Dose-Response Relationship

Dose-response is a term that describes a relationship between an exposure and the risk of an out come. A dose-response relationship is one in which increasing levels of exposure are associated with either increasing or decreasing risk of the outcome. Demonstration of a dose-response relationship is considered strong evidence for a causal relationship between the exposure and the outcome, although the absence of a doseresponse relationship does not eliminate the possibility of a causal relationship.

The increase in the exposure can be in its intensity or its duration. Exposure can be characterized in different ways such as the peak exposure; the duration of exposure at or above a set level; average exposure, which is a time-weighted average of exposure; or cumulative exposure, which is the sum of timeweighted exposures.

The time to response must be considered when examining the relationship of the exposure to the outcome as there may be a latent period between exposure and the outcome. If the effects of exposure are measured too soon after the exposure, no effect will be seen even in the case where the exposure causes the outcome. One example of this is the increased risk of leukemia after exposure to radiation.

Odds ratios or relative risks can be calculated for categories of increasing exposure each compared with a baseline exposure level, as shown in Table 1. The mathematical relationship of exposure to outcome may be linear, log linear, or follow some other pattern. There may be some level of risk even in the absence of exposure, or there may be a threshold dose below whichnoaffectofexposureonriskisseen(Figure1).

In some cases, the relationship between exposure and outcome may be U-shaped with high risk at both extremes of exposure, but lower with intermediate exposure (see Figure 2). One example of this is the relationship of vitamin A with birth defects. Increased risk of birth defects is seen not only with deficiency in vitamin A but also with excessive doses.

A statistical test for trend can be performed to verify that any apparent trend in the data is statistically significant. The Cochran-Armitage Test is one test for a trend in a binary outcome (ill or not ill) and applies to a linear relationship between exposure and outcome. Another is the Mantel-Haenszel Test, an extension of the chi-square test for trend (see Schlesselman, 1982, pp. 203–206, for details).

Inclusion of small numbers in the groups at the extreme ends of the exposure distribution may lead to statistically unstable rates in these groups. This may affect the validity of an apparent trend. Also, these end categories sometimes include extreme values, and the results can be sensitive to these extreme values.

Figure 1 Threshold Dose

[Page 283]

Figure 2 U-Shaped Relationship

For example, very few subjects may be included in the smoking exposure category ‘More than two packs per day,’ and this category may include a subject with exposures far in excess of anyone else in the study. For this reason, it is important to examine the effect of extreme values on the results.

...