Understanding the mechanisms of outcome according to the time frame can help optimize the therapeutic development in severe alcoholic hepatitis. We assessed short-term and long-term survival in severe alcoholic hepatitis based on baseline disease severity, extent of therapeutic improvement, long-term influence of alcohol relapse, and their interaction. Data and alcohol consumption were prospectively recorded in 398 patients treated with corticosteroids in the short term (from corticosteroid initiation to 6 months) and long term (from 6 months to maximum follow-up time). Cumulative incidence rate of first alcohol relapse was 25.2%, 33.7%, and 35.2% at 1, 3, and 5 years, respectively. Alcohol relapse (≥30 g/day) was not associated with mortality (P = 0.24) during the short-term period (1,606 patient-months at risk), but the Lille (P < 0.0001) and Model for End-Stage Liver Disease (P < 0.0001) scores were independent prognostic factors. In patients who were alive at 6 months (median follow-up, 42 months; interquartile range 11-88), corresponding to 10,413 patient-months at risk, alcohol consumption (≥30 g/day) was associated with mortality (hazard ratio, 3.9; P < 0.0001). Additional analysis with abstinent patients as a reference showed a dose effect of alcohol on the hazard ratio of death: 2.36 (P = 0.052) for 1-29 g/day, 3.2 (P = 0.003) for 30-49 g/day, 3.51 (P < 0.0001) for 50-99 g/day, and 5.61 (P < 0.0001) for ≥ 100 g/day. The baseline Model for End-Stage Liver Disease score was not predictive of long-term outcome, while Lille score (P = 0.02) and alcohol relapse (P < 0.0001) were independent prognostic factors. This study shows that new therapeutic development for severe alcoholic hepatitis must target liver injury in the short term and alcohol consumption in the long term; thus, health agencies can endorse future study designs adapted to the time frame of factors influencing mortality; with this in mind, drug-targeting mechanisms involved in liver injury should only be tested for the short-term period.