Exposure measurement error in time-series studies of air pollution: concepts and consequences.

Abstract

Misclassification of exposure is a well-recognized inherent limitation of epidemiologic studies of disease and the environment. For many agents of interest, exposures take place over time and in multiple locations; accurately estimating the relevant exposures for an individual participant in epidemiologic studies is often daunting, particularly within the limits set by feasibility, participant burden, and cost. Researchers have taken steps to deal with the consequences of measurement error by limiting the degree of error through a study's design, estimating the degree of error using a nested validation study, and by adjusting for measurement error in statistical analyses. In this paper, we address measurement error in observational studies of air pollution and health. Because measurement error may have substantial implications for interpreting epidemiologic studies on air pollution, particularly the time-series analyses, we developed a systematic conceptual formulation of the problem of measurement error in epidemiologic studies of air pollution and then considered the consequences within this formulation. When possible, we used available relevant data to make simple estimates of measurement error effects. This paper provides an overview of measurement errors in linear regression, distinguishing two extremes of a continuum-Berkson from classical type errors, and the univariate from the multivariate predictor case. We then propose one conceptual framework for the evaluation of measurement errors in the log-linear regression used for time-series studies of particulate air pollution and mortality and identify three main components of error. We present new simple analyses of data on exposures of particulate matter < 10 microm in aerodynamic diameter from the Particle Total Exposure Assessment Methodology Study. Finally, we summarize open questions regarding measurement error and suggest the kind of additional data necessary to address them.ImagesFigure 1Figure 2Figure 3

Exposure and measurement contributions to estimates of acute air pollution effects

“…We further assume that X it N is independent of X it A and a it is independent of X it C . Our model differs from an exposure model that weights the nonambient source component by (1Àa it ) Zeger et al, 2000), which does not coincide with the mass balance equation interpretation for a it (see below).…”

“…The bias comes from not using within-day exposure variation in estimating b and it occurs because E(exp{Xb})aexp{E(X)b}). Common wisdom suggests that there is no bias because the measurement X P ðtÀ'Þ has Berkson error structure (Kunzli and Tager, 1997;Zeger et al, 2000). However, in the case of a nonlinear disease model where the variance of the unmeasured part of the exposure is correlated with the mean (as would be common for a lognormally distributed exposure), there can be specification bias due to using an average of the exposure in the disease model (Sheppard, 2003).…”

“…While average personal exposure allows the parameter directly estimated in the time series model to be b, unless there are measurements on hundreds of individuals on each day, the large variation in nonambient source exposure and its independence of ambient exposure effectively induces a huge amount of classical measurement error when average personal exposure is used as the predictor in the time series design. Zeger et al (2000) suggest that risk-weighted average personal exposure is the desired exposure measurement in time series studies. Our work supports that conclusion from the perspective of directly estimating the target parameter b, but differs by further showing that the nonambient source exposure variability will induce a large amount of classical measurement error into the analysis (unless the daily samples for estimating X P t are huge).…”

“…They state that ''the three sources of measurement error are instrument error (the accuracy and precision of the monitoring instrument), error resulting from the nonrepresentativeness of a monitoring site (reflected by the spatial variability of the pollutant measured), and differences between the average personal exposure to a pollutant and the monitored concentration (influenced by microenvironmental exposures).'' This follows from work on measurement error in air pollution time series studies Zeger et al, 2000). Although these papers are specific to the time series design, the NRC report implies that the three sources of measurement error are important regardless of study design.…”

Ambient hydrogen sulfide, total reduced sulfur, and hospital visits for respiratory diseases in northeast Nebraska, 1998–2000

“…Misclassification of the outcome variable in studies of hospital visits constitutes another important concern, but it is likely to have occurred nondifferentially and thus have attenuated any association that may exist (Zeger et al, 2000). Another limitation is that some affected residents may not have sought medical treatment, or may have sought treatment in facilities other than the two local hospitals.…”