Background. Questionnaire-based information of occupational physical activities is extensively used, but susceptible to systematic errors. Calibration modeling may reduce errors and improve precision of questionnaire-based information by transforming the selfreported data into more correct estimates of “true” exposure. We aimed (1) to determine the ability of unadjusted ratings of Saltin and Grimby’s Occupational Physical Activity (SGOPA) question to estimate objectively measured sedentary behaviour, physical activity and cardiovascular load, and (2) to develop and evaluate statistical models calibrating SGOPA ratings into expected values of objectively measured exposures.
Methods. 214 blue-collar workers responded to a questionnaire comprising the SGOPA question and questions on several individual and work-related factors. They wore two accelerometers measuring time spent in sedentary and in physical activities, and one Actiheart monitoring cardiovascular load (eventually expressed as %Heart Rate Reserve) for one to four days. Least-squares linear regression models were developed to predict each objectively measured exposure from SGOPA and additional self-reported individual and work-related predictors.
Results. SGOPA alone explained 22% (R2 adjusted=21%) of the variance between individuals in sedentary behaviour and physical activities, and 8% (R2 adjusted =7%) of the variance in high cardiorespiratory load. When adding predictors related to individual and work to the regression model, explained variance increased to 51% (R2 adjusted=46%) for both sedentary behaviour and physical activities, and to 27% (R2 adjusted=19%) for high cardiorespiratory load. Bootstrap validation suggested that explained variance would be reduced by 9-15% for the three exposures when using the model on other data sets.
Discussion. SGOPA itself shows only limited ability to predict objectively measured sedentary behaviour, physical activities and cardiovascular load at work, but the performance of a calibration model can be considerably improved by adding further self-reported predictors.