OBJECTIVES: To apply different regression models to estimate rate ratios for temporary sick-leave (TSL) which may occur several times in the same individual during a period, and the frequency is not constant for the observation period. SUBJECTS AND METHODS: All workers employed more than 30 days between January 1st of 1992 and June 1st of 1995 were included into the population study. The following period was 1,259 days and the total number of workers included in the study was 2,306. During that period 2,649 TSL episodes were notified, which meant 85,947 lost days. Poisson regression, Generalised Estimating Equations (GEE) and Andersen-Gill modification of Cox regression modify by Wei (WLW) were applied. RESULTS: The highest TSL incidence rates were seen in women, lesser than 30 years old, cleaners, maintenance workers and auxiliary nurses, and those involved in shiftwork. This profile was not modified after applying GEE and WLW regression models, although confidence intervals were widened. CONCLUSIONS: TSL data does not fit Poisson regression assumptions, but GEE and WLW regression models do not appear as alternatives. Other conditional regression models would need to be explored to suitably analyse this data.
OBJECTIVES: To apply different regression models to estimate rate ratios for temporary sick-leave (TSL) which may occur several times in the same individual during a period, and the frequency is not constant for the observation period. SUBJECTS AND METHODS: All workers employed more than 30 days between January 1st of 1992 and June 1st of 1995 were included into the population study. The following period was 1,259 days and the total number of workers included in the study was 2,306. During that period 2,649 TSL episodes were notified, which meant 85,947 lost days. Poisson regression, Generalised Estimating Equations (GEE) and Andersen-Gill modification of Cox regression modify by Wei (WLW) were applied. RESULTS: The highest TSL incidence rates were seen in women, lesser than 30 years old, cleaners, maintenance workers and auxiliary nurses, and those involved in shiftwork. This profile was not modified after applying GEE and WLW regression models, although confidence intervals were widened. CONCLUSIONS: TSL data does not fit Poisson regression assumptions, but GEE and WLW regression models do not appear as alternatives. Other conditional regression models would need to be explored to suitably analyse this data.
Authors: A M García-Díaz; S Pértega-Díaz; S Pita-Fernández; C Santos-García; J Vázquez-Vázquez Journal: Aten Primaria Date: 2006-01 Impact factor: 1.137