Evaluate Known Predictors of Exposure

Smoke Free In One Hour

The Painless Stop Smoking Cure

Get Instant Access

Although the nature of the association between exposure and the disease of interest is uncertain to at least some extent, or there would be no motivation to conduct the study, there may be well-established predictors of exposure that are known with certainty. If accurately assessed exposure can safely be assumed to have certain such predictors, assessing whether the expected associations with exposure are present would help to indicate whether the exposure data are valid.

Exposure predictors are generally not of direct interest in relation to disease, so that this information will not always be collected unless the application of such information to the validation of exposure measurement is anticipated. The basis for the linkage between the antecedent and the exposure need not be causal, in that a well-established non-causal statistical predictor would serve the same purpose of helping to indicate that exposure has (or has not) been measured successfully. When such a predictor of exposure is available, and the expected relation to exposure is very strong, the predictor may even be a useful proxy measure of exposure.

For example, assessment of the use of illicit drugs is a great challenge. It is well known however, that a strong predictor, and perhaps a causal determinant, is the pattern of drug use among friends. Therefore, it would be expected that those who are using illicit drugs would also report having friends who do so. Thus, to avoid at least part of the sensitivity and stigma associated with such behavior, questionnaires might include items pertaining to drug use among friends, something that respondents may be more willing to admit to than their own drug use. Such information can be used to determine whether the expected positive association is found with self-reported drug use, and also to create a category of uncertain drug use when the individual reports not using drugs but having friends who do so.

Another illustration of ascertaining and using information on the predictors of exposure is often applied in the assessment of use of therapeutic medications. There are certain illnesses or symptoms that serve as the reasons for using those medications, and the credibility of reports of drug use (or even non-use) can be evaluated to some extent by acquiring information on the diseases that the drug is used to treat. When a respondent reports having an illness that is known to be an indicator for using a specific medication, along with recall of using that medication, confidence is enhanced that they are accurately reporting the medication use. Those who had an illness that should have resulted in use of the drug but did not report doing so, and those who reported using the medication but without having reported an illness for which that medication is normally used, would be assigned a less certain exposure status.

In a study of the potential association between serum selenium and the risk of lung and prostate cancer among cigarette smokers, Goodman et al. (2001) provided a rather detailed analysis of predictors of serum selenium concentrations (Table 8.5). In addition to addressing concerns with the comparability of collection and storage methods across study sites, they were able to corroborate the expected reduction in serum selenium levels associated with intensity and recency of cigarette smoking. Even though the background knowledge is limited to help anticipate what patterns to expect, confirming the inverse association with smoking adds confidence that the measurements were done properly and are more likely to be capturing the desired exposure.

Even when the linkage of antecedent to exposure is less direct, as in the case of social and demographic predictors, there may still be value in assessing exposure predictors as a means of evaluating the accuracy of exposure information. Weaker associations with exposure or those that are less certain will be less contributory but can help to provide at least some minimal assurance that the exposure information is reasonable. If assessing the consequences of otitis media in children on subsequent development, the known positive association of the exposure with attendance in day care and sibship size and patterns of occurrence by age (Hardy & Fowler, 1993; Zeisel et al., 1999) may be helpful in verifying that otitis media has been accurately documented. As always, when the data conflict with prior expectations, the possibility that prior expectations were wrong

Table 8.5. Adjusted Mean Serum Selenium (^g/dl) Concentration in Control Participants, Carotene and Retinol Efficacy Trial, 1985-1999






11.55 (0.07)


11.75 (0.17)




11.58 (0.07)

African American

11.57 (0.32)


11.64 (0.37)

Exposure Population



11.31 (0.16)

Heavy smokers

11.72 (0.09)

Study Center



9.89 (0.27)


10.76 (0.22)

New Haven

11.20 (0.28)


11.96 (0.13)

San Francisco

10.99 (0.30)


12.01 (0.10)

Blood Draw Smoking Status



11.34 (0.09)


11.86 (0.09)

*Adjusted means from model including all variables with p < 0.1 except years quit smoking for the heavy-smoker population. Adjusted means for blood draw year (p = 0.28) not given because this was a linear variable in the model.

tp for test of heterogeneity.

SE, standard error.

Goodman et al., 2001.

*Adjusted means from model including all variables with p < 0.1 except years quit smoking for the heavy-smoker population. Adjusted means for blood draw year (p = 0.28) not given because this was a linear variable in the model.

tp for test of heterogeneity.

SE, standard error.

Goodman et al., 2001.

needs to be considered as an alternative to the inference that the data are in error.

Was this article helpful?

0 0
5 Easy Ways To Stop Smoking

5 Easy Ways To Stop Smoking

Your first day without cigarettes can be difficult, but having a plan will make it easier! Learn what steps to take on the day you quit smoking.

Get My Free Ebook

Post a comment