Dementia is a serious, chronic, and costly public health problem. Prior studies have described dementia as increasing length of hospital stay, but so far no explanations have been proposed. To identify early markers for prolonged hospital stay in demented elderly inpatients, 178 community-dwelling or institutionalized subjects aged 75+, hospitalized through an emergency department in 9 teaching hospitals in France, were analyzed. Prolonged hospital stays were defined according a limit adjusted for Diagnosis Related Group. All patients underwent a comprehensive geriatric assessment at admission. Logistic regression multifactorial mixed model was performed. Center effect was considered as a random effect. Of the 178 stays, 52 were prolonged. Most concerned community-dwelling patients (86%). Multifactor analysis demonstrated that demographic variables had no influence on the length of stay, while diagnosis of delirium (OR 2.31; 95% CI 1.77 - 2.91), walking difficulties (OR 1.94; 95% CI 1.62 - 2.43) and report by the informal caregiver of moderate or severe burden (OR 1.52; 95% CI 1.19 - 1.86) or low social quality-of-life score (OR 1.25; 95% CI 1.03 - 1.40), according to the Zarit's Burden Inventory short scale (12 items) and the Duke's Health Profile respectively, were identified as early markers for prolonged hospital stays. At the time of the rising incidence of cognitive disorders, these results suggest that preventive approaches might be possible. In a hospital setting as well as in a community-dwelling population, more specific, specialized and coordinated care, using the expertise of multiple disciplines appears as a probable effective measure to limit prolonged hospital stay. Such approaches require (i) clear patient-oriented goal definition, (ii) understanding and appreciation of roles among various health care and social disciplines and, (iii) cooperation between partners in patient's management. However, the cost- and health-effectiveness of such approaches should be evaluated.