Toxicological risk assessment is a structured, science-based process used to estimate the likelihood that exposure to a chemical or other harmful substance will cause adverse health effects in humans. It follows a four-step framework originally formalized by the U.S. National Research Council in 1983 and still used by regulatory agencies worldwide: hazard identification, dose-response assessment, exposure assessment, and risk characterization. The process underpins nearly every chemical safety standard you encounter, from the allowable levels of pesticides on food to the limits on pollutants in drinking water.
The Four-Step Framework
Every toxicological risk assessment moves through the same four stages, each building on the one before it. The first step, hazard identification, asks a simple question: can this substance cause harm? Scientists review animal studies, human epidemiological data, and cell-based experiments to determine whether exposure to a chemical can increase the incidence of specific health problems like cancer, birth defects, or organ damage. The goal isn’t to quantify the risk yet. It’s to establish whether a real danger exists and how strong the evidence is.
The second step, dose-response assessment, documents the relationship between the amount of exposure and the severity or likelihood of harm. A tiny amount of a substance might be harmless while a larger amount causes serious effects. This step maps out that curve. The third step, exposure assessment, shifts focus to real-world conditions: how much of the substance are people actually encountering, through what routes, and for how long? The final step, risk characterization, pulls everything together into an overall judgment about whether a meaningful health risk exists, along with a transparent account of the assumptions and uncertainties involved.
How Dose-Response Data Is Gathered
Dose-response assessment relies on identifying a threshold, a dose below which no harmful effects are expected. Historically, scientists have used two key benchmarks from animal or human studies. The No-Observed-Adverse-Effect Level (NOAEL) is the highest dose tested at which no statistically significant harmful effect was detected compared to a control group. The Lowest-Observed-Adverse-Effect Level (LOAEL) is the lowest dose at which harmful effects were observed.
The NOAEL approach has a significant limitation: it depends entirely on the specific doses researchers chose to test. If the gap between doses is wide, the NOAEL could be masking real effects that were simply too small to detect in that particular study. For this reason, regulators increasingly favor a more sophisticated method called the Benchmark Dose (BMD) approach. Instead of relying on individual dose comparisons, BMD modeling uses the entire dataset to fit a dose-response curve and then calculates the dose associated with a predefined level of effect, typically the lower confidence bound of that dose (called the BMDL). This approach extracts more information from the same data, accounts for statistical uncertainty, and doesn’t depend on which specific doses happened to be tested.
Three Routes of Exposure
Exposure assessment focuses on three routes through which a substance can enter the body: inhalation (breathing it in), ingestion (swallowing it through food, water, or soil), and dermal absorption (absorbing it through the skin). For each route, assessors estimate the magnitude, frequency, and duration of exposure. A factory worker inhaling chemical fumes eight hours a day faces a very different risk profile than a homeowner who occasionally uses a product containing the same chemical.
The assessment also considers which populations are exposed. Children, pregnant women, and elderly individuals often metabolize chemicals differently or are more vulnerable to their effects. Exposure estimates try to capture the range of real-world scenarios, from average consumers to highly exposed subgroups, and they explicitly flag the uncertainties in those estimates.
Safety Factors and the Reference Dose
Once scientists identify a dose threshold from animal or human studies, they don’t simply declare it safe for the general public. They divide it by a series of uncertainty factors (sometimes called safety factors) to build in a margin of protection. The EPA’s formula for calculating a Reference Dose (RfD), the daily exposure level considered unlikely to cause harm over a lifetime, looks like this: divide the NOAEL by the combined uncertainty factors.
Each factor is typically 10-fold and addresses a specific gap in knowledge:
- Animal-to-human extrapolation: A 10-fold factor accounts for the possibility that humans are more sensitive than the test animals.
- Human variability: Another 10-fold factor protects sensitive individuals, including children and people with preexisting conditions, who may react at lower doses than healthy adults.
- Short-term to lifetime extrapolation: If the study only observed animals for weeks or months rather than a full lifespan, a 10-fold factor compensates for this shorter observation period.
- LOAEL to NOAEL conversion: If no true NOAEL was established and the calculation starts from a LOAEL instead, another 10-fold factor is applied.
These factors can stack. A chemical with limited data from a short-term animal study using a LOAEL could have a combined uncertainty factor of 10,000. In practice, regulators generally cap the cumulative factor at 3,000. If the data requires more than that, the evidence is considered too weak to derive a reliable safety threshold at all. When more specific data is available, such as detailed modeling of how the chemical behaves in both animal and human bodies, the default factors can be reduced. For example, studies in non-human primates use a smaller interspecies factor (roughly 3 instead of 10) because primates are biologically closer to humans.
Cancer Risk vs. Non-Cancer Risk
Risk assessment handles cancer-causing substances differently from non-carcinogens because of a key assumption: for many carcinogens, there may be no truly safe threshold. Even a very small dose could theoretically initiate the process that leads to cancer, so the math changes accordingly.
For non-cancer effects, the standard output is a Hazard Quotient (HQ). This is simply the estimated exposure dose divided by the Reference Dose. If the HQ is below 1, the exposure is considered unlikely to cause harm. If it’s above 1, the exposure exceeds the health guideline and warrants closer investigation. An HQ of 2 doesn’t mean harm is guaranteed; it means the margin of safety has been eroded and a deeper toxicological analysis is needed.
For carcinogens, risk is instead expressed as a probability, such as a one-in-a-million chance of developing cancer over a lifetime of exposure. Regulators use a cancer slope factor, derived from dose-response data, and multiply it by the estimated lifetime dose to arrive at this probability. Many regulatory programs consider a lifetime cancer risk between one in a million and one in ten thousand as the acceptable range for setting exposure limits.
International Standards for Testing
The data feeding into risk assessments comes from standardized studies. The OECD Guidelines for the Testing of Chemicals are the internationally recognized standard, covering everything from physical-chemical properties to health effects and environmental fate. These guidelines are accepted across member countries under a Mutual Acceptance of Data agreement, meaning a study conducted according to OECD protocols in one country is recognized by regulators in others. This avoids duplicating expensive, time-consuming studies and, importantly, reduces unnecessary animal testing.
The guidelines are organized into five sections: physical-chemical properties, effects on living systems, environmental behavior, health effects, and other specialized tests. They are regularly updated to incorporate scientific advances and evolving regulatory needs.
New Approach Methodologies
Traditional risk assessment has relied heavily on animal studies, but the field is shifting. New Approach Methodologies (NAMs) encompass a range of tools designed to reduce or replace animal testing while improving relevance to human biology. These include cell-based laboratory systems using human tissues, computer modeling that predicts how chemicals interact with biological pathways, and integrated testing strategies that combine multiple data streams.
The FDA has issued guidance on using computational models to assess the safety of medical devices, drugs, and biologics. For certain categories of drugs, like some monoclonal antibodies, recent guidance allows manufacturers to eliminate or reduce long-term primate toxicity studies by substituting data from computational models and human-relevant test systems. Computational tools are also already in regulatory use for predicting whether drug impurities have the potential to damage DNA and cause cancer, reducing the need for animal-based mutagenicity tests.
These methods don’t yet replace traditional testing across the board. They’re integrated through a weight-of-evidence approach, where multiple lines of evidence are evaluated together to build confidence in the safety assessment. As validation frameworks mature, NAMs are expected to play an increasingly central role in how chemicals and drugs are evaluated for safety.