Dialysis, Residual Kidney Function, and Drug Clearance

Why Dialysis Requires a Separate Clearance Term

For a patient receiving intermittent dialysis, drug disappearance may reflect more than the patient's own kidney function. Residual kidney function can contribute endogenous renal clearance between sessions, while the dialyzer can remove drug from the circulation during a session. These are distinct processes: residual renal clearance may operate continuously, whereas extracorporeal removal is generally active only during the recorded dialysis interval.

A model that represents both processes with a single renal-function effect can assign too much or too little clearance to residual kidney function, especially when samples are collected near a session. The central modeling task is to separate the source of clearance from its timing. The earlier material on time-varying renal function establishes how a changing renal-function measure can enter a longitudinal model; here, dialysis session records and residual kidney function must also be represented as separate predictors.

This additive representation assumes the parallel pathways remove drug from the same systemic amount, with each pathway's elimination rate proportional to the relevant drug concentration. It also assumes that the clearances are defined on compatible concentration and volume bases. For a simple model with central amount \(A_i(t)\), central concentration \(C_i(t)=A_i(t)/V_i\), and systemic input rate \(R_{\mathrm{in},i}(t)\), the amount balance is

Here, \(CL_{\mathrm{NR},i}\) is nonrenal clearance, \(CL_{\mathrm{R},i}(t)\) is residual renal clearance, \(CL_{\mathrm{D},i}\) is dialysis clearance while the session is operating, and \(D_i(t)\) is an indicator equal to 1 during dialysis and 0 otherwise. The indicator is tied to actual session start and end times, not merely to whether a patient is classified as receiving dialysis. For patients not receiving dialysis, \(D_i(t)=0\) throughout.

Representing Residual Function and Session Timing

A continuous residual-function measure can be related to renal clearance using a power model, as in earlier renal covariate material. For example,

In this expression, \(R_i(t)\) is the chosen residual kidney function measure for individual \(i\), \(R_{\mathrm{ref}}\) is a positive reference value in the same units, \(CL_{\mathrm{R,ref}}\) is the renal clearance at that reference value, and \(\gamma_{\mathrm{R}}\) describes the function-clearance relationship. The measure and its timing need clinical justification: a value obtained well before the PK observation may not represent function during that observation. If function is effectively zero, the model's behavior at zero must also be considered; a positive power gives zero renal clearance there, but measurement limits and other renal pathways can complicate that interpretation.

The dialysis term requires its own timing information. In an intermittent treatment, \(D_i(t)\) changes from 0 to 1 at session start and from 1 to 0 at session end. If clearance varies within a session—for example, because the procedure settings change—a time-varying \(CL_{\mathrm{D},i}(t)\) may be more appropriate. A single session indicator cannot describe such changes; session duration, settings, and any directly measured removal data may be needed.

Time since a session is not itself the same thing as active dialysis. The active interval determines when \(CL_{\mathrm{D}}\) contributes to the elimination rate. After the session, concentrations may rise or decline differently because drug redistributes between compartments, input continues, or other clearance pathways remain active. Such patterns should be assessed with the PK structural model and sampling schedule rather than represented automatically as a lingering dialysis clearance.

Concentration Change During a Session

A simple one-compartment calculation shows why session timing matters. Suppose there is no systemic input during a short interval, volume \(V_i\) is constant, and each clearance component is constant over that interval. Since \(C_i(t)=A_i(t)/V_i\), the amount equation implies

The second expression follows by separating variables, \(dC_i/C_i=-(CL_{\mathrm{total},i}/V_i)dt\), integrating from 0 to \(t\), and exponentiating. It applies over an interval on which the total clearance and volume can reasonably be treated as constant. If dialysis starts or stops, calculate the concentration change separately on each interval and carry the ending concentration forward as the next interval's starting value.

This derivation relies on linear elimination and a well-mixed one-compartment approximation. A multi-compartment drug can show a rapid fall during dialysis followed by rebound as drug moves back into the central compartment. In that case, a single exponential fitted across the session and post-session period may misattribute redistribution to a change in clearance. The model needs enough samples at informative times to distinguish these processes.

Estimating Dialysis Clearance and Checking the Data

If concentration measurements are available immediately before and at the end of a session, an effective dialysis clearance can sometimes be approximated from the concentration slope. This requires an interval with no input, constant volume and clearances, approximately one-compartment behavior, and no important redistribution. Under those conditions, the observed total clearance can be inferred from the log concentration ratio; subtracting the independently specified background clearance gives an estimate of dialysis clearance.

In population analysis, the same separation can be made by fitting the concentration data to a structural model with residual-function and dialysis terms. Individual variability may be represented on supported parameters, for example by \(CL_{\mathrm{D},i}=CL_{\mathrm{D,typ}}\exp(\eta_{\mathrm{D},i})\), where \(CL_{\mathrm{D,typ}}\) is the typical dialysis clearance and \(\eta_{\mathrm{D},i}\) represents an individual's deviation on the log scale. This does not solve a design problem: if session timing and residual function are not independently represented in the data, the model may not identify their effects separately.

Alternatives, Failure Modes, and Limits

One alternative is a single dialysis-status covariate on total clearance. It may be adequate if the scientific question concerns only a population-average difference between active-session and off-session periods and there is little interest in separating mechanisms. It is less informative when residual renal function varies substantially or when the model must predict different session schedules. Another alternative is to fit a separate clearance for each session; that can describe observed data but may be unstable and difficult to generalize unless there are repeated, informative observations.

A mechanistic additive model is useful when the pathways can be estimated with the available data. Its main limitation is identifiability: concentrations observed only during dialysis may not reveal how much removal is attributable to dialysis versus background clearance. Samples off-session, variation in residual function, informative session timing, or external information about one component can help. Conversely, if all patients have nearly identical function and session schedules, adding several separate parameters may produce imprecise estimates rather than useful detail.

• Misclassifying the predictor: Coding a patient as “on dialysis” for every time point makes dialysis clearance appear to operate between sessions. Use the active-session interval for the time-dependent term.
• Combining predictors: A low residual-function value does not represent the dialyzer's contribution. Keep \(CL_{\mathrm{R}}\) and \(CL_{\mathrm{D}}\) distinct in the model.
• Ignoring time alignment: A renal-function measurement from a different period may not describe function at the PK sample. Check measurement timing and represent meaningful change when supported.
• Forcing a constant session effect: Clearance may vary with procedure conditions, and concentration rebound may reflect distribution rather than clearance. Check session-period residuals and post-session samples before interpreting a fitted dialysis term.
• Overinterpreting a fitted term: A well-fitting model does not establish that each clearance component is separately identified. Evaluate uncertainty and whether the data contain observations that distinguish the pathways.