Table 17 A casecontrol study of Chlamydia infection and coronary atherosclerosis13

Chlamydia present Chlamydia absent (exposure = yes) (exposure = no)

Total

Atherosclerotic 71 (78.9%)

coronary disease (Outcome = Yes)

No atherosclerotic 1 (4.2%) coronary disease (Outcome = No)

example, the authors only included subjects who were documented as being seronegative at the time of traumatic exposure to HIV-contaminated material.8 Similarly, in the cohort study of DCC protein status, the outcome was death, an event which was clearly separated in time from the time of assessment of DCC protein status.9 In other studies, the situation may be somewhat ambiguous or in the extreme, such as in the case of the Chlamydia infection study, totally unclear. An adequate assessment of an epidemiologic study must always therefore include as an essential element a consideration of whether the temporal sequence was adequately characterized.

Precision

Subjects included in epidemiologic studies can generally be seen as a sample from a much larger, possibly infinite, population. This population may be conceptual (for example, all future cases of HIV infection) or real (for example, all actual cases on a given territory). For example, in the case-control study of primary pulmonary hypertension presented earlier, 95 cases of the disease were included while thousands of such cases are actually known.10 The estimate of the relative risk can therefore be considered as somewhat imprecise, being based on observations from a limited subset of the entire population of such cases. The assessment of precision of the estimates derived from epidemiologic studies relies on statistical methodology. Statistical issues in epidemiologic research are covered in Chapter 2 of this book.

Bias

The validity of results from an epidemiologic study may be affected for several reasons. Study subjects may not be comparable, with for example older patients among the exposed members of a study than among the unexposed. The method of selection of the study subjects may be biased in such a way as to lead for example to an exaggerated level of exposure in cases, and therefore an inflated odds ratio. Various errors may occur in the collection and recording of the data, leading to erroneous estimates of the measures of association. A standard approach in classifying these various types of biases is to regroup them under the categories of confounding, selection bias, and information bias.

Confounding

Confounding arises when exposed and unexposed subjects are not comparable for a variable which represents a risk factor for the outcome of interest. Tables 1.8 and 1.9 give an example. In Table 1.8, results of a cohort study of the association between radiotherapy to the chest for Hodgkin's disease and subsequent mortality from coronary artery disease are presented.14 The risk of coronary disease death was 2.0% in the exposed (70/3530), and 4.0% in the unexposed (54/1360). The relative risk was therefore 0.5 (= 2%/4%), indicating a reduced coronary disease mortality after radiation to the chest. This result was surprising, as the investigators' hypothesis, arising from previous research in other types of patients and also from animal models, was that exposure to radiation would increase, and not decrease, coronary disease mortality. Further analysis of these data revealed, however, that these results were biased because of the confounding effect of age. Table 1.9 presents results of the same study, stratified into three age groups. The relative risk of coronary disease death was estimated separately for each of these three age groups, and the direction of the association was now positive, with coronary disease mortality being larger in the exposed subjects than in the unexposed subjects in each of the three groups. The difference between the relative risk estimated from the data presented in Table 1.8 and those presented in Table 1.9 is due to a confounding bias arising from the unequal distribution of the exposed and unexposed subjects with respect to age. We can see in Table 1.9 that 76% (2,680/3,530) of the exposed subjects were 0-39 years old while only 35% (470/1,360) of the unexposed subjects were in the same age category. The exposed population was therefore much younger than the unexposed, and since coronary disease mortality is known to be lower in younger subjects, the comparison of exposed and unexposed subjects was biased by age. When results are considered within age strata, this bias is corrected, and this is why the relative risks in Table 1.9 show an increased risk rather than a decreased risk as suggested by the biased comparison based on Table 1.8. The effect of confounding variables may be

Table 1.8. A cohort study of coronary artery disease mortality in patients treated with radiation for Hodgkin's disease (adapted from Boivin et al)*14

Was this article helpful?

0 0

Post a comment