Number of publication:   089/2014-BGIA

Author:		Morfeld, P.; Ellegast, R.P.; Ditchen, D.; Kuß, O.; Schäfer, K.; Kersten, N.; Haufe, E.; Luttmann, A.; Jäger, M.
Title:		Estimation of cumulative dose models to analyse effects of physical exposure. Methodology of Multimodel Analysis within the EPILIFT Exposure Criteria Study
Source:		Zentralblatt für Arbeitsmedizin, Arbeitsschutz und Ergonomie 64 (2014) No. 3, pp. 169-182, 32 lit. refs., 6 tables, 13 figs. (Language:D)

Abstract

The EPILIFT exposure criteria study (DWS-Richtwertestudie) is an in-depth reanalysis of the EPILIFT case control study with the aim to estimate cumulative dosimetric models that can be used to derive dose-response relationships between physical workload and the occurrence of disc-related lumbar spine diseases and, in addition to suggest exposure criteria (thresholds) for men and women that may be applied in the German occupational disease No. 2108 (lumbar spine disease). Statistical analyses are confronted with three challenges. 1. Dosimetric modeling. The best dosimetric model is unknown. Candidate models vary considerably in thresholds for lumbar disc compressive forces, degree of trunk forward inclination, and daily doses (exposures were only considered for dose cumulation if they reached or exceeded the specific threshold, and if so all values were considered without any subtractions and with a squared weighted disc force in relation to action duration). 2. Epidemiological modeling. The structure of the best risk model is unknown and may be complex. The continuous odds ratio (OR) curve may show a W-form across the cumulative long-term dose. This leads to a large number of candidate models. 3. Exposure criteria: The risk analysis should try to derive an estimate of the long-term dose that is associated with a doubling of risk (doubling dose). The point estimate of the doubling dose should be presented with a 95% confidence interval. A multimodel analysis (MMA) was performed in two tiers, averaging models with the help of information criteria. Fractional polynomials (FPs) of the second and fourth degrees were fitted to all sensible dosimetric models. Goodness of fit was measured by the Akaike information criterion (AIC). In the first step of the MMA best thresholds were estimated separately for each property of interest by appropriate weighting. Weighting was defined by the relative information of each FP (Akaike weighting). These best thresholds were inserted simultaneously to yield combined dosimetric models that were used to define the reference dose in all further analyses. In the second step of the MMA all continuous regression curves and confidence bands were averaged across the reference dose by again applying Akaike weighting (by averaging model predictors within each individual, relative to OR = 1 at cumulative dose = 0). The OR estimates were smoothed by FPs of the fourth degree to yield continuous final OR curves across the reference dose with 95% confidence limits. These curves were used to derive a doubling dose with confidence limits by inversion. The MMA was applied to four sub-studies (gender: male/female, outcome: prolapse/chondrosis). The approach is demonstrated for the sub-study investigating the prolapse prevalence among men. The MMA method presents a potential solution for specification of the optimal dosimetric and/or epidemiological models if a priori criteria are missing.

Back to overview