What to expect from our Statistics course
What can you expect from joining our course “Statistics for Assessment of Occupational Exposures and Health Outcomes“?
Course leader Anne Straumfors gives an introduction to the topic, followed by a few words from each lecturer
Course: Statistics for Assessment of Occupational Exposures and Health Outcomes, 29th of November – 1st of December 2022, Radisson Blu Scandinavia Hotel, Oslo, Norway
More information: Course web page | Course registration | Last registration date: 27th of October 2022
Anne Straumfors, STAMI: Introduction to the course
Occupational exposure and health effect measurements are important both for evaluation of potential health risks and for their potential reduction through identifying and testing efficient control measures. Occupational exposure assessment plays an integral role in epidemiological studies and deepens our understanding of health hazards and risks. It also supports efforts to characterize exposure prevalence and levels in the working environment, and to estimate disease burden attributable to work across time periods, geographical regions, and worker populations.
To be meaningful for risk evaluation, the measurement strategies of both exposure and health effects must be properly designed for the aim of the evaluation. Should the exposure measurement data mirror the worst-case scenario or the distributions of exposure concentrations over a time? Should the health effect data give information of acute effects, chronic or long-term effects over a shift or over years? Are there any cause-effect relationships? There are many ways to collect data and to assess exposure levels and health responses. To make an informed decision in which method to apply, one needs to know the advantages and disadvantages of each method. Whether the data output will be representative for the study population or valid for the question asked, will depend on the data collection strategy and the statistical analyses of the data. Therefore, knowledge of the proper statistical methods is very important.
To put the importance in perspective; more than 2 million people die every year from work-related diseases and injuries, while many more suffer from non-fatal injuries or non-healthy conditions at work, according to WHO and United Nations Global Compact. Work-related health problems result in an economic loss of 4-5 % GDP for most countries. Occupational exposure and health risk assessments and knowledge on health and work is the fundament for primary health protection efforts at the global, national, and local levels, including policies, regulations, and other intervention measures that aim to reduce or eliminate hazards at work and increase health in the working population. This is a great motivation for keeping up with the necessary statistical methods for research in the field.
“More than 2 million people die every year from work-related diseases and injuries, while many more suffer from non-fatal injuries or non-healthy conditions at work”
Research in the occupational health field requires thorough knowledge of the statistical tools necessary to make the right decisions when planning studies, assessing exposure and health effects, assigning exposure estimates for epidemiological use and when analysing exposure-response relationships. This course will focus on relevant statistical approaches and considerations related to this and is shortly described in the following.
Hans Kromhout, Utrecht University: Statistical aspects of occupational exposures
Occupational exposures are known to vary tremendously. Even when measuring the exposure of a group of workers performing the same job at the same location one can expect to see large differences in exposure concentrations from day-to-day, but also between longer-term average exposures of individual workers. In this lecture statistical models to describe the distribution of exposure concentrations will be discussed. Examples of empirical statistical models that will allow to unravel determinants of the level and variability of exposure will also be discussed. Relevant routes of exposure can vary depending on the agent, but also the circumstances under which occupational exposure occurs. Different metrics of occupational exposures will be addressed, and guidance will be provided on which metric(s) to use for informative epidemiological studies and when needed for effective exposure control.
Anne-Kristin M. Fell, University Hospital of Telemark: Study design and planning
Cases or case-series are useful for hypothesis generation regarding the relationship between exposure and health outcome. A cross-sectional study with self-reported exposure and measurements of for example lung function for exposed and unexposed, is a common design in occupational epidemiology. There are advantages of this design, in particular regarding costs and time used. However, different study designs are needed to establish causality between the exposure and health outcome at question. In a cohort study, exposure is measured over time, preferably including levels and variation in exposure, and new cases of disease occurring during the same time-period are registered. This will be the classical cohort design which allows causal conclusions.
Nevertheless, this design will not be well suited for all health outcomes, such as musculoskeletal diseases or psychological problems that often vary over time. Further, how the health problem or disease is reported will be of importance as self-report may give different results than objective measurements or registry-based information. Similarly, self-report of exposure may lead to overestimation among those experiencing a health problem that they relate to the exposure, while objective measurements are to be preferred but not always feasible.
Øivind Skare, STAMI: Basic theory of mixed models
When data have dependencies due to repeated measurements being taken from the same individual or due to clusters, such as plants or geographical units, this need to be considered. This can be done using mixed models. Mixed models might be considered an extension of regression models. In addition to fixed effects (i.e., independent variables like gender, age) shared with regression models, random effects are introduced to model the inherent dependency in the data. We will discuss different form of variance structures that may arise, such as random intercept and random slope models. In addition to linear mixed models, generalized linear mixed effects models will be addressed, in particular logistic and Poisson mixed models, that are used to analyse binary responses and count data.
Mixed models offer a wide range of possible variance structure models, and we will give some guidelines on how to choose between these. Assumptions and properties of mixed models will be discussed. Real data examples will be given to illustrate how mixed models can be applied to different types of data.
Susan Peters, Utrecht University: Exposure assessment methods
There are many ways to collect data and to assess occupational exposure levels. Common methods include self-reported exposures, job-exposure-matrices (JEM) and job-specific modules (JSM), but one can also make use of (expert-based) decision rules and algorithms, or modelling of individual exposure measurements.
To make an informed decision in which method to apply, one needs to know the advantages and disadvantages of each method. Different research questions and differences in study design, as well as available resources may influence the choice for the best exposure assessment method.
Hans Kromhout, Utrecht University: Exposure assignment (individual- versus group-based approaches)
Assigning exposure to an individual can take place at the individual level or based on a common characteristic shared with others (e.g. job performed). To a large extent this will be depending on the actual exposure assessment method used. Individual (bio)monitoring data or case-by-case assessment by an expert might sound very precise and accurate, but the earlier mentioned large variability in occupational exposures might result in a considerable bias towards the null in an epidemiological analysis. On the other hand, group-based approaches like (semi-)quantitative JEMs linked to an individual solely based on a job history will in most cases result in unbiased estimates of exposure-response associations but will come at a loss of precision. The underlying theories of classical measurement error and Berkson-error will be illustrated.
Dagfinn Matre, STAMI: Sleep and pain measurements in occupational health (repeated measures in micro-longitudinal and experimental study designs)
Several factors at work are associated with musculoskeletal pain complaints. The best-known causal factor is probably mechanical exposures, but also psychosocial exposures and disturbed sleep are associated with pain. Shift work or night work are factors that commonly disturb sleep. The lecture will cover sleep and pain measurements in a micro-longitudinal design and in an experimental design perspective. In epidemiological studies, it is common to ask for e.g. average neck pain during the last 4 weeks. However, subjective health complaints typically fluctuate within such a time frame. Therefore, if the purpose is to determine causal associations, it makes sense to study pain in a micro-longitudinal day-to-day (repeated measures) design. The same holds for sleep, which may inherit considerable day-to-day variation.
When studying mechanisms, experimental study designs are appropriate. In pain research, this means applying controlled pain stimuli in the laboratory in some form of cross-over design. One may manipulate e.g. sleep and study its effect on pain in a repeated measures design across days, or over the duration of the pain stimulus. In experimental studies, subjects typically report pain on a visual analogue scale that is digitally sampled and stored. Examples of these designs and repeated measures approaches will be presented.
Alex Burdorf, Erasmus MC, University Medical Centre, Rotterdam: Statistical methods with repeated measurements for exposure-response associations (GEE models, adjustment for baseline health parameters)
In longitudinal studies on associations between self-reported occupational exposure, e.g. psychosocial factors at work, and self-reported (chronic) health problems, problems with reverse causality will arise. In addition, the repeated measurements of exposure and disease over time are not completely independent. Hence, generalized estimating equations (GEE) as statistical method can be used to take into account the correlation between the different waves during the study. Three different specifications of this statistical model will be used to illustrate analytical possibilities in a longitudinal study with annual waves. First, a time-lag model whereby exposure is linked to disease one year later. Second, an autoregressive model is used whereby the disease of interest is adjusted for its baseline value. Third, a fixed effects model is used to study the relation between changes in exposure with changes in disease.
Vivi Schlünssen, Århus University: Associations between exposures and health outcomes – does the quality of exposure assignment matters?
To provide trustworthy estimates of associations between occupational exposures and health effects valid estimates for both exposure and outcome is pivotal. Due to various reasons a spectrum of different exposure assessments and assignments are still applied in occupational epidemiology including self-reported exposure, expert assessed exposure, objectively measured exposure, statistical and deterministic exposure modelling, and combinations of those. This lecture will provide you with examples how the validity of the exposure assessment and assignment may influence the results in epidemiological studies
Lothar Lieck, EU-OSHA, Bilbao: The OSH barometer – a statistical visualisation tool for major OSH indicators
The OSH Barometer is the official source of information on the status of and trends in Occupational Safety and Health (OSH) in the EU-countries. It provides – mainly visualised – data and information for important OSH indicators at EU and national level.
The OSH Barometer contains information on OSH authorities and infrastructures, on sector and workforce profiles, on OSH strategies and social dialogue in the EU Member States, on working conditions, on outcomes (accidents at work, burden of diseases, well-being), and on relevant OSH institutions and enforcement capacities.
The tool allows you to:
- view the main indicators as diagrams
- compare indicators between two or three countries and with the EU-level
- generate and download graphics or export them as Excel files
- download detailed country reports with all quantitative and qualitative data
The OSH Barometer is the product of collaboration between EU-OSHA, DG Employment of the European Commission, and its National Contact Points.