The pressure on the Dutch healthcare system keeps growing. The increasing amount of total costs that is spent by the sector, requires more efficient usage of key-resources like capacity or medical professionals. Capacity management is frequently used as an approach to achieve higher efficiency and quality. By using the information streams that are generated by the processes within care institutions, Chipsoft offers IT-modules in its Electronic Health Record which support the implementation of capacity management. This research was conducted in the context of the Rivas Zorggroep, a care group located in Gorinchem, The Netherlands. Rivas, a customer of Chipsoft, offers healthcare in a transmural care chain as it consists of a general hospital and many respective aftercare institutions. The hospital struggles to manage the flow of patients from their wards to the desired aftercare institutions. When a patient is ready to be dismissed to an aftercare location, it regularly happens that no empty spot is available at the required location. As a result, patients are occupying hospitals beds for too long, which is termed bed blocking. The management of the hospital requires insights in the expected demand for aftercare to support tactical decision-making to control this phenomenon. The proposed research does initially define the current patient flows and profiles in the interinstitutional processes of the care group. Secondly, it investigates which tactical patient flows can be clustered on similar characteristics of its admissions that require aftercare. For three different levels of grouping admissions, we present a set of multivariate regressions models to predict the required demand for each type of aftercare. These models are dependent of the total distribution of similar admissions for patients who are treated at the hospital. Based on a set of metrics, we evaluate and define the best set of models for practical application. We finalize the research by improving the chosen set of multivariate regressions by applying regularization methods. This ensures a interpretable balance between accuracy and simplicity.