|Home | About | Journals | Submit | Contact Us | Français|
B. M. Blatter1, J. Heinrich1, J. R. Anema2, A. J. Van der Beek2. 1TNO Quality of Life Work and Employment; 2Department of Public and Occupational Health, VU University Medical Centre
ObjectivesSickness absence due to musculoskeletal disorders is high. Occupational physicians (OP) in The Netherlands often refer to second‐line medical care at a late stage in the sickness absence period, despite the presence of Dutch OP guidelines advising differently. In addition, the development of the internet has expanded enormously, but not in occupational medicine. Therefore, a website has been developed called www.snelbeter.nl, which counsels workers absent due to neck or back pain. The website starts with a questionnaire, followed by individually based instructions for exercises, education and coping tools. The purpose of this website is to stimulate daily activity and to improve coping with pain by increasing the worker's grip on the situation, which might result in earlier return to work. The objective of the study is to investigate the effectiveness of this website.
MethodsThe study is a randomised controlled trial (RCT) with a 6‐month follow‐up questionnaire. The study population consists of employees of KLM Royal Dutch Airlines and the Dutch National Railways sick‐listed due to non‐specific neck or back pain. Sickness absence may be partial, and has to last at least 2 weeks. “Red flags” (fever, weight loss, radiating pain, etc) are exclusion criteria. Randomisation takes place at the level of the OP, because the OPs play an important role implementing the web‐based counselling program. Since the 51 OPs in the study were assigned to the intervention and control group and not the sick‐listed employees, pre‐randomisation was carried out. This means that the employees are kept unaware of their intervention or control status, but that they are informed about the different approaches that are compared in the study. Outcome measures are time to return to work, grip on the situation (of employee) and performance (of the OP) regarding the presence of a rehabilitation plan and referral to second‐line medical care.
ResultsAfter 4 months, 32 employees have been included in the study.
ConclusionSince the objective was to include 128 employees within 6 months, the inclusion period has to be extended and the OP population has to be enlarged. Once more, enthusing OPs to include employees in scientific studies appears to be difficult.
Key wordsrehabilitation; web‐based intervention; neck and back pain
N. Plato. Division of Occupational Medicine, Department of Public Health Sciences, Karolinska Institute, Stockholm, Sweden
ObjectivesA congestion charging scheme (CCS) was temporarily introduced during the first half of the year 2006 in the city of Stockholm, Sweden (population 1 million inhabitants). The traffic intensity decreased by 23%. The aim was to study the effect of these changes on personal exposure to motor exhaust among bus drivers in central Stockholm, and among drivers in a nearby suburb.
MethodsPersonal sampling of nitrogen dioxide (NO2) was performed during 10 days on 10 bus drivers who drove in Stockholm city before the introduction of the CCS (period I). The measurements were repeated for the same drivers and busses after the introduction of the CCS in March 2006 (period II). Similar sampling was performed on 10 bus drivers in a nearby suburb outside the pay toll area during both period I and II. In total, 404 full shift samples were taken by passive diffusion samplers specific for NO2.
ResultsThe suburban bus drivers had 8‐h mean exposure during weekdays increased by 24% from 84.3 μg/m3 (SD 33 μg/m3) during period I to 104.8 μg/m3 (SD 41 μg/m3) during period II (CCS). City bus drivers showed 17% higher NO2 exposure levels, increasing from 67.1 μg/m3 (SD 23 μg/m3) during period I to 78.4 μg/m3 (SD 19 μg/m3) during period II, despite that fact that the number of vehicles decreased from 425000 to 375000.
ConclusionDiesel exhaust exposure, measured as NO2, was higher during the CCS than before, both within the pay toll area and in the nearby suburb. The NO2 concentration at street level is influenced by exhausts from vehicles, wind conditions and seasonal variations, which seem to have had a stronger effect on the NO2 level for the drivers than the CCS itself. Bus drivers in the suburb had higher NO2 exposure both during period I and II than the drivers in the city, possibly since the buses in the suburb were diesel powered while the city busses were ethanol powered. A follow‐up measurement period (period III) was performed in March/April 2007.
Key wordscongestion charging scheme; diesel exhausts exposure; bus drivers
H. S. Shannon, G. R. Norman. McMaster University
ObjectivesTo describe a problem in analyses of occupational health and safety data.
ResultsChoice of sample size and statistical analysis of data must both take into account all aspects of the study design, including the sampling method. In occupational studies especially, it is common to sample subjects in clusters – work groups, departments, or even workplaces and/or companies. As a result, standard analyses that treat data as independent are often incorrect. Account must be taken of the multi‐level nature of the data. Unfortunately, examples show that this does not always happen. An added complication occurs in the derivation of the factor structure of safety climate (SC). Safety climate refers to the shared perceptions of workers on their work environment, notably how supervisors and management deal with health and safety. The object of observation is the supervisor or management, and so several workers are providing ratings of the same “object”; differences result from variability between observers (the workers). In the extreme, all workers within each work group would rate the supervisor identically (indeed SC is intended to measure shared perceptions), and the effective sample size would be the number of work groups, not the number of workers. A correct analysis must decompose the variances into between group and within group estimates. Various authors have noted that the factors derived in different studies are not the same and several methodological issues have been noted that might affect which factors are identified. However, the fact that the data are multi‐level has apparently not been considered, even though the samples have typically been selected from a relatively small number of workplaces and workgroups. It thus appears that the analyses of data are incorrect.
ConclusionIncorrect analyses are still done in occupational health. It would be useful to re‐do factor analyses of SC measures and compare the results. Standard packages and even specialised commercial programs (MLWin, EQS, Mplus) do not do multilevel exploratory factor analysis. We will describe how to circumvent this problem.
Key wordsmethodology; safety climate; multi‐level analysis
L. Riklik1, D. Chung1, D. Verma2. 1Workplace Safety and Insurance Board, Ontario; 2McMaster University and Workplace Safety and Insurance Board
ObjectivesIn Ontario, Canada, occupational disease compensation is determined by lay adjudication supported by at least three pieces of scientific information, medical diagnosis, workplace exposure assessment and the epidemiological relationship between the diagnosis and exposure. Medical diagnosis has error estimates for false positives and false negatives and epidemiology has a complex statistical framework for error estimation. By contrast exposure assessment often only utilises sampling statistics to present estimates of error. Error estimation for measured biological dose (stored/recent/current) and environmental dosimetry (radiation/noise) is somewhat developed. Error estimation of the presence or absence of a substance in a historic workplace (or a claimant in that workplace) has been less clearly defined and we will describe our methods.
MethodsInformation is collected from the claimant, co‐workers, employers and others and compared against published data, reference works, professional experience, judgment and scientific principles. The gap between workplace information and technical documentation can be examined. Error estimates can be made utilising the physical parameters of the workplace such as production rates, physical properties of substances or by documenting worker location and activities.
ResultsThe NIOSH Office of Compensation Analysis and Support for “Dose Reconstruction under the Energy Employees Occupational Illness Compensation Act” will be examined as a standard that has been reviewed from a scientific and policy perspective. We will compare our methods against the NIOSH methodology that appears to be receiving recognition as a reference standard in North American dose reconstruction for disease compensation. By considering such factors as production volumes and control technologies, we will discuss the relevance of spotty compliance sampling data.
ConclusionHow we communicate the error estimates when providing retrospective exposure assessments for occupational disease adjudication will be described. As this error estimation for exposure assessment in occupational disease compensation adjudication resembles “applied epidemiology with a sample size of one”, some of these concepts may be useful in improving the exposure assessment for epidemiological studies or more explicitly stating error estimates.
Key wordsexposure assessment errors; retrospective exposures; disease compensation
F. C. Breslin1, P. Smith1, J. R. Dunn2. 1Institute for Work & Health; 2Research on Inner City Health, St. Michael's Hospital
ObjectivesThe investigation of geographical variation in occupational injuries has received little attention. Young workers 15–24 years of age are of particular concern because they consistently show elevated occupational injury rates compared to older workers. The present study sought: (1) to describe the geographical variation of work injuries; (2) to determine whether geographical variation remained after controlling for relevant demographic and job characteristics; and (3) to identify the region‐level factors that correlate with the geographical variation.
MethodsData for this project were taken from numerous sources including: administrative data from the Ontario Workplace Safety and Insurance Board; the 2000 Labour Force Survey; the 2001 Canadian Census; the Survey of Labour and Income Dynamics; and the Education, Quality and Accountability Office of Ontario. Using workers compensation claims and census data, we estimated claim rates per 100 full‐time equivalents for 15–24 year olds in 46 regions in Ontario. A total of 21 region‐level indicators were derived to reflect social and material deprivation of the region as well as demographic and employment characteristics of youth living in those areas.
ResultsDescriptive findings showed substantial geographical variation in young worker injury rates, even after controlling for several job and demographic variables. Region‐level characteristics such as greater residential stability were associated with low work injury rates. Also, regions with the lowest claim rates tended to have proportionally fewer cuts and burns than high claim rate regions.
ConclusionThe finding of substantial geographical variation in youth claim rates even after controlling for demographic and job factors can aid prevention resource allocation. The association between region‐level indicators such as residential stability and youth work injury suggests that work injury prevention strategies can be integrated with other local economic development measures. Also, it suggests that relevant authorities might examine whether work safety measures are unevenly distributed with respect to their socio‐economic environments. It would be worthwhile for future research to examine temporal trends in geographical variation in ecological factors and work injury risk.
Key wordsoccupational health; spatial epidemiology; age
P. C. Koopmans1, C. A. M. Roelen2, J. W. Groothoff3. 1ArboNed Groningen; 2ArboNed Zwolle; 3University Medical Center Groningen, University of Groningen
ObjectivesThis study investigates parametric transition rate models for long‐term absence inception and work resumption rate.
MethodsThe study population consisted of 53990 employees in three nation‐wide companies in The Netherlands, which were followed in a sickness absence register over a period of 5 years. Duration until the first long‐term absence (absence inception rate) and duration of long‐term absence (work resumption rate) were modelled by means of parametrical transition rate models; the first by an exponential model (constant transition rate) and the second by a Gompertz‐Makeham model (monotonically declining transition rate). Long‐term absent employees were diagnosed according to the International Classification of Diseases, Ninth Revision (ICD‐9). In the work resumption model, effect parameters of diagnostic categories were estimated.
ResultsThe presence of a malignant disorder, cardiovascular disorder, mental disorder or disorder of the nervous system increased the duration of the long‐term absence. Injuries, infectious disorders and digestive disorders showed a shorter long‐term absence duration. Females, elderly employees and employees in lower salary classes had a higher transition rate into long‐term absence. Females, elderly and higher salary classes had a lower work resumption rate from long‐term absenteeism. Employees with a history of long‐term or frequent absence had a higher transition rate into long‐term absence and longer absence duration.
ConclusionGenerally, research on duration of absence or recovery episodes is done by semi‐parametric survival analysis (Cox regression). In the Cox model the baseline rate is unspecified. From our study it appeared that parametric models are more accurate, that is, the exponential model for the long‐term absence inception rate and the Gompertz‐Makeham model for the work resumption rate. It is recommended that future absenteeism studies address the issue of time dependence of the transition rates. Parametric transition models offer challenging and promising possibilities for future sickness absence research.
Key wordslong‐term absence; Gompertz transition rate model; work resumption rate
J. Ashley‐Martin1, J. Van Leeuwen2, J. Guernsey1, A. Cribb3, P. Andreou1. 1Dalhousie University; 2Atlantic Veterinary College, University of Prince Edward Island; 3Faculty of Veterinary Medicine, University of Calgary
ObjectivesExposure assessment is a challenge for investigations of the impact of pesticide exposures on human health. A case‐control study of risk factors for breast cancer incidence in Prince Edward Island (PEI) was recently conducted, and it was subsequently hypothesised that fungicide exposure from potato production may be a risk factor. The objective was to create a geographical tool to estimate fungicide exposure for participants in a breast cancer study using historical databases.
MethodsTwo estimates of fungicide exposure were determined. Intensity of potato production was derived by (1) merging a validated, field‐based, land use map of PEI (2000) with a 2001 CSD boundary map to calculate the area of land planted in potatoes in each census subdivision (CSD), and (2) dividing this area by the total CSD area. Intensity of fungicide use was derived by (1) merging a 1991 consolidated census subdivision (CCS) boundary map with 1991 Agricultural Census data on land treated with fungicide, and (2) dividing the hectares of treated land by the total CCS area. Atmospheric drift of chemicals into non‐agricultural CSDs was accounted for by assuming that a non‐agricultural CSD was subject to 50% of the exposure of an adjacent CSD. As case‐control study participants provided information on postal code, exposure and postal code maps were merged to determine exposure (for the two variables) of individuals within each postal code. Categories of exposure (low, medium, high) were created to reduce the potential for exposure measurement error bias.
ResultsEstimates of historical fungicide exposure among cases and controls from a breast cancer study were determined, depending on the postal code in which they resided. It was not possible to validate the exposure measurements due to the lack of individual exposure data. Using the fungicide use variable, there were 70%, 20% and 7% of cases in low (<4.5%), medium (4.5%–13.5%), and high (>13.5%) fungicide exposure categories, respectively, whereas there were 68%, 22% and 10% of controls in these categories.
ConclusionThis exploratory study produced a feasible but unvalidated geographical approach for estimating fungicide exposure for participants in a breast cancer study using historical databases.
Key wordsexposure assessment; fungicide; ecological
J. Lavoué1, P. O. Droz2. 1Institute for Work and Health; 2Institute for Occupational Health Sciences
ObjectivesWhile still much present in the industrial hygiene literature, p value‐based model selection is currently being questioned by an increasing number of statisticians. A trend is emerging that emphasises making statistical inferences from a set of plausible models rather than from a single model regarded as “best”. The objective of this paper is to present a case study of modelling occupational exposure data using multi‐model inference, as introduced in the monograph by Burnham and Anderson, Model selection and multi model inference, 2nd edition (Springer, 2002), and applied in the context of linear mixed‐effect models.
MethodsRespirable dust exposure levels collected over the years by the Institute for Occupational Health Sciences (Lausanne, Switzerland) were selected for this study. A set of plausible models was defined a priori by taking into account the sample size and previous knowledge of variables influent on exposure levels. A finite sample version of the Akaike information criterion (AICc) was calculated to evaluate the relative support of the data for each model. Multi‐model inference was performed by averaging predictions and coefficients over the set of models, with weighing based on the values of AICc. Alternate weights were calculated using a bootstrap procedure. For comparison purpose, traditional forward stepwise selection was performed using p values, the Aikaike information criterion, and the Bayesian information criterion.
ResultsThe modelling datasets contained, respectively, 432 and 607 personal and area measurements taken between 1987 and 2006. The a priori set of models contained 96 models for both types of measurements. The relative weights of the different models showed that no single model had a clear support from the data at hand. As measured by cumulative relative weights, the five best models represented, respectively, 92% and 80% of evidence for the personal and area measurements. Model averaged predictions and accompanying confidence intervals, unconditional on a particular model, were calculated.
ConclusionMulti‐model inference represents a promising procedure that incorporates the notion that several models can be supported by the data and permits evaluation, to a certain extent, of model selection uncertainty, which is seldom mentioned in current practice.
Key wordsmulti‐model inference; linear mixed‐effect models; model selection
J. Cui1, M. Abramson2, N. de Klerk3, M. Dennekamp2, A. DelMonaco2, G. Benke2, B. Musk3, M. Sim2. 1Monash University; 2Department of Epidemiology and Preventive Medicine, Monash University; 3School of Population Health, University of Western Australia
ObjectivesOccupational health longitudinal studies often need to analyse the dose–response relationship between relevant exposure variables and health outcomes. However, exposure data are usually not normally distributed and not every individual is exposed to the contaminant of interest. The aim of this study was to use several different statistical methods to model longitudinal occupational exposure data and select the optimal approach to analyse the risk of exposure on a health outcome.
MethodsThe main outcome variable was FEV1 (forced expiratory volume in 1 s) measured on 446 individuals at their annual interviews in an inception cohort study in aluminium smelters. The main exposure variable was the cumulative exposure of fluoride (mg/m3‐year) since start of employment. A linear mixed effects model and GEE methods were used in the analysis of the longitudinal data. Different methods were used to analyse the exposure data. The quasi‐likelihood under independence model criterion (QIC) was used to select the best model.
ResultsThe model using tertiles of fluoride exposure had the lowest QIC and thus gave the best fit to the data. The next best model was to use the trend of the tertiles. The model using the original scale gave the worst fit to the data and the model using log‐transformation was slightly worse than the model using trend over tertiles.
ConclusionWe recommend using the tertile method in similar analyses of longitudinal data. The model using the original exposure data gave the worst fit and should be avoided. The model using the trend over tertiles can also be used in evaluation of a dose–response relationship.
Key wordslongitudinal; lung function; statistical models