It will probably surprise many readers of this Journal to know that occupational exposures to chemicals are rarely assessed with quantitative data, either for epidemiology or for control of hazardous conditions. Only about 13% of epidemiology studies published in the 1980s, on the aetiology of chronic diseases, used any exposure measurements,1 and even those tended to rely on trivial sample sizes (median n = 4).2 Thus, uncertainty in exposure levels is the largest source of error in defining exposure–response relationships. Likewise, industrial surveys of air contaminants typically obtain only a single air measurement per occupational group.3 Surprisingly, this was not always the case. In the period between 1920 and 1960, engineers and health professionals routinely collected many air samples from a given working population, despite the primitive state of measurement technology.4^–7 The likely motivation for such effort was the tremendous variability of air concentrations observed over time and among locations; thus, many measurements were needed to accurately assess exposures. The resulting databases allowed early investigators to delve into the sources and magnitudes of exposure, relationships between exposures and health outcomes, and strategies for comparing air levels with occupational exposure limits (OELs). Such applications are beautifully illustrated in a classic set of contributions to the British Journal of Industrial Medicine (BJIM), published more than 50 years ago.

I vividly recall discovering these papers in the Berkeley library shortly after joining the faculty of the University of California in 1976. Having been charged with developing a graduate course in exposure assessment, I reviewed the usual reference books of the time and found little inspiration. Then I sought guidance from journals devoted to occupational epidemiology and hygiene, searching manually year by year, starting with the most recent volumes. After reviewing about 20 past years of the BJIM, I discovered a single issue in which Wright, Roach, Oldham, and Long considered the health consequences of dust exposures in British coal mines after the Second World War.8^–11 Although these papers are not well known, they have inspired a substantial body of work over the past 30 years.

Wright discussed the relative importance of time and intensity of dust exposure as predictors of lung burden and respiratory disease.10 He considered two hypotheses. The first, which he termed the “average” hypothesis, posits that the burden should be proportional to the product of the average dust concentration and the duration of exposure (what we now call cumulative exposure). Wright appreciated the parsimony of the “average” hypothesis and found it disquieting that prominent physicians thought that lung damage was unduly influenced by transient “peak” exposures. He disdained the “peak” hypothesis because it relied on a nebulous “threshold mechanism” and encouraged inefficient and biased air sampling strategies. In fact, recent work by Buchanan et al suggests that the fraction of cumulative exposure to quartz above 2 mg/m³ is more toxic to the lungs than that representing lower air levels.12 While this lends support to the “peak” hypothesis, it is important to point out that Buchanan et al considered exposure “peaks” over months or years, whereas Wright envisioned them as short term excursions within a work shift. Thus, Wright’s “average” hypothesis remains relevant today.

Roach used Wright’s “average” hypothesis to relate the incidence of simple pneumoconiosis to cumulative dust exposure (in particle-years).9 He noted that, while both average exposure and duration of exposure had been shown to increase workers’ risks of pneumoconiosis, these two measures had not been combined into a single index for estimating exposure–response relationships. Since epidemiologists generally regard cumulative exposure as a reasonable predictor of long term health risks today,13^–15 it is ironic that Roach (a mining engineer) would be the first to apply the idea.

Roach developed a job–exposure matrix to quantify cumulative exposures of colliers (coal miners), based on current dust levels, individual work histories, and historical changes in mine ventilation. Current dust concentrations were estimated from repeated random samples using 3-minute breathing-zone measurements (24 random measurements per collier!). This “random colliers” method, which had been published the previous year in BJIM by Roach and statistician PD Oldham,7 offered the first application of analysis of variance of (log transformed) air levels. The authors observed that

“…significant variation was occurring in the dust concentrations from one collier’s experience to another’s, and from one day to another in the same collier’s experience.”

To say that Oldham and Roach were ahead of their time would be a vast understatement. Indeed, the revelation that exposure varied both within and between workers in a given job group was ignored by health professionals for about four decades. We now realise that these variance components are central to the design and interpretation of epidemiology studies, i.e. grouping workers and evaluating measurement error effects,16^–18 as well as the recognition and control of workplace hazards.19^–22

Perhaps the most significant paper published in Volume 10 of the BJIM, was Oldham’s statistical characterisation of dust levels in the mines.8 Oldham was surprised that the variation in 3-minute dust levels about the shift average was “…almost perfectly proportional to the value of the average”. Recognising that this is a property of the lognormal distribution, he concluded that

“…the relative frequency of the logarithms of individual levels of dust concentration appears to be governed by the Normal law. So far as we are aware, this observation has not been previously made.”

We now realise that virtually all sets of random measurements of airborne chemicals (both particulate and gaseous) are approximately log normally distributed, and we use this knowledge to make inferences about the magnitudes of air levels.20 Thus, the importance of Oldham’s observation cannot be overstated. He ended his paper by illustrating how the log-scale parameters of the exposure distributions can be used to optimise strategies for estimating average exposures and the frequencies of “peak” exposures. He noted that both “peaks” and “averages” can be evaluated with the same data as long as sufficient numbers of random air measurements are obtained.

Long elaborated on the statistical issues involved with setting and enforcing exposure limits for coal dust.11 Given the inherent variability of dust levels, he noted that the vagaries of OELs issued by the National Coal Board created problems. Long argued that health standards should be set and enforced using probabilistic schemes related to both “peaks” and “averages”, an argument that my colleagues and I extended to contemporary OELs several decades later.21 23

At the end of his paper, Long bemoaned the poor technology for measuring dust concentrations and envisioned an ideal sampler that would “…sample the air which each collier breathes…”. Of course, we now have such personal air samplers24 and can easily obtain data to accurately characterise exposures of large numbers of workers. Sadly, the promise of such devices has never been realised, and sample sizes have grown ever smaller since the early 1980s.25 Indeed, occupational hygienists complain that it is too difficult to obtain reasonable data for statistical analysis of exposure levels and have turned increasingly to predictive models of dubious validity.22 26^–29 Perhaps they should be reminded that Roach did not find it too difficult to randomly sample colliers’ exposures, even as he dragged heavy equipment through the bowels of the earth half a century ago.9 30

The fact that we do not routinely measure exposures today points to a particular failure of the occupational health community. I daresay that Wright, Roach, Oldham, and Long would be perplexed to discover that the remarkable advances in measurement technology we have witnessed since 1953 would not yield a trove of exposure data. I encourage readers to revisit these authors’ works and consider whether it might still be possible to heed their advice.