Aortic dissection is a severe cardiovascular disease caused by the occurrence of a tear in the aortic wall. As a result, the blood penetrates the wall and makes a new blood channel called false lumen. The hemodynamic conditions in the false lumen may contribute to the formation of thrombi, which influence the patient diagnosis and outcomes. In this study, the focus is on a hemodynamic-based model of thrombus formation. Since the model construction presents sources of uncertainties in the model parameter, a variance-based sensitivity analysis is performed. Thrombus formation at a backwards-facing step is considered as a benchmark for the numerical simulations and sensitivity analysis. This geometry is capable of representing the main contributions of the model in thrombus formation. The study aims not only at getting a better insight into the model's structure but also at preparing model simplifications with the aim of future patient-specific simulations. A polynomial chaos expansion is employed as a surrogate model, from which the derivation of quantitative sensitivity indices is enhanced. In this study, nine model parameters are selected, whose actual values are not well known. The model responses taken into account are the maximum volume fraction of thrombosis, its time development, and the thrombus growth rate. The results show that the model lends itself to model reduction since some of the model parameters show little to no influence on the model's outputs. A threshold value related to the concentration of bounded platelets is identified as the key input parameter dominating the model predictions in the current geometry.