12.2 Point of Departure (PoD) Determination: NOAEL, LOAEL & Benchmark Dose (BMD/BMDL) Modeling

From curve to a number you can divide

Handbook III.C.1–2 moves from “the curve has a shape” to “which point of departure (PoD) will we use for a reference value or a slope factor?” Independent OpenExamPrep teaching covers the no-observed-adverse-effect level (NOAEL), the lowest-observed-adverse-effect level (LOAEL), benchmark dose (BMD) modeling and the BMDL, drug-development cousins (HNSTD, MABEL, STD10), and why organ weights, clinical pathology, and histopathology must be read together. This section is not an ABT product and does not claim official approval, review, or partnership with ABT, EPA, or FDA.

A PoD is a dose on or near the observed range chosen as the starting point for extrapolation. It is not the reference dose (that is PoD divided by uncertainty factors, Chapter 13) and not the cancer slope factor (section 12.3).

NOAEL and LOAEL

NOAEL: the highest tested dose at which the chosen adverse effect is not observed, using biological judgment plus statistics.

LOAEL: the lowest tested dose at which that adverse effect is observed.

They are study properties, not molecular constants. Three design facts constrain them:

1. Listed doses only. If groups are 0, 10, 50, and 250 mg/kg/day and adversity appears somewhere between 10 and 50, the NOAEL can only be 10 and the LOAEL 50. The true biological breakpoint might be 18 or 40; the design cannot name those numbers because they were never given.
2. Spacing. Wide geometric ratios (10, 100, 1000) can park a NOAEL far below the true effect region: conservative, crude, and an artifact of the protocol. Tight spacing near the effect region is more informative and still can only report tested levels.
3. Sample size n. With 10 rats/sex, a 10% incidence (1/10) is often not statistically significant. The same biology at n = 50 might be. Underpowered studies inflate NOAELs (false-negative adversity calls). Over-calling a single incidental lesion deflates them.

Worked spacing and n example. Male rats, 10/group, oral gavage, 90 days.

If the pathologist and study director treat the single minimal focus at 30 as incidental (no enzyme pattern, not reproduced in females, within historical incidence), NOAEL = 30 mg/kg/day and LOAEL = 100. If they treat any necrosis plus the directionally higher ALT as adverse, NOAEL = 10 and LOAEL = 30. Same slides, two PoDs. If the protocol had used 0, 10, 12, and 15 instead of jumping to 100, you still could not declare a NOAEL of 12 unless 12 mg/kg/day was administered.

BMD, BMR, and BMDL

Benchmark dose (BMD) is the dose that produces a specified benchmark response (BMR) on a fitted model. It need not equal a tested dose.

BMR is the size of change you declare to be of concern:

• Quantal (dichotomous): commonly 10% extra risk (BMR = 0.10). Extra risk = [P(d) − P(0)] / [1 − P(0)]. If background P(0) = 0.02 and extra risk = 0.10, then 0.10 × (1 − 0.02) = 0.098, so P(d) = 0.02 + 0.098 = 0.118 (11.8% absolute incidence). Added risk of 0.10 would mean P(d) = 0.02 + 0.10 = 0.12. Extra and added are not interchangeable when background is not zero.
• Continuous: commonly a mean change of 1 control SD, or sometimes a 5% or 10% relative change in the mean (for example a 5% drop in body weight). 1 SD is a statistical convention, not a proof that 1 SD is “minimally adverse” in every organ.

BMDL is a one-sided lower statistical bound (often 95%) on the BMD. Risk assessment uses the BMDL as the PoD, not the central BMD, so fit uncertainty is taken toward lower dose. If BMD10 = 40 mg/kg/day and BMDL10 = 25 mg/kg/day, PoD = 25, not 40.

BMD advantages versus NOAEL: the fit uses the whole curve; the PoD is not forced to equal a listed dose; better dose placement and larger n usually tighten the BMDL; the BMR is explicit.

BMD does not remove judgment. A central limitation is that the BMR itself is a science-and-policy choice. Ten percent extra risk is a common reporting convention for quantal apical effects; it is not a law of biology and not a hidden constant in the likelihood function. Choosing 5% extra risk, 1% extra risk, or 1 SD versus 0.5 SD moves the PoD even when the same model and the same data are used. Two assessors can both fit the curve competently and still disagree because they disagreed on what change is of concern. That is a limitation to disclose in the assessment, not a software crash. It is also why “the BMD” is an incomplete sentence until you name the endpoint, the BMR, and whether you mean the central estimate or the BMDL.

Software, conceptually. EPA’s Benchmark Dose Software (BMDS) is widely used in U.S. chemical assessment and is a convenient way to picture the workflow: choose a BMR, screen models, look at residuals and information-criterion comparisons, report BMD and BMDL, and ask whether the BMR sits in a region the data actually constrain (a BMR far below the lowest positive group is poorly supported). Other academic and commercial packages exist. DABT items test whether you know BMD versus BMDL versus BMR, not whether a single vendor product is legally required. No package is a prerequisite for a conceptually valid PoD.

HNSTD, MABEL, and STD10 as drug-development PoDs

Chemical IRIS-style files rarely use these names. First-in-human and oncology programs do.

Healthy-volunteer maximum recommended starting dose (MRSD) under FDA 2005 starts from NOAEL → HED → safety factor (section 12.3). Oncology patients accept a different risk, so STD10/HNSTD logic applies. An immune agonist that is silent in cynomolgus monkeys at milligrams per kilogram can still need a microgram MABEL start in humans. Mixing those playbooks is a high-yield error.

Integration: organ weights, clinical pathology, histopathology

A PoD that uses only asterisks on a clinical-chemistry table is incomplete. Adversity is a package:

• Organ weights (absolute and relative to body weight): liver and kidney increases may be adaptive enzyme induction or injury. Pair with microscopy.
• Clinical pathology: magnitude, concordance (ALT with AST or glutamate dehydrogenase), and time course—not a lone p-value in a 40-parameter battery.
• Histopathology: the referee for injury versus adaptation versus noise.

If relative liver weight is +40%, ALT is 5× concurrent control, and centrilobular necrosis is present, that dose is a LOAEL for hepatic injury. If relative liver weight is +15%, enzymes are quiet, and histology is only centrilobular hypertrophy with smooth endoplasmic reticulum, many programs call that non-adverse enzyme induction—so the dose can remain a NOAEL for injury even though “something changed.” BMD modeling of “any liver-weight change” versus “necrosis incidence” will produce different BMDL values. The endpoint definition is part of the PoD, not an afterthought.

Scenario

A pesticide 90-day study lists NOAEL = 50 mg/kg/day because Dunnett’s test on ALT was not significant at 50 (n = 10) even though 1/10 animals had necrosis. A BMD model of necrosis incidence with BMR = 10% extra risk returns BMDL10 = 22 mg/kg/day. Using 50 as the PoD ignores both the n-limitation of the NOAEL and the explicit BMR. Using 22 still requires you to defend 10% extra risk rather than 5% as the BMR—the policy/science choice above, not a hidden BMDS default that cannot be questioned. An oncology FIH program on a cytotoxic would not use this NOAEL path at all; it would ask for STD10 or HNSTD. A CD28 superagonist would ask for a MABEL, not for the pesticide’s 50 mg/kg NOAEL.

Traps

• Believing the NOAEL is the biological threshold rather than a tested dose.
• Using the central BMD as the PoD when a BMDL is available.
• Treating EPA BMDS as the only lawful software, or treating any BMD printout as independent of the BMR.
• Applying MABEL to a low-potency industrial chemical IRIS file, or applying a healthy-volunteer NOAEL MRSD to an immune superagonist.
• Setting a hepatic NOAEL from organ weight without reading the slides, or ignoring a concordant necrosis-plus-enzyme package because n = 10 was underpowered.