|Home | About | Journals | Submit | Contact Us | Français|
H. J. Lipscomb. Division of Occupational and Environmental Medicine, Department of Community and Family Medicine, Duke University Medical Center, Durham, North Carolina, USA
The assumption that occupational epidemiology should be an applied science frames this presentation. With this premise in mind, arguments are made for why the occupational health and safety community should broaden our approaches to incorporate more qualitative methods with traditional quantitative epidemiological techniques. These research methodologies are viewed on a continuum rather than as two discrete entities. While the presentation is from the perspective of an American researcher using illustrations largely from injury epidemiology, a number of concepts are more broadly applicable.
In the 1960s when Sir Austin Bradford Hill outlined his criteria for causal inference, he first mentioned strength of association. It is not surprising that epidemiologists spend much effort seeking ways to make our effect measures more precise. We strive to clearly define our outcomes, exposures and other variables of interest, and we are enamoured with analytical methods that more precisely define the strength and precision of associations. This focus on strength of association has also taken us to meta‐analyses, and systematic reviews, in which studies are tediously limited/restricted based on strict inclusion criteria. There is certainly merit in these efforts and we should continue to work to refine effect measures. However, the precise measures lack important breadth, and efforts to improve our work should not be limited to the refinement of effect measures.
Many methods designed for the study of chronic disease fall short in the study of injury. As a science injury epidemiology has been underdeveloped and viewed by many as the step‐child of epidemiology. The goals are to understand why workers are injured, to prevent those events and the sequelae potentially associated with them; some aspects of injury epidemiology are more straightforward than chronic disease epidemiology. For example, traumatic injury occurs because of energy transfer; the proximal cause of injury is often easy to identify. Defining temporality is rarely an issue and we are not dealing with long latency periods. However, occupational injury results from a sometimes complex mix of factors including tool design, work norms, personal behaviour, and work environments to name only a few. There are also broad policy issues, both formal and informal, that extend beyond occupational health that influence the work that people do, how that work is done, and, consequently, their exposures as well as the acceptance of control measures. We need to measure effect; we also need to understand it, interpret it, and communicate what our work means. To do that effectively, as public health practitioners, we need to work with more than effect measures.
Examples are presented to demonstrate how our quantitative science can be enhanced through methods not always openly embraced by epidemiologists, and some challenges involved in that process. Specific examples focus on working with vulnerable worker populations, enhanced surveillance and improved intervention evaluation.
B. Armstrong. London School of Hygiene and Tropical Medicine, UK
Occupational and environmental epidemiology have closely related objectives. Indeed occupational epidemiology could be considered a sub‐specialty of environmental epidemiology, or environmental epidemiology an extension of occupational epidemiology. However, the two disciplines have evolved fairly separately, with some overlap but some important differences in predominant study designs. I have sought to identify study designs that are much better developed in one of these two disciplines, and explore possibilities for more exploitation of these developments in the other. An example of each is given below.
“Time series regression” studies have been enormously popular in investigations of the health effects of air pollution and more recently weather, but apparently little used in occupational contexts. Data comprise series, usually daily, of health outcomes (eg, mortality) and environmental exposures (eg, air pollution). Inference on health effects is drawn from investigating the association between the two: “Do high mortality days tend to be preceded by high pollution days?”. Interrupted time series, in which the “exposure” series comprises a before/after (eg, a regulatory change) dichotomy, are a special case. The very small relative risks (often less than 1%) detected by such studies met with initial scepticism by reviewers and regulatory bodies, but substantial consensus has since emerged that this design, when carefully applied, can deliver resolution around this level. Could these study designs be used in occupational epidemiology? They can only identify acute effects, and only when exposures vary quickly over time. This limits but surely does not exclude use of the design to investigate occupational hypotheses of interest. Day‐of‐week effects have been studied in occupational contexts, but might be improved using some of the insights from environmental applications of this design. In circumstances where production factors such as line speed vary or there is a change following regulation, health or injury events could be investigated in relation to these variations/changes.
Cohort and nested case‐control designs, especially for cancer end‐points, have been developed over decades, with occupational epidemiology often leading methodology. They are much less common in environmental epidemiology, although some that have been published (the “six‐cities” and American Cancer Society cohorts) have been very influential. Although the geographical dimension of the environmental cohort studies raises special problems less of concern in occupational contexts (spatial autocorrelation), other complexities are those well‐explored in occupational epidemiology. In particular “time window” analyses (identifying the time course of the effects of exposure) have been little used in environmental cohorts to date. This is in part due to data limitations, but as these become better resolved, I believe that the experience of occupational epidemiology can importantly inform this research.
In summary, although there are some issues specific to occupational or environmental epidemiology, there are also opportunities for more cross‐fertilisation between the two disciplines. In particular, there may be occupational uses for the “time series regression” design, and environmental uses for “time window” methodology used in occupational cohort studies.