With a precise definition of the optimal exposure measure, or more plausibly, a set of exposure measures, each addressing a particular hypothesis, we can compare candidate operational measures of exposure to the ideal ones. The exact means by which the exposure indicator is constructed needs to be scrutinized, focusing on the operational details that go into the final assignment of individual exposure. The goal is to reveal the compromises that have been made, many without explicit consideration, and the resulting potential for disparity between the operational and ideal exposure measures.
For example, we may be interested in total ethanol intake over a lifetime in relation to cardiovascular disease endpoints, such as angina or myocardial infarction. Obviously, we will not have installed an alcohol meter at birth or directly observed alcohol intake over a lifetime. We may instead have self-report of typical weekly ingestion of beer, wine, and liquor averaged over adulthood, or intake of those beverages for specific periods of life, and use that information to construct a quantitative estimate of lifetime exposure. There is an abundance of opportunities for this operational measure to deviate from the ideal exposure measure, including inaccurate recall and intentional deception. Also there may be error even if the self-report is perfect in that there is likely to be variability in alcohol consumption over time and variable alcohol content of beverages. The etiologic process may require consideration of the amount of alcohol consumed on each occasion or drinking at different ages or different intervals relative to disease onset, introducing additional forms of misclassification when comparing the operational to the ideal measure.
Thus there are two ways in which the operational and ideal measures of exposure deviate from one another. One arises from conceptual problems in the approach to exposure assessment, such that a perfectly executed data collection effort would still result in an imperfect match with the etiologically relevant exposure. The error arises in the very choice of the operational definition of exposure. Second, superimposed on any conceptual misclassification is the more traditional misclassification based on errors in implementing the chosen approach. Environmental measurements contain sampling error and technical imprecision in characterizing chemical and physical agents, for example. Self-reported information on exposure inevitably introduces erroneous recall, which would exacerbate the inherent imperfections in the operational exposure definition. Recall may be distorted due to faulty memory, intentional deception, or bias related to the occurrence of disease in studies in which exposure is reported after disease occurrence. Laboratory data are often less subject to error in the conventional sense of imprecise measurement, but often more susceptible to conceptual error in not reflecting the exposure of ultimate interest.
A nearly ubiquitous challenge in collecting accurate data on exposure is the difficulty of gathering information over the potential etiologic period of interest. That is, the ideal definition often includes a time dimension over which exposure needs to be integrated or monitored. Even our most elegant tools, whether based on self-report, environmental measurements, or biological markers, rarely capture the exposure of interest over the period of interest. If we are interested in several years or decades of dietary intake of a specific nutrient, our options for data collection are limited. We can ask for people to use their memories to integrate over the interval, we can obtain more precise measurements at a point or several points over the interval, or some combination, such as a precise measure at one point and self-report regarding stability over a longer interval. In many instances, the most sophisticated, detailed, and accurate exposure indicators are only applicable to a brief period around the time of measurement. A rare exception to the generalization that lifetime exposure markers are unavailable is the collection of occupational ionizing radiation exposure through the use of film badges. These instruments, deployed at the time of hire and used throughout the period of employment, provide quarterly or annual measurements of all external ionizing radiation encountered. Subject to compliance and an interest restricted to occupational as opposed to other sources of ionizing radiation, the desired temporal information will be available from longitudinal data collection.
A variety of biochemical markers of tobacco use, for example, urinary or salivary cotinine, or carboxyhemoglobin, are precise indicators that are reflective of hours or at most a day of exposure. The alternative approach to assessing tobacco exposure is the ostensibly cruder measure of self-report, subject to the ability and willingness of respondents to recall their smoking behavior. If the ideal measure is lifetime (or long term) exposure, however, self-report is likely to be superior even to a series of biochemical measures only because the latter cannot integrate over time the way the participants' memories can. If the ideal exposure measure were lifetime inhalation of tar from tobacco combustion, the operational definition based on self-report of cigarettes smoked daily over specified periods of life is likely to be far more strongly correlated with that ideal measure than any present or recent biochemical markers of recent exposure. If our "gold standard" definition were inhalation of tar from tobacco combustion in the past 24 hours, the biochemical indicators would likely be far superior to self-report. The hypothesized temporal course of the relationship between exposure and disease should guide the selection of the optimal marker.
For those who generate research on challenging exposures (and nearly all exposures are challenging to measure), sufficient information should be provided on both the ideal and operational definition to compare them. While researchers are trained or even forced to reveal exactly what was done in the study, i.e., the operational exposure measure, they often neglect to be specific about the ideal exposure measure for addressing the hypothesis under study. In reviewing research reports, the often implicit definitions of the "gold standard" need to be extricated so that the actual methods of exposure assessment can be compared to the ideal. Readers should be watchful for the temptation on the part of researchers to state their goals in modest, attainable terms whereas the more etio-logically appropriate index is less readily approximated. Problems can arise in the choice of the ideal exposure measure as well as in implementing that measure.
Was this article helpful?