Dansk Epidemiologisk Selskab og Dansk Selskab for Arbejds‐ og Miljømedicin afholdete videnskabeligt møde om: Composite variablesin medical research, generally, and with a focus on occupationalstress research


Dato: September 20th
Lokation: Bispebjerg University Hospital

1300 –1330 Sigurd Mikkelsen: Introduction and background for the meeting
1330 –1400 Mika Kivimäki: IPD‐Work main models for the analysis of job strain and effort‐reward-imbalance effects on an outcome
1400 –1430 Michael Ingre: Job strain and coronary heart disease
1430 –1500 Break, coffee/tea and cake
1500 –1530 David Coggon: Combined effects of two or more variables, epidemiological considerations
1530 –1600 Niels Keiding: Combined effects of two or more variables,statistical considerations
1600 –1700 Discussion (panel ofspeakers, discussion with the audience)

Background and motivation for the meeting

The combined effect of two factors on an outcome is usually examined in regression analysis by modelsthat include both factors as covariates. If an interaction between the two factorsis suspected the model may further include an interaction term defined as the product of the two factors. This model is informative with respect to the independent effects of each of the two factors and their interaction. The two dominantstress theories, Karasek’sjob strain theory and Siegrist’s effort‐reward‐imbalance (ERI) theory, are usually examined by a composite variable. Job strain is usually measured by the product of demands and control after dichotomizing the continuous variables by a median split into binary variables. ERI is usually measured by the ratio between efforts and rewards, and the effects of this ratio may be examined as a continuous variable and by categories, including dichotomization, often at unity. The difference between the two ways of examining the combined effect of two factors is if the main effects of the two factors are included in the regression model or not. Proponents of the standard regression model claim that an effect of a composite variable without adjustment for the main effects can lead to ambiguous or misleading inferences. Proponents of the

Slides som blev brugt til oplæg på mødet kan ses her