Dementia affects large number of elderly, manifested with memory impairment symptoms at the onset. This results in difficulties for elderly in scheduling and completing daily activities. In this respect a context-aware reminder system becomes an essential tool in helping elderly navigate their daily activities. Current work on reminder systems typically follows a set of pre-defined activities, organised into a plan. The plan is then used to prompt the elderly to execute specific activities or actions if they have not been completed. However, this is less than an ideal scheme, since the elderly sometimes choose to ignore the reminders due to being engaged in other activities that overlap with the scheduled activities. Therefore, a more flexible schedule is required, that delivers reminders in a context-appropriate manner. Such system must take into account not only the planned activities, but also the current activities of the elderly. We address this problem by monitoring the activities of the elder through our activity recognition system and using this information as a feedback to the reminder system. The reminder system can then decide whether a reminder prompt is appropriate to be delivered or should it be postponed for at a later time. We present the results of activity recognition and discuss how they affect the context-appropriate reminder system.