Optimal Biological Dose vs. Maximum Tolerated Dose

Introduction

Early-phase clinical trials have traditionally focused heavily on one question: How much drug can patients tolerate?

This question leads naturally to the concept of the Maximum Tolerated Dose (MTD). The MTD is generally understood as the highest dose that can be administered with an acceptable level of dose-limiting toxicity under the rules of the dose-finding study.

However, tolerability and biological activity are not the same thing. A dose can be tolerated but provide little additional pharmacologic or antitumor benefit compared with a lower dose.

Conversely, the dose associated with the strongest biological activity may also produce substantially greater toxicity.

What Is the Maximum Tolerated Dose?

The MTD is a toxicity-driven dose-selection concept. In a conventional Phase I dose-escalation study, patients receive increasing dose levels until the observed toxicity becomes sufficiently concerning that further escalation is not justified.

Let \(d\) denote dose and let \(q(d)\) represent the probability of a dose-limiting toxicity.

A conceptual definition of an MTD is the dose satisfying:

where \(q^*\) is the target toxicity probability specified by the dose-finding framework.

The exact definition depends on the clinical trial design. For example, model-based dose-finding methods may explicitly target a predefined probability of DLT, whereas rule-based designs such as the traditional 3+3 design use escalation and de-escalation rules rather than directly estimating a toxicity probability.

What Is the Optimal Biological Dose?

The Optimal Biological Dose is fundamentally different. Rather than selecting a dose primarily from observed toxicity, the OBD uses evidence that the drug is producing the intended biological effect.

That evidence might include:

• Pharmacodynamic biomarker modulation
• Target engagement
• Receptor occupancy
• Pathway inhibition or activation
• Changes in downstream signaling
• Immune-cell activation
• Tumor-response measures
• Other mechanism-based biological endpoints

The OBD therefore depends on a relationship between dose or exposure and biological activity.

The critical point is that biological activity does not necessarily continue to increase indefinitely with dose.

MTD and OBD Answer Different Questions

The Fundamental Problem: Efficacy Can Plateau Before Toxicity

Consider a hypothetical treatment whose biological activity increases as dose increases. At low doses, increasing the dose produces substantial increases in biological activity.

Eventually, however, the biological pathway may approach saturation.

Additional drug exposure beyond this point may produce little additional biological effect while continuing to increase toxicity.

This is the fundamental rationale for considering an OBD separately from an MTD.

A Conceptual Dose-Response Model

Suppose biological activity follows an Emax model:

where:

• \(E_0\) is baseline activity.
• \(E_{\max}\) is the maximum drug effect.
• \(EC_{50}\) is the dose producing 50% of the maximum effect.
• \(d\) is dose.

At sufficiently high doses:

Thus, increasing dose eventually produces diminishing biological returns.

Dose-Toxicity Relationships

Toxicity can also be modeled as a function of dose. For example, a logistic model may be written as:

where:

• \(d\) is dose.
• \(\alpha\) controls the baseline toxicity level.
• \(\beta\) controls the dose-toxicity relationship.

The biological-response curve and toxicity curve therefore describe two different aspects of dose selection.

Putting the Two Curves Together

The most useful conceptual picture is not a single dose-toxicity curve. Instead, consider both:

An ideal dose would produce substantial biological activity while maintaining acceptable toxicity.

If 50 mg provides nearly all of the biological activity obtainable at 100 mg, while toxicity rises substantially between the two doses, an MTD-based strategy could select a dose that is unnecessarily high.

Why the MTD Can Be Higher Than the OBD

Suppose the biological effect follows:

At 20 mg:

At 80 mg:

Increasing the dose from 20 mg to 80 mg therefore increases the modeled effect from 50 to 80 units. But consider what happens beyond 80 mg.

The dose has doubled, but the biological effect increases by only about 11%.

If toxicity rises sharply over the same dose range, the incremental benefit may not justify the additional toxicity.

The Role of Pharmacodynamics

Pharmacodynamic measurements can provide information that toxicity alone cannot.

For a mechanism-based drug, investigators may measure a biomarker that is directly connected to the drug's pharmacologic mechanism. Examples include:

• Reduction in a phosphorylated signaling protein
• Inhibition of an enzyme pathway
• Receptor occupancy
• Suppression of a downstream biomarker
• Immune-cell activation
• Changes in circulating tumor DNA
• Changes in disease-associated molecular markers

Suppose the biomarker response is represented by:

The investigator can then ask whether the desired biological effect has already been achieved at a dose substantially below the MTD.

Target Engagement Is Not the Same as Clinical Benefit

A rigorous dose-selection strategy may therefore integrate:

Exposure May Be More Informative Than Dose

Dose is not always the most appropriate predictor of biological activity. Patients receiving the same nominal dose can have substantially different drug exposures because of differences in:

• Absorption
• Bioavailability
• Clearance
• Body size
• Organ function
• Drug-drug interactions
• Genetic variability

For this reason, pharmacokinetic analysis can replace or supplement a simple dose-response analysis.

Potential exposure metrics include:

• \(C_{\max}\)
• \(C_{\min}\)
• \(AUC\)
• Time above a pharmacologic threshold
• Steady-state concentration

Worked Example: MTD Versus OBD

Consider a hypothetical Phase I/II oncology study evaluating five dose levels. The investigators collect both dose-limiting toxicity information and a pharmacodynamic biomarker measuring pathway inhibition.

Suppose the dose-finding procedure identifies 100 mg as the MTD. That does not automatically imply that 100 mg is the best dose for subsequent development.

Step 1: Examine Biological Activity

The PD biomarker shows:

The largest increase occurs between 25 mg and 50 mg. The incremental biological benefit after 50 mg is substantially smaller.

The increase from 50 mg to 75 mg is only:

The increase from 75 mg to 100 mg is only:

This suggests that the biological response is approaching a plateau.

Step 2: Examine Toxicity

Toxicity moves in the opposite direction. At 50 mg the observed DLT rate is approximately 8.3%. At 75 mg it increases to approximately 16.7%. At 100 mg it increases to approximately 33.3%.

Thus, the dose escalation produces diminishing biological returns while toxicity continues to increase.

Step 3: Compare the Candidate Doses

A reasonable biological interpretation is that approximately 50 mg may represent a strong OBD candidate even though the MTD is 100 mg.

OBD Is Not Simply "The Dose Below the MTD"

A common oversimplification is:

That is not a general definition.

The OBD should instead be derived from the relationship between dose or exposure, biological activity, and toxicity.

For example, if biological activity continues increasing meaningfully up to the MTD, then the OBD may be close to the MTD.

Conversely, if biological activity plateaus far below the MTD, the OBD may be substantially lower.

A Utility-Based View of Dose Selection

One way to formalize the concept is to define a utility function combining benefit and toxicity. Let:

where:

• \(B(d)\) represents biological or clinical benefit.
• \(T(d)\) represents toxicity or treatment burden.
• \(w_B\) is the weight assigned to benefit.
• \(w_T\) is the weight assigned to toxicity.

The optimal dose could then be conceptualized as:

This equation is useful conceptually because it makes the underlying trade-off explicit.

OBD and the Therapeutic Window

The relationship can also be viewed through the therapeutic window.

A drug with a wide therapeutic window may achieve near-maximal biological activity before significant toxicity appears.

A drug with a narrow therapeutic window may have its biological and toxicity curves occurring over a similar exposure range.

When MTD and OBD Are Similar

The MTD and OBD may be similar when:

• Biological activity continues to increase substantially with dose.
• Target engagement is not saturated at lower doses.
• Higher exposure produces meaningful additional clinical benefit.
• Toxicity remains acceptable near the biologically active plateau.
• The therapeutic index is relatively narrow but the higher dose provides meaningful benefit.

When OBD Is Well Below MTD

The OBD may be substantially below the MTD when:

• Target engagement approaches saturation at relatively low exposure.
• PD activity plateaus early.
• Higher doses provide little incremental biological activity.
• Toxicity increases steeply with dose.
• Chronic administration makes cumulative toxicity important.
• Treatment tolerability is important for long-term adherence.

The Difference Between Phase I and Phase II Objectives

Traditional Phase I dose escalation is often heavily focused on safety. The fundamental question may be:

An integrated OBD strategy asks a broader question:

This distinction becomes particularly important when moving from a purely cytotoxic paradigm toward targeted therapies, immunotherapies, molecularly targeted agents, and other treatments in which biological activity may occur well below doses associated with severe toxicity.

Using Exposure-Response Information

Suppose the pharmacodynamic effect depends on exposure rather than nominal dose. An Emax model might be written:

where \(C\) represents a relevant exposure metric.

The investigator can then estimate the exposure required to achieve a specified fraction of the maximum effect. For example, the exposure corresponding to 90% of the maximal effect can be derived from:

Solving:

This illustrates why large increases in exposure may be required to obtain small additional increases in effect once the system approaches saturation.

R Implementation: Modeling the Biological Response

The Emax relationship can be implemented directly in R.

dose <- seq(1, 120, by = 1)

E0 <- 10
Emax <- 90
EC50 <- 20

effect <- E0 +
  (Emax * dose) /
  (EC50 + dose)

head(data.frame(
  dose = dose,
  effect = effect
))

Calculate a Dose-Toxicity Relationship

A logistic toxicity model can be evaluated using:

alpha <- -4
beta <- 0.05

p_dlt <- exp(alpha + beta * dose) /
  (1 + exp(alpha + beta * dose))

head(data.frame(
  dose = dose,
  p_dlt = p_dlt
))

The resulting values represent the modeled probability of DLT across the dose range.

Combine Biological Activity and Toxicity

The two relationships can be combined into a single data frame.

dose_data <- data.frame(
  dose = dose,
  effect = effect,
  p_dlt = p_dlt
)

head(dose_data)

This creates the fundamental dataset needed to examine whether biological activity is still increasing meaningfully when toxicity begins to rise rapidly.

Identify a Candidate Biological Plateau

One simple exploratory approach is to calculate the incremental biological benefit associated with increasing the dose.

dose_data$incremental_effect <-
  c(
    NA,
    diff(dose_data$effect)
  )

head(dose_data)

A more clinically meaningful approach is to define a target biological effect. For example, suppose investigators consider 85% of the modeled maximum effect sufficient.

target_effect <- 0.85 *
  max(dose_data$effect)

candidate_doses <-
  dose_data[
    dose_data$effect >= target_effect,
  ]

candidate_doses

This identifies doses reaching the prespecified biological target.

Apply a Toxicity Constraint

Suppose investigators additionally require the modeled DLT probability to remain below 20%.

acceptable_doses <-
  candidate_doses[
    candidate_doses$p_dlt <= 0.20,
  ]

acceptable_doses

The resulting dose range represents a candidate region in which the desired biological activity is achieved without exceeding the specified toxicity constraint.

Selecting the OBD From a Candidate Region

Rather than treating OBD as a single automatically generated number, it is often more informative to think of the analysis as defining a region of acceptable doses.

Why Small Sample Sizes Matter

Early-phase dose-finding studies are usually relatively small. Consequently, observed toxicity and biological response rates can be highly uncertain.

For example, observing zero DLTs in six patients does not demonstrate that the true DLT probability is zero. Likewise, observing a strong biomarker response in a small cohort does not guarantee that the effect will persist in a larger population.

Observed Response Versus Model-Based Response

Suppose 10 patients receive a dose and eight demonstrate the desired biological response. The observed response proportion is:

The estimate is 80%, but the underlying population response probability remains uncertain.

This distinction becomes particularly important when comparing several dose levels. A dose with an observed 90% response in five patients may not be superior to a dose with an observed 80% response in 20 patients.

MTD-Based Versus OBD-Based Dose Selection

The 3+3 Design and the MTD

The traditional 3+3 design is a toxicity-based dose-escalation procedure. Its rules are driven primarily by observed DLTs.

For example, a simplified structure is:

The design can therefore be useful for identifying a tolerability boundary. However, toxicity-based escalation alone does not necessarily identify the dose at which biological activity is optimized.

Model-Based Dose-Finding

Model-based methods can explicitly model the dose-toxicity relationship. Examples include:

• Continual Reassessment Method (CRM)
• Bayesian logistic models
• Bayesian optimal interval designs
• Joint toxicity-efficacy models
• Pharmacologically informed dose escalation approaches

When pharmacodynamic information is available, dose selection can become more informative because the model can incorporate evidence beyond toxicity alone.

Joint Modeling of Efficacy and Toxicity

A particularly useful framework considers both efficacy and toxicity. Let:

represent the probability of the desired biological or clinical effect and:

represent the probability of unacceptable toxicity.

The preferred dose may then be defined as a dose satisfying: