The COVID-19 virus has been raging around the world for months and killing more than a million people. It is extremely infectious due to its easy transmission and long incubation period. Until now, the number of people diagnosed with COVID-19 infection has been increasing dramatically each day. At this stage, fast, accurate, and early clinical assessment of the disease severity is vital. For this purpose, the machine learning tool is an effective way to diagnose it. To support decision-making and logistical planning in healthcare systems, inspired by earlier works, we study the application of Broad Learning System (BLS) to predict the mortality of COVID-19 patients via their blood samples. We evaluated three models on the 375 patients' blood samples, and the BLS model achieves a sensitivity of 94.50% while having a specificity of 94.80%. Except for specificity and sensitivity rates, the area of under receiver operating characteristic (AUC), prediction accuracy, confusion matrix, and precision are also presented in this paper. The performance of BLS performs obviously better than the other models compared in this work. The encouraging results are inspired us to do a future analysis on a large set of COVID-19 blood samples to have a more reliable prediction.