Using Accumulation Data to Set a Phase I Dosing Interval

Why accumulation informs an interval decision

In a Phase I study, a tolerable single dose does not by itself establish that the same dose can be given repeatedly at a chosen interval. Residual drug from earlier doses contributes to later concentrations. If elimination is slower than expected, the resulting accumulation can raise steady-state peaks, troughs, or both. Multiple-dose observations therefore provide a useful check on the disposition model used to choose the next cohort’s interval.

The calculations below are an illustrative decision exercise, not clinical guidance. They assume a fixed dose and a simple linear, one-compartment model with first-order elimination after the post-distribution peak. The model is useful for a transparent comparison of intervals, but its predictions are only as credible as those assumptions and the data supporting its half-life estimate.

Model and assumptions: Doses are given at equal intervals \(\tau\), with the same dose and bioavailability each time. After each dose, the concentration increment at the post-distribution peak is \(C_1\), and concentration then declines by first-order elimination with rate constant \(k\). Elimination is linear and time-invariant: the same fraction is eliminated per unit time regardless of concentration or dose history. The model ignores absorption and distribution phases, nonlinear clearance, time-varying pharmacokinetics, and between-person variability. Predictions are for the modeled individual or representative profile, not guarantees for every participant.

The terminal half-life \(t_{1/2}\) and elimination rate constant are related by \(k=\ln(2)/t_{1/2}\). Between doses, the fraction remaining after an interval is \(r=e^{-k\tau}\). Since \(0<r<1\), each prior dose contributes less to a later peak than the dose before it.

From repeated doses to a steady-state prediction

After \(m\) doses, the post-dose peak under this model is the sum of the current dose increment and the residual contributions from earlier doses. The geometric series gives the accumulation factor and, as the number of doses increases, its steady-state limit.

C_{\max,m}=C_1(1+r+r^2+\cdots+r^{m-1}) =C_1\frac{1-r^m}{1-r}, \qquad R_{\mathrm{ss}}=\lim_{m\to\infty}\frac{C_{\max,m}}{C_1} =\frac{1}{1-r}.

At steady state, the trough immediately before the next dose is the post-dose peak multiplied by the fraction remaining over one interval. Thus \(C_{\min,\mathrm{ss}}=C_{\max,\mathrm{ss}}r\). These are model-based concentrations at specified points in the interval; they are not estimates of total exposure such as area under the concentration-time curve.

C_{\max,\mathrm{ss}}=C_1R_{\mathrm{ss}} =\frac{C_1}{1-e^{-k\tau}}, \qquad C_{\min,\mathrm{ss}}=C_{\max,\mathrm{ss}}e^{-k\tau} =\frac{C_1e^{-k\tau}}{1-e^{-k\tau}}.

The derivation depends on the model assumptions: linearity allows each dose’s contribution to be added, and the constant elimination rate makes successive contributions a geometric sequence. If clearance changes during treatment, or concentrations are not proportional to dose, that sequence need not describe accumulation.

Worked Example: Checking the Accumulation Pattern

Suppose a hypothetical participant receives the same dose every 24 hours. The post-distribution peak after a single dose is 5.20 mg/L. A terminal-phase analysis gives a nominal half-life of 26 hours. After the seventh dose, the observed post-dose peak is 10.9 mg/L and the predose concentration is 5.70 mg/L. Are these observations broadly consistent with the simple accumulation model?

First calculate the fraction remaining over 24 hours:

r=e^{-(\ln 2)(24/26)}\approx 0.5274.

The predicted peak after seven doses uses the finite-dose factor, not the steady-state factor:

C_{\max,7}=5.20\frac{1-(0.527383)^7}{1-0.527383} \approx (5.20)(2.092) \approx 10.88\ \mathrm{mg/L}.

The predicted predose concentration is approximately \((10.88)(0.527383)=5.738\ \mathrm{mg/L}\). These predictions are close to the hypothetical observations of 10.9 mg/L and 5.70 mg/L. The agreement supports using the model as a working approximation for this example; it does not establish that the model is correct or that the half-life is known precisely. Sampling error, assay error, and deviations from the assumptions could also explain small differences.

Worked interval comparison at the nominal half-life

Consider a hypothetical next-cohort decision with \(C_1=5.20\ \mathrm{mg/L}\), a nominal half-life of 26 hours, and three candidate intervals: 24, 30, and 36 hours. For this exercise, the team has specified a modeled steady-state peak ceiling of 11.0 mg/L and a trough floor of 2.50 mg/L. These are assumed decision limits for the example, not general clinical thresholds. A candidate interval meets the stated exposure criteria only if its predicted peak is at or below the ceiling and its predicted trough is at or above the floor.

Worked Example: Calculating Nominal Steady-State Peaks and Troughs

For each interval, calculate \(r=e^{-(\ln 2)\tau/26}\), then use the steady-state equations. Retain unrounded values during the calculations and round the reported concentrations to three decimal places.

For a 24-hour interval:

r=e^{-(\ln 2)(24/26)}\approx 0.527383,\quad R_{\mathrm{ss}}=\frac{1}{1-0.527383}\approx 2.1159,

C_{\max,\mathrm{ss}}=(5.20)(2.1159)\approx 11.003\ \mathrm{mg/L}, \qquad C_{\min,\mathrm{ss}}=(11.0026)(0.527383)\approx 5.803\ \mathrm{mg/L}.

For a 30-hour interval, \(r=e^{-(\ln 2)(30/26)}\approx0.449425\) and \(R_{\mathrm{ss}}\approx1.8163\). Therefore, \(C_{\max,\mathrm{ss}}=(5.20)(1.8163)\approx9.445\ \mathrm{mg/L}\), and \(C_{\min,\mathrm{ss}}=(9.4447)(0.449425)\approx4.245\ \mathrm{mg/L}\).

For a 36-hour interval, \(r=e^{-(\ln 2)(36/26)}\approx0.382992\) and \(R_{\mathrm{ss}}\approx1.6207\). Therefore, \(C_{\max,\mathrm{ss}}=(5.20)(1.6207)\approx8.428\ \mathrm{mg/L}\), and \(C_{\min,\mathrm{ss}}=(8.4277)(0.382992)\approx3.228\ \mathrm{mg/L}\).

At the nominal half-life, the 30- and 36-hour intervals meet both example limits. The 24-hour interval has a predicted peak of 11.003 mg/L, just above the 11.0 mg/L ceiling, so it does not meet the specified peak criterion. This comparison alone does not resolve the choice: the half-life estimate is uncertain, and a longer half-life increases accumulation.

Interval	Predicted peak at 26 h half-life	Predicted trough at 26 h half-life	Meets both example limits?
24 h	11.003 mg/L	5.803 mg/L	No; peak exceeds 11.0 mg/L
30 h	9.445 mg/L	4.245 mg/L	Yes
36 h	8.428 mg/L	3.228 mg/L	Yes

Allowing for half-life uncertainty

A single nominal half-life can make a decision look more certain than the evidence warrants. For a fixed interval, a longer half-life means a larger \(r\) and a larger accumulation factor, raising both steady-state peak and trough. A shorter half-life reduces both. This creates two relevant stress points: the longest plausible half-life for the peak ceiling, and the shortest plausible half-life for the trough floor.

Suppose sparse terminal-phase observations support a plausible sensitivity range of 20 to 32 hours around the nominal 26-hour estimate. In this example, that range is used as a decision sensitivity range, not as a formal confidence interval or a probability statement about the true half-life. Check the 24-hour interval at 32 hours for the highest peak; check the 36-hour interval at 20 hours for the lowest trough. The same checks can be made for the 30-hour interval.

Worked Example: Selecting an Interval Across the Sensitivity Range

At a 32-hour half-life and a 30-hour interval, \(r=e^{-(\ln 2)(30/32)}\approx0.522137\), so \(R_{\mathrm{ss}}=1/(1-r)\approx2.0927\). The predicted peak is \((5.20)(2.0927)\approx10.882\ \mathrm{mg/L}\), and the trough is \((10.8818)(0.522137)\approx5.682\ \mathrm{mg/L}\). The peak remains below the 11.0 mg/L ceiling.

At a 20-hour half-life and a 30-hour interval, \(r=e^{-(\ln 2)(30/20)}\approx0.353553\) and \(R_{\mathrm{ss}}\approx1.5470\). The predicted trough is \((5.20)(1.5470)(0.353553)\approx2.844\ \mathrm{mg/L}\), above the 2.50 mg/L floor.

The 30-hour interval therefore meets both limits at these stress points. For comparison, the 24-hour interval at a 32-hour half-life has \(r\approx0.594604\), \(R_{\mathrm{ss}}\approx2.4667\), and a predicted peak of about 12.827 mg/L, above the ceiling. The 36-hour interval at a 20-hour half-life has \(r\approx0.287175\), \(R_{\mathrm{ss}}\approx1.4029\), and a trough of about 2.095 mg/L, below the floor. Under the stated assumptions and sensitivity range, 30 hours is the only candidate interval that meets both example limits across the range.

This decision follows from the specified objective: meet both a peak ceiling and a trough floor across the stated half-life range. It is not a universal rule to choose the interval that passes two calculations. The team would also need to consider the quality and timing of the concentration samples, the relevance of the exposure limits, participant safety observations, and whether the simple model is appropriate. If the uncertainty range, dose, or decision limits change, the interval comparison should be recomputed.

Diagnostics, alternatives, and limitations

Before using accumulation predictions to support a cohort decision, examine whether observed data behave as the model expects. With repeated dosing at a fixed interval, post-dose peaks and predose troughs should approach a plateau in a pattern compatible with the estimated elimination rate. A continuing rise after the predicted time to near steady state, unexpectedly high troughs, or a peak-to-trough pattern inconsistent with the terminal slope can signal model misspecification, poor sampling, nonadherence, assay problems, or changing pharmacokinetics. Review actual dose and sample times: a nominal “predose” sample taken after the scheduled dose is not a trough for that interval.

A simple alternative is to use observed accumulation ratios directly, for example, the multiple-dose peak divided by the single-dose peak. That approach is intuitive but can be misleading if measurements are not at comparable times, steady state has not been approached, or the single-dose and repeated-dose conditions differ. A half-life-based model makes the time dependence explicit and permits prediction at intervals not yet observed, but adds assumptions. If absorption is slow, the post-distribution peak assumption may be poor; if disposition is nonlinear or time-varying, the geometric accumulation formula may fail. A more mechanistic model may be needed, but its added complexity must be supported by informative data.

These calculations also do not quantify every source of uncertainty. A range of half-lives alone does not account for uncertainty in \(C_1\), measurement error, between-participant variability, or uncertainty in the exposure limits. The outputs are conditional predictions, not a guarantee that every participant will remain within the limits. Where the clinical decision depends on individual risk, relevant variability and uncertainty should be incorporated rather than hidden behind a single representative profile.

Common pitfall: It is incorrect to conclude that an interval is safe because its nominal predicted peak is below a ceiling. The correction is to state the model, check both peak and trough criteria, and assess how predictions change across a defensible uncertainty range. Similarly, a trough above a limit does not establish that the peak is acceptable.

Key takeaway: Observed multiple-dose accumulation can check a half-life-based prediction, while the steady-state equations translate that model into interval-specific peaks and troughs. A defensible Phase I interval decision makes the exposure criteria and uncertainty range explicit and remains conditional on the model assumptions.

Check Your Understanding

Use the model and hypothetical limits in this tutorial. The solutions are included below each question so the calculations can be checked directly.

If \(r=0.40\), what is the steady-state accumulation factor? Solution: \(R_{\mathrm{ss}}=1/(1-0.40)=1.667\).
For \(C_1=4.00\ \mathrm{mg/L}\) and \(r=0.50\), calculate the steady-state peak and trough. Solution: \(R_{\mathrm{ss}}=2\), so the peak is \(4.00(2)=8.00\ \mathrm{mg/L}\); the trough is \(8.00(0.50)=4.00\ \mathrm{mg/L}\).
Why can the longest plausible half-life be used to stress-test a peak ceiling? Solution: At fixed \(\tau\), a longer half-life gives a larger \(r=e^{-k\tau}\), hence a larger \(1/(1-r)\) and a higher predicted peak.
Why is the shortest plausible half-life relevant to a trough floor? Solution: A shorter half-life reduces residual concentration and accumulation, producing the lowest trough at a fixed interval; that is the conservative direction for checking a minimum exposure criterion.
What does a close match between observed and predicted accumulation establish? Solution: It shows that the observations are compatible with the model at the sampled conditions. It does not prove the assumptions, eliminate uncertainty, or establish that every participant’s exposure will meet a limit.