Assess Rates for Diseases Known Not to Be Affected by the Exposure

The Big Heart Disease Lie

How I Healed my Cardiovascular Disease

Get Instant Access

For most exposures of possible health relevance, we have sufficient background knowledge to delineate some health outcomes that are likely to be affected (typically the ones that motivate the study and similar diseases) and other health outcomes that are highly unlikely to be affected. The conventional wisdom is fallible, of course. There are notable historical examples of erroneous assumptions about health outcomes that were certain to be unaffected by a given exposure. For example, men with chronic bronchitis were selected as controls in early case-control studies of lung cancer because chronic bronchitis was believed to be unaffected by tobacco smoking. Humans have a way of surprising epidemiologists with unanticipated associations, but in general, we can specify some diseases that are very likely to be affected by the exposure of interest based on current knowledge and diseases that are very unlikely to be affected by that exposure. Within the bounds of random error, and in the absence of selection bias, we would expect rates of disease that are not causally related to the exposure to be similar among exposed and unexposed groups. In other words, should differences be found in the rates of certain diseases in relation to exposure, and the possibility that such differences result from a causal effect of the exposure is remote, random error and selection bias become the most plausible candidate explanations.

For example, in a study of the effects of sunscreen use on risk of developing melanoma (an illustration from an oral presentation by Diana Petitti, Kaiser Permanente of Southern California), we would not expect sunscreen use to influence the risk of myocardial infarction, breast cancer, or motor vehicle injury. To determine whether our group of nonusers of sunscreen is a good counterfactual comparison group to the sunscreen users, reflecting the risk that the sunscreen users would have had in the absence of sunscreen use, we might find it useful to examine an array of causes of death that include some that should not differ due to a causal impact of sunscreen use. Even if our assumptions are incorrect in some of the diseases thought to be unrelated to sunscreen use, examination of the overall pattern of results across a range of presumably unrelated diseases would reveal whether a systematic tendency is present for exposed and unex-posed groups to differ. If, for example, we observed consistently depressed disease rates across a series of causes of death thought not to be causally related to sunscreen use, the comparability of the groups for studying melanoma might be called into question. We may well find that users experience a lower risk of my-ocardial infarction, for example, due to other manifestations of the health consciousness that led them to be sunscreen users and may have lower rates of mo tor vehicle injury due to seat belt use, likely correlated with sunscreen use as a preventive health measure. We would be reminded to look carefully for other, correlated preventive health measures that may lead to more (or less) favorable patterns of melanoma incidence among sunscreen users, such as more frequent examination by a physician. If the sunscreen users had disease patterns similar to nonusers, except for the one of interest, i.e., melanoma, the potential for selection bias would be reduced.

A recent report on the impact of fine particulate air pollution on mortality from respiratory and cardiovascular disease, plausible consequences of such exposure, also considered a residual set of deaths from other causes (Pope et al., 2002). The extraordinarily large study of volunteers enrolled by the American Cancer Society into the Cancer Prevention II Study, 1.2 million adults, provided the basis for this investigation. As is often the case with studies of this issue, the measures of association between pollutants and mortality are modest in magnitude but highly precise, given the large population (Table 4.3). The categories of particular interest and plausibility, lung cancer and cardiopulmonary disease, showed increments in risk of 6% to 13% per 10 ¡xgjm3 over the time intervals examined, contributing to an association with all-cause mortality that was present but lower in magnitude. Once deaths from lung cancer and cardiopulmonary disease are removed, the residual category showed essentially no association, as one might expect from a conglomeration of other cancers, infectious diseases, injury mortality, etc. That is, observing an association between fine particular air pollution and deaths from causes other than those most plausible would raise the serious possibility that some selection bias for persons living in high exposure communities was operating and would suggest that the apparent effect of particulates on lung cancer and cardiopulmonary diseases might be due to some non-specific aspect of living in more highly exposed communities.

Table 4.3. Adjusted Mortality Relative Risk Associated with a 10 ¡¡g/m3 Change in Fine Particles Measuring Less Than 2.5 ¡m in Diameter, American Cancer Society Cancer Prevention II Study

Adjusted RR (95% CI)*

CAUSE OF MORTALITY

1979-1983

1999-2000

AVERAGE

All-cause Cardiopulmonary Lung cancer All other cause

1.04 (1.01-1.08) 1.06 (1.02-1.10) 1.08 (1.01-1.16) 1.01 (0.97-1.05)

1.06 (1.02-1.10) 1.08 (1.02-1.14) 1.13 (1.04-1.22) 1.01 (0.97-1.06)

1.06 (1.02-1.11) 1.09 (1.03-1.16) 1.14 (1.04-1.23) 1.01 (0.95-1.06)

♦Estimated and adjusted based on the baseline random-effects Cox proportional hazards model, controlling for age, sex, race, smoking, education, marital status, body mass, alcohol consumption, occupational exposure, and diet.

RR, relative risk; CI, confidence interval.

♦Estimated and adjusted based on the baseline random-effects Cox proportional hazards model, controlling for age, sex, race, smoking, education, marital status, body mass, alcohol consumption, occupational exposure, and diet.

RR, relative risk; CI, confidence interval.

Like all criteria for assessing selection bias, this approach can also be misleading. As already noted, diseases thought to be unrelated to exposure may turn out to be causally related to the exposure, so that we would erroneously infer selection bias when it is not present. Many if not all known causes of disease affect more than one specific entity. Conversely, comparability for diseases other than the one of interest is only indirectly pertinent to whether the exposure groups are comparable for the disease of interest. A selection bias may be present or absent solely for the health outcome of interest, so that reassuring patterns for other outcomes are misinterpreted as indicative of valid results for the outcome of interest. The patterns of disease other than the one of interest are a flag to examine the issue further, not a definitive marker of the presence or absence of bias.

Was this article helpful?

0 0
A Disquistion On The Evils Of Using Tobacco

A Disquistion On The Evils Of Using Tobacco

Among the evils which a vitiated appetite has fastened upon mankind, those that arise from the use of Tobacco hold a prominent place, and call loudly for reform. We pity the poor Chinese, who stupifies body and mind with opium, and the wretched Hindoo, who is under a similar slavery to his favorite plant, the Betel but we present the humiliating spectacle of an enlightened and christian nation, wasting annually more than twenty-five millions of dollars, and destroying the health and the lives of thousands, by a practice not at all less degrading than that of the Chinese or Hindoo.

Get My Free Ebook


Post a comment