Weak associations in occupational epidemiology: adjustment for exposure estimation error

Epidemiological studies often estimate the health effects of occupational exposures by multiple regression techniques. The standard theory of regression analysis is based on the assumption that the explanatory variables are known without error, and it has long been realised that departures from this assumption will lead to underestimation of the true regression coefficients. In reality, there may be considerable imprecision in the measurement of individuals’ exposures to hazards in the workplace, but this is seldom taken into account in analyses. We therefore studied the effect of allowing for imprecision in the exposure estimate with more sophisticated statistical methods, using lung function data from a sample of 348 British miners exposed to mixed coal dust, over an eight-year period. Change in lung function over an eight-year period was regressed on cumulative dust exposure, weight, age and smoking habits. The error in the exposure estimation was assumed to be up to 30% of the total varience of the distribution of the exposure. Adjustment of the regression coefficients of lung function change on dust exposure for the estimation error using linear structural relationships increased the regression coefficient more than threefold compared with those calculated by standard regression analysis. The adjustment led to a change of the coefficients of age as well. The results indicate that a serious underestimation of the relation between lung function change and occupational exposure may occur, which might lead to false interpretations about the relative importance of the occupational exposure as a determinant of disease.