The ability of limited exposure sampling to detect effects of interventions that reduce the occurrence of pronounced trunk inclination
Introduction
Extensive research has been performed to evaluate workplace interventions designed to reduce employee risks of contracting work-related musculoskeletal disorders (WMSD). This research includes numerous field and laboratory studies that have evaluated the effects of changes in tools, work methods, work environment or work organization. A number of recent reviews discuss different aspects of this large body of literature (e.g. Westgaard and Winkel, 1997, Silverstein and Clark, 2004, van der Molen et al., 2005, Brewer et al., 2006, Boocock et al., 2007, Denis et al., 2008). While the ultimate goal of the intervention is to reduce the risk of contracting or maintaining WMSD, changes in biomechanical exposure accompanying the intervention are often used as a proxy for its eventual effect on health (Westgaard and Winkel, 1997, Lötters and Burdorf, 2002, Cole et al., 2003). Due to constraints on time and resources, safety and ergonomics practitioners are particularly compelled to judging intervention effects by how they affect exposure.
There are significant challenges to quantifying the exposure effects of ergonomics interventions (Burdorf and van Riel, 1996, Burdorf and van Tongeren, 2003, Dempsey, 2007). One important issue appears from the growing literature demonstrating that biomechanical exposures often vary considerably across people within the same job or task, and over time for the same individual. Thus, postures and muscle activity of the back or upper extremities have been shown to vary between and within individuals in several settings, including dairy factory work, cyclic assembly work, cleaning and construction (Burdorf, 1993, Burdorf et al., 1994, van der Beek et al., 1995, Burdorf and van Riel, 1996, Paquet et al., 1999, van der Beek et al., 1999, Hoozemans et al., 2001, Mathiassen et al., 2002, van Dieën et al., 2002, Mathiassen et al., 2003a, Mathiassen et al., 2003b, Möller et al., 2004, Nordander et al., 2004, Paquet et al., 2005, Hansson et al., 2006, Jackson et al., 2009). This inherent exposure variability leads to uncertainties when limited samples of subjects and/or exposure data within a subject are collected and compared as a means of deciding whether an intervention is effective or not. With an insufficient sample size, the investigation will have a low ability to detect a true intervention effect; i.e. it lacks statistical power. The issue of determining a suitable measurement strategy is a challenge, both to researchers operating at the level of groups of workers with the aim of generalizing their results to other workers or other similar work situations, and to practitioners, who are often more interested in knowing whether or not specific individuals benefit from, for instance, modifications of their work station or a course in work technique.
Analytical methods are available by which the ability of an exposure assessment strategy to detect changes in the exposure mean can be estimated on the basis of data on exposure variability and allocated resources (Mathiassen et al., 2002, Mathiassen et al., 2003b). These methods are generally applicable for situations in which the compared pre- and post-intervention exposure distributions are both normal. Normality is a reasonable assumption if the compared mean exposures are both based on a large number of samples, since the distribution of a large-sample mean will approach normal, irrespective of the underlying distribution of the individual samples (Cochran, 1977). However, both in research (Mathiassen et al., 2002) and, in particular, in ergonomics practice (Paquet et al., 2006), exposure samples are often both few and short. While rarely addressed in intervention studies, pre- and post-intervention mean values cannot in this case be expected to be normally distributed without further study.
In ergonomics practice, improvements aimed at reducing the overall mean exposure are often focused on those parts of the job that are considered the worst or those that can be most easily improved upon, while the rest of the job is left unattended. A successful intervention will then change only parts of the overall exposure distribution. For an example, an intervention that reduces the upper 1/4 of a worker's exposures by 50% will, indeed, change the overall mean exposure, but it will also have a pronounced effect on the shape of the exposure distribution (Fig. 1). Thus, with interventions focusing only on the “worst” exposures, pre- and post-intervention exposures are unlikely both to be normally distributed. For instance, the occurrence of trunk flexion of at least 20° illustrated in Fig. 1 is normally distributed across the job before the “intervention”, while afterwards it is not. Examples of intervention concepts with a scope of reducing large exposures while leaving lower ones unattended are courses in lifting technique intended to reduce biomechanical peak loads on the back (Martimo et al., 2008), and biofeedback devices returning an auditory or visual signal when a certain exposure level is exceeded (Madeleine et al., 2006; the Virtual Corset™ (www.microstrain.com/virtual-corset.aspx)).
In this practitioner's case of a successful intervention changing only a part of the exposure distribution, and the intervention effect being assessed on the basis of few pre- and post-intervention exposure samples from one or a few workers, standard analytical procedures cannot be used to evaluate the performance of exposure assessment strategies. Non-parametric approaches then offer a viable alternative. Statistical re-sampling techniques that do not rely on assumptions about the shape of the exposure distribution have been used previously to empirically test the performance of different measurement strategies for quantitative assessment of biomechanical exposures in the field (e.g., Burdorf and van Riel, 1996, Hoozemans et al., 2001, Mathiassen et al., 2002, Mathiassen et al., 2005, Paquet et al., 2005, Fethke et al., 2007), but these methods have not previously been used to investigate ergonomics intervention assessment practices.
The objective of this empirical simulation study was to evaluate the relationship between the number of measurements collected before and after ergonomics interventions of varying effectiveness in a self-paced manufacturing job, and the probability of correctly determining whether and to which extent the interventions reduced the overall occurrence of pronounced trunk inclination, defined as inclination of at least 20°. Taking a practitioner's rather than researcher's perspective, the study was designed to evaluate this relationship for interventions carried out on individual workers, and assessed by few measurements.
Section snippets
Exposure data
Three workers performing the same physically demanding job were randomly selected from a convenience sample of nine workers at a manufacturing plant to provide realistic biomechanical exposure distributions at an individual level. The job was a self-paced automotive upsetting manufacturing job that required the following tasks: lift a steel rod weighing between 45 and 70 N from a pallet (depending on the batch), place rod in an electrical heater, place the rod into a die casting machine, operate
Detecting a reduction in the overall job exposure
As expected, the ability of a sampling strategy to correctly detect that the intervention reduced the overall occurrence of pronounced trunk inclination generally increased with an increasing number of pre- and post-intervention measurements (Fig. 6). For all three workers, this ability was acceptable (p > 0.8) for sampling strategies with three or more pre- and post-intervention measurements when the most extreme half of the exposure distribution was reduced by 50%. The probability of detecting
Job exposure assessment for interventions
While ergonomics interventions are usually aimed to reduce the risk of MSD, intervention effects on biomechanical exposure are often used as proxies of health effects, assuming that exposure is predictive of risk (Westgaard and Winkel, 1997, Lötters and Burdorf, 2002, Cole et al., 2003). One major reason is that the exposure effects of an intervention have a much shorter latency (if any) than health outcomes, making them more readily accessible, in particular to practitioners. Thus, a large
Conclusions
For interventions in self-paced manufacturing that lead to substantial reductions of trunk inclination in those parts of the job where it occurs the most, as few as three pre- and post-intervention exposure samples of ten work cycles each can be sufficient to reach a fair probability of detecting that the intervention has an effect at all on the overall occurrence of pronounced trunk inclination in the job for an individual worker. Larger samples are needed to reasonably estimate the magnitude
Acknowledgements
This work was supported by the Swedish Council for Working Life and Social Research, (grant #2004-0600) and the National Institute for Occupational Safety and Health (grant #R03 OH04105-02).
References (89)
- et al.
Increasing the degree of automation in a production system: consequences for the physical workload
Int J Ind Ergon
(2006) - et al.
Ergonomic effects of a management-based rationalization in assembly work – a case study
Appl Ergon
(1996) - et al.
Musculoskeletal exposure of manual spray painting in the woodworking industry – an ergonomic study on painters
Int J Ind Ergon
(2000) - et al.
PATH: a work sampling-based approach to ergonomic job analysis for construction and other non-repetitive work
Appl Ergon
(1996) - et al.
Design of strategies to assess lumbar posture during work
Int J Ind Ergon
(1996) Invited point of view: the art of conducting workplace intervention studies
Int J Ind Ergon
(2007)- et al.
The effects of mechanised equipment on physical load among road workers and floor layers in the construction industry
Int J Ind Ergon
(2007) Effectiveness of ergonomics interventions to prevent musculoskeletal disorders: beware of what you ask
Int J Ind Ergon
(2007)- et al.
Intervention practices in musculoskeletal disorder prevention: a critical literature review
Appl Ergon
(2008) - et al.
A portable ergonomic observation method (PEO) for computerized on-line recording of postures and manual handling
Appl Ergon
(1995)