• environmental health
• personal exposure assessment

The paper by Harrison and colleagues¹ and the accompanying editorial by Cherrie² in the October 2002 issue of Occupational and Environmental Medicine address the important issue of personal exposure assessment (of air pollutants) in environmental epidemiology. After reading both papers we would like to make some comments with regard to the design, conduct and statistical analysis of the study by Harrison et al and at the same time answer the question raised by Cherrie in his editorial.

Coming from the occupational exposure assessment arena it is interesting to see that our environmental colleagues are still relying to a large extent on static (microenvironmental) sampling and even rely on shadowing to represent personal exposure. The latter brought back memories of old occupational hygiene textbooks with pictures of technicians standing with a sampling probe in the breathing zone of a worker (clearly hindered while carrying out his work task). It is interesting to note that Dr Cherrie’s very relevant earlier work³ on whether wearing sampling pumps affects exposure (it hardly did) was not mentioned in both papers.

The paper by Harrison and colleagues¹ clearly states as one of its goals to answer the question “Does modelling through the use of microenvironment measurements and activity diaries produce reliable estimates of personal exposure to air pollutants”. However, in the only setting where personal exposures were actually measured (phase 1, volunteers; with regard to phase 2 we do not think that shadowing results can be seen as equivalent to personally measured exposure) it is hard to grasp from both fig 1 and table 2 which exposure was actually modelled (1 hour averages, 2–3 day averages) and how (a formula was only provided for measurements within the susceptible groups).

When comparing direct personal measurements for CO and PM₁₀ with the modelled results, the authors exclude all data which are not directly comparable—that is, when the volunteer spent most of their time out of house, and all the data for smokers. It is therefore not surprising that good correlations were found between personal and static measurement results. Why were smokers excluded? Was their measured CO exposure representing a different kind of CO leading to a different health effect? We know that excluding smokers or people with unventilated gas heaters is common practice in the statistical analyses of environmental exposures, but this would only make sense if we were expecting different risks from the same exposure originating from different sources.

In fig 1 the authors present 120 comparable data points for 11 individuals; given the repeated nature of the sampling these data points cannot be seen as statistically independent. Putting a simple regression line through these points is therefore not correct and application of a mixed effects model would have been more appropriate. Besides that, when estimating environmental exposure, for instance, for a panel study, we are interested in the full range of exposures both in the temporal and spatial sense (not only for the room with the static sampler). However, Harrison et al conclude, “... modelled personal exposure is unable to reflect the variability of measured personal exposures occasioned by the spread of concentrations within given microenvironments”.

Both Cherrie and Harrison et al claim that microenvironmental sampling would be a good alternative for direct personal exposure measurements that supposedly are “costly and time consuming”. However, the costs for sampling microenvironments in a general population study will be far greater if we want to measure all the microenvironments people end up in (for instance, in table 1 seven environments are indicated, and most of them will most likely be different for each study participant). In addition, it will be practically impossible to measure some of these environments as the authors point out. In their study, it was not possible to collect data for all appropriate microenvironments, even for a comparatively small number of subjects.

Recently, a very insightful paper was presented at the X2001 conference in Gothenburg. Seixas and colleagues⁴ showed that in a study to assess occupational noise exposure, a task based methodology (analogous to microenvironmental sampling in environmental exposure assessment) could only account for 30% of variability in daily exposures. They even considered this estimate somewhat optimistic since their estimated noise exposures were derived from the same data on which the daily average exposures were estimated. In addition they clearly pointed out that using simple task based averages that artificially compress exposure variability resulted in a very substantial negative bias in the estimated daily exposure.

In our opinion, we should aim to collect personal exposure measurements when estimating exposure for epidemiological studies. We agree that smaller and lighter sampling instruments will need to be developed, as was suggested by Cherrie in his editorial. Recent studies in both the occupational and environmental arenas have shown that study subjects are capable of carrying out personal measurements themselves (and by doing so, cutting out the costs of the technician).^5–9 In all these studies except one,⁷ far more than 100 personal measurements were generated, which shows that studies of this size are not exceptional as was suggested in the editorial by Cherrie.

The question raised by Cherrie, “How important is personal exposure assessment in the epidemiology of air pollution?”, can only be answered with a firm “very important”, if we want to capture the full range of personal exposures experienced in the general environment. In addition, given the relatively low concentrations in the general environment, we will need to measure these accurately. Microenvironmental monitoring and consequent modelling based on diaries will not provide sufficient resolution and accuracy.