Disease May Distort Reporting or Perception of Exposure. Assuming a given amount or level of exposure has occurred, there must be some mechanism by which information on that exposure is ascertained. Questionnaires or interviews are usually at least a component of the exposure assessment process, whether done in a clinical setting and recorded in a medical record, elicited by an interviewer exclusively for the study, or through a mailed questionnaire. The respondent is asked to tell about his or her exposure, either directly or indirectly. Therefore, the opportunity for their representation of that exposure to be influenced by current or past health experience needs to be scrutinized.
One classic mechanism by which disease distorts reporting of exposure is through conscious concerns with the potential exposure-disease association under study. If an individual has heard or believes that exposure causes a disease, and he or she is suffering from that disease, he or she may be more inclined to believe and report that he or she was exposed. What is only a vague memory may be reported as fact, or rumination about whether he or she did or did not have the putative causal exposure may result in erroneously reporting that he or she did. While this phenomenon is appropriately invoked as a concern in studies that require reporting of exposure after disease onset, usually case-control studies, the circumstances under which it would occur are limited. Although there is much speculation about disease-inspired reporting of exposure because it seems reasonable to anticipate it, there are few instances in which this biased reporting has been documented.
One of the issues of concern has been the reporting of exposure to medications during pregnancy in relation to the risk of having a child with a congenital defect (Klemetti & Saxen, 1967). Mothers who have suffered the emotional trauma of having an affected child are postulated to overreport use of medications that they did not truly take. A number of studies indicate that recall of medications and many other exposures during pregnancy is often incomplete, but overreporting is quite rare (Werler et al., 1989). In addition, the patterns of erroneous underreporting tend to be similar among those with and without the disease. The susceptibility to differentially incomplete reporting depends in large part on how susceptible the exposure is to incomplete reporting more generally. Recall of relatively inconsequential events, such as taking cold medications during pregnancy, is highly vulnerable to faulty, often incomplete, recall. In contrast, recall of whether medications to control epilepsy were taken during pregnancy is likely to be much more comprehensively recalled and thus is less likely to differ for those with and without an adverse health outcome.
In order for the completeness of recall to be affected by the health outcome, the respondent must consciously or unconsciously establish a link between the exposure and outcome. Therefore, the potential for biased reporting of exposure depends in part on perception among respondents of a link between exposure and disease, and those relations that have a high level of media attention and public interest are therefore more vulnerable. Many putative associations gather widespread public interest and awareness, making it difficult to elicit information without some contamination based on the participants' expectations. Investigators sometimes will ask directly if the participant has heard about or believes that the exposure of interest can cause the disease under study, allowing for stratification by those who do and do not believe it does. Although it can be argued that those lacking suspicion would generate the more accurate responses, it could also be hypothesized that those who are most attentive to media reports are also the most attentive to their own exposures and provide the more accurate data.
An additional determinant of susceptibility to biased recall is the degree of subjectivity in defining the exposure. Events such as prior surgery or level of education, for example, are unambiguous and not subject to varying interpretation. In contrast, history of diseases based on self-diagnosis, such as frequent headaches or conditions that are inherently subjective such as perceived stress are much more vulnerable to differential misclassification in that the reporting contains an element of judgment.
There are several approaches to evaluating how likely it is that such biased recall has occurred and if it has occurred, how much it has affected the study results. As always, the opportunity to validate reported exposure is optimal, in this case requiring validation of reports from a sufficient number of persons with and without the disease to contrast the two groups. Both the absolute level of accuracy, as an indicator of potential bias due to exposure misclassification generally, and relative level of accuracy among those with and without disease, as an indicator of potential bias due to differential exposure misclassification, are of interest. The closer the validation measure is to the gold standard, the better, but even a measure of similar quality that is less vulnerable to distortion due to disease may be helpful. That is, an equally fallible exposure indicator that could not possibly be affected by development of disease, such as record-based exposure ascertained long before disease development, can be compared to self-reported exposure among cases and non-cases to determine whether there is evidence of differential error, even if the absolute level of error is not known.
Another approach is to examine some exposures that are very unlikely to affect the occurrence of disease but otherwise meet all the criteria for biased reporting. Exposures that are widely perceived as being related to disease causation are most vulnerable to reporting error. If data on such an exposure or ideally, multiple exposures, are available, and show no evidence of differing in relation to disease status, this would provide reassurance that it is unlikely to be present for exposures that are less widely perceived as being related to disease or more objective in nature. If there is evidence of recall bias for those exposures most susceptible to biased reporting, then further examination of the potential for bias in the measurement of exposures of interest is warranted. Data on a spectrum of exposures can be helpful in looking for patterns of this sort.
Often, this can be readily incorporated into study of multiple specific agents, e.g., medications, dietary constituents. Even if the true interest is in a single medication or a narrow class of medications, for example, the investigators might obtain a somewhat longer list to use the other drugs as markers of susceptibility to differential misclassification. That is, if medications thought to be very unlikely to cause disease are nevertheless reported more commonly by those with disease than those without, differential exposure misclassification for those drugs as well as the ones of primary interest might be inferred as being more probable. Application of this strategy requires placebo exposures that are perceived by the public to be potential causes but unlikely to actually affect disease.
An example from the literature serves to illustrate several of these pathways for differential exposure misclassification. The possibility that head trauma may affect risk of developing brain tumors has been evaluated rather extensively (Ahlbom et al., 1986; Burch et al., 1987), and always in case-control studies. This is the sort of hypothesis that is more intuitively appealing to the public than to most investigators, who are skeptical about whether a plausible mechanism might link physical trauma to tumor development. In a series of case-control studies reporting a positive association, there is a persistent concern with over-reporting by cases. In order to address that possibility, more objective measures of head trauma have been sought by restricting the severity. Whereas recall of minor bumps on the head over one's life may be highly vulnerable to distortion, memory of head trauma resulting in loss of consciousness or medical treatment is less susceptible to individual interpretation and selective recall. The ratio of reports of major, verifiable head trauma to unverifiable minor head trauma might be examined for cases and controls. Other ostensibly significant exposures in the head area might be queried as markers of biased reporting, e.g., cuts, insect bites. The proportion of participants who believe that head trauma may play a role in the development of brain tumors would be of interest, giving some sense of the level of vulnerability to biased reporting.
Disease May Distort the Measure of Exposure. Analogous to the phenomenon in which the occurrence of disease may alter perception or self-report of exposure, disease may alter biological measures of exposure if those measures are taken after or even near the time of disease onset. Despite many attractive features of biological markers of exposure, the question of whether their accuracy has somehow been altered by the disease of interest deserves careful appraisal. For diseases that develop over extended periods of time before becoming clinically apparent, including many chronic diseases such as neurodegenerative disease, cancer, and atherosclerotic heart disease, measures taken months or even several years before the manifestation of disease could reflect, in part, the evolving disease itself. Early stages of disease or disease precursors could alter the measure of exposure and produce an association between measured exposure and disease that has no etiologic significance.
Research on chlorinated hydrocarbons and breast cancer illustrates the concerns that can arise about whether the disease or its treatment might affect measured serum levels of DDT, DDE, and other stored compounds (Ahlborg et al., 1995). If, in fact, early stages of breast cancer result in the release of some of the persistent organochlorines that are stored in fat tissue, then women who go on to develop clinically apparent breast cancer will have had some period prior to diagnosis during which blood levels of those compounds were elevated, solely as a result of the early disease process. A prospective study that uses measurements obtained in the period shortly before case diagnosis could be affected. In a case-control study in which blood levels of organochlorines are measured after diagnosis and treatment of disease, there is even greater susceptibility to having case-control differences arise as a result of metabolic changes among cases. The problem arises when the measure of exposure is also, in part, reflecting the consequence of the disease itself.
One way to evaluate the potential influence of disease on measures of exposure is through a thorough understanding of the biological determinants of the exposure marker, allowing assessment of whether the disease process would be expected to modify the measure. Obviously, this requires substantive knowledge about the often complex pathways affecting the measure and a thorough understanding of the biologic effects over the course of disease development. In the case of breast cancer and chlorinated hydrocarbons, the pathways are quite complex and make it difficult to predict the combined effect of the disease on metabolism, influence of disease-associated weight loss, etc.
A better approach is to empirically assess the influence of disease on the exposure marker, through obtaining measurements before the disease has occurred, ideally even before the time when disease precursors would have been present, as well as after disease onset. The interest is in exposure during the etiologic period before the disease has begun to develop, so the measure prior to the onset of the disease can be viewed as the "gold standard" and the accuracy of the measure taken after onset can be evaluated for its adequacy as a proxy. This requires having stored specimens for diseases that are relatively rare and develop over long periods of time, so that a sufficient number of prediagnosis measurements are available. One study was able to use specimens obtained long before the manifestation of disease (Krieger et al., 1994), but did not report any comparisons of prediagnosis with postdiagnosis measures from the same women. One study did evaluate the effect of treatment for breast cancer on such markers (Gammon et al., 1996), and verified that the values just prior to initiation of treatment were quite similar to those observed shortly after treatment began, addressing at least one of the hypothesized steps at which distortion might arise.
Even if the ideal data are not available, strategies exist for indirectly assessing the likely impact, if any, of disease on the measure of exposure. The disease may have varying degrees of severity, with influences on exposure measures more likely for severe than mild forms of the disease. For example, breast cancer ranges from carcinoma in situ, which should have little if any widespread biologic effects, to metastatic disease, with profound systemic consequences. If the association between chlorinated hydrocarbons and breast cancer were similar across the spectrum of disease severity, it would be unlikely to merely reflect metabolic changes associated with disease given that the severity of those metabolic changes would be quite variable among breast cancer cases. Examining the pattern of results across the spectrum of disease severity would reveal the extent to which the disease process had altered measurements, with results for the least severe disease (carcinoma in situ) most valid and the results for late-stage disease least valid.
The timing of exposure ascertainment relative to the onset of disease has been examined as well to indicate the likely magnitude of distortion. A series of studies in the 1970s and early 1980s had suggested that low levels of serum cholesterol were related to development of a number of types of cancer, with cholesterol assessed prior to the onset of disease (Kritchevsky & Kritchevsky, 1992). Nonetheless, there was great concern with the possibility that even cancer in its early, preclinical stage may have affected the serum cholesterol levels. Under this scenario, low levels of cholesterol followed within a limited period, say 6 to 12 months, by the diagnosis of cancer, may have been low due to the developing cancer itself. The approach taken to assess this problem has been to examine risk stratified by time since the measurement in longitudinal studies, in order to determine whether the pattern of association with disease differs across time. It would be more plausible that preclinical cancer that became manifest in the first 6 months following cholesterol measurement had affected the cholesterol measure than it would for cancers diagnosed 5 years or 10 years after cholesterol measurement.
Disease May Cause Exposure. For certain types of exposure, it is possible for early disease not just to distort the measure of exposure, as described above, but also to actually cause the exposure. This is especially problematic for exposures that are closely linked to the early symptoms of the disease of concern, which may include medications taken for those symptoms or variants of the symptoms themselves. Prior to being diagnosed, early disease may lead to events or behaviors that can be mistakenly thought to have etiologic significance. The causal sequence is one in which preclinical disease results in exposure, and then the disease evolves to become clinically recognized.
Among the candidate influences on the etiology of brain tumors are a number of diseases or medications, including the role of epilepsy and drugs used to control epilepsy (White et al., 1979; Shirts et al., 1986). It is clear that epilepsy precedes the diagnosis of brain tumors, and medications commonly used to treat epilepsy are more commonly taken prior to diagnosis by brain tumor cases than controls in case-control studies. What is not clear is whether the early symptoms of brain tumors, which are notoriously difficult to diagnose accurately in their early stages, include epilepsy, such that the disease of interest (brain tumor) is causing the exposure (epilepsy and its treatments). Similar issues arise in studying medications used to treat early symptoms of a disease, e.g., over-the-counter medications for gastrointestinal disturbance as a possible cause of colon cancer.
Similarly, undiagnosed chronic disease has the potential to distort certain exposures of interest as potential causes, illustrated in a recent study of depression as a potential influence on the risk of cancer. In a large cohort study in Denmark, Dalton et al. (2002) evaluated the association between depression and other affective disorders in relation to the incidence of cancer. By stratifying their results by years of follow-up (Table 8.8), they were able to consider the time course of depression and cancer incidence to better understand the etiologic significance of the results. Focusing on the results for the total cohort, note that brain cancer risk was substantially elevated for the first year of follow-up only, returning to near baseline thereafter. Although it could be hypothesized that depression is causally related to brain cancer with a short latency, it seems much more likely that undiagnosed brain cancer was a cause of the depressive symptoms given what is known about the time course of cancer development. That is, the disease of interest, brain cancer, undiagnosed at the time of entry into the cohort, may well have caused the exposure of interest, depression. The likelihood that undi-agnosed brain cancer becomes manifest many years after entry, distorting measures of association years later, is much less plausible so that the overall pattern is suggestive of reverse causality.
Was this article helpful?
Learning About 10 Ways Fight Off Cancer Can Have Amazing Benefits For Your Life The Best Tips On How To Keep This Killer At Bay Discovering that you or a loved one has cancer can be utterly terrifying. All the same, once you comprehend the causes of cancer and learn how to reverse those causes, you or your loved one may have more than a fighting chance of beating out cancer.