A decision support tool is provided for predicting a patient's length of stay within a healthcare unit, such as the emergency department. The predicted length of stay may be determined using patient information and facility information, which includes information about the department and the facility in which the department is located. The patient information and the facility information may be used to determine feature values input into a plurality of machine learning models. The models may be used sequentially such that initial models are used to predict variables relating to the patient's length of stay and output of such models is used with a length-of-stay model to determine the predicted length of stay. One or more response actions may be automatically initiated based on the predicted length of stay. Such response actions may include resource management actions to allocate resources within the healthcare unit.