Power Models for Continuous Renal Function Effects

Why Normalize Renal Function?

A continuous renal-function measure can describe gradations in kidney function without first dividing patients into groups. A power model is one way to express how those gradations relate to drug clearance. Normalizing the measure to a reference value makes the model’s coefficient interpretable: the reference clearance is the predicted typical clearance for an individual whose renal-function value equals that reference.

This is a model for a covariate relationship, not a claim that renal function alone determines clearance. Other covariates and individual differences may also matter. The renal-function metric should be chosen and prepared for modeling before specifying this relationship; the model uses the resulting positive, consistently scaled covariate.

Definition: A normalized renal-function power model for typical clearance is $$ CL_{\mathrm{typ}}(R) = CL_{\mathrm{ref}} \left(\frac{R}{R_{\mathrm{ref}}}\right)^{\theta}, $$ where $R$ is the individual’s renal-function covariate, $R_{\mathrm{ref}}>0$ is its chosen reference value, $CL_{\mathrm{ref}}>0$ is typical clearance at $R=R_{\mathrm{ref}}$, and $\theta$ is the power exponent. The ratio $R/R_{\mathrm{ref}}$ is dimensionless; $CL_{\mathrm{typ}}$ and $CL_{\mathrm{ref}}$ have the same clearance units.

For example, if $R$ is estimated glomerular filtration rate (eGFR) recorded in $\mathrm{mL/min/1.73\,m^2}$, then $R_{\mathrm{ref}}$ must be in the same units. A reference value is a modeling anchor, not a universal definition of normal renal function. Its choice affects the numerical value of $CL_{\mathrm{ref}}$, but, as shown below, does not by itself change the predictions of an exactly reparameterized model.

What the Exponent Means

The exponent describes relative change. At the reference value, the model predicts $CL_{\mathrm{ref}}$, because $R/R_{\mathrm{ref}}=1$ and $1^\theta=1$. Away from the reference, the exponent determines how strongly the clearance prediction responds to a proportional change in renal function.

\frac{CL_{\mathrm{typ}}(R_2)}{CL_{\mathrm{typ}}(R_1)} = \left(\frac{R_2}{R_1}\right)^\theta, \qquad \frac{d\log CL_{\mathrm{typ}}}{d\log R}=\theta.

The first equation compares predictions at any two positive renal-function values, holding other model inputs fixed. The second states that $\theta$ is the elasticity: locally, a 1% increase in $R$ corresponds to an approximately $\theta$% change in typical clearance. This approximation concerns small changes; for a finite change, use the exact ratio in the first equation. A positive exponent gives higher predicted clearance at higher $R$; an exponent of zero gives no renal-function effect in this model. A negative exponent would imply lower predicted clearance at higher $R$, so its direction would need careful scrutiny in context.

Worked Example: Predicting Clearance at Two Renal-Function Values

Suppose a hypothetical model uses $R_{\mathrm{ref}}=90\ \mathrm{mL/min/1.73\,m^2}$, $CL_{\mathrm{ref}}=7.20\ \mathrm{L/h}$, and $\theta=0.60$. Calculate typical clearance at $R=45$ and $R=135\ \mathrm{mL/min/1.73\,m^2}$.

At $R=45$. The value is half the reference, so

CL_{\mathrm{typ}}(45) = 7.20\left(\frac{45}{90}\right)^{0.60} = 7.20(0.5^{0.60}) \approx 7.20(0.659754) = 4.750\ \mathrm{L/h}.

At $R=135$. This value is 1.5 times the reference:

CL_{\mathrm{typ}}(135) = 7.20\left(\frac{135}{90}\right)^{0.60} = 7.20(1.5^{0.60}) \approx 7.20(1.275425) = 9.183\ \mathrm{L/h}.

The predictions are below and above $7.20\ \mathrm{L/h}$, respectively, because the exponent is positive. The ratio of the second prediction to the first is $(135/45)^{0.60}=3^{0.60}$, approximately $1.933$. This also equals $9.183/4.750$, up to rounding.

Log-Scale Form and Reference-Value Changes

Taking logarithms exposes the structure of the model. This step requires positive $R$, $R_{\mathrm{ref}}$, and clearance, as specified above. The resulting expression is linear in the logarithm of the normalized covariate, with slope $\theta$.

\log CL_{\mathrm{typ}}(R) = \log CL_{\mathrm{ref}} + \theta\log\left(\frac{R}{R_{\mathrm{ref}}}\right).

This form justifies the log-log interpretation: a change in $\log R$ produces a change in $\log CL_{\mathrm{typ}}$ at rate $\theta$. It also gives a direct result for changing the reference. Suppose the same relationship is written using a new reference $R_{\mathrm{ref}}^*$. To preserve predictions, set the new reference clearance equal to the old model’s prediction at $R_{\mathrm{ref}}^*$:

CL_{\mathrm{ref}}^* = CL_{\mathrm{ref}} \left(\frac{R_{\mathrm{ref}}^*}{R_{\mathrm{ref}}}\right)^\theta.

Substituting this into the model with the new reference gives

CL_{\mathrm{ref}}^* \left(\frac{R}{R_{\mathrm{ref}}^*}\right)^\theta = CL_{\mathrm{ref}} \left(\frac{R_{\mathrm{ref}}^*}{R_{\mathrm{ref}}}\right)^\theta \left(\frac{R}{R_{\mathrm{ref}}^*}\right)^\theta = CL_{\mathrm{ref}}\left(\frac{R}{R_{\mathrm{ref}}}\right)^\theta.

The cancellation proves that changing the reference changes the reported reference clearance but not the predictions, provided the exponent and covariate definition remain unchanged and the reference clearance is transformed accordingly.

Worked Example: Re-expressing the Same Model at a New Reference

A hypothetical model has $CL_{\mathrm{ref}}=6.50\ \mathrm{L/h}$ at $R_{\mathrm{ref}}=100\ \mathrm{mL/min/1.73\,m^2}$, with $\theta=0.40$. Re-express it at $R_{\mathrm{ref}}^*=50\ \mathrm{mL/min/1.73\,m^2}$, then compare predictions at $R=75$.

First calculate the clearance at the new reference:

CL_{\mathrm{ref}}^* = 6.50\left(\frac{50}{100}\right)^{0.40} = 6.50(0.757858) \approx 4.926\ \mathrm{L/h}.

Using the original reference, the prediction at $R=75$ is

6.50\left(\frac{75}{100}\right)^{0.40} = 6.50(0.891302) \approx 5.793\ \mathrm{L/h}.

Using the new reference and its transformed clearance gives

4.9260788\left(\frac{75}{50}\right)^{0.40} = 4.9260788(1.176079) \approx 5.793\ \mathrm{L/h}.

The calculations use unrounded values for the transformed reference clearance before the final rounding. The agreement is expected from the algebraic identity above; small discrepancies from calculations using only the displayed rounded values are rounding effects, not changes in the underlying model.

Assumptions and Model Checks

The equation is simple, but its use depends on substantive and mathematical conditions. First, $R$ must be positive because the model uses a power of $R/R_{\mathrm{ref}}$ and the log-scale form uses its logarithm. Second, the covariate must have a consistent definition and units across individuals. Changing the renal-function metric or its scale changes what the exponent describes. Third, the power relationship should be a reasonable approximation over the range where predictions will be used. A good fit in a narrow observed range does not guarantee reliable extrapolation far beyond it.

In a population pharmacokinetic model, this equation describes the typical clearance relationship with renal function, conditional on the other included model inputs. It does not assert that every individual lies exactly on that curve. Individual variability and unexplained differences are handled by the broader model; they do not alter the meaning of the typical prediction defined here.

Useful checks focus on whether the assumed form is compatible with the data and intended predictions:

Check the covariate scale and support. Confirm positive values, consistent units, and adequate data coverage around the values for which predictions matter.
Inspect the log-scale relationship. Where typical clearance estimates or relevant residual summaries can be examined, assess whether the relationship with $\log(R/R_{\mathrm{ref}})$ is approximately linear. Curvature can indicate that one constant exponent is inadequate.
Look for systematic residual patterns. Trends remaining with renal function suggest the fitted relationship has not captured the observed pattern. A small overall residual spread does not rule out a systematic trend.
Consider precision near the extremes. Sparse observations at very low or high $R$ can make the exponent and predictions there uncertain. Do not treat an extrapolated curve as equally supported across all possible values.
Check biological and clinical plausibility. Confirm that the direction and magnitude of the estimated relationship make sense for the drug and population, while recognizing that plausibility alone does not establish model adequacy.

Comparison With Other Continuous Forms

A power model is not the only way to relate a positive covariate to clearance. Its particular feature is a constant elasticity: proportional changes in renal function have the same modeled relative effect at different points on the scale. The following comparison concerns the mathematical form and interpretation, not a claim that one form is always preferable.

Form	Relative interpretation	Key distinction
Power: $CL_{\mathrm{typ}}=CL_{\mathrm{ref}}(R/R_{\mathrm{ref}})^\theta$	A proportional change in $R$ changes clearance by a ratio determined by $\theta$.	Constant elasticity on the log-log scale; positive clearance for positive parameters and covariate.
Linear: $CL_{\mathrm{typ}}=a+bR$	A fixed absolute change in $R$ corresponds to a fixed absolute change in clearance.	Not a constant percentage effect; depending on $a$ and $b$, predictions may become nonpositive outside the fitted range.
Log-linear in $R$: $\log CL_{\mathrm{typ}}=a+bR$	A fixed absolute change in $R$ corresponds to a fixed multiplicative change in clearance.	Its effect is not constant per proportional change in $R$; it is exponential in the untransformed covariate.

These forms encode different assumptions about how effects change across the renal-function range. Model choice should be guided by the scientific question, observed coverage, predictive performance, and plausibility. The power form is especially useful when a relative interpretation is desired, but it can still be inadequate if the data show curvature or if its implied behavior outside the observed range is implausible.

Common Mistakes and Limitations

Common pitfall: Treating $\theta$ as the percentage change in clearance for every fixed increase of one renal-function unit is incorrect. The exponent is a relative-effect measure: a doubling of $R$ multiplies typical clearance by $2^\theta$, while a one-unit increase has a different relative effect depending on the starting value of $R$. Calculate the exact ratio for finite changes.

Another mistake is to change $R_{\mathrm{ref}}$ while leaving $CL_{\mathrm{ref}}$ unchanged and then assume the predictions are preserved. The reference clearance is defined at the reference covariate value; moving that value requires transforming the clearance as shown above. Finally, the power model does not establish a causal effect of renal function, guarantee accurate predictions beyond observed data, or capture all mechanisms of drug elimination. It is a parameterization of a covariate relationship that must be assessed in the context of the model and data.

Worked Example: Interpreting a Finite Change Correctly

Suppose a hypothetical model has $\theta=0.75$. Compare predicted typical clearance at $R=60$ and $R=90\ \mathrm{mL/min/1.73\,m^2}$, holding other inputs fixed. The ratio is

\frac{CL_{\mathrm{typ}}(60)}{CL_{\mathrm{typ}}(90)} = \left(\frac{60}{90}\right)^{0.75} = \left(\frac{2}{3}\right)^{0.75} \approx 0.7378.

Thus, predicted clearance at $R=60$ is about 73.78% of its prediction at $R=90$, or about 26.22% lower. This is not a 30-unit increase or decrease interpreted through a fixed percentage-per-unit rule; it is the model’s exact prediction for a change from 90 to 60.

Check Your Understanding

Answer each question using the power model and show the relevant calculation. Solutions follow the questions.

A model has $CL_{\mathrm{ref}}=6.00\ \mathrm{L/h}$ at $R_{\mathrm{ref}}=80\ \mathrm{mL/min/1.73\,m^2}$, with $\theta=0.50$. What is predicted typical clearance at $R=20$?
For $\theta=0.40$, by what percentage does predicted clearance change when renal function doubles?
A model has $CL_{\mathrm{ref}}=5.00\ \mathrm{L/h}$ at $R_{\mathrm{ref}}=75$, with $\theta=0.80$. What reference clearance preserves predictions if the reference is changed to $R_{\mathrm{ref}}^*=100$, in the same units?
In one sentence, interpret $\theta=0.60$ as an elasticity, and state whether it describes an exact percentage change for every finite change in $R$.

Solutions. (1) The covariate ratio is $20/80=0.25$, and $0.25^{0.50}=0.500$. Therefore $CL_{\mathrm{typ}}=6.00(0.500)=3.00\ \mathrm{L/h}$. (2) The clearance ratio is $2^{0.40}\approx1.3195$, so doubling $R$ corresponds to an increase of $(1.3195-1)\times100\%\approx31.95\%$. (3) The transformed reference clearance is $5.00(100/75)^{0.80}\approx5.00(1.2589)=6.295\ \mathrm{L/h}$. (4) A 1% increase in $R$ corresponds locally to an approximately 0.60% increase in typical clearance; this is a local approximation, and finite changes require the exact ratio $(R_2/R_1)^{0.60}$.

Form	Relative interpretation	Key distinction
Power: \(CL_{\mathrm{typ}}=CL_{\mathrm{ref}}(R/R_{\mathrm{ref}})^\theta\)	A proportional change in \(R\) changes clearance by a ratio determined by \(\theta\).	Constant elasticity on the log-log scale; positive clearance for positive parameters and covariate.
Linear: \(CL_{\mathrm{typ}}=a+bR\)	A fixed absolute change in \(R\) corresponds to a fixed absolute change in clearance.	Not a constant percentage effect; depending on \(a\) and \(b\), predictions may become nonpositive outside the fitted range.
Log-linear in \(R\): \(\log CL_{\mathrm{typ}}=a+bR\)	A fixed absolute change in \(R\) corresponds to a fixed multiplicative change in clearance.	Its effect is not constant per proportional change in \(R\); it is exponential in the untransformed covariate.