The use and misuse of surrogate variables in environmental epidemiology

Abstract

AbstractThis paper discusses some common statistical problems that are often encountered in the specification and interpretation of regression models used in environmental epidemiology; such models have been used to establish new or modified ambient standards intended to protect public health. These statistical problems include: collinearity (identifying the ‘correct’ pollutant), confounding (omission of other variables that may be correlated with both response and putative dose), the ‘ecological fallacy’ (aggregating individual doses and responses over space or time), measurement error (uncertainties in data, applicability and measurement per se) and linearity (identifying curvature or thresholds in dose‐response function). These problems occur in both time‐series and cross‐sectional studies. Although none of these potential problem areas is new, they have rarely been considered together or comprehensively. This paper considers them as specific instances of the general problem of surrogate variables, for which an analytical framework is presented together with some examples of their practical consequences and some guidelines for interpreting environmental epidemiology studies. Findings of the analysis include: single‐pollutant regression models are likely to overstate effects; although aggregation results in loss of information, it biases the estimates only when confounding is present; the traditional approaches to correcting for measurement errors implied by the difference between personal exposures and ambient air quality do not apply, but estimates may be based on consideration of the ‘error’ term as an additional source of exposure; it may not be possible to deduce the correct shape of a dose‐response function in the presence of measurement error and correlated covariates. These findings are intended to be descriptive rather than definitive; the main purpose is to stimulate the detailed research required to develop practical remedies that would allow epidemiology to be used appropriately in setting environmental standards.Copyright © 1999 John Wiley & Sons, Ltd.