Abstract
BACKGROUND: Inconsistent findings in previous studies of the association between sleep duration and changes in body mass index (BMI) may be attributed to misclassification of sleep duration fluctuations over time and unmeasured confounders such as genetic factors. The aim of the present study was to overcome these failings by using repeated measurements and panel data analysis to examine the sleep-BMI association. METHODS: Panel data were derived by secondary use of the data from mandatory health checkups at a Japanese gas company. Male non-shift workers aged 19-39 years in 2007 were annually followed until 2010 (n = 1687, 6748 records). BMI was objectively measured, and sleep duration was self-reported. RESULTS: Compared with 7-hour sleepers, panel analysis with the population-averaged model showed a significant increase in BMI among 5-hour (0.11 kg/m(2), P = 0.001), 6-hour (0.07 kg/m(2), P = 0.038), and ≥8-hour (0.19 kg/m(2), P = 0.009) sleepers. On the other hand, after adjustment for unobserved time-invariant confounders using the fixed-effects model, the magnitude of the association was considerably attenuated and no longer significant (5-hour, 0.07 kg/m(2), P = 0.168; 6-hour, 0.02 kg/m(2), P = 0.631; ≥8-hour sleepers, 0.06 kg/m(2), P = 0.460). CONCLUSIONS: The longitudinal association between sleep duration and changes in BMI may be upwardly biased by unobserved time-invariant confounders rather than misclassified sleep duration. The net effect of sleep duration on weight gain may therefore be less than previously believed.