[BDSP. Notice produite par INIST-CNRS 8snlmR0x. Diffusion soumise à autorisation]. To identify predictive factors for 2-year mortality in frail elderly patients after acute hospitalisation, and from these to derive and validate a Mortality Risk Index (MRI). A prospective cohort of elderly patients was set up in nine teaching hospitals. This cohort was randomly split up into a derivation cohort (DC) of 870 subjects and a validation cohort (VC) of 436 subjects. Data obtained from a Comprehensive Geriatric Assessment were used in a Cox model to predict 2-year mortality and to identify risk groups for mortality. A ROC analysis was performed to explore the validity of the MRI. Five factors were identified and weighted using hazard ratios to construct the MRI : age 85 or over (1 point), dependence for the ADL (1 point), delirium (2 points), malnutrition risk (2 points), and comorbidity level (2 points for medium level, 3 points for high level). Three risk groups were identified according to the MRI. Mortality rates increased significantly across risk groups in both cohorts. In the DC, mortality rates were : 20.8% in the low-risk group, 49.6% in the medium-risk group, and 62.1% in the high-risk group. In the VC, mortality rates were respectively 21.7,48.5, and 65.4%. The area under the ROC curve for overall score was statistically the same in the DC (0.72) as in the VC (0.71). The proposed MRI appears as a simple and easy-to-use tool developed from relevant geriatric variables. Its accuracy is good and the validation procedure gives a good stability of results.