Combining epidemiology and toxicology in environmental health risk assessment
- The text on this page is taken from an equivalent page of the IEHIAS-project.
The purpose is to offer a systematic method to quantitatively combine epidemiological and toxicological data.
Usually in environmental health risk assessment there is only animal or in vitro data available on the substance under study. However, for the most important exposures causing the main public health concerns and the most difficult risk management decisions, there is typically also epidemiological data on humans available. To make more valid risk management decisions, we need better methods to combine animal and human data.
Animal and human data contribute both to hazard identification and to dose-response assessment. In hazard identification, combining these two sources of data is slightly less difficult, as the outcome of hazard identification is typically qualitative, and several attempts have been made to achieve comprehensive combinations. In contrast, animal and human data have rarely been combined in quantitative dose-response assessment.
The ultimate aim of both animal and human studies in environmental health risk assessment is to contribute to the estimation of the true slope and shape of the dose-response function of the substance under study in humans. The main difference is the sources of uncertainty, e.g. animal-to-human, high-to-low-dose, and endpoint extrapolations in animal studies; and multiple biases in human studies. The only valid possibility to quantitatively combine animal and human studies in dose-response assessment would be to estimate the size and direction of these uncertainties using a common metrics.
Contents
Scope
Purpose
The purpose is to offer a systematic method to quantitatively combine epidemiological and toxicological data.
Usually in environmental health risk assessment there is only animal or in vitro data available on the substance under study. However, for the most important exposures causing the main public health concerns and the most difficult risk management decisions, there is typically also epidemiological data on humans available. To make more valid risk management decisions, we need better methods to combine animal and human data.
Animal and human data contribute both to hazard identification and to dose-response assessment. In hazard identification, combining these two sources of data is slightly less difficult, as the outcome of hazard identification is typically qualitative, and several attempts have been made to achieve comprehensive combinations. In contrast, animal and human data have rarely been combined in quantitative dose-response assessment.
The ultimate aim of both animal and human studies in environmental health risk assessment is to contribute to the estimation of the true slope and shape of the dose-response function of the substance under study in humans. The main difference is the sources of uncertainty, e.g. animal-to-human, high-to-low-dose, and endpoint extrapolations in animal studies; and multiple biases in human studies. The only valid possibility to quantitatively combine animal and human studies in dose-response assessment would be to estimate the size and direction of these uncertainties using a common metrics.
Boundaries
The method is basically applicable in all situations where epidemiological and toxicological data is being integrated.
Method description
Input
Dose-responses from single studies on a certain exposure.
Output
The product of the method is a combined dose-response estimate, received by quantitatively integrating animal and human data, for a given risk assessment situation.
Rationale
A good method for this purpose notifies all the possible sources of uncertainty and gives them a numerical value. Such methods are very few (or do not exist).
Method
The method consists of a checklist of possible corrections (below) that have to be considered when using data from single animal and human studies to estimate the slope of the dose-response curve for a given risk assessment situation. An estimate of the importance of the specific correction in each study type is given: + = not so important; ++ = in-between; +++ = very important. Short descriptions of the corrections are given below the table.
| Possible corrections | Animal, experimental | Human, observational |
|---|---|---|
| Single study: Internal validity | ||
| Shape of dose-response curve in measured range | ++ | ++ |
| Measurement error in treatment/exposure (exposure misclassification) | + | +++ |
| Measurement error in outcome | + | + |
| Systematic differences in other exposures (confounding factors) | + | +++ |
| Selection of participants (selection bias) | (+) | ++ |
| Single study: Extrapolation | ||
| Shape of dose-response curve outside measured range (e.g. high-to-low-dose extrapolation) | +++ | ++ |
| Not correct outcome (endpoint extrapolation) | ++ | + |
| Not correct dose timing | + | ++ |
| Not correct dose scaling | ++ | + |
| Not correct route of administration | ++ | + |
| Not correct toxicodynamics | +++ | + |
| Not correct toxicokinetics | +++ | + |
| Within human variation | +++ | ++ |
| Pooling studies | ||
| Heterogeneity between studies | ||
| Publication bias | ||
| Plausibility | ||
Short descriptions of the corrections
Some terminology:
- ideal population: target population, for which the dose-response curve is being developed
- study population: the population in a single study
- outcome: health effect or its indicator
1. Shape of dose-response curve in measured range
Based on the data and other information, is it possible to tell the shape of the dose-reponse curve? Is it linear, log-linear, exponential or something else? Does it have a threshold exposure below which no effects occur?
2. Measurement error in treatment/exposure (exposure misclassification)
Describe and assess the size of error in treatment or exposure level. E.g. dosing error in animal studies = true exposure vs. the exposure that is assumed to be administrated.
3. Measurement error in outcome
Describe the sources and size of the possible errors in measuring the outcome, e.g. problems in precision and accuracy of an analytical procedure.
4. Systematic differences in other exposures (confounding factors)
Confounders will mess up or confuse an effect of an exposure with an effect of another variable. Confounder is a risk factor of the outcome (health effect) and is also associated with the exposure. Assess all the confounders and their effect on dose-response.
5. Selection of participants (selection bias)
- Describe and assess how the study population is dissimilar than the basic population and how well can the study results be generalized to an ideal population.
- Describe and assess the effect of drop-out on study result. The drop-out from a study population is the number of study subjects who have dropped out (by themselves or have been dropped-out by an author) from the selected study population.
6. Shape of dose-response curve outside measured range (e.g. high-to-low-dose extrapolation)
Based on the estimated shape of the dose response curve in the measured range and other information, like mode of action etc., is it possible to extrapolate outside the measured range? If yes, even roughly, what is the best estimate for the shape of the dose-response curve outside the measured range, which is needed for the extrapolation and which are the main uncertainties?<o p="">
7. Not correct outcome (endpoint extrapolation)
Because of the restricted possibilities of measuring health effects and outcomes in the study, different and/or indirect indicators must often be used in order to measure the outcome of exposure.
→ Describe how the measured results of health effect can be extrapolated to the ideal population.
8. Not correct dose timing
Describe how correct is the timing of exposure of the study population in relation to the probable/known timing of exposure of the ideal population.
9. Not correct dose scaling
If the dose is scaled differentially in study population vs. ideal population, then assess the effect of this difference on the dose-response curve.
10. Not correct route of administration
If the route of administration is different between the study and ideal population, then assess the effect of this difference on the dose-response curve. Often it is useful to consider the target organ dose.
11. Not correct toxicodynamics
Describe the differences between the ideal population and the study population in toxicodynamics.
12. Not correct toxicokinetics
Describe the differences between the ideal population and the study population in toxicokinetics.
13. Within human variation
Describe and assess the variation in outcome between individuals within the population.
14. Heterogeneity between studies
After the correction of potential confounding within studies (compared with the population of interest), the studies may still show more than random variation of results. Describe possible reasons to this. Is there publication bias (e.g. studies showing no association have not been published)? Are the studies different in some major respect, i.e. age or health status of the study population, differences in design, differences in outcome definitions etc. Are there reasons to believe that there are differences in the quality of the studies?
15. Publication bias
The publication bias is the difference of the dose-response curve between all the studies that have been made and the studies that have been published. The unpublished studies have more likely negative results than published studies.
16. Plausibility
Given all information, not only the studies included in the analysis, is it plausible that the association between the exposure and the disease is really causal? In the assessment, any available information on the exposure or related exposures or on the disease and other factors should be taken into account. This assessment is typically done through expert elicitation.
References
- Niittynen M., Tuomisto J.T., Karjalainen, A., Miettinen H., Viluksela M., and Pekkanen J. A case study of combining epidemiology and toxicology in environmental health risk assessment: TCDD-induced dental aberrations in humans and rats. Manuscript.
- Pekkanen J., Niittynen M., Viluksela M., Tuomisto J.T. Combining animal and human data in dose-response assessment: towards evidence-based environmental health risk assessment. Manuscript.
