Cross-sectional studies show that activity fluctuations in healthy young adults possess robust temporal correlations that become altered with aging, and in dementia and depression. This study was designed to test whether or not within-subject changes of activity correlations (i) track the clinical progression of dementia, (ii) reflect the alterations of depression symptoms in patients with dementia, and (iii) can be manipulated by clinical interventions aimed at stabilizing circadian rhythmicity and improving sleep in dementia, namely timed bright light therapy and melatonin supplementation. We examined 144 patients with dementia (70-96 years old) who were assigned to daily treatment with bright light, bedtime melatonin, both or placebos only in a 3.5-year double-blinded randomized clinical trial. We found that activity correlations at temporal scales <~2 hours significantly decreased over time and that light treatment attenuated the decrease by ~73%. Moreover, the decrease of temporal activity correlations positively correlated with the degrees of cognitive decline and worsening of mood though the associations were relatively weak. These results suggest a mechanistic link between multiscale activity regulation and circadian/sleep function in dementia patients. Whether temporal activity patterns allow unobtrusive, long-term monitoring of dementia progression and mood changes is worth further investigation.