INTRODUCTION: Early prediction of posthospital disposition is crucial for counseling and planning, particularly for adults away from age extremes, given the greater uncertainty about returning home or requiring postacute care among these patients. We aimed to create a prediction model for discharge disposition using data from the first 24 h of admission. METHODS: We conducted a retrospective cohort study using data from the National Trauma Data Bank encompassing patients treated from 2007 to 2016, focusing on individuals aged 35-70, categorized by discharge disposition. Our objective was to predict discharge outcomes - home, rehabilitation, skilled nursing facility, or mortality - employing machine learning techniques based on patient factors, including demographics, comorbidities, injuries, and early resource utilization. Each base model underwent training and parameter tuning to optimize the F1 score and was then evaluated on unseen data. The top three base models were chosen to build a stack ensemble model, and performance was assessed using area under the receiver operating characteristics (AUC), F1, recall, and precision metrics through the one-versus-rest approach, comparing each class. External validation was conducted using data from the Pennsylvania Trauma Outcomes Study. RESULTS: A total of 2,342,703 patients were included. A stack ensemble model was built using the three top performers, which yielded AUCs from 0.73 to 0.97 for each class on held out National Trauma Data Bank data. This stacked model demonstrates strong generalizability across Pennsylvania Trauma Outcomes Study, with AUCs spanning from 0.71 to 0.97. CONCLUSIONS: We created a stacked ensemble model that predicts discharge disposition for adults outside of the extremes of age with injuries within 24 h of admission. Further validation is warranted to show the potential benefits of this model for planning and patient and family guidance.