To study the effects of exercise on falls, fractures, hospitalizations, and death in people with dementia. We conducted an individual-level patient data meta-analysis of 7 randomized controlled trials (RCTs). We looked for studies from the reference list of previous systematic reviews and undertook an electronic search for articles published between 2013 and 2019 in Ageline, CENTRAL, PsycINFO, PubMed, and SportsDiscus. Main (binary) outcome measures were the risk of mortality, hospitalization, faller, multiple faller, injurious faller, and fractures. Secondary (count) outcomes were the incident rates of hospitalizations, falls, and injurious falls. From the 1314 participants, 771 were allocated to the exercise group and 543 to the control group. The number of cases regarding the main outcome measures in exercisers and controls were, respectively: 45 (5.8%) and 31 (5.7%) deaths; 102 (14.4%) and 65 (13.4%) participants hospitalized; 221 (34.4%) and 175 (41.3%) had at least 1 fall; 128 (20.2%) and 92 (21.7%) had multiple falls; 78 (24.8%) and 92 (29.3%) had injurious falls; and 19 (2.9%) and 15 (3.5%) had suffered a fracture. Two-step meta-analysis found no effects of exercise on any outcome. One-step meta-analysis found exercise reduced the risk of falls (odds ratio 0.75; 95% CI: 0.57-0.99). Exploratory analysis showed exercise decreased the rate of incident falls in participants with the lowest functional ability (incident rate ratio 0.48; 95% CI: 0.30-0.79). Although the 2-step meta-analysis suggests exercise does not have an effect on the outcomes, 1-step meta-analysis suggested that exercise may reduce fall risk. Data from further high-quality RCTs are still needed.< Réduire