Sampling strategies for occupational exposure assessment under generalized linear model

Occupational exposure assessment is a major task in industrial hygiene studies. Although statistical analyses for magnitudes and variations of exposures to various types of working populations based on existing data sets are extensive, relatively few discussions on study designs appear in the literature, especially for sample size determination and number of repeated measurements.

In this paper, we propose a general framework of sampling strategies on sample size requirement together with the number of repeated measurements using the mixed-effects generalized linear model (GLM). As illustrative examples, we discuss sampling strategies separately under the log-normal assumption for hypotheses testing on (i) mean exposure differences of multiple worker groups and (ii) presence of a long-term exposure trend.

Given a specified alternative hypothesis, the desired significance level and statistical power, the number of repeated measurements, within-worker and between-worker variances, and a correlation structure, we have derived and tabulated an explicit sample size requirement of the two hypothetical cases under log-normal distribution assumption.

On the basis of the tabulated outcomes, the sample size requirement is much more dominant than the number of repeated measurements for a group exposure comparison. Thus, in this case, recruiting more workers with fewer repeated measurements may be more economical than the opposite approach. For testing the presence of a long-term exposure trend, the sample size required decreases substantially with the number of repeated measurements. Also, equally spaced sampling times would be optimal because the effect of between-worker variance is algebraically cancelled out in sample size calculations.