Examine Multiple Indicators of Exposure

Smoke Free In One Hour

The Painless Stop Smoking Cure

Get Instant Access

It is preferable to have a single, perfect "gold standard" to serve as the referent for the routine exposure measure, or at least a measure that better approximates the "gold standard." It may still be informative however, to have another exposure proxy of similar quality to the routinely applied measure but differing in character, and thus in the nature of the error it contains. Neither the routine exposure measure nor the alternative measure in isolation is necessarily a better approximation of the exposure of interest, but we would expect them to be associated with one another to at least some extent because both are associated with the true exposure.

For example, in measuring exposure to environmental tobacco smoke over long periods of time, there are two basic strategies available: biochemical markers of short-term exposure and self-reported information on proximity to active smokers. The biochemical markers are precise indicators over short periods of time. Assuming that the measurement period represents a sample from the long-term period of etiologic interest, the accurate short-term measure has some value as a marker of long-term exposure. The self-report of proximity to smokers is capable of integration over the extended time period of etiologic interest, given that questions can be focused on specific periods of interest and the respondent can presumably reflect on and recall the period of interest. Self-report is vulnerable to the uncertainties of perception and recall, however, including potential biases in perceiving the presence of tobacco smoke and faulty memory in recalling those experiences. What type of association might we expect to observe between these markers, both of which are imperfect relative to the "gold standard," but for very different reasons? How do we interpret measures of their relationship to one another and the relationship of each marker to disease status?

First, it is important to remember that the accurate indicator of short-term exposure does not serve as the "gold standard" for the imprecise indicator of long-term exposure. Both are approaches to estimating the long-term exposure of eti-ologic interest, and both are inferior to concurrent measurement of exposure throughout the etiologic period as might be accomplished in a true prospective

Table 8.2. Odds Ratios and 95% Confidence Intervals for Magnetic Field Exposure and Acute Lymphocytic Leukemia from Different Residences, Categorised According to Initial Cutoff Points of Magnetic Field Exposure, National Cancer Institute Childhood Leukemia Study

Relative Risk for Acute Lymphoblastic Leukaemia Calculated With: TWA from Measurement from Longer Measurement from Former Measurement from Currently

Table 8.2. Odds Ratios and 95% Confidence Intervals for Magnetic Field Exposure and Acute Lymphocytic Leukemia from Different Residences, Categorised According to Initial Cutoff Points of Magnetic Field Exposure, National Cancer Institute Childhood Leukemia Study

Relative Risk for Acute Lymphoblastic Leukaemia Calculated With: TWA from Measurement from Longer Measurement from Former Measurement from Currently

EXPOSURE

Two Homes Measured

Lived in Home Only

Lived in Home Only

Lived in Home Only

CATEGORIES

MEAN

MEAN

MEAN

MEAN

(MT)

(MT)

CASES

OR

95% CI

(MT)

CASES

OR

95% CI

(MT)

CASES

OR

95% CI

(MT)

CASES

OR

95% CI

< 0.065

0.047

53

1.00

0.042

64

1.00

0.042

59

1.00

0.042

66

1.00

> 0.065-

0.082

33

0.97

0.52 to

0.080

27

1.23

0.62 to

0.083

22

0.85

0.43 to

0.079

31

1.28

0.68 to

< 0.099

1.81

2.39

1.68

2.44

> 0.100-

0.137

40

1.14

0.63 to

0.137

35

1.28

0.69 to

0.140

39

1.35

0.73 to

0.133

32

1.09

0.59 to

< 0.199

2.08

2.38

2.47

2.02

> 0.200

0.350

23

1.81

0.81 to

0.374

23

1.15

0.57 to

0.370

29

1.65

0.82 to

0.322

20

1.47

0.67 to

4.02

2.33

3.32

3.20

Ptrend =

0.2

ptrend

= 0.3

ptrend =

0.1

ptrend =

0.4

OR, odds ratio; CI, confidence interval; TWA, time-weighted average Baris et al., 1999.

OR, odds ratio; CI, confidence interval; TWA, time-weighted average Baris et al., 1999.

study. Because both are believed to be related to long-term exposure, however, one expects some association between the two measures and it would be troubling if they were completely unrelated. The magnitude of the association between two measures, even if both are associated with a third (e.g., the true value), however, can be quite small. Two variables can show a correlation coefficient as high as 0.7 with a third variable, for example, yet have a correlation of 0 with one another (Berkson, 1946). Thus, it would be possible for each of two proxies to be rather strongly related to the "gold standard," yet not related strongly to each other. In the absence of a "gold standard" measure, interpretation of the association between two imperfect indicators is thus of limited value in assessing the quality of either one.

Ideally, substantive understanding of each of the exposure measures could be used to create an integrated measure of exposure that takes advantage of the information provided by each. That is, with recognized strengths and deficiencies in each of two or more markers of exposure, those information sources might be combined into a single measure that is expected to be better than any of them would be in isolation. There is no generic approach to integrating these sources of exposure data because it depends on the way in which the information they provide is complementary. It might be known that when certain combinations of results occur, one measure should override the other. For exposures in which measurement is known to be highly insensitive, e.g., illicit drug use, we might accept that any of self-report, metabolites in urine, or metabolites in hair constitutes exposure. We might know that one measure is more accurate in a certain range of exposure or for certain respondents and another measure is better under other circumstances. For example, in comparing a biological measure with self-report, there may be known sources of interference from certain medications, unusual dietary habits, or illness. If self-report were negative, and the biological measure were positive but with an indication of a metabolic disease that can create false positive readings, one might simply override the biological marker for that individual and classify the person as negative. Similarly, there may be persons in whom the quality of self-report is thought to be fallible due to poor cooperation according to the interviewer, and we would then rely instead on the biological marker.

In the absence of substantive understanding of how errors in one measure or the other arises, one simplistic approach is to combine their information empirically into a composite variable, i.e., define exposed as positive on both markers or as positive on either of two markers. If the major problem is known to be one of sensitivity, then an algorithm might be applied in which being positive on either marker is used to infer that exposure is present. This will enhance the sensitivity relative to using either measure in isolation, and decrease the specificity. Alternatively, if specificity is the problem one wishes to minimize, classification as exposed may require being positive on both indicators, with only one being positive resulting in assignment as unexposed. Preferable to either extreme would be an examination of the pattern of results across all combinations of the two ex posure measures, i.e., both positive, one measure positive and one measure negative, and both negative. Unless one or both of the measures is behaving very strangely, it would be expected that these levels would correspond to a monotonic gradient of true exposure, less subject to misclassification than use of either measure in isolation.

When one of the data sources on exposure can be viewed as a "gold standard," it provides an opportunity to better understand and ultimately refine the routine measure. For example, with a combination of self-reported exposure to environmental tobacco smoke over long periods in the past, and short-term biochemical markers, there is an opportunity to integrate the information to validate the self-report. Self-reported exposure can be generated over the time frame of ultimate interest, as well as for the brief time period reflected by the biochemical measure of exposure, i.e., the recent past. With that information and accepting the biochemical marker as a gold standard for the recent past, predictive models can be developed in which the self-reported information is optimized to estimate actual exposure. In the example of environmental tobacco smoke, self-report of exposure in the preceding 24 or 48 hours might be queried, for which the biochemical indicator would be a legitimate gold standard. With that quantitative prediction model now in hand, the questionnaire components for the period of etiologic relevance, typically a prolonged period in the past, can be weighted to generate a more valid estimate of historical exposure. The data would be used to determine which self-reported items are predictive of measured exposure to environmental tobacco smoke and the magnitude of the prediction, through the use of regression equations. Although there is no direct way to demonstrate that this extrapolation from prediction over short periods to prediction over long periods is valid, the availability of a "gold standard" for brief periods offers some assurance. The development of the predictive model linking self-reported exposure data to biological markers need not include all the study subjects and could be done on similar but not identical populations. The relationship between perceived experiences and actual exposure may well differ among different populations, however, suggesting that the validation be done on the study population or a group that is quite similar to the actual study subjects.

Multiple exposure indicators also may be used when it is unclear which is the most influential on the health outcome of interest. A sizable body of research has addressed particulate air pollution and health, particularly morbidity and mortality from cardiovascular and respiratory disease. As the research has evolved, there has been an increasing focus on the small particles, those < 10 fig/m3 or < 2.5 fig/m3 in diameter. In a large cohort study of participants in the American Cancer Society's Cancer Prevention II Study, air pollution measures in metropolitan areas were examined in relation to mortality through 1998 (Pope et al., 2002). To examine the nature of the relationship between particulate air pollution and mortality, a range of indices were considered, defined by calendar time of measurement, particle size, and sulfate content (Fig. 8.1). These results suggest once

[A All-Cause Mortality

Cardiopulmonary Mortality i I

Was this article helpful?

0 0
Quit Smoking Today

Quit Smoking Today

Quit smoking for good! Stop your bad habits for good, learn to cope with the addiction of cigarettes and how to curb cravings and begin a new life. You will never again have to leave a meeting and find a place outside to smoke, losing valuable time. This is the key to your freedom from addiction, take the first step!

Get My Free Ebook


Post a comment