Background/Aims: We aimed to develop realistic definitions of clinical worsening in advanced Alzheimer's disease (AD) and to use them in a post hoc responder analysis of memantine. Methods: 2,340 patients with moderate to severe AD (Mini-Mental State Examination <20) were included from 9 multicentre, 16- to 28-week, randomised, double-blind, placebo-controlled studies of memantine 20 mg/day versus placebo. Responder meta-analyses were performed, with definitions of response based on minimally important differences (MIDs) on cognitive, functional, and global assessment scales. Validated or established MIDs were used where available; otherwise, MIDs were estimated by a data-driven approach, using data from our moderate to severe AD population. Results: Patients with moderate to severe AD treated with memantine had a lower incidence of worsening from baseline to endpoint than patients treated with placebo, in cognition [24.4 vs. 35.0%; odds ratio (OR) = 0.60; p < 0.001], function (38.1 vs. 43.4%; OR = 0.81; p = 0.01), global status (39.8 vs. 48.6%; OR = 0.70; p < 0.001), and in a combined ‘triple' worsening measure (9.4 vs. 16.1%; OR = 0.54; p < 0.001). Conclusions: New definitions of clinical worsening based on MIDs represent a more realistic functional decline in advanced stages of AD. Results of this new analysis show that memantine reduces the incidence of clinical worsening in key symptomatic domains in moderate to severe AD.