50 m². What is the de-indexed estimate?
At fixed BSA 1.50 m², a model uses indexed eGFR with slope 0.300 L/h per mL/min/1.73 m². What slope gives the same predictions when absolute eGFR is used?
Why is a serum creatinine-based estimate potentially misleading when serum creatinine is changing rapidly?

Solutions. (1) \(\mathrm{CrCl}_{\mathrm{CG}}=[(140-70)(70)]/[72(1.00)]=4900/72\approx68.06\ \mathrm{mL/min}\); no female multiplier applies. (2) \(90.0(1.50/1.73)\approx78.03\ \mathrm{mL/min}\). (3) Here \(c=1.50/1.73\), so the absolute-scale slope is \(0.300/c=0.300(1.73/1.50)=0.3460\ \mathrm{L/h\ per\ mL/min}\). (4) The estimate assumes creatinine is sufficiently stable to reflect filtration; during rapid change, creatinine can lag the change in renal function, so the calculated value may not represent current filtration.