Stable Calculations at Extreme Extraction Ratios

Why Extreme Extraction Ratios Need Careful Calculation

Hepatic clearance calculations can be numerically fragile at the same limits where their physiological interpretation is often clearest. When extraction is very small, a formula that subtracts two nearly equal numbers can lose most or all meaningful digits. When extraction is nearly complete, clearance may round to hepatic blood flow, hiding a small but important residual fraction of drug leaving the liver.

This tutorial focuses on numerical evaluation, not on choosing among liver models. The standard model inputs used here are hepatic blood flow \(Q_h\), in L/h, and the effective intrinsic clearance \(x=f_{u,b}CL_{int}\), also in L/h, where \(f_{u,b}\) is the unbound fraction in blood and \(CL_{int}\) is intrinsic clearance on a compatible basis. Define the extraction ratio \(E_h\) as the fraction removed during one passage through the liver, and hepatic availability \(F_h=1-E_h\) as the fraction that escapes that passage.

The model assumptions still matter: these expressions describe a steady-state, linear liver model with the specified flow and effective intrinsic clearance. They do not, by themselves, account for concentration-dependent metabolism, changing flow during passage, or other departures from the chosen model. The earlier material on extraction ratios and liver-model limits provides the physiological interpretation; here the concern is whether the computer evaluates the equations faithfully.

Stable Forms for the Standard Liver Equations

For the well-stirred model, extraction and availability can be computed as complementary ratios:

When \(x\) is much larger than \(Q_h\), a useful equivalent clearance form is

This avoids forming a very large sum \(Q_h+x\) when the inputs differ greatly in scale. It also makes clear that clearance approaches, but does not exceed, \(Q_h\). The gap from flow is often more informative than the rounded clearance itself:

For the parallel-tube model, let \(z=x/Q_h\). Its outlet fraction is \(F_h=\exp(-z)\), so the direct expression for clearance is \(Q_h[1-\exp(-z)]\). At small \(z\), evaluating the bracket by ordinary subtraction is vulnerable to cancellation. Use the standard computational function expm1, which returns \(\exp(u)-1\) accurately when \(u\) is near zero:

At high extraction, retain \(F_h\) as its own quantity rather than trying to recover it by subtracting a rounded \(E_h\) from one. If the exponent is so negative that \(\exp(-z)\) underflows to zero in the computing system, that is a numerical limit on the representable outlet fraction—not evidence that the true fraction is exactly zero.

Why the Reformulations Work

The small-extraction problem is a direct consequence of the series for the exponential. For \(z\) near zero,

The true difference is approximately \(z\). But a computer first rounds \(\exp(-z)\) to a number close to one, then subtracts it from one. If the rounded exponential is exactly one, the computed difference is zero even though \(z>0\). expm1 evaluates the small difference directly, retaining the terms that subtraction can discard. Multiplying that difference by \(Q_h\) then gives clearance without first losing the extraction signal.

The same principle applies to the well-stirred model. For \(r=x/Q_h\), its clearance is \(Q_h r/(1+r)\). When \(r\) is very large, calculate the small complementary ratio \(1/r\) and use \(CL_h=Q_h/(1+1/r)\), rather than relying on a rounded extraction ratio and then subtracting it from one. These algebraic rearrangements do not change the model; they change the route used to evaluate it.

Stable Use of Outlet Fractions in Other Model Calculations

The dispersion model can be evaluated by a model-specific calculation of the outlet-to-inlet ratio \(F_h\). Whatever solver supplies that ratio, the conversion to clearance is still \(Q_h(1-F_h)\). If the solver can return the logarithm of the outlet ratio, \(\log F_h\), evaluate the extraction as \(-\operatorname{expm1}(\log F_h)\), rather than calculating \(\exp(\log F_h)\) and subtracting the result from one. This is particularly useful when \(F_h\) is very close to one. Conversely, when \(F_h\) is tiny, retain \(F_h\) directly if it is needed for first-pass availability or another downstream calculation.

A log-domain calculation is only as accurate as the underlying model solution. If a solver has already rounded a near-one outlet ratio to exactly one, taking its logarithm cannot recover the lost digits. Prefer a solver that exposes a log ratio, a directly calculated extraction, or sufficient internal precision. Check convergence by tightening solver tolerances or increasing precision; the resulting \(F_h\) and \(CL_h\) should stabilize at the precision being reported.

Diagnostics, Comparisons, and Limitations

Stable algebra improves numerical evaluation; it does not validate the biological inputs or select the correct liver model. A computed result should still pass basic checks. For positive \(Q_h\) and \(x\), extraction should lie between zero and one, availability should lie between zero and one, and clearance should lie between zero and \(Q_h\). In the low-extraction limit, well-stirred and parallel-tube clearance approach \(x\). At high extraction, their clearance approaches \(Q_h\), although the predicted residual availability can differ by model. A result outside the physical bounds, a negative clearance, or a value that changes materially with tighter solver tolerances signals an input, implementation, or numerical problem.

The main alternatives are to compute extraction first and multiply by flow, to compute clearance directly from a model-specific formula, or to calculate the complementary outlet fraction first. These are algebraically equivalent in exact arithmetic, but not equally robust in finite precision. Near zero extraction, a direct formula using \(1-\exp(-z)\) is less reliable than an expm1 form. Near complete extraction, a rounded extraction ratio can hide the remaining availability; computing and retaining \(F_h\) separately is safer. For well-stirred clearance near flow, the scaled ratio and direct shortfall are preferable when the distance from flow matters.

No algebraic rearrangement can restore information that was lost before the calculation, such as an input rounded to zero, an outlet ratio rounded to one, or a quantity below the computer’s representable range. Nor should extra digits be reported merely because software prints them. Round the final result to a precision justified by the inputs, but preserve small complementary quantities at adequate precision if they will be used in a later calculation.

Common pitfall: “The computed extraction is 1, so availability is exactly zero and clearance equals flow.” The correction is that a displayed extraction of 1 may be a rounded value. Calculate and retain availability directly, check whether the solver preserved the outlet fraction, and distinguish a numerical zero from a true model prediction of zero.