BACKGROUND: given the complex relationship between sleep and neurodegenerative processes, it is important to examine whether changes in sleep patterns occur prior or close to dementia onset. OBJECTIVE: to examine the relationship between sleep parameters and dementia incidence and, to characterize trajectories of sleep patterns before dementia diagnosis. DESIGN: a 14-year longitudinal study including a nested case-control study. SETTING: the French Three-City Study. SUBJECTS: overall, 1,749 cognitively healthy participants (>/=65 years) for the longitudinal study and, 182 incident dementia cases and 719 controls matched by sex, age and educational level for the case-control study. METHODS: dementia cases were assessed at each visit and self-reported sleep parameters at baseline, 2, 8, 10, 12 and 14 years. Cox models were used to estimate the risk of dementia associated with baseline sleep parameters (sleep duration, time in bed (TIB), sleep timing, sleepiness and insomnia). Latent-process mixed models were performed to compare sleep trajectories according to the case-control status. RESULTS: long baseline nighttime and 24-h sleep durations (>/=9 h) as well as being persistent or becoming long sleepers during follow-up were associated with dementia incidence. Trajectories of sleep durations and TIB showed faster increases in cases compared with controls up to 12 years before dementia. The mean differences [95%CI] for 24-h sleep duration between cases and controls were: 0.27 h [0.01;0.52], 0.34 [0.09;0.58] and 0.67 [0.44;0.90] at -12, -8 and -2 years, respectively. Bedtime trajectories showed an earlier bedtime in cases up to -8 years. CONCLUSION: long sleep duration and earlier bedtime may impact dementia incidence.< Réduire