OBJECTIVES:  The doctor-patient relationship is shifting towards empowered patients who want to be informed about their hospitalisation. In this context, prediction models on health care utilisation can help support conveying information. As road traffic victims are often young and employed patients, who have a lot of questions about their health care use, we will focus our analysis on this subgroup. The aim of this study is to make personalized predictions of length of stay within acute hospitalisation for road traffic victims. METHODS:  As there are multiple endpoints and time-dependent variables we chose to perform dynamic predictions by landmarking (LM) in competing risks. The competing risks were death and hospital discharge. Median length of stay LM were set every day, up until percentile 90. LM’s were marked every 3 days, leading to the following landmarks: day 0,1,2,3,6,9,12,15,18 and 21. Baseline-variables taken into account were socio-demographics, type of roadway-user, place and type of injury and comorbidities. Time-dependent-variables were surgeries, residence at intensive care and blood transfusion. Cox proportional hazards models are made for both outcomes at each landmark. RESULTS:  Over the different LM-points the following trend is seen: in predicting time to death being male, being transferred to another hospital, acute circulatory diseases and blood-transfusions show a significantly higher probability of dying. In predicting time to hospital discharge having higher age and being male leads to a significantly higher probability of discharge while chronic diseases, blood transfusions and surgery are significant predictors with lower probabilities of discharge. Roadway user types had both an influence on deceased and being discharged. CONCLUSIONS:  Over the different landmarks for the same endpoint the predictors behave consistent. Significance and direction can differ between the endpoints. We are still working on the dynamic supermodel which averages the estimates over the different landmarks.